Homologs that fail to synapse during mammalian meiosis are transcriptionally inactivated via meiotic silencing, an epigenetic mechanism critical to fertility. Checkpoint protein ATR localizes to unsynapsed chromosomes, but its role in meiotic silencing is poorly defined. Using a strategy to temporarily delete ATR in the germline, Royo et al. show that ATR first regulates unsynapsed chromosome sensing and later transduces signals to the surrounding chromatin. This study reveals multiple roles for ATR in meiotic silencing and presents a novel gene ablation technique.
In mammals, homologs that fail to synapse during meiosis are transcriptionally inactivated. This process, meiotic silencing, drives inactivation of the heterologous XY bivalent in male germ cells (meiotic sex chromosome inactivation [MSCI]) and is thought to act as a meiotic surveillance mechanism. The checkpoint protein ATM and Rad3-related (ATR) localizes to unsynapsed chromosomes, but its role in the initiation and maintenance of meiotic silencing is unknown. Here we show that ATR has multiple roles in silencing. ATR first regulates HORMA (Hop1, Rev7, and Mad2) domain protein HORMAD1/2 phosphorylation and localization of breast cancer I (BRCA1) and ATR cofactors ATR-interacting peptide (ATRIP)/topoisomerase 2-binding protein 1 (TOPBP1) at unsynapsed axes. Later, it acts as an adaptor, transducing signaling at unsynapsed axes into surrounding chromatin in a manner that requires interdependence with mediator of DNA damage checkpoint 1 (MDC1) and H2AFX. Finally, ATR catalyzes histone H2AFX phosphorylation, the epigenetic event leading to gene inactivation. Using a novel genetic strategy in which MSCI is used to silence a chosen gene in pachytene, we show that ATR depletion does not disrupt the maintenance of silencing and that silencing comprises two phases: The first is dynamic and reversible, and the second is stable and irreversible. Our work identifies a role for ATR in the epigenetic regulation of gene expression and presents a new technique for ablating gene function in the germline.
ATRmeiosissex chromosomes
During mammalian meiosis, homologous chromosomes synapse and recombine, generating crossovers that are essential for correct chromosome segregation (Handel and Schimenti 2010). In order to prevent the generation of aneuploid gametes, meiotic cells exhibiting defects in synapsis or recombination are eliminated from the germ cell pool by meiotic surveillance mechanisms (Nagaoka et al. 2012). The molecular pathways that constitute these surveillance mechanisms are currently not well understood.
One proposed mechanism by which unsynapsed chromosomes trigger germ cell arrest is meiotic silencing (Baarends et al. 2005; Turner et al. 2005; Ichijima et al. 2012). At the onset of pachytene, genes located on unsynapsed chromosomes are inactivated, remaining so for the rest of meiosis and, in the male, after this time (Greaves et al. 2006; Namekawa et al. 2006; Turner et al. 2006). Meiotic silencing has been proposed to cause arrest by starving germ cells of essential gene products (Burgoyne et al. 2009). Other suggested functions include silencing of transposable elements and inhibition of transcription at sites of meiotic DNA double-strand break (DSB) formation (Inagaki et al. 2010). During wild-type male meiosis, the heterologous X and Y (sex) chromosomes remain largely unsynapsed, and this triggers meiotic silencing of X- and Y-linked genes. The resulting phenomenon, known as meiotic sex chromosome inactivation (MSCI) (McKee and Handel 1993; Yan and McCarrey 2009), affects protein-coding but not microRNA genes (Song et al. 2009) and results in the formation of the heterochromatic sex body (Solari 1974). MSCI is essential for male fertility (Royo et al. 2010) and represents an ideal paradigm for studying the epigenetics of meiotic silencing.
Meiotic silencing involves two sets of proteins: “sensors,” which localize to axial elements (AEs) and sense asynapsis, and “effectors,” which localize to the chromatin loops associated with unsynapsed AEs, causing gene silencing over a considerable distance. The AE component synaptonemal complex protein 3 (SYCP3) (Kouznetsova et al. 2009), HORMA (Hop1, Rev7, and Mad2) domain proteins HORMAD1 (Daniel et al. 2011) and HORMAD2 (Wojtasz et al. 2012), and breast cancer I gene BRCA1 (Turner et al. 2004) have been identified as sensors: BRCA1 accumulates along unsynapsed AEs in an SYCP3-, HORMAD1-, and HORMAD2-dependent manner, and mice deficient in any of these four proteins exhibit MSCI defects. In contrast, the mediator of DNA damage checkpoint 1 (MDC1) (Ichijima et al. 2011) and histone variant H2AFX (Fernandez-Capetillo et al. 2003) are silencing effectors: In MDC1- and H2AFX-nulls, gene silencing within the chromosome loops does not occur. Ser139 phosphorylation of H2AFX (γH2AFX) is the key epigenetic event triggering MSCI, with spreading of this modification along chromosome loops being MDC1-dependent (Ichijima et al. 2011). Ago4−/− mice also exhibit a defect in MSCI, associated with defective γH2AFX localization on the XY bivalent and altered small RNA profiles (Modzelewski et al. 2012).
H2AFX Ser139 phosphorylation is catalyzed by the PI3K-like kinases ataxia telangiectasia mutated (ATM), ATM and Rad3-related (ATR), and DNA-dependent protein kinase (DNA-PK) (Sedelnikova et al. 2003). Which kinase generates γH2AFX on unsynapsed meiotic chromosomes is unclear. H2AFX phosphorylation is preserved on the XY bivalent in DNA-PK and Atm single nulls (Bellani et al. 2005). This implies that either ATR is the responsible kinase or the PI3K-like kinases act redundantly in this context in a manner already documented in mitotic cells (Stiff et al. 2004). Indeed, both ATR (Turner et al. 2004) and Ser1981-phosphorylated ATM (Hamer et al. 2004) have been found to localize to the XY bivalent during pachytene. Attempts to discriminate between these possibilities have been hampered by the fact that Atr ablation causes embryonic lethality (Brown and Baltimore 2000).
A number of other questions concerning the mechanisms underlying meiotic silencing remain unresolved. First, it is unclear how sensing of asynapsis at AEs is transduced into gene silencing along associated chromatin loops. Interestingly, ATR exhibits the unusual property of localizing to both the unsynapsed AEs and the surrounding chromatin of the XY bivalent at the onset of MSCI (Turner et al. 2004). This raises the possibility that it connects the sensing and effector steps in the silencing pathway. Furthermore, the interrelationships between ATR and other silencing factors—e.g., HORMAD1, HORMAD2, BRCA1, MDC1, and H2AFX—have not been examined. In mitotic cells, ATR is recruited to sites of RPA-coated ssDNA through its binding partner, ATR-interacting peptide (ATRIP), with ATR subsequently being activated by DNA topoisomerase 2-binding protein 1 (TOPBP1) (Burrows and Elledge 2008). ATRIP and TOPBP1 localize to the sex chromosomes in male meiosis (Refolio et al. 2011), but how they are recruited there is unclear. Finally, ATR localizes to the XY bivalent from early pachytene until late diplotene; i.e., for some 8–9 d after the initiation of MSCI (Turner et al. 2004). This raises the questions of whether meiotic silencing is a dynamic process, involving continuous H2AFX phosphorylation, and whether ATR is required for the maintenance of the inactive state.
To interrogate these points, we generated two forms of conditional Atr mutant mice. The first uses a Cre-recombinase-driven approach to delete ATR at the initiation of silencing, and the second implements a novel, MSCI-driven silencing strategy to deplete ATR levels at later stages, after silencing has been established.
ResultsATR ablation causes meiotic arrest and defective H2AFX phosphorylation
To examine the role of Atr in MSCI, we used a previously described tamoxifen-inducible Cre-ERT2 approach in which Cre-ERT2 expression is driven from the human ubiquitin C promoter (Ruzankina et al. 2007). Atrflox/− males treated with tamoxifen and killed 3 mo later exhibit spermatogonial stem cell loss and no germ cells in the seminiferous epithelium (Ruzankina et al. 2007). We therefore used a modified strategy in which Atrflox/− males and Atrflox/+ controls received 1 wk of tamoxifen treatment and were then killed at earlier time points after treatment cessation (see the Materials and Methods for further details on treatment strategy). Here we focus only on the meiotic silencing phenotypes.
Beginning at 3 d after cessation of tamoxifen treatment, we observed abnormal nuclear morphology in spermatocytes from Atrflox/− males, extending from stages I to VI of the seminiferous cycle, as assayed by testis histology (Fig. 1A). This was associated with meiotic germ cell loss between stages III and V, corresponding to mid-pachytene. The level of ATR protein, determined by Western blotting (Fig. 1B) and immunostaining of testis sections (Fig. 1C), was globally reduced in Atrflox/− testes 3-d after treatment cessation relative to Atrflox/+ controls. This time point was therefore chosen for further characterization of the Atr-null phenotype.
ATR ablation in Atrflox/− mice causes pachytene spermatocyte losses. (A) PAS-stained stage IV tubule sections from Atrflox/+ and Atrflox/− mice showing pachytene germ cell death in Atrflox/− males. Arrows point to healthy (Atrflox/+ panel) versus dying (Atrflox/− panel) pachytene cells. Note that post-meiotic cells are present in the Atrflox/− section because of the transient nature of the Cre-ERT2-mediated knockdown. (Bottom) Pachytene spermatocytes at stages I–IV and VI. Abnormal morphology is visible from stages I to VI, but a major wave of cell death happens at stages III and IV. Bars, 5 μm. (B) Western blot showing reduction of ATR protein level in Atrflox/− testis compared with Atrflox/+. β-ACTIN is used as a loading control. (C) Immunofluorescence for ATR on control and Atrflox/− testis sections at low magnification (color), with tubules marked with an asterisk also shown under high magnification (black and white). In control spermatocytes, ATR is present as a diffuse signal in the nucleoplasm and as a bright signal in the sex body. These signals are dramatically reduced in Atrflox/− sections. Bar, 100 μm. (D) Surface spread images of Atrflox/− and control spermatocytes showing substantial depletion of ATR at zygotene (top row; n = 50 nuclei) and early pachytene (middle row; n = 100 nuclei) in Atrflox/− males. Residual ATR can be seen at the PAR of Atrflox/− spermatocytes (arrow). (Bottom row) Those late pachytene cells that have survived mid-pachytene cell death retain ATR staining (n = 100 nuclei). X and Y AEs are indicated by arrowheads. Bar, 5μm.
To assess the degree of ATR depletion in Atrflox/− testes in greater detail, we performed immunostaining on surface spreads in combination with an antibody to the AE marker SYCP3. In Atrflox/+ control males, we first observed ATR staining at zygotene, where it appeared as foci on AEs (Fig. 1D). Later, at pachytene, ATR was observed exclusively on the AEs and the chromatin of the XY bivalent (Fig. 1D). In Atrflox/− males, the ATR staining patterns at zygotene and pachytene were lost (Fig. 1D); in the latter cell type, traces of ATR were observed at the pseudoautosomal region (PAR) (Fig. 1D, and see the figure legends for number of cells examined for this and later experiments). Normal ATR staining was observed in Atrflox/− late pachytene and diplotene cells due to the fact that these more advanced cell types had not undergone Atr deletion (Fig. 1D; see below for further discussion).
ATM is required for H2AFX phosphorylation at DNA DSBs during leptotene, while ATR has been proposed to regulate H2AFX phosphorylation at DNA DSBs during zygotene (Bellani et al. 2005; Turner et al. 2005). ATR and Ser1981-phosphorylated ATM have both been localized to the XY chromatin at early pachytene (Hamer et al. 2004; Turner et al. 2004), and it is therefore unclear what the relative contributions of these kinases are to MSCI-related H2AFX phosphorylation. We observed normal γH2AFX staining patterns during leptotene in Atrflox/− males, but H2AFX phosphorylation at zygotene was dramatically reduced, consistent with ATR being the principle H2AFX kinase at this latter time point (Fig. 2A). In early pachytene Atrflox/− spermatocytes, H2AFX phosphorylation on the unsynapsed regions of the XY chromatin was also attenuated, and γH2AFX was observed only at the PAR (Fig. 2B), presumably due to the presence of residual ATR at this site (Fig. 1D). We subsequently found that the Ser1981-phosphorylated ATM antibody previously shown to exhibit XY chromatin staining (Hamer et al. 2004) gave the same localization pattern in Atm−/− mice (Supplemental Fig. 1A). Thus, although this antibody gives irradiation-dependent staining in spermatogonia and Sertoli cells (Hamer et al. 2004), the XY-staining pattern is nonspecific. We observed no localization of ATM to the XY chromatin or the PAR in either Atrflox/+ or Atrflox/− males using two additional ATM antibodies (Supplemental Fig. 1B,C). We conclude that ATR is required for the bulk of H2AFX phosphorylation during zygotene and early pachytene.
ATR is required for H2AFX phosphorylation at zygotene and pachytene and for meiotic silencing. (A) γH2AFX immunofluorescence of Atrflox/+ and Atrflox/− spermatocytes showing normal γH2AFX patterns at leptotene (n = 50 nuclei) and loss of γH2AFX staining at zygotene (n = 50 nuclei) in Atrflox/− males. (B) Surface spread image of Atrflox/+ and Atrflox/− pachytene spermatocytes showing loss of γH2AFX staining at the XY bivalent in Atrflox/− spermatocytes, with γH2AFX observed only at the PAR. Arrowheads point to the X and Y AEs. (C) RNA FISH images for Atrflox/+ (n = 100 nuclei) and Atrflox/− (n = 100 nuclei) early pachytene cells showing Scml2 misexpression in Atrflox/− spermatocytes. The RNA FISH signal (arrow) is found close to the γH2AFX-stained PAR. The graph shows the proportion of spermatocytes exhibiting Scml2 RNA FISH signals in each genotype. Bar, 5 μm.
H2AFX phosphorylation at early pachytene results in meiotic silencing, which can be assayed by RNA FISH. To examine the role of ATR in silencing, we studied expression of the X-linked gene Scml2 in Atrflox/− male pachytene cells. Scml2 is silenced by MSCI, and mice defective in MSCI show inappropriate Scml2 expression at early pachytene (Wojtasz et al. 2012). We observed Scml2 expression in 97% of Atrflox/− early pachytene cells, easily identifiable by their characteristic PAR γH2AFX staining (Fig. 2C). Thus, ATR is essential for the initiation of meiotic silencing.
ATR also regulates localization of silencing factors at unsynapsed AEs
HORMAD1, HORMAD2, and BRCA1 are essential for ATR localization to unsynapsed AEs (Turner et al. 2004; Daniel et al. 2011; Wojtasz et al. 2012). Nevertheless, since the PI3K-like kinases are required for accumulation of various repair proteins at DNA DSBs in mitotic cells, we questioned whether ATR could also reciprocally regulate HORMAD1, HORMAD2, and BRCA1 localization and, as such, function in the sensing step of meiotic silencing.
We observed grossly normal HORMAD1 and HORMAD2 localization to the XY AEs in Atrflox/− males (Fig. 3A), which was confirmed quantitatively (Supplemental Fig. 1D). Interestingly, however, BRCA1 localization at the XY AEs was disrupted (Fig. 3A). Consistent with a role for ATR in regulating BRCA1 localization, we observed ectopic BRCA1 accumulation at the PAR in Atrflox/− males (Fig. 3A). We conclude that BRCA1 and ATR are interdependent in the meiotic silencing pathway.
ATR regulates silencing factor localization to unsynapsed axes. (A) XY AEs from Atrflox/+ and Atrflox/− surface spread pachytene spermatocytes showing normal HORMAD1 (top row; n = 50 nuclei) and HORMAD2 (middle row; n = 50 nuclei) and defective BRCA1 (bottom row; n = 50 nuclei) localization in Atrflox/− males. In Atrflox/− pachytene cells, BRCA1 accumulates at the PAR. The insets show HORMAD1, HORMAD2, and BRCA1 staining only. (B) XY AEs from Atrflox/+ and Atrflox/− surface spread pachytene spermatocytes showing disrupted HORMAD1Ser375 (top row; n = 50 nuclei) and HORMAD2Ser271 (bottom row; n = 50 nuclei) localization in Atrflox/− males. In Atm−/− spermatocytes, HORMAD1Ser375 (n = 50 nuclei) and HORMAD2Ser271 (n = 50 nuclei) localization are grossly normal. The insets show HORMAD1Ser375 and HORMAD2Ser271 only. (C) XY AEs from Atrflox/+ and Atrflox/− surface spread pachytene spermatocytes showing disrupted ATRIP (n = 50 nuclei) and TOPBP1 (n = 50 nuclei) localization in Atrflox/− males. Both proteins accumulate at the PAR. Arrowheads point to the X and Y AEs. Bar, 5 μm.
Hop1, the yeast HORMAD1/2 ortholog, is a known phosphotarget of Mec1, the yeast ATR ortholog (Carballo et al. 2008). It has therefore been proposed that ATR phosphorylates HORMAD1/2 and that this may be essential in mediating certain HORMAD1/2 functions, including meiotic silencing (Fukuda et al. 2012). Indeed, a recent study has shown that HORMAD1 and HORMAD2 are phosphorylated, with Ser375-phosphorylated HORMAD1 (pHORMAD1Ser375) being enriched at unsynapsed AEs (Fukuda et al. 2012). We confirmed localization of pHORMAD1Ser375 to the unsynapsed XY AEs (Fig. 3B) and, using a newly synthesized antibody, found that Ser271-phosphorylated HORMAD2 (pHORMAD2Ser271) is also enriched at these sites (Fig. 3B). To test whether these SQ motif HORMAD1 and HORMAD2 modifications were ATR-dependent, we analyzed their localization in Atrflox/− males. Both phospho-epitopes were barely detectable on the XY AEs in Atrflox/− pachytene cells (Fig. 3B). Consistent with ATR being the principal HORMAD1Ser375/HORMAD2Ser271 kinase, we observed normal localization of both modified forms to the XY AEs in Atm−/− cells (Fig. 3B).
Next, we wished to investigate the relationship between localization of ATR and that of its cofactors, ATRIP and TOPBP1, to unsynapsed AEs. Interestingly, enrichment of both ATRIP and TOPBP1 to the XY AEs was disrupted in Atrflox/− males, and both proteins accumulated instead at the PAR (Fig. 3C). We conclude that, in addition to its role in H2AFX phosphorylation, ATR regulates the localization of multiple silencing factors at unsynapsed axes.
A positive feedback loop between ATR, MDC1, and H2AFX amplifies the silencing response
In mitotic cells, ATM, MDC1, and H2AFX participate in a positive feedback loop in which MDC1 binds phosphorylated H2AFX and subsequently recruits ATM, which catalyzes further H2AFX phosphorylation, thereby amplifying the DNA damage response (Lou et al. 2006). A recent study has found that in the absence of MDC1, spreading of ATR and γH2AFX into the chromatin of unsynapsed AEs is defective, suggesting that an analogous feedback system operates during meiotic silencing (Ichijima et al. 2011). Under this model, the loss of ATR, MDC1, or γH2AFX should affect the localization of the other two silencing factors.
To test this prediction, we analyzed MDC1 immunostaining in Atrflox/− males. MDC1 is a silencing effector and is therefore normally seen in the chromatin of the XY bivalent (Ichijima et al. 2011). As observed for γH2AFX (Fig. 2B), MDC1 localization to the XY chromatin was disrupted in the absence of ATR (Fig. 4A). We subsequently analyzed MDC1 and ATR localization in H2AFX−/− males (Fernandez-Capetillo et al. 2003). We observed defective spreading of both silencing proteins into the XY chromatin in this mutant (Fig. 4B). ATR localization to the XY axes was unaffected in H2AFX−/− males (Fig. 4B) as in MDC1−/− males (Ichijima et al. 2011), showing that neither MDC1 nor H2AFX is required for the sensing of unsynapsed chromosomes. We conclude that the effector step of meiotic silencing requires interdependence between ATR, MDC1, and H2AFX.
ATR, MDC1, and H2AFX phosphorylation are interdependent in the meiotic silencing pathway. (A) XY AEs from Atrflox/+ and Atrflox/− surface spread pachytene spermatocytes showing disrupted MDC1 localization in Atrflox/− males (n = 50 nuclei). (B) XY AEs from surface spread control H2afx+/− and H2afx−/− pachytene spermatocytes showing disrupted MDC1 (top row,; n = 50 nuclei) and ATR (bottom row; n = 50 nuclei) localization in H2afx−/− males. (C) XY AEs from Atrflox/+ and Atrflox/− surface spread pachytene spermatocytes showing disrupted SUMO-1 (top row; n = 50 nuclei) and ubiquitinated H2A (uH2A) (bottom row; n = 50 nuclei) localization in Atrflox/− males. Ubiquitinated H2A accumulates only at the PAR. (A–C) Arrowheads point to the X and Y AEs. Bar, 5 μm.
Finally we examined the effect of ATR loss on the localization of two other epigenetic modifications—SUMO-1 (Fig. 4C; Vigodner and Morris 2005) and ubiquitinated H2A (Fig. 4C; Baarends et al. 1999)—to the XY bivalent. SUMO-1 accumulation has been proposed to precede H2AFX phosphorylation on the sex chromosomes (Vigodner 2009; see also Ichijima et al. 2011), and H2A ubiquitination by the ligase UBR2 is thought to function independently of BRCA1/ATR in MSCI (An et al. 2010). In contrast to Atrflox/+ males, in Atrflox/− males, we did not observe enrichment of either mark at the XY bivalent during pachytene. We conclude that acquisition of both SUMO-1 and ubiquitinated H2A on unsynapsed chromosomes is ATR-dependent.
Atr is dispensable for the maintenance of meiotic silencing
In Atrflox/− males, defective H2AFX phosphorylation on the XY bivalent was a highly penetrant phenotype, affecting all early pachytene cells in testis sections and giving rise to germ cell apoptosis at stages III and IV (Fig. 1A). However, ATR localization and H2AFX phosphorylation were grossly normal during late pachytene (Fig. 1D), and germ cells from stage VII onward appeared healthy (Fig. 1A), indicating that these later cells had evaded Cre-recombinase-mediated excision. Thus, Atrflox/− males could not be used to address the role of ATR in late meiosis.
To overcome this problem, we implemented a novel meiosis-specific gene knockdown strategy that uses MSCI to silence transcription of a chosen gene in a temporally precise manner (Fig. 5A). In this approach, a bacterial artificial chromosome (BAC) containing the Atr gene was targeted to the X chromosome. This was achieved by homologous recombination in mouse XY embryonic stem (ES) cells carrying BAC acceptor sites within the Hprt1 locus (Prosser et al. 2008). Male “XAtrY” mice derived from these ES cells were then crossed onto an Atr-null background, generating complemented XAtrY−/− males whose only source of Atr originated from the X-inserted BAC. Expression of Atr from the X chromosome in XAtrY−/− embryos rescues the embryonic lethality; however, as meiotic cells undergo MSCI, the X-inserted copy of Atr is silenced, the result being that pachytene cells are Atr-null (Fig. 5A).
ATR is not required for the maintenance of meiotic silencing. (A) Cartoon showing the MSCI-mediated gene knockdown strategy. The Atr knockout lethality is overcome by an X-integrated Atr transgene in XAtrY−/− males. In germ cells, Atr is expressed from the transgene until it is shut down by MSCI at pachytene. (B) RNA FISH showing Atr expression (arrows) during pachytene in control XY Atr+/− (n = 100 nuclei) but not XAtrY−/− (n = 100 nuclei) males. Bar, 10 μm. (C) Western blot showing drastic reduction of ATR protein level in XAtrY−/− testis compared with XY Atr+/− testis. Two different exposure times are displayed for the ATR blot: Only with a long exposure time (shown at the top) is ATR detectable in XAtrY−/− testis. β-ACTIN is used as a loading control. (D) XY axes from XY Atr+/− and XAtrY−/− pachytene spermatocytes immunostained for ATR. ATR is present during early pachytene in both genotypes but is absent at late pachytene in XAtrY−/− males. Arrowheads point to the X and Y AEs. Bar, 5 μm. (E) Immunofluorescence of ATR on XY Atr+/− and XAtrY−/− testis sections. (Left) Low-magnification view of tubule sections showing drastic reduction of ATR signal in XAtrY−/− testis. (Right) At stage I, ATR is detected in the sex body and nucleoplasm of control and XAtrY−/− pachytene spermatocytes. In control pachytene cells, the same pattern persists until stage XI. In XAtrY−/− pachytene spermatocytes (arrows), the ATR signal is diminished even at stages II and III, and ATR is undetectable from stage VIII (ATR-depleted nuclei are delineated by dotted lines). Note that in stages VIII and XI, ATR is detected in XAtrY−/− prepachytene cells (asterisks), as expected. Bars: left, 100 μm; right, 5 μm. (F) PAS-stained stage IX tubule sections from XAtrY−/− and XY Atr+/− mice showing normal spermatogenesis. Arrows point to pachytene spermatocytes. Bars, 20 μm. (G, left) Surface spread image of pachytene control and XAtrY−/− cells showing normal H2AFX phosphorylation in the sex body of XAtrY−/− spermatocytes (n = 100 nuclei). (Right) RNA FISH image for control and XAtrY−/− pachytene cells (arrows). Scml2 is silent at pachytene in both genotypes (n = 100 nuclei). Arrowheads point to Scml2 RNA FISH signals in expressing prepachtyene cells. Arrows point to pachytene cells. Bars, 5 μm.
We generated complemented XAtrY−/− males from four independently targeted ES cell clones, which gave equivalent results. In control Atr+/− males, Atr expression, assayed by RNA FISH, was observed throughout pachytene, but in XAtrY−/− males, the X-inserted copy of Atr was silenced in early pachytene cells and remained inactive thereafter (Fig. 5B). As a result, a dramatic reduction in the level of ATR protein could be detected in XAtrY−/− relative to XY Atr+/− testes by Western blotting (Fig. 5C). We noted that in wild-type males, the majority of ATR expression occurs during late pachytene (Fig. 1C). Since Atr silencing was observed at this stage in XAtrY−/− males (Fig. 5E) but not in Atrflox/− males (Fig. 1D), total ATR levels were lower in XAtrY−/− (Fig. 5C) than in Atrflox/− testes (Fig. 1B). Thus, MSCI can be used to silence gene expression during pachytene.
To establish when during spermatogenesis in XAtrY−/− males ATR protein levels were depleted, we analyzed ATR immunostaining in surface spreads. As expected, ATR was present on the XY chromatin at the point of initiation of MSCI but was undetectable during late pachytene (Fig. 5D). Subsequent immuno-analysis of XAtrY−/− testis sections confirmed normal ATR localization to the XY chromatin during MSCI initiation; i.e., at stage I of the seminiferous cycle (Fig. 5E). However, from stage II onward, the level of ATR on the XY bivalent was diminished, and no ATR staining could be observed by stage VIII (Fig. 5E). We conclude that, in wild-type males, ATR recruitment to unsynapsed chromatin is ongoing and dynamic and that Atr expression can be disrupted by MSCI-mediated knockdown.
Despite ablation of Atr expression, testis weights in XAtrY−/− males (mean 93 ± 14 mg, n = 4 males) were similar to those of XY Atr+/− control siblings (mean 91 ± 10 mg, n = 9 males), and XAtrY−/− males were fertile, with grossly normal testis histology (Fig. 5F). Moreover, XY H2AFX phosphorylation and MSCI, assayed by Scml2 RNA FISH, were established and maintained normally in XAtrY−/− males (Fig. 5G). Interestingly, silencing factors (e.g., BRCA1, HORMAD1Ser375, and HORMAD2Ser271) localized normally at the XY AEs in late pachytene XAtrY−/− cells (Supplemental Fig. 1E) despite the absence of ATR at this stage (Fig. 5D,E). We observed no localization of ATM to the XY chromatin in late pachytene XAtrY−/− cells, suggesting that this kinase does not maintain H2AFX phosphorylation in the absence of ATR (Supplemental Fig. 1F). We conclude that ATR is dispensable for the maintenance of H2AFX phosphorylation and XY silencing and that, once established early in pachytene, meiotic silencing is remarkably stable.
Discussion
Here we demonstrate a role for ATR in the initiation of meiotic silencing. Our data show that Atr ablation cannot be compensated for by Atm and DNA-PK and are consistent with ATR being the major kinase that phosphorylates H2AFX to initiate XY inactivation. In addition, we show that ATR functions earlier in meiotic silencing by operating in conjunction with BRCA1, ATRIP, and TOPBP1 and regulating HORMAD1/2 phosphorylation at unsynapsed axes.
A recurrent theme that emerges from our work is that meiotic silencing involves a complex interplay between ATR and other silencing effectors. This can be understood in the context of a model that integrates current and existing observations with the sequential steps in the silencing process (Fig. 6). Initial sensing of unsynapsed axes during zygotene is HORMAD1-, HORMAD2-, and BRCA1-dependent but ATR-independent (Fig. 6A). This is consistent with findings in Spo11−/− mice that show that HORMAD1, HORMAD2, and BRCA1 localize along unsynapsed axes during zygotene independently of ATR accumulation (Mahadevaiah et al. 2008; Wojtasz et al. 2009). Subsequently, ATR is recruited to sites of asynapsis in a HORMAD1-, HORMAD2-, and BRCA1-dependent manner (Fig. 6B). Once recruited, ATR–ATRIP then facilitates further enrichment of BRCA1 as well as other factors involved in ATR activation; e.g., TOPBP1 and phosphorylation of HORMAD1/2. These events maintain asynapsis signaling at chromosome axes during late zygotene (Fig. 6C). Our data (Fig. 3) show that this ATR-dependent phase of silencing factor recruitment is dynamic. This flexibility may ensure that silencing factors can be rapidly evicted from AEs should homologs eventually achieve synapsis. However, if asynapsis persists at early pachytene, ATR subsequently translocates into chromatin loops of unsynapsed regions, leading to induction of repressive post-translational modifications (e.g., γH2AFX) and irreversible gene silencing over many megabases (Fig. 6D). Thus, ATR can be viewed as an adaptor protein, linking the sensor and effector arms of the silencing response. Once established, repressive chromatin marks do not require continued ATR activity for their maintenance. Likewise, at this stage, the maintenance of silencing factors at unsynapsed AEs also becomes ATR-independent (Fig. 6D).
Model explaining the stage-specific requirements of ATR in meiotic silencing. (A) During zygotene, HORMAD1, HORMAD2, and BRCA1 associate with SYCP3-positive unsynapsed AEs independently of ATR. (B) This sensing complex is essential for subsequent ATR, ATRIP, and TOPBP1 recruitment at these sites. (C) ATR then facilitates ongoing asynapsis signaling through continued BRCA1 recruitment. (D) At the onset of pachytene, ATR translocates into chromatin loops, catalyzing H2AFX phosphorylation in a positive feedback loop with H2AFX and MDC1, resulting in recruitment of other silencing factors (e.g., ubiquitinated H2A) and irreversible gene silencing.
In this study, we also present a novel genetic strategy whereby the expression of a gene of interest can be silenced in the germline using MSCI. We used this to demonstrate that loss of ATR after the initiation of XY inactivation does not impair H2AFX phosphorylation or the maintenance of silencing. Our data suggest that H2AFX phosphorylation is not an ongoing process but is restricted to early in pachytene, a property that contrasts with the more dynamic behavior of other histone modifications (e.g., H3K9me3) during the maintenance phase of MSCI (van der Heijden et al. 2007). Notably, the X-targeting strategy described here leads to dramatic and sustained depletion of gene expression. It may therefore prove superior to Cre recombinase-based approaches, which exhibit mosaic Cre-expression and varying excision efficiencies, in interrogating the functions of genes expressed during late prophase or spermiogenesis.
Curiously, the residual ATR protein that we observed in early pachytene Atrflox/− cells localized to the PAR. This finding is surprising because synapsed regions are normally devoid of ATR staining and because ATR was not observed on other synapsed bivalents in this model (Fig. 1D). Although currently difficult to reconcile, this abnormal localization may hint at defects in synapsis and/or recombination in Atrflox/− meiosis that are more pronounced at the PAR. The PAR exhibits unusual properties in meiosis, including delayed DNA DSB formation and synapsis (Kauppi et al. 2011). Work in other organisms has shown that ATR does indeed regulate synapsis and recombination as well as other fundamental aspects of meiosis, including interhomolog bias, crossover formation, and checkpoint responses (Carballo and Cha 2007). Although dispensable for the maintenance of H2AFX phosphorylation, the persistent ATR observed on unsynapsed chromosomes may be related to one or more of these processes. The application of the conditional approaches for ablating Atr described here will provide a unique opportunity to examine whether these functions are conserved in mammals.
Materials and methodsMice
H2afx−/− (Fernandez-Capetillo et al. 2003) and Atm−/− (Barlow et al. 1996) mice were maintained on an MF1 background and according to United Kingdom Home Office regulations. Atrflox/− Cre-ERT2 males were as described (Ruzankina et al. 2007). Targeting of the Atr BAC into the X-linked Hprt locus by recombination-mediated cassette exchange in ES cells was performed as described previously (Prosser et al. 2008). All mice used in this project were maintained according to United Kingdom Home Office regulations.
Tamoxifen treatment
Atrflox/− Cre-ERT2 males were treated with tamoxifen at doses already described (Ruzankina et al. 2007). Males culled immediately after treatment cessation did not exhibit meiotic phenotypes, while those culled 3 d after treatment cessation showed mid-pachytene arrest. This implies that ATR excision occurred during an early, premeiotic germ cell type some 7–10 d prior to mid-pachytene. This is most likely to be a late stage A spermatogonial or early intermediate spermatogonial stage. We noted that as early as 1 mo after treatment cessation, Atrflox/− Cre-ERT2 males exhibited agametic tubules resulting from spermatogonial stem cell depletion. This is consistent with previous findings (Ruzankina et al. 2007) and meant that Atrflox/− Cre-ERT2 males cannot regain fertility after tamoxifen treatment.
RNA FISH, immunofluorescence, and Western blotting
Scml2 RNA FISH was carried out with digoxigenin-labeled probes as described (Mahadevaiah et al. 2009) using BAC RP24-204O18 (CHORI). Immunofluorescence experiments on surface spread spermatocytes were carried out as described in Turner et al. (2005) with antibody dilutions for SYCP3, BRCA1, ATR, and BRCA1 described in Turner et al. (2004); HORMAD1, HORMAD2, and TOPBP1 (gift from J. Chen) described in Wojtasz et al. (2009); and Ser375-phosphorylated HORMAD1 described in Fukuda et al. (2012). The ATRIP antibody (gift from K. Cimprich), MDC1 antibody (ABDSerotec), and SUMO-1 antibody (Abcam; ab11672) were used at 1:100. ATM antibodies pS1981-ATM (Rockland), 07-1286 (Miilipore), and H-248 (Santa Cruz Biotechnology) were used at 1:50. The Ser271-phosphorylated HORMAD2 antibody was made using the peptide EEEACG-S(PO3H2)-QVQRMN to immunize two rabbits. The same peptide was used to affinity-purify antibodies from the serum of immunized rabbits. EEEACGSQVQRMN peptide-coupled affinity resin was used to deplete antibodies that recognize the nonphosphorylated form of the immunizing peptide. Testis section immunofluorescence and Western blots (with ATR and β-ACTIN antibodies used at 1:1000 and 1:50 000, respectively) were carried out as described in Hamer et al. (2004). HORMAD1 and HORMAD2 quantitation was performed as described in Mahadevaiah et al. (2008).
Microscopy
For imaging, an Olympus IX70 inverted microscope with a 100-W mercury arc lamp using a 100× 1.35 U-PLAN-APO oil immersion objective (Olympus) or a 40× objective was used. Images were captured using a computer-assisted (Deltavision), liquid-cooled CCD camera (photometrics CH350L; sensor: Kodak KAF1400, 1317 × 1035 pixels). Each channel was captured separately as a 12-bit source image, and captured images were processed using ImageJ 1.46a.
Acknowledgments
We thank Karlene Cimprich and Junjie Chen for providing the ATRIP and TOPBP1 antibodies, respectively; Grzegorz Polikiewicz for genotyping; and Paul Burgoyne and members of Turner laboratory for critical reading of the manuscript. This work was funded by the Medical Research Council (U117588498), the National Institute on Aging (2R01AG027376), the Deutsche Forschungsgemeinschaft (TO 421/3-1, 421/3-2, 421/4-1, and 421/5-1), the Wellcome Trust (WT098051), the Swedish Cancer Society, the Swedish Research Council, and the Karolinska Institutet.
Supplemental material is available for this article.
Article is online at http://www.genesdev.org/cgi/doi/10.1101/gad.219477.113.
New information about the rules that guide the formation of multimolecular chromatin-bound complexes has helped to delineate gene-regulating pathways and describe how these pathways are altered in various pathological conditions. In this review, Tarakhovsky and Prinjha discuss approaches to therapeutically interfere with chromatin function for the purpose of cancer treatment and immunomodulation.
Recent advances in the enzymology of transcription and chromatin regulation have led to the discovery of proteins that play a prominent role in cell differentiation and the maintenance of specialized cell functions. Knowledge about post-synthetic DNA and histone modifications as well as information about the rules that guide the formation of multimolecular chromatin-bound complexes have helped to delineate gene-regulating pathways and describe how these pathways are altered in various pathological conditions. The present review focuses on the emerging area of therapeutic interference with chromatin function for the purpose of cancer treatment and immunomodulation.
cancerchromatinimmunity
The immune system plays a pivotal role in the control of cancer growth and metastasis (Vesely et al. 2011); the failure of the immune system to detect and eliminate cancer cells that are phenotypically different from the surrounding tissues is a major cause of cancer. Therefore, therapeutic agents that negatively affect immune cell development or activation may have a negative effect on immune surveillance of malignant cells and thereby promote cancer growth (Schreiber et al. 2011). However, most of the currently licensed anti-cancer drugs have a negative impact on hematopoiesis and hence on the immune system. Chemotherapeutic-induced immunosuppression increases susceptibility to infections and may prevent the successful usage of cell-based therapies that rely on cancer cell killing by activated lymphocytes (Mellman et al. 2011). Cancer immunotherapy is poised to become increasingly relevant due to the ascent of potent approaches that rely on rampant T-cell activation caused by suppression of the inhibitory signaling pathways in these cells (Sharma et al. 2011).
Immune cell function is regulated by a large number of specialized transcription factors, which overlap only partially with transcription factors that operate in normal and malignantly transformed cells (Busslinger 2004; Naito et al. 2011; Smale 2012). However, the basic chromatin processes that control DNA accessibility to transcription factors or support the chromatin-coupled processes such as RNA polymerase II (Pol II) pausing, elongation, or splicing as well as many others are likely to be common among every cell type in the human body. Consequently, the systemic administration of drugs that target chromatin regulators in cancer cells is likely to affect the same set of proteins in immune cells and vice versa. However, the degree of dependence of various immune cell subsets on the activity of specific transcription factors and chromatin regulators is likely to be different than that of tumor cells. For example, the dependence of rapidly dividing and actively metabolizing tumor cells on transcriptional regulators of the Myc family distinguishes malignant and nonmalignant cells (Shaughnessy 2008; Dang 2012). Distinct requirements for transcriptional circuits in immune versus tumor cells may present an opportunity for the therapeutic targeting of chromatin processes involved in tumor cell growth or pathologically activated immune cells. Ultimately, the source for the selective drug effect may lie not in the nature of the target chromatin protein, which could be common among cell types, but in the differential dependence of individual genes on a particular chromatin regulator. In this context, knowledge of chromatin-based mechanisms of immune cell regulation may serve as a blueprint for the rational design of therapies to selectively target cancer cells with little impact on immune cell function. Furthermore, information about the negative effects of anti-cancer drugs on immunity may contribute to the development of comprehensive supplementary therapies aimed at alleviating immunological side effects.
A brief overview of chromatin processes associated with immune cell responses
The ultimate aim of the immune system is to achieve a degree of diversity that mirrors the complexity of pathogens in the environment. In cells of the adaptive immune system, represented by B and T lineage cells, this diversity is achieved largely through the generation of highly selective antigen receptors, which collectively cover the complexity of environmental antigens (Rajewsky 1996; Abbas and Janeway 2000). The interaction of antigens with the corresponding receptors on B or T lineage cells results in cell activation followed by eventual elimination of the antigen.
During antigen-driven immune responses, cells of the adaptive immune system engage in multiple intercellular interactions that involve nonimmune cells as well as cells of the innate immune system (Iwasaki and Medzhitov 2010). The latter are represented largely by the migrating or tissue-specific myeloid lineage cells (e.g., macrophages, glia cells, or neutrophils) which possess the ability to recognize pathogens or damaged tissues through the pattern recognition receptor system (Palm and Medzhitov 2009). Initially described by Medzhitov and Janeway (1997), the pattern recognition system senses chemically distinct pathogen components such as nucleic acids, membrane lipids, or, in the case of damaged host tissues, cellular proteins such as high-mobility group protein 1 (Palm and Medzhitov 2009). The specificity of pattern recognition relies on membrane-bound receptors such as Toll-like receptors (TLRs), intracellular nucleic acid sensors such as RIG-I and MDA5, or NOD-like receptors (Medzhitov 2001; Pasare and Medzhitov 2005; Meylan et al. 2006).
The diverse nature of immune cell receptor signaling is further augmented by the existence of a cytokine-driven signaling system that modulates immune and nonimmune cell responses (Bezbradica and Medzhitov 2009). This cytokine signaling is functionally intertwined with antigen and pattern receptor signaling and is essential for optimal immune responses. Cytokine involvement also allows for the generation of whole-body awareness of local inflammatory processes, whereby changes in the cytokine concentration in the blood can affect the function of the nonimmune organs such as the heart, liver, or brain (Medzhitov 2008).
Immune system diversification uses signaling-coupled and RNA transcription-coupled processes to increase cell-to-cell variability in individual gene expression within a range imposed by the immune cell lineage constraints. In turn, the efficacy of the immune response requires the existence of mechanisms that can markedly and coordinately activate a large number of genes (Zak and Aderem 2009; Smale 2012). Execution of transcription factor activity at the chromatin template largely depends on post-translational histone modifications, which define the level of DNA accessibility and contribute to the recruitment of proteins that support RNA transcription (Jenuwein and Allis 2001; Kouzarides 2007; Ruthenburg et al. 2007). Similar to other cell types, active genes in immune cells are associated with nucleosomes enriched for histone H3 trimethylated at Lys 4 (H3K4me3) and acetylated histones H3 and H4 (Wei et al. 2009; Cuddapah et al. 2010). Opposite to these “positive” histone marks, histone H3 dimethylation at Lys 9 (H3K9me2) or H3K27me3 has been implicated in gene silencing in immune cells (De Santa et al. 2007; Wei et al. 2009; Kruidenier et al. 2012), which contributes to cytokine production by activated macrophages and other cell types. Early studies by the Natoli group (Saccani and Natoli 2002) show rapid demethylation of H3K9me2 at promoters of genes activated by macrophage treatment with bacteria-derived lipopolysaccharide (LPS). An important role for H3K9me2 in the regulation of cytokine expression was confirmed by findings showing lower levels of H3K9me2 at inflammatory gene promoters in innate immune cells as compared with nonimmune cells (Fang et al. 2012).
Pathogen-derived proteins that target various arms of the immune response frequently antagonize immune cells' drive to achieve maximum diversity and efficiency. Both bacteria and viruses express proteins that can enter the nucleus and interfere with distinct processes, including transcription and chromatin regulation (Bierne et al. 2012). One of the most glaring examples is human adenovirus, which employs viral E1A protein to hijack host transcription by binding to the histone acetyltransferase CBP (Ferrari et al. 2008) or the ubiquitin ligase hBre1 (Fonseca et al. 2012). Bovine and human papilloma viruses express E2 protein that binds to the C-terminal portion of the bromodomain-containing BRD4 protein, coupling histone lysine acetylation to RNA Pol II activation and RNA elongation (You et al. 2004). Similarly, HIV Tat protein employs Brd4 for the activation of virus transcription (Bisgrove et al. 2007; Sobhian et al. 2010; Rice 2013). Pathogen interference with the host epigenome may also involve proteins that carry histone-like sequences or “histone mimics”—short amino sequences that are similar to the amino acid sequences within the N-terminal histone portions (Sampath et al. 2007). Histone mimics are present in numerous mammalian and pathogen-derived proteins (Lee et al. 2012). In some cases, such as the histone methyltransferase G9a, the histone mimic can fully recapitulate the protein-binding capacity of its histone H3 counterpart (Sampath et al. 2007). In other cases, histone mimics serve as recognition modules that enable post-translational modification of nonhistone proteins for purposes not directly linked to chromatin function (Lee et al. 2010, 2012; Donlin et al. 2012).
Pathogen-derived histone mimics can compete with host histones for common binding partners required for transcriptional activation. For example, the histone mimic within the C-terminal portion of the immunosuppressive NS1 protein of the H3N2 influenza virus binds directly to the transcriptional elongation regulator PAF1 (polymerase-associated factor 1) (Zhou et al. 2012). By binding to the Paf1, which represents an essential part of the PAF1C (PAF1 complex) elongation complex (Krogan et al. 2003; Kim et al. 2010), NS1 inhibits elongation of virus-induced genes and attenuates the antiviral response (Marazzi et al. 2012). Accordingly, siRNA-induced suppression of Paf1 also attenuates the antiviral response and increases viral replication (Marazzi et al. 2012).
The ability of pathogen-derived proteins to control gene expression by interfering with gene transcription is reminiscent of drug-induced suppression of transcription in mammalian cells. Proteins such as BRD4 (see below) have been the focus of studies aimed at suppressing tumor growth and even reversing dedifferentiation associated with malignant transformation. However, the rise of novel anti-cancer drugs targeting transcriptional circuits in tumor cells requires an assessment of the potential effect of these drugs on similar transcriptional processes in immune cells.
Recent reviews have comprehensively described the multitude of chromatin processes that could be targeted for the purpose of cancer treatment. Here, we focus exclusively on transcription circuits that involve the histone methyltransferase Ezh2 and double-bromodomain-containing transcriptional regulators (BET [bromodomain and extraterminal]). Compounds that target these two classes of proteins display anti-cancer activity in vitro and in vivo and are considered potential candidates for cancer treatment in humans. Additionally, BET protein inhibitors have a marked impact on proinflammatory gene expression and can attenuate systemic inflammatory processes in mice (Nicodeme et al. 2010; Belkina et al. 2013).
Targeting Ezh2 for cancer treatment and its effect on immunity
The histone methyltransferase Ezh2, which catalyzes HK27me3 (Cao et al. 2002; Czermin et al. 2002), could be legitimately considered as a poster child of modern cancer “epigenetics,” which largely studies the transcriptional response and has little in common with “epigenetics” in Waddington's sense of it. At the time of writing this review, a PubMed search for “Ezh2 and cancer” yielded 620 references; the initial prominence of Ezh2 as a potential target for cancer therapy has been largely driven by findings of Ezh2 overexpression in metastatic prostate cancer cells as well as in rapidly progressing cancers of other types (van Kemenade et al. 2001; Varambally et al. 2002; Kleer et al. 2003).
Ezh2 operates within the PRC2 protein complex, where the presence of other PRC2 components such as Eed, Suz12, and RbAp48 is essential for the catalysis of H3K27me3 (Cao and Zhang 2004; Margueron and Reinberg 2011). In addition to the core components of PRC2, targeting PRC2 to specific gene loci in mammalian cells involves auxiliary proteins such as Jarid2 (Peng et al. 2009; Landeira et al. 2010; Li et al. 2010; Pasini et al. 2010). Most of the studies that deal with Ezh2 overexpression in tumors, including the initial description of Ezh2 overexpression in prostate cancer cells (Varambally et al. 2002), omit a detailed analysis of Ezh2-associated PRC2 components. Therefore, it is not clear how Ezh2 overexpression itself could be sufficient to affect H3K27me3, which requires the stochiometrically assembled PRC2.
It is generally assumed that Ezh2 expression controls cell division (van Kemenade et al. 2001; Pasini et al. 2004). The expression levels of PRC2 increase in dividing cells, and suppression of the Ezh2 in vitro by siRNA or specific inhibitors frequently leads to growth arrest (Varambally et al. 2002; Martinez-Garcia and Licht 2010). However, the mechanism of this suppression as well as the nature of genes or signaling processes that mediate it remain largely elusive. Confusion about the role of H2K27me3 and Ezh2 in cancer is compounded by studies showing the potential oncogenic function of both the loss-of-function and gain-of-function Ezh2 mutant proteins (Ernst et al. 2010; Martinez-Garcia and Licht 2010; Morin et al. 2010; Guglielmelli et al. 2011; Jankowska et al. 2011; Vainchenker et al. 2011; Yap et al. 2011; Ntziachristos et al. 2012). Recent work by Allis' group (Lewis et al. 2013) raised the question about the overall role of H3K27me3 in the regulation of tumor phenotypes. In highly malignant human gliomas, interactions between Ezh2 and a mutated histone H3 variant (in which the lysine at position 27 is substituted by the methionine) inactivate Ezh2 (Lewis et al. 2013). Tumors that carry H3K2K7-M mutations display negligible levels of H3K27me3, but this does not affect either cell division or the degree of malignancy. Essentially, these studies, along with the independent observation of Ezh2 inactivation by H3K2K7-M (Chan et al. 2013), show that neither Ezh2 activity nor H3K27me3 is required to support tumor growth and maintenance of the malignant phenotype.
The tumor-suppressive effect of Ezh2 inhibitors may also reflect the growth-promoting effect of Ezh2 up-regulation of specific targets. A limited set of data suggests that PRC2 can directly stimulate gene expression. In muscle cells, the Ezh2 homolog Ezh1 positively supports transcriptional elongation (Mousavi et al. 2012). Similarly, PRC2 has been implicated in the positive regulation of actively transcribed cytokine genes (Jacob et al. 2008, 2011). Similarly, the oncogenic function of EZH2 in castration-resistant prostate cancer is independent of its role as a transcriptional repressor but involves the ability of EZH2 to act as a coactivator for critical transcription factors, including the androgen receptor (Xu et al. 2012).
Despite the lack of sufficient clarity about the mechanism of Ezh2 involvement in carcinogenesis and tumor progression, a significant effort has been put into Ezh2 inhibition for the purpose of cancer therapy (Melnick 2012). Treatment of diffuse large B-cell lymphoma (DLBCL) that harbors Ezh2-activating mutations with a potent small-molecule inhibitor of EZH2 methyltransferase activity decreased global H3K27me3 levels, reactivated silenced PRC2 target genes, and inhibited the proliferation of these EZH2 mutant DLBCL cells (McCabe et al. 2012).
Before Ezh2 inhibitors gain further support as anti-tumor drugs, the potential side effect for relatively broad immunosuppression should be considered. Ezh2 deficiency in developing B lineage cells diminishes the antibody repertoire (Su et al. 2003) due to the selective impairment of VH gene rearrangement residing at the 5′ end of the 2.3-Mb-long immunoglobulin heavy chain (igH) gene locus (Malin et al. 2010). These so-called “distal” VH genes comprise a large portion of the immunoglobulin repertoire in mice and humans (Ebert et al. 2011). Perturbations in distal VH rearrangement cause partial humoral immunodeficiency associated with a paucity of certain antibody classes. In addition to VH gene rearrangement, Ezh2 is also involved in rearrangement of immunoglobulin light chain genes via a mechanism involving IL-7 receptor signaling (Mandal et al. 2011). Ezh2 is expressed at high levels in germinal center (GC) B cells, which are involved in the generation of antibodies with high affinities to their antigen (Velichutina et al. 2010). In GC B cells, EZH2 targets a large number of GC-specific targets, thus suggesting a key role for Ezh2 in GC B-cell function. Consequently, Ezh2 deficiency in GC B cells may reduce the efficacy of the long-lasting humoral immune response that relies on the presence of B cells expressing high-affinity antibodies.
In addition to its impact on B cells, Ezh2 deficiency also has a negative effect on T-cell immunity. Lack of Ezh2 in early hematopoietic progenitors prevents expansion of early T-cell precursors in the thymus (Su et al. 2005). Furthermore, Ezh2 deficiency reduces the T-cell antigen receptor (TCR)-driven proliferation of T cells in vitro and abrogates antigen-driven T-cell responses in vivo (He et al. 2012). Administration of the histone methylation inhibitor 3-deazaneplanocin A (DZNep) arrests ongoing T-cell-induced graft rejection in mice after allogeneic bone marrow transplantation and selective apoptosis in alloantigen-activated T cells mediating host tissue injury.
The nonnuclear function of Ezh2 in T-cell activation has been linked to the ability of cytosolic Ezh2 to control signal-driven actin polymerization in an H3K27me3-independent but methyltransferase-dependent fashion (Su et al. 2005). A cytosolic presence of PRC2 components has been reported for Eed (Witte et al. 2004) and Ezh1 (Ogawa et al. 2003) in addition to Ezh2 (Su et al. 2005; Bryant et al. 2008; Gonzalez et al. 2011). In T cells, cytosolic Ezh2 binds to the signaling protein Vav1, which plays an essential role in actin polymerization (Hobert et al. 1996a,b; Nolz et al. 2005; Su et al. 2005). Binding to Vav1 may explain the ability of overexpressed Ezh2 to activate the phosphoinositide 3-kinase (PI3K)/Akt pathway in breast cancer cells (Gonzalez et al. 2011) and control actin polymerization in the prostate cancer cells (Bryant et al. 2008). The exact nature of cytosolic Ezh2 substrates is not known, but the likelihood of these substrates' existence is highlighted by the recent identification of several nonhistone targets of Ezh2, such as RORα (Lee et al. 2012). It is tempting to speculate that Ezh2 involvement in actin polymerization—a process that plays a prominent role in tumor cell invasion (Kim et al. 2009)—may contribute more to tumor spreading than do putative changes in gene expression caused by altered Ezh2 expression or activity levels.
Ezh2 has been indirectly implicated in the regulation of the inflammatory gene expression in the innate immune cells. Activation of macrophages by LPS is associated with a selective increase in the H3K27me3-specific demethylase JMJD3 (De Santa et al. 2007). Conversely, JMJD3 deficiency or pharmacological inhibition affects macrophage gene expression (Kruidenier et al. 2012). In light of these findings, one would expect to see an increase in proinflammatory gene expression following suppression of Ezh2.
Potential effects of Ezh2 on the immune system have to be viewed differently in the context of solid tumors and blood malignancies. Unless delivered in a targeted fashion, systemically applied Ezh2 inhibitors may have a more immediate impact on the immune cells than on tumor cells. In such a case, immunodeficiency may precede a potential anti-tumor effect. In fact, in the case of blood malignancies, the immunosuppressive function of Ezh2 inhibition could be advantageous for the treatment of B- or T-cell lymphomas that rely on Ezh2-mediated cytosolic and nuclear signaling networks.
Targeting bromodomain-containing transcriptional regulators and effects on immunity
Lysine acetylation on various histone molecules as well as on nonhistone nuclear proteins is a hallmark of and prerequisite for transcription in mammalian cells (Cheung et al. 2000; Bannister and Kouzarides 2011). Acetyl-lysines bind to highly conserved bromodomains present in numerous cellular proteins and play an essential role in the assembly of protein networks that control gene expression (Zeng and Zhou 2002; Mujtaba et al. 2007). Recently, the tandem BET domain-containing BET proteins BRD2, BRD3, BRD4, and BRDT became the focus of studies on the pharmacological control of gene expression (Arrowsmith et al. 2012; Prinjha et al. 2012). Before gaining prominence as potential drug targets, BET proteins were known as regulators of gene expression in vitro and in vivo (Belkina and Denis 2012). Some of the BET proteins, such as BRD4, were found in association with the Mediator complex to play a prominent role in transcription (Jiang et al. 1998). In addition to its association with Mediator, BRD4 gained additional importance as the only ubiquitously expressed BET protein that can bind directly to p-TEFb through an extended C-terminal domain (Dey et al. 2003; Peterlin and Price 2006; Zhou et al. 2012). Connection to p-TEF-B provided a direct link between BRD4 binding to the acetyl-lysines and transcriptional elongation (Zhou et al. 2012). BRD2 and BRD3 can also participate in elongation through their association with the RNA Pol II-associated elongation complex PAF1C (Dawson et al. 2011). The PAF1C/BET interaction likely relies on bromodomain-unrelated sequences such as ET or other not yet identified motifs (Rahman et al. 2011). The interaction of the ET domain of BETs with numerous effector proteins such as NSD3 (a SET domain-containing histone methyltransferase), JMJD6 (a histone arginine demethylase), and CHD4 (a catalytic component of the NuRD nucleosome remodeling complex) points to active BET involvement in chromatin modifications that might be required for transcriptional elongation (Rahman et al. 2011). The complexity of BET-mediated regulation is further increased by the ability of BET proteins such as BRD4 or BRD2 to bind directly to acetylated lysines within transcription factors such as NF-κB or GATA1 (Huang et al. 2009; Lamonica et al. 2011). It is important to note that the nucleus may not be the only place where BET functions. Brd2 translocation from the cytoplasm to the nucleus is controlled by serum factors (Guo et al. 2000), which suggests a signal-dependent control of Brd2-mediated gene regulation. In the developing mouse neural tube and dorsal root ganglia, Brd2 localized to the nucleus during proliferation but was predominantly cytoplasmic when cells were terminally differentiated (Crowley et al. 2004).
The involvement of BET proteins in transcriptional regulation and the specificity of BET bromodomain binding to acetyl-lysines provided a foundation for the development of synthetic compounds that control gene expression by inhibiting BET binding to acetylated lysines. The first generation of BET inhibitors was developed independently by several groups, including GlaxoSmithKline (I-BET) in collaboration with our group (Nicodeme et al. 2010) and Bradner's group (Filippakopoulos et al. 2010) (JQ1) in collaboration with the Structural Genomics Consotium (SGC). The lion's share of studies that use BET inhibitors describe the effect of the inhibitors on tumor growth (Dawson et al. 2011; Delmore et al. 2011; Zuber et al. 2011; Dawson and Kouzarides 2012; Loven et al. 2013). It appears that MYC-overexpressing tumors are especially sensitive to BET inhibitors, which suppress MYC expression followed by a dampening of the magnitude of the MYC-driven transcriptional response (Delmore et al. 2011; Ott et al. 2012; Loven et al. 2013; Puissant et al. 2013). In addition, BET inhibitors act effectively against tumors that express a rare form of an oncogenic BRD4-NUT fusion protein (Filippakopoulos et al. 2010). The ability of BET inhibitors to suppress tumor growth is not entirely unexpected in light of the earlier studies of Ozato's group (Maruyama et al. 2002; Dey et al. 2003), which demonstrated a crucial role for BRD4 during mitosis.
One of the first studies of BET inhibitors (GSK525762 compound, designated as I-BET) revealed a potent impact on inflammatory gene expression (Nicodeme et al. 2010). This finding was partly expected due to previous studies revealing an important role for BRD4 in the regulation of proinflammatory gene expression. BRD4-supported cotranscriptional mRNA splicing is important for controlling LPS-inducible inflammatory gene expression in macrophages (Hargreaves et al. 2009). In the absence of stimulation, RNA Pol II generates low levels of full-length but unspliced and untranslatable transcripts at many of the LPS-induced response genes (Hargreaves et al. 2009). Compared with BRD4, which appears to be a generic regulator of elongation, BRD2, which also binds to I-BET, could play a more selective role in the regulation of immune response genes. The positioning of the BRD2 gene within the myosin heavy chain (MHC) class II gene cluster on human chromosome 6 or in syntenic regions of other organisms (Belkina and Denis 2012) could be seen as a sign of specialized BRD2 involvement in immune responses. In support of this model, low levels of BRD2 are associated with the reduced cytotoxic cytokine production by in vitro triggered macrophages (Belkina et al. 2013).
One of the surprising outcomes of studies on the effect of I-BET on LPS-triggered macrophages was the rather selective impact of I-BET on gene expression (Nicodeme et al. 2010). A common theme that has emerged from studies of LPS-inducible genes is a connection between the timing of gene expression and the state of the chromatin associated with the promoter and transcriptional start site of the inducible gene. Different temporal patterns of gene expression in response to LPS appear to be embedded within the CpG content of inducible gene promoters (Natoli et al. 2011; Smale 2012). In macrophages, CpG island-rich promoters are prevalent among primary and weakly induced secondary response genes, while CpG-low promoters are much more prevalent among more highly induced secondary response genes (Bhatt et al. 2012). BET proteins are associated with both primary and secondary response genes at relatively similar levels before LPS induction of macrophages (Nicodeme et al. 2010). However, treatment of macrophages in vitro with I-BET resulted in the strong and selective attenuation of secondary response gene expression while leaving the expression of primary response genes largely unaffected (Nicodeme et al. 2010). Most significantly, the selective effect of I-BET on secondary response genes holds true for macrophage as well as fibroblast responses to not only LPS, but secondary mediators of the inflammatory responses such as TNF or type I interferon (IFN).
The ability of I-BET to profoundly suppress numerous proinflammatory genes in cells of the innate as well as adaptive immune system (Bandukwala et al. 2012) should be considered, as it might affect the host response to pathogens. The negative impact of I-BET on immunity is particularly relevant in cancer patients who may additionally suffer from the effects of tumor- and/or chemotherapy-induced immunosuppression.
Summary
Rapidly emerging information about the transcriptional control of tumor growth and activated immune cells is likely to continue to fuel excitement about the therapeutic promise of drugs that target catalytic activity of the chromatin-modifying enzymes as well as protein–protein interactions within chromatin-bound regulatory complexes. Additional data on the aberrant structure of tumor-expressed chromatin regulators may further the rational design of therapeutics to achieve a therapeutic effect in tumor cells but spare healthy tissues. The effort to affect tumor growth by interfering with overexpressed—but structurally unaltered—chromatin regulators may cause potential side effects, including immunosuppression. However, the likely differences in gene regulation in tumor cells and cells of the immune system may provide an opportunity for the selective targeting of gene circuits involved in a particular type of tumor cell. Ultimately, however, using chromatin targeting drugs to control systemic immune disorders without causing broad immunosuppression or persistent damage to nonimmune cells will require a better understanding of the transcriptional pathways that govern pathological functions of immune cell subsets that drive the disease.
Acknowledgments
This work has been supported by the Lupus Research Institute and the Starr Cancer Consortium (A.T.).
Article is online at http://www.genesdev.org/cgi/doi/10.1101/gad.221895.113.
The interdependence of p53 and MDM2 is critical for proper cell survival and cell death. Zhu et al. find that TAB1, an activator of TAK1 and p38α, inhibits the E3 ligase activity of MDM2 toward p53 and its homolog, MDMX. Cisplatin-induced cell death is mitigated by TAB1 knockdown. TAB1 stabilizes MDMX and activates p38α to phosphorylate p53, allowing p53 target induction. TAB1 levels are relatively low in cisplatin-resistant clones of ovarian cells and in ovarian tumors, implicating TAB1 as a tumor suppressor.
The interdependence of p53 and MDM2 is critical for proper cell survival and cell death and, when altered, can lead to tumorigenesis. Mitogen-activated protein kinase (MAPK) signaling pathways function in a wide variety of cellular processes, including cell growth, migration, differentiation, and death. Here we discovered that transforming growth factor β-activated kinase 1 (TAK1)-binding protein 1 (TAB1), an activator of TAK1 and of p38α, associates with and inhibits the E3 ligase activity of MDM2 toward p53 and its homolog, MDMX. Depletion of TAB1 inhibits MDM2 siRNA-mediated p53 accumulation and p21 induction, partially rescuing cell cycle arrest induced by MDM2 ablation. Interestingly, of several agents commonly used as DNA-damaging therapeutics, only cell death caused by cisplatin is mitigated by knockdown of TAB1. Two mechanisms are required for TAB1 to regulate apoptosis in cisplatin-treated cells. First, p38α is activated by TAB1 to phosphorylate p53 N-terminal sites, leading to selective induction of p53 targets such as NOXA. Second, MDMX is stabilized in a TAB1-dependent manner and is required for cell death after cisplatin treatment. Interestingly TAB1 levels are relatively low in cisplatin-resistant clones of ovarian cells and in ovarian patient's tumors compared with normal ovarian tissue. Together, our results indicate that TAB1 is a potential tumor suppressor that serves as a functional link between p53–MDM2 circuitry and a key MAPK signaling pathway.
p53MDM2MDMXTAB1p38αcisplatin
The protein product of the p53 gene that is frequently mutated in human tumors plays a central role in coordinating the cellular responses to a wide variety of stress signals (Vousden and Prives 2009). In those tumors that harbor wild-type p53 protein, the functions of p53 are often compromised by overexpression of MDM2 or its homolog, MDMX (Marine et al. 2006). Both MDM2 and MDMX are crucial negative regulators of p53. They bind to the N-terminal transactivation domain of p53 and can thereby inhibit p53 target gene transcription. MDM2 also functions as an E3 ubiquitin ligase promoting ubiquitination and proteasomal degradation of p53, MDMX, and itself (Marine and Lozano 2010; Wade et al. 2010). MDMX is not active as an E3 ligase but can form heteroligomers with MDM2 and regulate the E3 ligase function of MDM2 (Shadfan et al. 2012). Both MDM2 and MDMX are crucial for keeping p53 in check under nonstressed conditions in mice, since deletion of either gene leads to embryonic lethality unless mice also lack functional p53 alleles (Parant et al. 2001; Migliorini et al. 2002; Marine et al. 2006). This suggests that MDM2 and MDMX play nonredundant roles in the regulation of p53. Interplay between p53, MDM2, and MDMX may determine whether a cell responds to p53 activation with growth arrest or apoptosis (Wade et al. 2010). In response to various sources of cellular stress, multiple post-translational modifications, including phosphorylation, regulate the association between MDM2/MDMX and p53, and, in turn, such modifications lead to p53 stabilization and activation (Kruse and Gu 2009; Waning et al. 2010). Most relevant to this study, numerous protein kinases—including members of the mitogen-activated protein kinase (MAPK) family such as p38α, JNK (c-Jun N-terminal kinase), and ERK—have been shown to functionally interact with as well as phosphorylate and activate p53, leading to p53-mediated cellular responses in response to stress stimuli (Wu 2004). Modulation of MDM2 and MDMX phosphorylation also affects their respective functions, especially toward p53 (Chen 2012). Furthermore, several members of the dual specificity phosphatase family—such as Wip1, MKP1, PAC1, and DUSP5—that negatively regulate MAPK signaling have been shown to be p53 targets (Janicke et al. 2008), while MDM2 and MDMX have also been shown to be substrates of MAPKs as well as phosphatases such as Wip1 (Lu et al. 2007b; Malmlof et al. 2007; Gilkes et al. 2008; Zhang et al. 2009). Both p53 and MAPK signaling pathways regulate a wide variety of cellular processes, and alterations in either are often associated with cancer (Dhillon et al. 2007; Vousden and Prives 2009). A better understanding of the interplay between these two critical networks would provide valuable insights for tumorigenesis and therapy resistance.
Transforming growth factor β (TGFβ)-activated kinase 1 (TAK1)-binding protein 1 (TAB1, also known as MAP3K7IP1) is a modular adapter protein that was initially described as an activator of TAK1 (also known as MAP3K7) in response to TGFβ (Shibuya et al. 1996). TAK1 acts as an upstream kinase for NF-κB and MAPK activation in response to multiple stress signals (Ono et al. 2001; Johnson 2002). TAB1 also regulates p38α autoactivation through direct binding (Ge et al. 2002; Johnson 2002). Furthermore, AMP-activated protein kinase (AMPK) binds to TAB1, increasing recruitment and activation of p38α in the ischemic heart (Li et al. 2005). It was also documented that TAB1 and IKKβ (IκB kinase β) form complexes albeit solely in breast cancer cells or normal mammary epithelial cells following the induction of the epithelial–mesenchymal transition by TGF-β (Neil and Schiemann 2008). TAB1 possesses several seemingly distinct protein-binding regions. In addition to the extreme C-terminal 68 amino acids that are sufficient for binding to and activation of TAK1, the C-terminal portion of TAB1 also contains a p38α-binding domain that precedes the TAK1-binding region. Within its N-terminal portion, TAB1 contains a PP2C (protein phosphatase 2C)-like domain, and structural and biochemical analysis suggests that TAB1 may function as a pseudophosphatase (Conner et al. 2006). This PP2C-like region was also found to interact with XIAP, a member of the inhibitor of apoptosis proteins (IAP) family, which has been shown to play roles in signaling to NF-κB and MAPK activation (Yamaguchi et al. 1999; Lu et al. 2007a). Therefore, through its association with several protein complexes, TAB1 is involved in multiple signaling pathways.
Although TAB1 and TAK1 are believed to be functionally entwined, each has distinct roles in the MAPK signaling cascade, and they may not always function together as a complex. TAK1-deficient mouse embryos die at around embryonic day 10 (E10) and exhibit abnormal development of the neural tube, while TAB1-deficient mice die at a later stage of gestation due to abnormal cardiovascular and lung morphogenesis (Komatsu et al. 2002; Shim et al. 2005; Inagaki et al. 2008). Additional observations have disputed an essential role for TAB1 in TAK1 activation (Shim et al. 2005; Bertelsen and Sanfridson 2007; Mendoza et al. 2008).
Here, we identified TAB1 as an MDM2-binding partner that modulates the E3 ligase activity of MDM2, leading to stabilization of MDM2, MDMX, and p53. We discovered that TAB1 regulates p53-mediated outcomes such as cell cycle arrest under some conditions and cell death initiated by cisplatin treatment. Furthermore, our data implicate both TAB1 activation of p38α and stabilization of MDMX in facilitating apoptosis in response to cisplatin. We hypothesize that TAB1 serves as a functional link to regulate cross-talk between p53 and MAPK signaling pathways.
ResultsTAB1 interacts with MDM2
A high-throughput yeast two-hybrid screen was carried out to identify MDM2-interacting proteins (see Zhu et al. 2009 for more information), and multiple interacting clones encoding TAB1 were identified from three different cDNA libraries. The association of endogenous TAB1 and MDM2 in human cells was then demonstrated using cell lysates from U2OS cells (Fig. 1A) and HCT116 cells (Supplemental Fig. S1A); the specificity of the interaction was confirmed by the diminished signal seen when the two proteins were coimmunoprecipitated from extracts of U2OS cells treated with either TAB1 or MDM2 siRNA (Fig. 1A).
TAB1 stabilizes p53 by inhibiting MDM2. (A) Specificity of binding between endogenous MDM2 and TAB1. (Left panel) Whole-cell lysates (500 μg) from U2OS cells were immunoprecipitated with a rabbit polyclonal anti-TAB1 antibody (α-T) or control rabbit IgG and then subjected to immunoblotting with anti-MDM2 (3G5+4B11+5B10) and anti-TAB1 antibodies. (Right panel) U2OS cells were transfected with siRNAs targeting luciferase (C), TAB1 (T1), or MDM2 (M). After 48 h, whole-cell lysates (500 μg) were prepared and subjected to immunoprecipitation with rabbit polyclonal anti-TAB1 (α-T) followed by immunoblotting with anti-MDM2 (3G5+4B11+5B10), anti-TAK1, and anti-TAB1 antibodies. Short (SE) and longer (LE) exposures of the TAB1 immunoblot are shown. (B) TAB1 blocks degradation of p53 by coexpressed MDM2. U2OS cells were transfected with Myc-TAB1 (1.5 μg), Flag-MDM2 (1.5 μg), and HA-p53 (0.35 μg) constructs as indicated. Cell lysates were used for immunoblotting with anti-HA, and anti-Flag antibodies. A coexpressed GFP construct was added in each case to control for transfection efficiency and loading. (C) Ectopic expression of TAB1 stabilizes endogenous MDM2 and p53. U2OS cells were transfected with increasing amounts of Myc-TAB1 (0.5, 1, 2, and 3 μg) as indicated. Cell lysates were used for immunoblotting with anti-MDM2 (3G5+4B11+5B10), anti-p53 (DO-1), anti-p21, anti-Myc, and anti-Actin antibodies. (D) TAK1 cooperates with TAB1 to inhibit MDM2-mediated p53 degradation. U2OS cells were transfected with HA-p53 (0.3 μg) and combinations of Flag-MDM2 (1.5 μg), Myc-TAB1 (0.5 or 1 μg), pc-TAK1 (0.5 or 1 μg), or kinase-defective pc-TAK1 K63W (TAK1KW; 0.5 or 1 μg) as indicated. Cell lysates were used for immunoblotting with anti-Flag, anti-HA, anti-Myc (9E10), and anti-TAK1 antibodies. A GFP construct was cotransfected in each case to control for transfection efficiency and loading.
To map the region of MDM2 that is required for TAB1 binding, a series of Flag-tagged MDM2 deletion and truncation mutants was constructed and transiently overexpressed in H1299 cells together with Myc-tagged TAB1. Cell extracts were subjected to immunoprecipitation with an anti-Myc antibody followed by immunoblotting with an anti-Flag antibody (Supplemental Fig. S1). Our results indicate that amino acids 223–339 span the primary binding sites for TAB1. However, the MDM2 N terminus (1–222) and MDM2 C-terminal RING-containing regions also interacted with TAB1 albeit weakly, indicating that either TAB1 interacts with multiple surfaces on MDM2 or its interactions are dependent on the tertiary structure of MDM2.
TAB1 inhibits MDM2-mediated p53 degradation and ubiquitination
To determine the functional consequences of the TAB1–MDM2 association, we coexpressed TAB1 with MDM2 and p53 in U2OS cells. MDM2-mediated degradation of p53 was markedly inhibited by TAB1 (Fig. 1B). More significantly, the levels of endogenously expressed p53 protein and two p53 targets (MDM2 and p21) in U2OS cells were elevated following ectopic TAB1 expression (Fig. 1C). Ectopic TAK1 and TAB1 cooperated to inhibit MDM2-mediated p53 degradation (Fig. 1D), and TAB1 and TAK1 mutually stabilized each other (Fig. 1D, cf. lanes 3 and 5 for TAK1 stabilization of TAB1 and lanes 5 and 6 for TAB1 stabilization of TAK1). It is therefore possible that the inhibitory effect of TAK1 on MDM2 was mediated by TAB1. A kinase-dead mutant of TAK1 (K63W) (Yamaguchi et al. 1995) that stabilized TAB1 to a lesser extent (Fig. 1D, cf. lanes 5 and 8) also inhibited p53 degradation to a correspondingly reduced extent. Note that both wild-type TAK1 and its kinase-dead derivative by themselves had no inhibitory effect on MDM2 degradation of p53.
In keeping with its ability to stabilize p53, TAB1 expression inhibited the ability of MDM2 to ubiquitinate p53 (Fig. 2A). Ectopic TAB1 also modestly repressed MDM2 autoubiquitination (Fig. 2B) and, more significantly, inhibited MDM2 ubiquitination of MDMX, one of its well-known E3 ligase substrates (Fig. 2C). These results suggest that TAB1 functions as a general inhibitor of the E3 ligase activity of MDM2 albeit to varying extents. Note that TAB1 is not a universal inhibitor of E3 ligase/proteasome machinery, as it had no inhibitory effect on E2F1 ubiquitination, which has been shown to be mediated by the E3 ligase Skp2 (Supplemental Fig. S2; Marti et al. 1999).
Ectopic expression of TAB1 inhibits E3 ligase activity of MDM2. (A) TAB1 inhibits MDM2-mediated p53 polyubiquitination. H1299 cells were transfected with the indicated combinations of Myc-TAB1 (0.6 or 1.2 μg), Flag-MDM2 (1.2 μg), and p53 (0.3 μg) plasmids along with an HA-ubiquitin (HA-Ub; 1.2 μg) plasmid as indicated. The cells were treated with MG132 (20 μM) for 4 h before harvesting. Whole-cell lysates were subjected to immunoprecipitation with anti-p53 (FL-393-G) antibody followed by immunoblotting with anti-HA antibody to detect ubiquitinated p53. (B) TAB1 inhibits MDM2 autoubiquitination. H1299 cells were transfected with combinations of Myc-TAB1 (1.2 μg) and Flag-MDM2 (1.2 μg) plasmids in the presence of an HA-ubiquitin (HA-Ub; 1.2 μg) plasmid as indicated. The cells were treated with MG132 as in A before harvesting. Whole-cell lysates were subjected to immunoprecipitation with anti-Flag antibody followed by immunoblotting with anti-HA antibody to detect ubiquitinated MDM2. (C) TAB1 inhibits MDM2-mediated MDMX ubiquitination. H1299 cells were transfected with combinations of HA-MDMX (0.6 μg), Flag-MDM2 (0.8 or 1.6 μg), and Myc-TAB1 (0.5 or 1.0 μg) plasmids in the presence of a His-ubiquitin (His-Ub; 1.4 μg) plasmid as indicated. The cells were treated with MG132 as in A before harvesting. Cells were lysed in denaturing buffer and subjected to Ni-NTA bead binding as described in the Materials and Methods. Ubiquitinated MDMX was detected by an anti-MDMX antibody.
TAB1 ablation attenuates p53 activation that results from knockdown of MDM2
To evaluate how TAB1 regulates MDM2 when expressed at normal endogenous levels, we used two different siRNAs to examine the effect of TAB1 down-regulation on the functions of MDM2 and p53. Depletion of TAB1 in U2OS cells did not affect the cellular levels of p53 and MDM2 (Fig. 3A). When we introduced MDM2 siRNA into U2OS cells, as expected, p53 was stabilized, and p21 expression was increased. Interestingly, upon down-regulation of TAB1, MDM2 knockdown-mediated p53 stabilization was attenuated, and the levels of p21 protein (Fig. 3A) and RNA (Fig. 3B) were also markedly decreased. In line with this, ablation of TAB1 partially rescued cell cycle arrest resulting from MDM2 knockdown in U2OS cells (Fig. 3C). Similar restoration of the cell cycle was observed when MDM2 and/or TAB1 were ablated in HCT116 or RKO cells (Supplemental Fig. S3). Since p53 stabilization and p21 expression induced by Nutlin (a small molecule that disrupts the p53–MDM2 interaction) were not attenuated upon TAB1 ablation, we surmise that TAB1 modulates p53 function through MDM2 (Fig. 3D).
TAB1 is required for full p53 activation upon ablation of MDM2. (A) Down-regulation of TAB1 attenuates p53 activation in cells depleted of MDM2. U2OS cells were transfected with 25 nM each control luciferase siRNA (C), two different TAB1 siRNAs (T1 and T2), or MDM2 siRNA (M) as indicated to a total siRNA concentration of 50 nM in all samples (balanced with control luciferase siRNA). Forty-eight hours after transfection, cells were harvested, and cell lysates were prepared and then subjected to immunoblotting with anti-MDM2 (3G5+4B11+5B10), anti-MDMX, anti-TAB1, anti-p53 (DO-1), anti-p21, or anti-Actin antibodies. (B) Down-regulation of TAB1 attenuates p53-induced p21 transcription in response to MDM2 ablation. U2OS cells were transfected with 25 nM each control luciferase siRNA (C), MDM2 siRNA (M), TAB1 siRNA (T1 or T2), or both MDM2 and TAB1 siRNAs (T1M or T2M) as indicated to a total siRNA concentration of 50 nM as in A. Forty-eight hours after transfection, cells were harvested, and total mRNAs were extracted. cDNAs were then reverse-transcribed, and quantitative real-time PCR was performed using p21 gene-specific primers. (C) Reducing TAB1 levels attenuates cell cycle arrest induced by MDM2 ablation. U2OS cells were transfected with siRNAs as in A. Cells were trypsinized and fixed followed by cell cycle analysis. The plot was obtained from three separate experiments. (D) Attenuation of p53-induced response by ablation of TAB1 requires p53–MDM2 interaction. U2OS cells were transfected with 25 nM each control luciferase siRNA (C) or two different TAB1 siRNAs (T1 and T2). Forty-two hours after transfection, cells were treated with DMSO or Nutlin (5 μM). Six hours later, cells were harvested, and cell lysates were prepared and then immunoblotted with anti-MDM2 (3G5+4B11+5B10), anti-p53 (DO-1), anti-MDMX, anti-TAB1, anti-p21, or anti-Actin antibodies.
TAB1 mediates the apoptotic response of p53 to cisplatin by selective induction of NOXA expression
Based on its ability to prevent MDM2 from degrading p53, we anticipated that TAB1 would be required to regulate the MDM2–p53 circuit under a variety of stress conditions. To test this, we examined the impact of TAB1 knockdown on a variety of different genotoxic assaults—including 5-fluorouracil, actinomycin D, cisplatin, daunorubicin (Dauno), deferoxamine mesylate, doxorubicin, etoposide (ETP), hydoxyurea, neocarzinostatin (NCS), and taxol—in U2OS cells. Unexpectedly, reduction of TAB1 levels had little or no effect on cell cycle arrest or cell death caused by all but one of these various agents (Supplemental Table S1). The one striking exception was cisplatin, where ablation of TAB1 markedly attenuated cell death induced by this agent as measured by analysis of cells with sub-G1 DNA content (Fig. 4A) or PARP cleavage (Fig. 4B). In both apoptosis assays, siRNA-mediated knockdown of TAK1 also impacted cisplatin-mediated cell death (Fig. 4A,B). The effects of cisplatin were completely abolished by the caspase inhibitor z-VAD-fmk, indicating that in this setting, cell death occurred by the process of apoptosis (Supplemental Fig. S4). As a control, ETP-induced cell death, which, as mentioned above, was not significantly attenuated by ablation of TAB1, was also abolished by z-VAD-fmk treatment (Supplemental Fig. S4). As previously reported (Bragado et al. 2007), cisplatin treatment resulted in substantially more cell death in HCT116 (p53+/+) than in HCT116 (p53−/−) cells, indicating that the drug requires p53 to produce apoptosis (Fig. 4C). Correspondingly, ablation of TAB1 significantly reduced cisplatin-induced cell death in HCT116 (p53+/+) cells but not in HCT116 (p53−/−) cells, indicating that TAB1 is involved in the p53-mediated apoptotic response (Fig. 4C). TAB1 knockdown also modulated cisplatin-induced cell death in SkHep1 liver cancer cells containing wild-type p53 but not in the p53-null Saos2 osteosarcoma cell line (Supplemental Fig. S5). Consistent with these findings, a colony formation assay confirmed that TAB1-ablated cells have a higher survival rate after cisplatin treatment compared with control cells (Fig. 4D).
TAB1 ablation attenuates the cellular p53 response to cisplatin treatment. (A) TAB1 knockdown reduces cisplatin-mediated cell death. U2OS cells were transfected with 25 nM each control luciferase siRNA (C), TAB1 siRNAs (T1 and T2), or TAK1 siRNA (TK). Forty-eight hours after transfection, cells were treated with vehicle (DMSO) or cisplatin (CDDP; 5 μg/mL) for 24 h and then trypsinized and fixed followed by cell cycle analysis. The plot was obtained from three separate experiments. (B) Cisplatin-induced Parp cleavage is inhibited by TAB1 ablation. U2OS cells were transfected with siRNAs and treated with either vehicle (DMSO) or cisplatin (CDDP) as in A. Total cell lysates were prepared and then immunoblotted with anti-Parp antibody and anti-Actin antibodies. (C) Cisplatin-induced cell death is a p53-mediated response. HCT116 (p53+/+) and HCT116 (p53−/−) cells were transfected with 25 nM each control luciferase siRNA (C) or TAB1 siRNA (T1). Forty-eight hours after transfection, cells were treated with vehicle (DMSO) or cisplatin (CDDP; 10 μg/mL) for 24 h. Cells were then trypsinized and fixed, followed by FACS analysis. The plot was derived from three separate experiments. (D) TAB1 depletion enhances cell survival in response to cisplatin treatment. U2OS cells were transfected with 25 nM each control luciferase (C) or TAB1 siRNA (T1). Eighteen hours after transfection, cells were trypsinized, and 8 × 104 cells were seeded in a 35-mm dish. Twenty-four hours later, cells were treated with cisplatin for 24 h. Cells were then washed with PBS three times and incubated in fresh Dulbecco modified Eagle medium (DMEM) with 10% fetal bovine serum (FBS). Six days later, viable cells were assayed for colony formation by cystal violet staining. Stained plates were photographed, and the number of colonies formed was manually scored using a 1-cm × 1-cm grid system and graphed as described in the Materials and Methods. (E) TAB1 ablation inhibits NOXA expression. U2OS cells were transfected with 25 nM each luciferase control siRNA (C), TAB1 siRNAs (T1 and T2), or TAK1 siRNA (TK) as indicated to a total siRNA concentration of 50 nM and then treated with either vehicle (DMSO) or cisplatin (CDDP) as in A. Total cell lysates were prepared and immunoblotted with anti-MDM2 (3G5+4B11+5B10), anti-p53 (DO-1), anti-TAK1, anti-TAB1, anti-p21, anti-NOXA, or anti-Actin antibodies. (F) TAB1 ablation has a selective effect on expression of p53 target genes. U2OS cells were transfected with 25 nM each luciferase control siRNA (C), TAB1 siRNAs (T1 and T2), or TAK1 siRNA (TK) and treated with vehicle (DMSO) or cisplatin (CDDP) as in E. Total mRNAs were extracted and reverse-transcribed, and then cDNAs were analyzed by quantitative real-time PCR using primers targeting NOXA, PUMA, and p21 genes. (G) NOXA ablation reduces cisplatin-induced apoptosis. U2OS cells were transfected with 25 nM each luciferase control siRNA (C), NOXA siRNAs (N1 and N2), or TAB1 siRNA (T1). Forty-eight hours after transfection, cells were treated with vehicle (DMSO) or cisplatin (CDDP; 5 μg/mL) for 24 h and then trypsinized and fixed, followed by cell cycle analysis.
Not only was TAB1 regulation of the p53 response stimulus-specific, but we also found that p53 targets varied in their response to cisplatin. In particular, NOXA protein levels were increased after cisplatin treatment in a manner that required full expression of TAB1, while protein levels of p21 protein remained unchanged in the absence or presence of the drug with or without TAB1 knockdown (Fig. 4E). The inability of cisplatin treatment to increase levels of p21 protein has been previously reported (Cuadrado et al. 2007). The inhibitory effect of TAB1 ablation also selectively affected NOXA mRNA levels, since it had no impact on either p21 or Puma (another p53 proapoptotic target) mRNA induction by p53 (Fig. 4F). Similar inhibition of NOXA induction in response to cisplatin treatment upon TAB1 ablation was observed in SkHep1 cells (Supplemental Fig. S6). Importantly, NOXA ablation by siRNA reduced cisplatin-induced U2OS cell death to an extent similar to TAB1 knockdown (Fig. 4G). p21 induction has been correlated with protection from cell death (Gartel and Tyner 2002), and we cannot rule out that unchanged p21 protein levels upon cisplatin treatment may be an additional contributing factor to the apoptotic response. Taken together, we delineated a pathway whereby TAB1 leads to apoptosis in cells treated with cisplatin that requires selective induction of the p53 target NOXA.
TAB1 is required for p38α-mediated p53 phosphorylation
Intriguingly, despite the impact on p53-mediated cell death in cisplatin-treated cells, ablating TAB1 (or TAK1) had little effect on p53 protein levels under the conditions used; i.e., after 24 h of treatment with this drug (Fig. 4E). Since our data in Figures 1 and 2 implicated TAB1 in regulating p53 stability and ubiquitination, we examined the kinetics of p53 induction after cisplatin treatment and found that at an earlier time point (4 h after cisplatin treatment), TAB1 or TAK1 ablation led to a modest but reproducible attenuation of steady-state p53 levels (Supplemental Fig. S7). In line with this, p53 in cells with TAB1 depletion had a shorter half-life in the presence of cisplatin compared with cells transfected with control siRNA (Fig. 5A). Moreover, more ubiquitinated p53 species were identified in TAB1-ablated cells in both in vivo ubiquitination (Fig. 5B) and in vitro degradation (Fig. 5C) assays. At later time points after drug treatment when p53 levels are not affected, there may be other processes affected by cisplatin, such as levels or translation of p53 mRNA. Importantly, despite its only subtle effect on p53 protein levels, there was a significant impact of TAB1 knockdown on p53 post-translational modifications, specifically on phosphorylation of S15 and S46 (Fig. 5D).
TAB1 regulates p53 phosphorylation via p38α signaling in cisplatin-treated cells. (A) TAB1 ablation accelerates p53 turnover. U2OS cells were transfected with 25 nM each luciferase control siRNA (C) or TAB1 siRNA (T1). Forty-eight hours after transfection, cells were treated with vehicle (DMSO) or cisplatin (CDDP; 5 μg/mL) for 5 h. Next, cyclohexamide (CHX; 100 μg/mL) was added, and cells were harvested at the indicated times. (Top panel) Cell lysates were subjected to immunoblotting with the indicated antibodies. (Bottom panel) Quantification of the immunoblot data was carried out using Image J software. (B) Endogenous p53 ubiquitination is increased upon TAB1 depletion in response to cisplatin treatment. U2OS cells (∼8 × 105 cells in 10-cm dishes) were transfected with 25 nM each luciferase control siRNA (C), TAB1 siRNA (T), or MDM2 siRNA (M). Six hours later, each plate of cells was trypsinized and replated into three 60-mm dishes. Twenty-four hours after the initial siRNA transfection, cells were transfected with His-tagged ubiquitin (His-Ub; 4 μg) for 24 h and treated with vehicle (DMSO), cisplatin (CDDP; 5 μg/mL), or ETP (15 μM) for 5 h before harvesting. Cells were lysed in denaturing buffer and bound to Ni-NTA beads as described in the Materials and Methods. Ubiquitinated endogenous p53 was detected by an anti-p53 antibody (DO-1). (C) TAB1 plays a role in ubiquitination of endogenously expressed p53 in response to cisplatin treatment. U2OS cells were transfected with 25 nM each luciferase control siRNA (C), TAB1 siRNA (T1), or MDM2 siRNA (M). Forty-eight hours after transfection, cells were treated with cisplatin (CDDP; 5 μg/mL) for 5 h before harvesting. In vitro degradation assays were performed as described in the Materials and Methods, and aliquots were removed from reaction mixtures at the indicated time points and subjected to immunoblotting with anti-p53 (DO-1) antibody. (D) A p38α inhibitor attenuates cisplatin-mediated p53 phosphorylation in U2OS cells. U2OS cells were transfected with 25 nM each luciferase control siRNA (C) or TAB1 siRNAs (T1 and T2). Forty-eight hours after transfection, cells were pretreated with the p38α inhibitor SB203580 (SB; 5 μM) or DMSO for 1 h and then treated with vehicle (DMSO), cisplatin (CDDP; 5 μg/mL), or both cisplatin and SB203580 (CDDP+SB) for 24 h. Total cell lysates were prepared and immunoblotted with anti-p53 (S15-p), anti-p53 (S46-p), anti-p53 (DO-1), anti-p38α (T180/Y182-p), and anti-p38α antibodies. (E) A p38α inhibitor attenuates cisplatin-mediated cell death in U2OS cells. U2OS cells were transfected with 25 nM each luciferase control siRNA (C) or TAB1 siRNAs (T1 and T2) and treated with p38α inhibitor SB203580 and cisplatin as in D. Cells were then trypsinized and fixed, followed by cell cycle analysis.
Phosphorylation of p53 at S46 has been well documented to facilitate the apoptotic response (Oda et al. 2000; D'Orazi et al. 2002; Mayo et al. 2005), and activation of p38α has been shown to mediate p53 phosphorylation at N-terminal sites, including S15 and S46 (Bulavin et al. 1999; Kim et al. 2002). Consistent with previous reports that TAB1 activates p38α through either activating TAK1 or direct binding to p38α (Ge et al. 2002; Johnson 2002), TAB1 knockdown prevented full phosphorylation of p38α at T180 and Y182. Since treatment with a p38α inhibitor led to reduced p53 phosphorylation at S15 and S46 (Fig. 5D), we believe that the activity of p53 is altered due to its modification by phosphorylation by p38α in a TAB1-dependent manner.
TAB1 has been shown to activate p38α through either direct binding or stimulation of TAK1 (Ge et al. 2002; Johnson 2002). To further investigate which kinase signaling pathway involves TAB1 to affect p53 phosphorylation status and mediate a p53 response, we treated cells with inhibitors of TAK1 or p38α together with cisplatin treatment. A TAK1 kinase inhibitor (5Z-7-oxozeaenol) (Ninomiya-Tsuji et al. 2003) had little effect on cisplatin-induced cell death in U2OS cells treated with control or TAB1 siRNAs (Supplemental Fig. S8), suggesting that the kinase activity of TAK1 is not essential for cisplatin-activated signaling pathways. Relevantly, inhibition of TAK1 downstream pathways (CAY10512 for NF-κB and SP100625 for JNK) in U2OS cells also had little effect on or even enhanced cisplatin-induced cell death (Supplemental Fig. S9). Furthermore, TAB1 knockdown was still able to attenuate cisplatin-induced cell death in the presence of those inhibitors (Supplemental Fig. S9). Therefore, it is not likely that TAB1 works through TAK1 to regulate p53 phosphorylation and activation after cisplatin treatment. On the other hand, when we preincubated U2OS cells with a p38α inhibitor (SB203580), this compound counteracted cisplatin-induced cell death (Fig. 5E). However, since TAB1 knockdown could further reduce the percentage of sub-G1 cells treated with the p38α inhibitor (Fig. 5E), this suggests that in addition to its functional association with p38α, TAB1 employs one or more additional mechanisms to contribute to cisplatin-induced cell death.
TAB1 contributes to p53-mediated intrinsic apoptosis through modulating cellular levels and localization of MDMX in response to cisplatin treatment
Previous studies reported that in cisplatin-treated cells, MDMX has an unexpected function as a mitochondrially associated proapoptotic factor in the p53-intrinsic cell death pathway (Mancini et al. 2004, 2009; Mancini and Moretti 2009). Since these investigators provided evidence that MDMX plays a role in recruiting p53 phosphorylated at S46 to mitochondria in cisplatin-treated cells, we tested whether TAB1 might be involved in this process. Indeed, our results implicate MDMX in the regulation of p53 by TAB1. First, we observed reduced cellular levels of MDMX in cells treated with cisplatin when TAB1 (or TAK1) was ablated by siRNA (Fig. 6A,E). This is consistent with our data (shown in Fig. 2C) that TAB1 inhibited MDM2-mediated ubiquitination of MDMX. Second, it is well established that some DNA damage-inducing agents (via ATM activation) elicit MDM2 degradation of MDMX (Kawai et al. 2003; Chen et al. 2005; Okamoto et al. 2005; Pereg et al. 2005; Jin et al. 2006). Strikingly, we found that MDMX was significantly more resistant to degradation upon cisplatin treatment when compared with other stress stimuli such as Dauno, ETP, and NCS (Fig. 6B). Note that we found that Dauno and NCS induced cell cycle arrest while ETP induced apoptosis in U2OS cells (data not shown), suggesting that the changes in cellular levels of MDMX do not correlate with any specific cell cycle profile. Third, consistent with the above-mentioned reports as well as our data that TAB1 is required for maximal cell death upon cisplatin treatment, ablation of MDMX by different siRNAs attenuated cisplatin-induced cell death similarly to ablation of TAB1, suggesting the possibility that TAB1 may participate in the cisplatin-mediated cellular response through modulating the cellular levels of MDMX (Fig. 6C). Finally, less mitochondrially localized MDMX was detected in U2OS cells treated with TAB1 siRNA compared with control siRNA (Supplemental Fig. S10). Thus, our results suggest that TAB1 contributes to p53-mediated intrinsic apoptosis through maintaining cellular levels of MDMX and modulating its localization in response to cisplatin treatment.
TAB1 contributes to p53-mediated intrinsic apotptosis through modulating cellular levels of MDMX in response to cisplatin treatment. (A) TAB1 ablation reduces cellular levels of MDMX in response to cisplatin treatment. U2OS cells were transfected with 25 nM each luciferase control siRNA (C), TAB1 siRNAs (T1 and T2), or TAK1 siRNA (TK). Forty-eight hours after transfection, cells were treated with vehicle (DMSO) or cisplatin (CDDP; 5 μg/mL) for 4 h. Total cell lysates were prepared, followed by immunoblotting with anti-MDMX and anti-Actin antibodies. (B) MDMX is resistant to degradation in cisplatin-treated cells. U2OS cells were treated with DMSO (−), cisplatin (CDDP; 5 μg/mL), ETP (30 μM), Dauno (0.22 μM), NCS (320 ng/mL), or actinomycin D (ActD; 5 nM) for 5 h. Cell lysates were prepared for immunoblotting with anti-MDMX and anti-Actin antibodies. (C) MDMX ablation attenuates cisplatin-mediated cell death. U2OS cells were transfected with 25 nM each luciferase control siRNA (C), TAB1 siRNA (T1), or MDMX siRNAs (X4 and X11). Forty-eight hours after transfection, cells were treated with vehicle (DMSO) or cisplatin (CDDP; 5 μg/mL) for an additional 24 h and then trypsinized and fixed for cell cycle analysis. The plot was obtained from three separate experiments.
TAB1 is a potential tumor suppressor, and its ablation contributes to cisplatin resistance
Cisplatin is widely used in treating solid tumors, including testicular, ovarian, cervical, head and neck, and small-cell lung cancers (Basu and Krishnamurthy 2010). However, patients who initially respond to cisplatin therapy often develop resistance to the drug during the course of the treatment (Kelland 2007). To examine the relevance of the pathway that we delineated, we analyzed publicly available data sets. As shown in Figure 7A, levels of TAB1 were markedly lower in clonally derived cisplatin-resistant A2780 ovarian cancer cell lines than in their cisplatin-sensitive counterparts (Gene Expression Omnibus [GEO] data set no. GSE33482). A similar trend was observed in another data set using both cisplatin-sensitive and cisplatin-resistant HeLa cells (data not shown). Furthermore, TAB1 levels in human ovarian serous cystadenocarcinoma tumor samples were relatively low compared with normal ovarian tissue samples (Fig. 7B, The Cancer Genome Atlas [TCGA] data matrix). Interestingly, when we stratified the p53 status in samples, TAB1 levels also tended to be even lower in samples with wild-type p53 compared with samples with p53 having missense mutation (Fig. 7C). Therefore, our results both suggest that TAB1 ablation contributes to cisplatin resistance and implicate TAB1 as a potential tumor suppressor.
TAB1 levels are relatively low in cisplatin-resistant clones of ovarian cancer cells and in ovarian tumors. (A) Cisplatin-resistant ovarian cancer cell line clones have lower cellular levels of TAB1 compared with their cisplatin-sensitive counterparts. Cellular levels of TAB1 mRNA in clonally derived cisplatin-resistant A2780 ovarian cancer cell lines (A2780-cis) and their cisplatin-sensitive counterparts (A2780) were analyzed based on a GEO publicly available data set (GSE33482). (B) TAB1 levels in human ovarian serous cystadenocarcinoma tumor samples compared with normal ovarian tissue samples. (Top panel) TAB1 levels (TCGA Agilent G4502A; N = 589) were analyzed comparing tumor samples with normal ovarian tissue samples. (Bottom panel) The Student's t-test (two-tailed, unpaired heteroscedastic t-test) was performed to verify statistically significant difference between the tumor (N = 589) and normal (N = 8) samples (P-value < 0.0005). The bars within the box represent the median gene expression. Round circles indicate outliers. The asterisk indicates statistical significance. (C) Comparison of TAB1 levels in ovarian serous cystadenocarcinoma tumor samples harboring wild-type p53 or p53 with missense mutations. Relative expression levels of TAB1 in tumor samples were analyzed based on their p53 status (wild-type, N = 77; mutant, N = 253). (D) Regulation of the p53/MDM2/MDMX circuit by TAB1. TAB1 inhibits MDM2 E3 ligase activity toward p53. (Left panel) Stabilization of p53 leads to cell cycle arrest. In response to cisplatin treatment, TAB1 also activates p38α, which in turn phosphorylates p53 to mediate an apoptotic response. (Right panel) At the same time, TAB1 modulates the cellular levels of MDMX and facilitates MDMX mitochondrial localization, which contributes to the p53-mediated intrinsic apoptotic response.
Discussion
We report here a functional link between p53/MDM2/MDMX circuitry and MAPK signaling through a newfound interaction between TAB1 and MDM2. Our results demonstrate that TAB1, a scaffold protein with multiple binding partners that are involved in different signaling pathways, is critical for p53 activation under specific conditions. TAB1, an inhibitor of MDM2 E3 ligase activity, is required for p53 up-regulation and cell cycle arrest when MDM2 is ablated. TAB1 is also a key mediator of p53-dependent cell death albeit uniquely in cisplatin-treated cells. In response to cisplatin, TAB1 both modulates p53 phosphorylation and activation through its functional interaction with p38α and regulates the cellular level of MDMX to facilitate p53-intrinsic apoptosis (modeled in Fig. 7D). Several aspects of this pathway merit further discussion.
TAB1 regulates p53 levels when MDM2 is ablated
In unstressed cells, p53 is rapidly turned over due in large part to the E3 ligase activity of MDM2. We discovered that reducing TAB1 levels via siRNA affects p53 levels and activity only in the context of codepleted MDM2. One speculation is that the reduced cellular pool of MDM2 upon siRNA knockdown may be a better inhibitory target for TAB1. Alternatively, p53 stabilization upon MDM2 depletion may set the cells into a “stressed” mode, which may trigger downstream events that somehow facilitate the interaction between TAB1 and MDM2. If the association between p53 and MDM2 were not disrupted under this kind of “stressed” condition, ablation of TAB1 would enhance MDM2 E3 ligase function.
TAB1 works through p38α to activate p53 in cisplatin-treated cells
Since TAB1 interacts with two major cellular signaling proteins (TAK1 and p38α) and plays important roles in their respective activations, we tested which of these two proteins is required for TAB1 regulation of p53 after cisplatin treatment. TAK1 is a member of the MAPKKK family that, through activation of downstream kinases, including IKK, JNK and p38α, activates key transcription factors such as AP-1 and NF-κB (Landstrom 2010). TAK1 was also reported to play a role in activation of the Snf1/AMPK family, which is important for metabolic regulation in cells (Momcilovic et al. 2006; Herrero-Martin et al. 2009; Chang et al. 2010). The outcome of the TAK1 signaling cascade is most likely tissue-specific and context-dependent, as TAK1 has been implicated in both tumor suppression (Konishi et al. 2003; Thakur et al. 2009; Bettermann et al. 2010; Inokuchi et al. 2010) and tumorigenesis (Melisi et al. 2011). Unlike TAB1, both wild-type TAK1 and its kinase-dead derivative do not inhibit MDM2-mediated p53 degradation by themselves. However, ectopic TAK1 enhances the inhibitory effect of coexpressed TAB1, most likely due to its ability to stabilize TAB1 (see Fig. 1). TAB1 is itself a substrate of TAK1 kinase (Prickett et al. 2008), and our results also suggest that TAK1 kinase activity is required for its ability to stabilize TAB1 (Figs. 1, 4). Therefore, we surmise that TAK1 modulates the p53–MDM2 circuit through its regulatory effect on TAB1. However, the fact that a TAK1-specific inhibitor (5Z-7-oxozeaenol) has no apparent effect on cisplatin-induced cell death in U2OS cells implies that signaling via this drug does not involve TAK1 activation (Supplemental Fig. S8).
In contrast to TAK1, our data strongly implicate p38α as a TAB1 target in regulating p53 after cisplatin treatment. p38 MAPKs, also known as stress-activated protein kinases, comprise four family members—p38α, p38β, p38γ, and p38δ—that possess both overlapping and specific functions (Cuenda and Rousseau 2007). TAB1 has been shown to specifically interact with p38α but not other p38 family members (Ge et al. 2002). The roles of p38α are complex (Wagner and Nebreda 2009). It can serve as a negative regulator of cell cycle progression and can facilitate induction of apoptosis, indicating that it may function as a tumor suppressor. However, its involvement in invasion, inflammation, and angiogenesis suggests that it may also have oncogenic functions. In response to DNA damage, activation of p38α can mediate apoptosis or induce a G2/M cell cycle checkpoint through p53-dependent and p53-independent mechanisms (Han and Sun 2007). Selective nuclear accumulation of p38α in response to DNA damage has been suggested to increase the phosphorylation of p38α nuclear targets (Wood et al. 2009). Additionally, it has been reported that TAB1 modulates intracellular localization of p38α and its downstream signaling (Lu et al. 2006). Our observations further support the notion that p38α actively participates in the DNA damage-induced cellular response. However, inhibiting p38α catalytic activity by SB203580 reduces but does not completely abolish the inhibitory effect of TAB1 on p53 activation, implying that TAB1 works through additional mechanisms to regulate p53 function.
Cisplatin is distinct among genotoxic agents in requiring TAB1 for activation of p53-mediated apoptosis
It is remarkable that cisplatin treatment-associated cell death appears to be unique among the several stress-inducing agents that we tested in being mitigated by knockdown of TAB1. At this stage, we can only speculate on scenarios that might explain this finding. First, two distinct TAB1-regulated mechanisms (p38α activation and stabilized MDMX levels) are required for p53 to produce apoptosis in cisplatin-treated cells. Genotoxic agents such as ETP that activate p38α (Kurosu et al. 2005) may not be able to mediate MDMX-facilitated p53-intrinsic apoptotic events through TAB1 because they induce MDMX degradation. On the other hand, while MDMX is more resistant to degradation by some agents, such as actinomycin D (Biderman et al. 2012), those agents may not be able to activate p38α. Second, cisplatin but not other treatments may lead to phosphorylation and activation of TAB1 by either p38α or an unidentified protein kinase such that it can prevent MDM2-mediated MDMX degradation. Third, cisplatin may be unique in inhibiting the abililty of the Wip1 phosphatase (a p53 target that negatively regulates p53, ATM, and p38α) to dephosphorylate and thereby inactivate p38α or TAB1. Fourth, cisplatin may be unique in promoting specific modifications on MDMX or MDM2, which affect their interactions with TAB1 and prevent degradation of MDMX, and/or facilitating its mitochondrial function. Finally, TAB1/TAK1 involvement in HIPK2 activation in response to Wnt-1 signaling (Kanei-Ishii et al. 2004) may facilitate the p53 response to cisplatin. HIPK2 has been shown to phosphorylate p53 at S46, and its function and stability are regulated by MDM2 depending on the dosage of the DNA damage agents (Rinaldo et al. 2007). It would be interesting to investigate the interplay between TAB1, MDM2, and HIPK2 in the cellular apoptotic response upon cisplatin treatment as well as determine whether cisplatin regulates relevant modifications in TAB1, MDM2, or MDMX.
NOXA is selectively impacted by TAB1 knockdown in cisplatin-treated cells
Our results suggest that in response to cisplatin, specific genes (such as NOXA) are activated to promote apoptosis. Furthermore, our results suggest that NOXA plays a role in apoptosis in cells treated with this drug. It was previously shown that ERK-dependent but p53-independent NOXA induction is critical to cisplatin-induced cell death in some cell types and that a MEK inhibitor (U0126) attenuates cisplatin-mediated apoptosis (Sheridan et al. 2010). When we tested this inhibitor, we found that it has little effect on cisplatin-induced cell death in U2OS cells, although it greatly attenuates cell death in SkHep1 cells (Supplemental Fig. S11). Thus, the involvement of the ERK signaling cascade in cisplatin-associated cytotoxicity is cell type-specific, and this pathway is not the major player in cisplatin-mediated cell death in U2OS cells. Further studies are needed to investigate how TAB1 selectively regulates p53 transactivation of NOXA. It is noteworthy that a p38α-regulated transcription coactivator, Hamlet (p18), stimulates p53-dependent apoptosis in response to certain types of stresses such as UV and cisplatin but not in response to γ-irradiation (Cuadrado et al. 2007). Also in that study, it was shown that down-regulation of Hamlet specifically prevents NOXA induction. It would be interesting to test whether TAB1 affects the levels or function of Hamlet in response to cisplatin treatment, which in turn would affect p53 transcription.
Cisplatin is a common and effective chemotherapeutic agent for some forms of cancer (Basu and Krishnamurthy 2010). That TAB1 levels are inversely correlated with cisplatin sensitivity (at least in A2780 ovarian cancer cell line-derived clones) suggests that loss of TAB1 may contribute to cisplatin resistance in some cases. The interplay between p53/MDM2/MDMX and TAB1 that we uncovered here may provide new insights into resistance to this drug that occurs in a significant number of cisplatin-treated patients, in particular those whose tumors harbor wild-type p53. It should be noted, however, that some tumors commonly treated with cisplatin, such as ovarian cancers, sustain a very high frequency of p53 mutation. While our analysis revealed significantly lower levels of TAB1 in ovarian tumors with wild-type p53 when compared with those with mutant p53, there was still reduced TAB1 in a subset of the latter tumors as well. This implies that either mutant p53 can be activated by TAB1 to produce cell death, which is unlikely, or TAB1 is playing additional p53-independent roles in cisplatin-treated cells. Further studies will hopefully clarify these points and provide more insight into the many roles of TAB1 in cells.
Materials and methodsPlasmids and cell culture
Flag-MDM2 (wild-type), HA-p53, Myc-MDMX, and HA-ubiquitin were described previously (Poyurovsky et al. 2003; Zhu et al. 2009). HA-MDMX was a gift from Dr. A. Jochemsen (Leiden University Medical Center, Netherlands). His-ubiquitin (pcBH2Ub) was kindly provided by Dr. R. Baer (Columbia University). Construction of a Myc-tagged TAB1 plasmid (Myc-TAB1) and an untagged TAK1 plasmid (pcDNA-TAK1) is described in the Supplemental Material. U2OS cells (osteosarcoma cells expressing wild-type p53), H1299 cells (p53-null lung epithelial carcinoma cells), HCT116 (p53+/+), and derivative HCT116 (p53−/−) cells (colon carcinoma cells; a kind gift from Dr. B. Vogelstein, Johns Hopkins University) were grown in Dulbecco modified Eagle medium (DMEM) containing 10% fetal bovine serum (FBS) at 37°C.
Antibodies and drugs
A complete list of the antibodies and drugs used in this study is provided in the Supplemental Material.
Transfection, Western blot, and cell cycle analysis
Transfection, Western blot, and cell cycle analysis were performed as previously described (Zhu et al. 2009). Detailed procedures are provided in the Supplemental Material.
Ubiquitination and degradation assays
In vivo ubiquitination assays and in vitro degradation assays were carried out as previously described (Zhu et al. 2009). Detailed protocols are provided in Supplemental Material.
siRNA interference
siRNA oligonucleotides (T1, J-004770-07 and T2, J-004770-08) targeting human TAB1 were purchased from Dharmacon. Reported siRNA oligonucleotides targeting luciferase (Urist et al. 2004), human TAK1 (Bertelsen and Sanfridson 2007), MDM2 (Jin et al. 2003), and NOXA (N1 [Sheridan et al. 2010]) were obtained from Invitrogen. Second siRNA oligonucleotides targeting NOXA (N2, D-005275-07) were purchased from Dharmacon. siRNA oligonucleotides targeting MDMX (X11 [Chen et al. 2005] and X4 [Hs_MDM4_2 FlexiTube siRNA]) were obtained from Qiagen. Transfection was performed using Dharmafect 1 (Dharmacon) according to the manufacturer's instructions.
RNA extraction and quantitative RT–PCR analysis
RNA was extracted using a Qiagen RNeasy minkit, and cDNA was synthesized with the QuantiTect reverse transcription kit (Qiagen). Samples were analyzed by quantitative real-time PCR on an Applied Biosystems Step One Plus instrument using the SYBR Green dye (Applied Biosystems). RNA expression was normalized to RPL32 mRNA expression. Relative levels were calculated by the comparative Ct method (ΔΔCT method). Graphs are representative of multiple independent experiments, with error bars representing technical PCR replicates. Primer sequences are available on request.
Colony formation assay
U2OS cells (8 × 104) transfected with either control or TAB1 siRNAs were seeded in a 35-mm dish. Twenty-four hours later, cells were treated with cisplatin for 24 h. Cells were then washed with PBS three times and incubated in fresh DMEM medium with 10% FBS. Six days later, cells were fixed and subjected to cystal violet staining. Stained dishes were photographed, and the number of colonies formed in each dish was manually scored using a 1-cm × 1-cm grid system and graphed. A more detailed protocol is provided in the Supplemental Material.
Gene expression analysis
Microarray gene expression data from a GEO publicly available data set (GSE33482) and TCGA human ovarian serous cystadenocarcinoma data matrix (Agilent G4502A, N = 589) were obtained for determining cellular levels of TAB1 in clonally derived cisplatin-resistant A2780 ovarian cancer cell lines (A2780-cis) and their cisplatin-sensitive counterparts (A2780) and comparing TAB1 expression between tumor samples with normal ovarian tissue samples, respectively. The somatic mutation data sets for gene expression of the ovarian serous cystadenocarcinoma tumor samples in the TCGA data matrix were downloaded from the Broad Genome Data Analysis Center (GDAC) Firehose Analytics Platform (https://confluence.broadinstitute.org/display/GDAC/Home) and analyzed to stratify tumor samples on the basis of p53 mutation status (wild-type or missense mutations), and relative TAB1 expression was analyzed. A detailed description of the statistical analysis of these data sets is provided in the Supplemental Material.
Acknowledgments
We thank Ella Freulich for expert technical assistance, and members of the Prives' laboratory for helpful suggestions. This work was supported by grant CA58316 from the NIH.
Supplemental material is available for this article.
Article published online ahead of print. Article and publication date are online at http://www.genesdev.org/cgi/doi/10.1101/gad.212258.112.
Poly-ADP-ribosylation post-translational modifications play important roles in many biological processes, including the DNA damage response. Yu and colleagues now identify a set of Forkhead-associated (FHA) and BRCA1 C-terminal (BRCT) domains that recognize poly(ADP-ribose) (PAR). They show that the interaction between PAR and the BRCT domain of the Nijmegan breakage syndrome protein NBS1 is important for the activation of ATM in response to DNA damage. This study thus demonstrates how two novel PAR-binding modules mediate the DNA damage response.
Poly-ADP-ribosylation is a unique post-translational modification participating in many biological processes, such as DNA damage response. Here, we demonstrate that a set of Forkhead-associated (FHA) and BRCA1 C-terminal (BRCT) domains recognizes poly(ADP-ribose) (PAR) both in vitro and in vivo. Among these FHA and BRCT domains, the FHA domains of APTX and PNKP interact with iso-ADP-ribose, the linkage of PAR, whereas the BRCT domains of Ligase4, XRCC1, and NBS1 recognize ADP-ribose, the basic unit of PAR. The interactions between PAR and the FHA or BRCT domains mediate the relocation of these domain-containing proteins to DNA damage sites and facilitate the DNA damage response. Moreover, the interaction between PAR and the NBS1 BRCT domain is important for the early activation of ATM during DNA damage response and ATM-dependent cell cycle checkpoint activation. Taken together, our results demonstrate two novel PAR-binding modules that play important roles in DNA damage response.
poly(ADP-ribose)PAR binding domainDNA damage
Post-translational modifications, such as protein phosphorylation, ubiquitination, acetylation, and methylation, are important for numerous biological processes. Protein ADP-ribosylation is a unique post-translational modification that has been shown to play critical roles in many cellular events, especially DNA damage response (Schreiber et al. 2006; Gibson and Kraus 2012).
Poly-ADP-ribosylation is catalyzed by poly-ADP-ribosylation polymerases (PARPs) (Ame et al. 2004; Hottiger et al. 2010). Using NAD+ as the donor, ADP-ribose is covalently linked to the side chains of arginine, aspartic acid, and glutamic acid residues in target proteins. After catalyzing the first ADP-ribose on the target, other ADP-riboses can be covalently linked onto the first ADP-ribose to form both liner and branched polymers, known as poly(ADP-ribose) (PAR) (Schreiber et al. 2006; Luo and Kraus 2012). Following DNA damage, massive poly-ADP-ribosylation is synthesized at DNA lesions within a few seconds (D'Amours et al. 1999; Kim et al. 2005). To date, two major nuclear substrates of poly-ADP-ribosylation are known: the PARP1 enzyme itself and histones (Schreiber et al. 2006; Messner and Hottiger 2011). Once PAR is synthesized at the DNA damage site, it is also quickly recognized and hydrolyzed by PAR glycosylase (PARG) (D'Amours et al. 1999; Gagne et al. 2006; Kim et al. 2007a). Thus, the half-life of PAR at DNA damage sites is very short, and the biological function of this dynamic post-translation modification at DNA damage sites remains elusive.
Recently, several PAR-binding proteins have been identified as the “readers” to recognize PAR signals (Karras et al. 2005; Ahel et al. 2008; Wang et al. 2012), suggesting that PAR is likely to function as recruiting signals to induce DNA damage response factors to DNA damage sites. To identify other PAR-binding modules, we examined both the Forkhead-associated (FHA) domain and the BRCA1 C-terminal (BRCT) domain. The FHA and BRCT domains are known as phospho-protein-binding domains. Many FHA domain- or BRCT domain-containing proteins are involved in DNA damage response (Li et al. 2002; Glover et al. 2004; Mahajan et al. 2008; Mohammad and Yaffe 2009). It has been shown that the FHA domains recognize phospho-Thr (pThr) motifs (Sun et al. 1998; Durocher et al. 1999, 2000; Li et al. 2002; Mahajan et al. 2008). For example, the FHA domain of Rad53 recognizes the pThr motif of Rad9 in budding yeast (Sun et al. 1998), the FHA domain of fission yeast NBS1 recognizes the pThr of Ctp1(Williams et al. 2009), and the FHA domain of human NBS1 and RNF8 recognizes the pThr motifs of MDC1 (Chapman and Jackson 2008; Spycher et al. 2008; Wu et al. 2008; Lloyd et al. 2009). While the BRCT domains have been shown to recognize phospho-Ser (pSer) motifs (Manke et al. 2003; Yu et al. 2003), the BRCA1 BRCT domain binds pSer motifs in several downstream partners (Yu et al. 2003; Yu and Chen 2004; Kim et al. 2007b; Liu et al. 2007; Wang et al. 2007). The MDC1 BRCT domain recognizes the pS139 site of H2AX (Stucki et al. 2005). These phospho-protein-dependent interactions are important for DNA damage checkpoint activation and DNA damage repair. However, based on the peptide screening, not all of the FHA and BRCT domains have high affinity to phospho-proteins (Durocher et al. 2000; Rodriguez et al. 2003). In particularly, our recent study suggests that the BARD1 BRCT domain recognizes ADP-ribose (Baer 2013; Li and Yu 2013). Thus, we asked whether these domains have other binding partners besides phospho-proteins.
Following the DNA damage, PAR is quickly synthesized at the DNA damage sites. Interestingly, one ADP-ribose residue contains two phosphate groups. Thus, massive PAR synthesized upon DNA damage brings huge amounts of phosphate moieties at DNA damage sites in a very short period. We therefore wondered whether the FHA and BRCT domains could recognize PAR. In this study, we screened 19 FHA and BRCT domains and found that five of them bind PAR both in vitro and in vivo. Moreover, the interaction with PAR facilitates the fast recruitment of these FHA or BRCT domain-containing proteins to DNA lesions and the relevant DNA damage repair process.
ResultsA set of BRCT and FHA domains binds PAR
Since CHFR and APLF are known PAR-binding proteins, which bind PAR via their PBZ motifs (Ahel et al. 2008; Li et al. 2010; Oberoi et al. 2010), we used GST, recombinant CHFR, and APLF PBZ motifs as negative and positive controls, respectively, to screen PAR-binding domains. We examined 19 FHA or BRCT domains and found that two FHA domains (from PNKP and APTX), two BRCT domains (from Ligase4 and XRCC1), and an FHA–BRCT fusion domain (from NBS1) interacted with PAR (Fig. 1A). Reverse pull-down assays confirmed the direct interaction between PAR and these FHA/BRCT domains (Fig. 1B). Using isothermal titration calorimetry (ITC) assays, we measured the affinity between PAR and these FHA/BRCT domains (Fig. 1C), which is in the physiologically relevant range and is similar to that between PAR and other PAR-binding domains (Karras et al. 2005; Wang et al. 2012). Since PAR is the ADP-ribose polymer with mixed length and contains both linear and branched forms, the affinity between the PAR-binding domain and PAR could not be accurately measured. We then measured the affinity between these FHA/BRCT domains and ADP-ribose, the basic unit of PAR. The ITC results show that the BRCT domains of Ligase4 and XRCC1 and the FHA–BRCT fusion domain of NBS1 recognize ADP-ribose. However, the FHA domains of PNKP and APTX do not interact with ADP-ribose (Fig. 1D). Since the FHA domains of PNKP and APTX recognize PAR, we wonder whether these FHA domains interact with the linkage between each ADP-ribose in PAR. We used phosphodiesterase to digest PAR into iso-ADP-ribose, the linkage between two individual ADP-riboses, and found that the FHA domains of PNKP and APTX have high affinity with iso-ADP-ribose (Fig. 1E). Following DNA damage, PAR is heavily synthesized at DNA lesions (D'Amours et al. 1999; Kim et al. 2005). It has been reported that PNKP, APTX, Ligase4, XRCC1, and NBS1 all participate in DNA damage response (Su 2006; Polo and Jackson 2011). We examined the in vivo interactions between PAR and PNKP, APTX, Ligase4, XRCC1, or NBS1. With ionizing radiation (IR) treatment, PAR was significantly synthesized in the cells and interacted with PNKP, APTX, Ligase4, XRCC1, or NBS1 (Fig. 1F). Taken together, these results demonstrate that a set of FHA/BRCT domains interacts with PAR.
The BRCT and FHA domains bind PAR. (A) PAR-binding screening among the BRCT and FHA domains. The interactions between PAR and GST (negative control), GST-CHFR (positive control), or the indicated domains was examined by dot blot using anti-PAR antibody. (B) The interactions between biotin-PAR and recombinant proteins were examined by pull-down with Streptavidin beads and blotted with anti-GST antibody. (C) The affinity between PAR and the recombinant proteins was measured by ITC. (D) The affinity between ADP-ribose and the recombinant proteins in B was measured by ITC. (E) The affinity between iso-ADP-ribose and the recombinant PNKP FHA or APTX FHA was measured by ITC. Titration of ligand (PAR, ADP-ribose, or iso-ADP-ribose) into a solution containing the purified protein was performed. The fit of the data to an equilibrium-binding isotherm is shown. (C–E) The fit provides an equilibrium dissociation constant (KD) for the binding of the ligand to the protein. (F) The in vivo interaction between PAR and PNKP, APTX, Ligase4, NBS1, or XRCC1 was measured by coimmunoprecipitation (co-IP) with the indicated antibodies. Whole-cell lysates were blotted and are shown as the input.
PAR-binding pockets are conserved in the BRCT and FHA domains
Next, we examined the PAR-binding sites in these FHA/BRCT domains. Different FHA or BRCT domains are predicted to fold into similar secondary structures, respectively, with binding pockets that can recognize pThr or pSer motifs (Fig. 2A; Glover et al. 2004; Mahajan et al. 2008; Mohammad and Yaffe 2009). We asked whether the similar binding pockets recognize iso-ADP-ribose and ADP-ribose, respectively. For the FHA domains of PNKP and APTX, two conserved arginine residues in each binding pocket were mutated into alanines. Both mutants abolished the interaction with PAR in the pull-down and reciprocal pull-down assays (Fig. 2B,C). Moreover, we did not detect any affinity between the mutant FHA domains and iso-ADP-ribose using ITC assays (Fig. 2D). Following IR treatment, the FHA domain mutations abolished the interaction between PAR and PNKP or APTX in vivo (Fig. 2E). Single mutation in the binding pocket of the BRCA1 BRCT domain abolished the interaction between the pSer motif and the BRCA1 BRCT domain (Shiozaki et al. 2004). Based on the similarity of the secondary structure in the BRCT domain, we mutated the conserved Ser or Lys residues to disrupt the putative binding pocket in the BRCT domain of Ligase4 and XRCC1 (Fig. 2A). Like those FHA domain mutants, these BRCT domain mutants abolished the interaction with PAR in vitro (Fig. 2B and C). The full-length proteins bearing these BRCT domain mutations failed to interact with PAR in vivo (Fig. 2E). In addition to these FHA domains and BRCT domains, NBS1 has a FHA domain and a BRCT domain that are fused together. This FHA–BRCT domain also recognizes PAR. To study which subdomain of this FHA–BRCT fusion domain recognizes ADP-ribose, we mutated the conserved residues in each binding pocket. As shown in Figure 2, B and D, mutation of conserved Lys residue in the FHA fold did not significantly affect the interaction between the FHA–BRCT fusion domain and PAR or ADP-ribose, whereas mutation in the BRCT fold abolished the interaction (Fig. 2B,D), suggesting that the BRCT fold in NBS1 mediates the interaction with PAR. This result is further confirmed by the in vivo coimmunoprecipitation (co-IP) assay (Fig. 2E). Moreover, when cells were treated with olaparib to suppress the DNA damage-induced PAR synthesis, these FHA/BRCT domain-containing proteins no longer interacted with PAR in vivo (Fig. 2E).
PAR-binding pockets in the FHA and BRCT domains. (A) Based on the secondary structure, amino acid sequences of FHA (left) and BRCT (right) are compared to show the conserved residues (similarity to the R28 of the NBS1 FHA domain and the S1655/K1702 of the BRCA1 BRCT domain) that form the potential PAR-binding pocket. The β strand is shown as arrow, and the α helix is shown as box. Conserved residues are marked in red. (B) Mutations of conserved residues (Lig4 BRCT domain S668A, XRCC1 BRCT domain K369A, APTX FHA R29A/R42A, PNKP FHA R35A/R48A, and NBS1 FHA+BRCT domain K160A) in the potential binding pockets of the FHA and BRCT domains abolish the PAR binding. Dot blot was used to examine the interaction with anti-PAR antibody. The intact domains were used as the positive control. (C) The interactions between biotin-PAR and the indicated GST tagged FHA and BRCT domains were examined by pull-down with Streptavidin beads and blotting with anti-GST antibody. (D) The affinity between iso-ADP-ribose or ADP-ribose and the indicated mutant domains was measured by ITC. (E) The in vivo interaction between PAR and wild-type or mutants (Lig4 S668A, XRCC1 K369A, APTX R29A/R42A, PNKP R35A/R48A, and NBS1 K160A) was examined by co-IP with the indicated antibodies in the presence or absence of olaparib pretreatment. Whole-cell lysates were blotted and are shown as the input.
Computational analysis of the PAR-binding pockets in the FHA and BRCT domains
The binding between these FHA domains and iso-ADP-ribose was further analyzed by computational modeling. The structure of several FHA domains, including the FHA domains of PNKP, APTX, CHK2, and RNF8, has been solved (Li et al. 2002; Huen et al. 2007; Ali et al. 2009; Becherel et al. 2010). We examined the binding pockets of these FHA domains. The electrostatic potential of these four proteins, calculated by the MM-PBSA program, indicated that the peptide-binding sites are primarily positively charged that are complementary and primed for the recognizing negatively charged molecules such as phosphate moieties. The FHA domain of PNKP has been shown to recognize an adjacent pS–pT motif derived from XRCC1 (Whitehouse et al. 2001; Loizou et al. 2004; Ali et al. 2009). However, the affinity between the PNKP FHA domain and iso-ADP-ribose is even slightly higher than that between the PNKP FHA domain and the pS–pT peptide (Supplemental Fig. 1). Consistently, the binding pocket formed by the R35 and R48 of the PNKP FHA domain well accommodates iso-ADP-ribose, in which two phosphate groups mimic the two phosphate groups in the pS–pT peptide and form salt bridges with Arg residues (Supplemental Fig. 1A). The binding model was further validated by a rigorous binding free energy calculation; namely, the MM-PBSA method (Rastelli et al. 2010), which has been employed for estimating the relative binding free energy of protein with peptide and protein with small molecules (Kollman et al. 2000), including the BRCA1 with phospho-peptides (Anisimov et al. 2011) and viruses with ADP-ribose (Rungrotmongkol et al. 2010) reported recently. Of note, such calculations provide estimates of the relative binding free energy but not the absolute binding free energy of the protein with a ligand. Thus, the calculated binding free energies are ideal to be compared with experimental binding affinities of different ligands with the same protein. The binding free energy between the PNKP FHA domain and iso-ADP-ribose, calculated by the MM-PBSM method, is lower than that between the PNKP FHA domain and pS–pT peptide (Supplemental Fig. 1B), indicating a stronger binding affinity between the PNKP FHA domain and iso-ADP-ribose. These results are consistent with the binding affinities measured by ITC assays (Supplemental Fig. 1D). The binding pocket formed by R29 and R42 in the FHA domain of APTX is similar to that of PNKP (Supplemental Fig. 2A). It has been reported that the FHA domain of APTX recognizes the pS-D–pT-D motif of MDC1 (Becherel et al. 2010). The affinity between the FHA domain of ATPX and the pS-D–pT-D peptide is consistent with previous published results. However, the binding between the FHA domain of ATPX and iso-ADP-ribose is much stronger. Again, based on the structure of the FHA domain of APTX (Becherel et al. 2010), the analyses by the MM-PBSM method suggest that the binding free energy between the APTX FHA and iso-ADP-ribose is much lower than that between the APTX FHA domain and the pS-D–pT-D peptide, which is consistent with the binding affinity measured by ITC assays (Supplemental Fig. 2B–D). Moreover, the R44 in PNKP is replaced by the more flexible K38 in APTX. The mobility of the K38 in APTX may contribute to a less favorable binding to the phospho-peptide than iso-ADP-ribose. As indicated by the models in Supplemental Figure 2A, K38 in APTX moves away from the peptide, whereas it cooperates with R29 and R42 to interact with the phosphate group in iso-ADP-ribose. Moreover, the two phosphate groups in iso-ADP-ribose are on the surface of FHA domains in PNKP and APTX and can connect with additional units of poly-(ADP-ribose) in PAR (Supplemental Figs. 1A, 2A).
The structure of the FHA domain of RNF8 and CHK2 is quite different from that of PNKP and APTX (Li et al. 2002; Huen et al. 2007). Both FHA domains failed to interact with iso-ADP-ribose or PAR (Supplemental Fig. 3). The FHA domain of RNF8 recognizes the pT-Q motifs in MDC1 (Huen et al. 2007; Kolas et al. 2007; Mailand et al. 2007). The binding pocket in the FHA domain of RNF8 includes two hydrophobic residues (L57 and L82), creating a narrower and electro-neutral binding site that favors binding Glu residue in the peptide. Although the first phosphate group in iso-ADP-ribose could form salt bridge with R61 of RNF8, the second phosphate group does not mimic the Glu of the binding peptide, which can be attributed to the steric hindrance caused by L57 in RNF8 (Supplemental Fig. 3A). Similarly, only one phosphate group could be fitted into the binding pocket in the FHA domain of CHK2 (Supplemental Fig. 3A). The calculated binding free energy suggests that the FHA domains of RNF8 and CHK2 favor interacting with pT motifs over iso-ADP-ribose, which is also consistent with our ITC analyses (Supplemental Fig. 3B,C).
We also examined the binding pocket in the BRCT domains of Ligase4, XRCC1, BRCA1, and MDC1 because the structure of these BRCT domains has been solved (Zhang et al. 1998; Sibanda et al. 2001; Shiozaki et al. 2004; Stucki et al. 2005; Wu et al. 2009; Campbell et al. 2010; Cuneo et al. 2011). However, the binding pockets in the BRCT domains of BRCA1 and MDC1 are much larger compared with those in the FHA domain. Although one phosphate group in ADP-ribose can form salt bridges with K1936 and T1898 in MDC1 or with K1702 and S1655 in BRCA1, the pockets form few contacts with ribose surges and the other phosphate in ADP-ribose. Once the adenine of ADP-ribose is docked into the binding pockets, one ribose sugar could not be linked with other unit of PAR (Supplemental Fig. 4A). Moreover, the binding free energy suggests that both BRCT domains favor binding to their phospho-protein partners over ADP-ribose, which once again is consistent with our results from ITC assays (Supplemental Fig. 4B,C). The phospho-peptide-binding partners of the BRCT domains of Ligase4 and XRCC1 have not been identified. However, both BRCT domains recognize ADP-ribose. Thus, we could not compare their binding affinities with phospho-proteins as well as the binding models. Based on the structure of the unbound BRCT domains and mutation analyses in Supplemental Figure 5, we showed that the phosphate groups in ADP-ribose interact with K675 in Ligase4 and K369 in XRCC1. Based on the binding models, the ADP-ribose adopts an extended conformation to bind to the binding site, allowing two sugar groups to connect with the remaining units of PAR. Due to lacking the structure of the human NBS1 FHA–BRCT domain, we are currently unable to examine the details of the interaction between the NBS1 BRCT domain and ADP-ribose.
Collectively, the model analyses allow us to examine the details of the interaction between the FHA/BRCT domains and PAR, which might provide the molecular basis of the interactions.
PAR mediates the early recruitment of NBS1 during DNA damage response
Massive poly-ADP-ribosylation occurs at DNA lesions, suggesting that PAR could provide critical signals to recruit PAR-binding proteins to DNA lesions and allow these PAR-binding proteins to fulfill their DNA damage repair missions. To study the biological function of these novel PAR-binding domains, we used laser microirradiation to examine whether PAR recruits the FHA and BRCT domains to DNA damage sites. The advantage of the laser microirradiation approach is that we can monitor the early DNA damage response in live cells, since the half-life of PAR is very short at DNA lesions. The times of cell exposure to the laser beam and the pulse energy were strictly equal in every operation (see the Materials and Methods for details), which ensured the same laser microirradiation treatment to all the cells. We first examined NBS1, since NBS1 is an important subunit in the MRN complex that activates ATM in response to DNA damage (Horejsi et al. 2004; Lee and Paull 2004; Stracker and Petrini 2011). As shown in Figure 3A, NBS1 was quickly recruited to DNA damage sites within 20 sec. The quick loading of NBS1 at DNA damage sites is consistent with the quick PAR synthesis at DNA damage sites (Fig. 3B), suggesting that PAR at DNA damage sites mediates the fast recruitment of NBS. Next, we treated cells with olaparib, the PARP inhibitor, to suppress PAR synthesis. Olaparib treatment abolished the early recruitment of NBS1 to DNA damage, although NBS1 still slowly accumulated at DNA damage sites (Fig. 3A), confirming that PAR is required for the fast relocation of NBS to DNA damage sites. The half-life of PAR is relatively short at DNA damage sites because PAR is quickly hydrolyzed by PARG at DNA damage sites. However, NBS1 was still retained at DNA damage sites even after PAR was digested (Fig. 3B). These results suggest that PAR mediates the fast recruitment of NBS1 to DNA damage sites, while other signals at DNA damage sites are important for the prolonged retention of NBS1 at DNA damage sites. Following DNA damage, γH2AX plays an important role in retaining DNA damage response factors at DNA damage sites (Celeste et al. 2003; Bonner et al. 2008; Lukas et al. 2011; Polo and Jackoson 2011). Thus, we examined whether γH2AX was associated with the prolonged retention of NBS1. Interestingly, in H2AX−/− cells, the stable retention of NBS1 at DNA damage sites was impaired. Following the hydrolysis of PAR, NBS1 was dropped from DNA damage sites (Fig. 3A). Moreover, additional olaparib treatment in H2AX−/− cells abolished both the fast recruitment and slow accumulation of NBS1 to DNA damage sites. These results suggest that there are two stages of the recruitment of NBS1 to DNA damage sites. The quick relocation of NBS1 to DNA damage sites is mediated by PAR at DNA damage sites, whereas the slow accumulation or stable retaining of NBS1 at DNA damage sites is regulated by H2AX. The FHA–BRCT fusion domain is essential for the relocation of NBS1 to DNA damage sites (Kobayashi et al. 2004; Stracker and Petrini 2011). The BRCT fold of NBS1 recognizes PAR, whereas the FHA fold of NBS1 could recognize the pThr motifs in MDC1 (Chapman and Jackson 2008; Spycher et al. 2008; Wu et al. 2008; Lloyd et al. 2009), a functional partner of γH2AX (Stucki et al. 2005; Lou et al. 2006). Again, in MDC1−/− cells, the stable retention of NBS1 to DNA damage sites is impaired, and olaparib treatment abolished the relocation of NBS1 to DNA damage sites in MDC1−/− cells (Fig. 3A). Similar results were observed when we examined the relocation of the FHA+BRCT domain of NBS1 to DNA damage sites (Supplemental Fig. 6), and olaparib treatment did not affect the relocation kinetics of γH2AX or MDC1 at DNA damage sites in wild-type cells (Supplemental Fig. 7A). Thus, these results suggest that PAR is the initial signal that induces the recruitment of NBS1 to DNA damage sites, while MDC1 retains NBS1 at DNA damage sites for the prolonged period. To validate this model, cells were treated with wortmannin, a PI3 kinase inhibitor, to suppress γH2AX and the recruitment of MDC1 to DNA lesions (Paull et al. 2000; Goldberg et al. 2003) but not affect the DNA damage-induced PAR synthesis at DNA damage sites (Supplemental Fig. 7B). With wortmannin treatment, NBS1 could still be recruited to DNA damage sites. However, it was dropped off from DNA damage sites when PAR was hydrolyzed (Fig. 3C). When cells were treated with both wortmannin and olaparib to suppress both the γH2AX-dependent pathway and PAR synthesis, NBS1 failed to be recruited to DNA damage sites (Fig. 3C). Moreover, with the treatment of gallotannin (GLTN), a cell-permeable PARG inhibitor (Ying et al. 2001; Fathers et al. 2012), to suppress PARG-dependent PAR hydrolysis, the half-life of PAR at DNA damage sites was significantly prolonged (Supplemental Fig. 7C). With both wortmannin and GLTN treatment to suppress γH2AX and prolong the half-life of PAR, NBS1 was still stably retained by PAR at DNA damage sites for the prolonged time (Fig. 3C). Collectively, these results demonstrate that PAR synthesis at DNA damage sites is critical for the recruitment of NBS1.
PAR mediates the function of NBS1 during early DNA damage response. (A) PAR is required for the fast recruitment of NBS1 to DNA damage sites. The NBS1-GFP was expressed in wild-type, H2AX−/−, or MDC1−/− MEFs treated with or without olaparib. The relocation of NBS1-GFP to DNA damage sites was monitored in a time course following laser microirradiation. (B) The kinetics of the accumulation of PAR and endogenous NBS1 at DNA damage sites was examined. (C) The effect of wortmannin, olaparib, or GLTN on the recruitment of NBS1 to DNA damage sites. NBS1-GFP was expressed in U2OS cells. The relocation of NBS1-GFP to DNA damage sites was monitored in a time course following laser microirradiation. GFP/RFP fluorescence at the laser line was converted into a numerical value using Axiovision software (version 4.5). Normalized fluorescent curves from 50 cells from three independent experiments were averaged. (A–C) The error bars represent the standard deviation.
The interaction between PAR and the BRCT domain of NBS1 is important for the recruitment of NBS1 to DNA damage sites
Since the BRCT fold of NBS1 is a PAR-binding motif and the FHA fold of NBS1 is a phospho-protein-binding motif, we further dissected the functions of these two motifs in vivo. We mutated Arg28 into alanine (R28A) in the FHA fold of NBS1, which does not affect the interaction between NBS1 and PAR. As shown in Figure 4A, in the wild-type mouse embryonic fibroblasts (MEFs), the R28A mutant could still be recruited to DNA damage sites but were not able to stay at DNA damage sites for the prolonged time, since this mutant abolished the interaction with MDC1. Olaparib treatment abolished the recruitment of NBS1 to DNA damage, since there is no PAR signal at DNA damage sites to recruit the R28A mutant. GLTN treatment that kept PAR at DNA damage sites for the prolonged time stabilized NBS1 at DNA damage sites. Moreover, the status of MDC1 did not affect the fast recruitment of the R28A mutant of NBS1 (Fig. 4A). In contrast, the K160A mutant of NBS1 that disrupts the interaction with PAR abrogated the early recruitment of NBS1 to DNA damage sites in wild-type cells. Loss of MDC1 further abolished the slow accumulation of NBS1 to DNA damage sites (Fig. 4B). Taken together, the FHA–BRCT fusion domain of NBS1 recognizes both PAR and MDC1. PAR is important for the initial recruitment of NBS1 to DNA damage sites, whereas MDC1 is required for the retention of NBS1 at DNA damage sites for the prolonged time.
The relocation kinetics of NBS1 mutants during DNA damage response. (A) The R28A mutation in the FHA domain of NBS1 abolishes the stable retention but not the early recruitment of NBS1 to DNA damage sites. The NBS1 R28A-GFP was expressed in wild-type or MDC1−/− MEFs treated with or without olaparib or GLTN. The relocation of the R28A-GFP to DNA damage sites was monitored in a time course following laser microirradiation. (B) K160A mutation in the BRCT domain of NBS1 abolishes the early recruitment but not the stable retention of NBS1 at DNA damage sites. The NBS1 K160A-GFP was expressed in wild-type or MDC1−/− MEFs treated with or without olaparib. The relocation of NBS1 K160A-GFP to DNA damage sites was monitored in a time course following laser microirradiation. GFP fluorescence at the laser line was converted into a numerical value using Axiovision software (version 4.5). Normalized fluorescent curves from 50 cells from three independent experiments were averaged. (A,B) The error bars represent the standard deviation.
PAR is important for the NBS1-mediated early ATM activation during DNA damage response
Since the MRN complex is important for the ATM activation in response to DNA damage (Horejsi et al. 2004; Lee and Paull 2004), we asked whether PAR-mediated relocation of NBS1 to DNA damage sites is important for the early activation of ATM. First, we examined the in vivo binding between PAR and wild-type NBS1 or the mutants following IR treatment. Cells with endogenous NBS1 depletion were reconstituted by siRNA-resistant wild-type NBS1, the K160A mutant, or the R28A mutant (Supplemental Fig. 8). As shown in Figure 5A, both wild-type NBS1 and the R28A mutant interacted with PAR within 5 min following IR. Since PAR is quickly digested by PARG in vivo, the interaction was disrupted at ∼15 min after DNA damage. As the K160A mutation abolishes the binding with PAR, we could not detect any interaction between PAR and the K160A mutant in vivo. With this interaction kinetics, we examined the activation of ATM. In the presence of wild-type NBS1 or the R28A mutant, we found that ATM was activated within 5 min following DNA damage response by examining the ATM Ser1981 phosphorylation (Fig. 5B), a surrogate maker of ATM activation (Bakkenist and Kastan 2003; Pellegrini et al. 2006). Moreover, we examined the phosphorylation of Chk2, the downstream functional partner of ATM (Falck et al. 2001; Smith et al. 2010), for the detection of ATM activation. Although the R28A mutant could not be stably retained at DNA damage sites, the mutant does not affect the late ATM activation. It is possible that activation of other PI3 kinases such as ATR and DNA-PKcs facilitates the late ATM activation (Yang et al. 2003; Stiff et al. 2006). However, in the presence of the K160A mutant, the activation of ATM was significantly delayed (Fig. 5B,C). Moreover, after IR treatment, the interaction between PAR and NBS1 and the early activation of ATM was also impaired when cells were pretreated with olaparib (Fig. 5D). Thus, these results show that PAR is important for the MRN complex-mediated early activation of ATM.
PAR is important for the NBS1-mediated early activation of ATM in response to DNA damage. (A) NBS1 binds PAR during the early DNA damage response. U2OS cells with endogenous NBS1 knockdown were reconstituted by siRNA-resistant wild-type NBS1, the R28A mutant, or the K160A mutant. Cells were lysed at the indicated time points after IR. The in vivo interaction between PAR and wild-type NBS1 or the mutants was measured by co-IP. Whole-cell lysates were blotted and are shown as the input. (B,C) PAR is important for the NBS1-mediated early activation of ATM and Chk2 during DNA damage response. U2OS cells with endogenous NBS1 knockdown were reconstituted by siRNA-resistant wild-type NBS1, the R28A mutant, or the K160A mutant. Following IR treatment, cells were lysed at the indicated time points and subjected to Western blot detected by anti-pATM (S1981) and anti-pChk2 (T68) antibodies. (D) The NBS1–PAR interaction and the early activation of ATM and Chk2 during DNA damage response were abolished by the PARP inhibitor olaparib. Following IR treatment, cells pretreated by olaparib were lysed at 5 min (left) or the indicated time points (right). (Left) The NBS1–PAR binding was detected by co-IP. (Right) The early activation of ATM and Chk2 was detected by Western blot.
The interaction between PAR and NBS1 mediates the G2/M checkpoint activation and early DNA damage repair
ATM kinase is a master kinase that phosphorylates numerous substrates and governs DNA damage-induced checkpoint activation and DNA damage repair (Kurz and Lees-Miller 2004; Shiloh and Ziv 2013). Thus, we continued to examine the role of PAR in ATM-dependent cell cycle checkpoint activation and DNA damage repair. During DNA damage response, 53BP1 is a key downstream effector of the ATM pathway (Anderson et al. 2001; Rappold et al. 2001; Ward et al. 2003). Accumulated evidence suggests that ATM-dependent 53BP1 phosphorylation activates both the G2/M checkpoint and DNA damage repair (DiTullio et al. 2002; Fernandez-Capetillo et al. 2002; Callen et al. 2013; Chapman et al. 2013). Thus, we asked whether PAR regulates the ATM-dependent 53BP1 phosphorylation. Cells pretreated with olaparib to suppress PAR synthesis were irradiated to induce double-strand breaks (DSBs). Olaparib treatment clearly suppressed early phosphorylation of 53BP1 during DNA damage response. To examine whether early phosphorylation of 53BP1 is also mediated by the interaction between PAR and the BRCT domain of NBS1, we examined the cells only expressing the K160A mutant, in which the early activation of ATM was suppressed. Consistently, the early phosphorylation of 53BP1 was suppressed in the presence of the K160A mutant of NBS1 compared with that in the presence of wild-type NBS1 (Fig. 6A). 53BP1 has been shown to mediate the G2/M checkpoint activation and DNA damage repair. (DiTullio et al. 2002; Fernandez-Capetillo et al. 2002; Callen et al. 2013; Chapman et al. 2013) Here, we examined both the G2/M checkpoint activation and DNA damage repair. Following DSBs, cells are transiently arrested before entering mitosis to provide enough time for DNA damage repair. This short and transient cell cycle arrestment at the G2/M boundary is named as the G2/M checkpoint (Lukas et al. 2004). To examine the G2/M checkpoint, we monitored mitotic population by examining the phospho-histone H3 population following DSBs, which is a standard assay for studying the transient G2/M checkpoint (Wang et al. 2002; Wu et al. 2011). As shown in Figure 6B, with IR treatment, normal cells were arrested before mitosis, as phospho-histone H3 positively stained cells were significantly reduced. However, with olaparib treatment, cells could not be fully arrested at the G2/M boundary, suggesting the loss of the transient G2/M checkpoint. Moreover, compared with wild-type NBS1, the K160A mutant also abrogated the G2/M checkpoint following IR-induced DSBs. Next, we examined the early DNA damage repair using comet assays. Again, DNA damage repair was significantly impaired when cells were treated with olaparib or only expressed the K160A mutant of NBS1 (Fig. 6C). Taken together, these results suggest that the PAR-mediated ATM activation is likely to be critical for the early checkpoint activation and DNA damage repair.
The interaction between PAR and NBS1 is important for the G2/M checkpoint activation and early DNA damage repair. (A) The DNA damage-induced early phosphorylation of 53BP1 is dependent on the interaction between PAR and NBS1. U2OS cells with endogenous NBS1 knockdown were reconstituted by siRNA-resistant wild-type NBS1 or the K160A mutant and were pretreated with or without olaparib. Following IR treatment, cells were lysed at the indicated time points and subjected to immunoprecipitation with anti-53BP1 antibody and Western blot with anti-pSQpTQ antibody. Western blot with anti-53BP1 was used as the input control. (B) Olaparib treatment or the K160A mutant abrogates IR-induced G2/M checkpoint activation. The phospho-histone 3-positive population was examined by flow cytometry, and mean values were calculated from three independent experiments. Error bars represent standard error of the mean. (C) Representative images of comet assay at the indicated time points following 4 Gy of IR treatment. Cells were subjected to neutral comet assays. Tail moments were summarized from three independent experiments with at least 30 cells in single time point per sample. The error bars represent the standard deviation.
PAR mediates the early recruitment of PNKP, APTX, Ligase4, and XRCC1 to DNA damage sites
Next, we examined the role of PAR for the recruitment of other FHA and BRCT domain-containing proteins. For PNKP, it was recruited to DNA damage sites, and olaparib treatment impaired the early recruitment of PNKP (Supplemental Fig. 9A). Like NBS1, PNKP could not be stably retained at DNA damage sites in the H2AX−/− cells. Lacking both PAR and H2AX totally abolished the relocation of PNKP to DNA damage sites. Interestingly, the R35A/R48A mutant PNKP also abolished the relocation of PNKP to DNA damage sites, suggesting that the FHA domain of PNKP is important not only for the early recruitment of PNKP, but also for the PNKP retention at DNA damage sites. A similar phenomenon was observed on the relocation of the PNKP FHA domain to DNA damage sites (Supplemental Fig. 9B). Thus, it is likely that the FHA domain of PNKP recognizes phosphate groups in other molecule besides PAR, which is important for the stability of PNKP at DNA damage sites. It has been reported that PNKP interacted with XRCC1 (Whitehouse et al. 2001; Mani et al. 2007), which also recognizes PAR at DNA damage sites. However, the recruitment of PNKP to DNA damage sites is independent of XRCC1 (Supplemental Fig. 9C). Moreover, when cells were treated with wortmannin to suppress DNA damage–induced phosphorylation signal (Sarkaria et al. 1998), the retention of PNKP at DNA damage sites was significantly impaired. Wortmannin and olaparib treatment together additively abolished the recruitment of PNKP to DNA damage sites (Supplemental Fig. 9D). Thus, it is likely that the recruitment of PNKP to DNA damage sites is via the direct interaction between the FHA domain of PNKP and PAR, and the retention of PNKP at DNA damage sites is mediated by the interaction between the FHA domain and its other phospho-binding partners. Moreover, we observed a very similar phenomenon on APTX (Supplemental Fig. 10A).
For DNA Ligase4, the olaparib treatment impaired the early recruitment of DNA Ligase4 to DNA damage sites. Lacking H2AX significantly affected the stability of Ligase4 at DNA damage sites (Supplemental Fig. 10B). However, the S668A mutant in the BRCT domain did not affect the slow accumulation of Ligase4 at DNA damage sites, suggesting that other motifs of Ligase4 mediate the H2AX-dependent retention of Ligase4 at DNA damage sites. Moreover, olaparib treatment did not show an additional delay for the recruitment of the S668A mutant (Supplemental Fig. 10B). It has been shown that Ligase4 forms a complex with other partners, such as XRCC4 and XLF, via other regions in the BRCT domain (Chen et al. 2000; Sibanda et al. 2001; Riballo et al. 2009). It is possible that other partners of Ligase4 facilitate the prolonged retention of Ligase4 at the sites of DNA damage.
For XRCC1, both the olaparib treatment and the BRCT domain mutation abolished fast recruitment of XRRC1 to DNA damage sites. Although lacking H2X mildly impaired the retention of XRCC1 at DNA damage sites, the olaparib treatment in H2AX−/− cells did not totally abolish the slow accumulation of XRCC1 to DNA damage sites (Supplemental Fig. 10C). These results suggest that PAR and the BRCT domain of XRCC1 are essential for the fast recruitment of XRCC1 to DNA damage sites. However, DNA damage-induced signals other than γH2AX are required for the retention of XRCC1 at DNA damage sites (Supplemental Fig. 10C). Collectively, our results demonstrate that PAR is a bona fide signal for the fast recruitment of various DNA damage response factors to DNA lesions. The retention of these DNA damage response factors at DNA damage sites is mediated by different mechanisms.
To examine the biological significance of PAR in response to DNA damage, we treated wild-type or H2AX−/− MEFs with olaparib followed by a low dose of IR. Lacking either PAR or H2AX, cells could still be resistant to the IR-induced DNA damage. However, after loss of both PAR and H2AX, cells were hypersensitive to a low dose of IR (Supplemental Fig. 10D). These results suggest that PAR synergizes with H2AX to recruit a set of DNA damage factors to DNA lesions for damage repair.
Discussion
Taken together, we identified two novel classes of PAR-binding module that are involved in DNA damage response. Although both the FHA and BRCT domains are known as phospho-protein-binding domains, here we found that a set of FHA and BRCT domains recognize PAR. Interestingly, the PAR-binding pocket coincides with the phospho-amino acid-binding pocket in the FHA or BRCT domain. Since both ADP-ribose and iso-ADP-ribose have two phosphate groups, it is likely that the PAR-binding pockets of the FHA or BRCT domains recognize the phosphate groups in the ADP-ribose or iso-ADP-ribose. In particular, the FHA domain of PNKP is also known to recognize the pS–pT peptide. The two phosphate groups in iso-ARP-ribose might mimic the two phosphate groups on the Ser and Thr residues in the pS–pT peptide and mediate the interaction with the FHA domain of PNKP (Supplemental Fig. 1). Moreover, the FHA domain of APTX can interact with the pS-D–pT-D peptide, in which the Asp residues are also negatively charged and mimic the phosphate group. Thus, the binding mode between the APTX FHA domain and iso-ADP-ribose could be very similar to that between the FHA and phospho-peptide.
Different from the FHA domains of PNKP and APTX, the BRCT domains of Ligase4, XRCC1, and NBS1 recognize ADP-ribose. It is possible that these BRCT domains might also recognize site-specific mono-ADP-ribose. However, in PARP1−/− cells where PAR synthesis is largely suppressed in response to DNA damage, the relocation of these BRCT domain-containing proteins to DNA lesions is also significantly suppressed (Supplemental Fig. 11). It suggests that, at least in the DNA damage context, these BRCT domains recognize PAR at DNA damage sites. Moreover, the binding pockets in these BRCT domains are associated with protein PARylation and PARPs during evolution. For example, PARylation and PARPs only existed in multicellular eukaryotes. Coincidently, XRCC1 does not exist in prokaryotes and yeast. Although both NBS1 and Ligase4 exist in yeast, the key residues in the binding pockets of the BRCT domains are missing in their yeast orthologs (Supplemental Fig. 12), and the function of the BRCT fold of yeast NBS1 is likely to recognize other signals or facilitate the FHA fold recognizing the phospho-amino acid (Lloyd et al. 2009). Thus, PAR synthesis at DNA damage sites could be very important for the recruitment of these BRCT domain-containing proteins to DNA lesions for DNA damage repair in other multicellular eukaryotes.
A previous study indicated that a region within XRCC1 (amino acids 379–400) interacted with PAR (Pleschke et al. 2000). We generated the peptide of this region but did not detect any binding between the peptide with PAR in a dot blot, pull-down, or ITC assay (data not shown). Indeed, this 22-amino-acid region in XRCC1 is too short to be correctly folded based on the structural analysis (Zhang et al. 1998) and does not directly contribute to the PAR-binding pocket of XRCC1. Moreover, not all of the FHA and BRCT domains could interact with PAR. In particular, the FHA domain of RNF8 and CHK2 and the BRCT domain of MDC1 and BRCA1 do not have the affinity with PAR. Computational model analyses also do not support the binding between the phospho-peptide-binding pockets in these domains and ADP-ribose/iso-ADP-ribose. Additional structural analysis will reveal the details of these PAR-mediated interactions.
Following DNA damage, PAR is massively synthesized at the DNA damage sites, which provides the platform to recruit DNA damage response proteins to lesions (D'Amours et al. 1999; Kim et al. 2005; Gibson and Kraus 2012). Here, we analyzed the biological function of the interaction between PAR and NBS1 because it may partially explain the molecular mechanism by which the MRN complex recognizes the DSBs and induces the early ATM activation. Our study demonstrates that both DNA damage-induced PAR synthesis and the ADP-ribose-binding pocket in NBS1 are important for early ATM activation, which controls the transient G2/M checkpoint and early DNA damage repair via the ATM-dependent signal transduction pathway. Accumulated evidence suggests that 53BP1 is a key downstream effector in the ATM-dependent pathway and could be phosphorylated by ATM and CHK2 (Anderson et al. 2001; Rappold et al. 2001; Ward et al. 2003). Phosphorylated 53BP1 is known to regulate its functional partners to govern the transient G2/M checkpoint and DNA damage repair (DiTullio et al. 2002; Fernandez-Capetillo et al. 2002; Callen et al. 2013; Chapman et al. 2013), which is consistent with our observations. Moreover, the PAR-dependent early ATM activation controls the transient G2/M checkpoint, which plays an important role in maintaining genomic stability. The transient G2/M checkpoint allows the completion of quick DNA damage repair before entering into mitosis so that the DNA lesions would not be transmitted from mother cells to daughter cells (Chen et al. 2000; Abraham 2001). However, the G2/M checkpoint only transiently exists. Prolonged arresting at the G2/M boundary will cause the mitotic exit and genomic instability (Hirose et al. 2001; Chiu et al. 2005; Yang et al. 2010). Meanwhile, DNA damage repair should also be completed quickly if cells are at the G2/M transition, since the G2/M checkpoint only transiently exists. Thus, the early activation of ATM and the ATM-dependent pathway at least plays an important role in maintaining the genomic stability of cells during the G2/M transition period. Moreover, DNA damage-induced PAR synthesis may regulate multiple layers of DNA damage repair, since PAR recruits many other repair machineries—including these FHA and BRCT domain-containing proteins—to DNA damage sites. These repair machineries may function together with ATM in a complicated network for early DNA damage repair.
Interestingly, PAR is also quickly degraded within a few minutes following DNA damage (D'Amours et al. 1999; Gagne et al. 2006; Kim et al. 2007a). Without other DNA damage response signals or other functional partners such as γH2AX, these DNA damage response factors could not stay at DNA lesions for the prolonged time (Su 2006; Polo and Jackoson 2011). Thus, DNA damage response signals such as γH2AX provide the selection for retaining DNA damage response factors at the DNA lesions for different repair mechanisms. Loss of both PAR and γH2AX, a portion of DNA damage repair proteins would not be able to reach DNA lesions, which causes cell lethality. Similar mechanism has been implicated in the cancer clinical trials, which combined PARP inhibitors and PI3 kinase inhibitors to boost the efficacy of cancer chemotherapy (Ibrahim et al. 2012; Juvekar et al. 2012).
Materials and methodsPlasmids and antibodies
For GST fusion proteins, the BRCA1 BRCT domain, 53BP1 BRCT domain, Ligase4 BRCT domain, MCPH1 BRCT2+3, MDC1 BRCT domain, PARP1 BRCT, PTIP BRCT3+4, REV1 BRCT, TdT BRCT, XRCC1 BRCT domain, APLF FHA, APTX FHA, CHK2 FHA, FOXK1 FHA, Ki67 FHA, NBS1 FHA+BRCT domain, PNKP FHA, RNF8 FHA, and MDC1 FHA were cloned into the pGEX-4T1 vector, respectively. CHFR with an N-terminal GST tag and the NBS1 FHA+BRCT domain with a C-terminal GST tag were cloned into pFastBac1 vector. For the constructs used in microirradiation experiments, PNKP, APTX, Ligase4, NBS1, XRCC1, and their indicated domains were cloned into pEGFP vector to generate plasmids encoding GFP fusion proteins (or domains). The mutations in the proteins or domains above were generated using the QuikChange site-directed mutagenesis kit (Stratagene). For the constructs used to establish stable cell lines, the siRNA-resistant full-length cDNA of NBS1 and NBS1 K160A were cloned into pCMV-Tag 4A vector. The siRNA duplexes were purchased from Dharmacon Research. The sequences of NBS1 and XRCC1 siRNA used were 5′-GTACGTTGTTGGAAGGAAAdTdT-3′ and 5′-GGGAAGAGGAAGTTGGATTdTdT-3′, respectively. siRNAs were transfected into cells using Oligofectamine (Invitrogen) according to the manufacturer's instructions. Anti-pATM, anti-pCHK2, anti-pSQpTQ, anti-XRCC1, and anti-NBS1 antibodies were purchased from Cell Signaling; anti-phosphorylated histone H3 antibody was purchased from Upstate Biotechnology; anti-Flag and anti-β-actin antibodies were purchased from Sigma; and anti-PAR antibody was purchased from Trevigen.
Immunoprecipitation and Western blot
U2OS cells were lysed with NETN-100 buffer (20 mM Tris-HCl at pH 8.0, 100 mM NaCl, 1 mM EDTA, 0.5% nonidet P-40) on ice. Soluble fractions were subjected to immunoprecipitation and Western blot and probed with antibodies as indicated.
Immunofluorescence
Cells were fixed in 3% paraformaldehyde for 25 min and permeabilized in 0.5% Triton X-100 for 20 min at room temperature. Samples were blocked with 5% goat serum and then incubated in primary antibody for 60 min. Samples were then washed with PBS three times and incubated with secondary antibody for 30 min. After PBS wash, the nuclei were stained by DAPI. The signals were visualized by fluorescence microscope.
Generation and purification of PAR and iso-ADP-ribose
PAR (or biotin-PAR) was synthesized and purified in vitro according to the previous work as described (Fahrer et al. 2007) with some modifications. Briefly, PAR was synthesized in a 15-mL incubation mixture comprising 100 mM Tris-HCl (pH 7.8), 10 mM MgCl2, 1 mM NAD+, 10 mM DTT, 60 mg/mL histone H1, 60 mg/mL histone type IIa, 50 mg/mL octameric oligonucleotide GGAATTCC, and 150 nM human PARP-1. The reaction was stopped after 60 min by addition of 20 mL of ice-cold 20% TCA. Following precipitation, the pellet was washed with ice-cold 99.8% ethanol. Polymer was detached using 0.5 M KOH/50 mM EDTA and was purified by phenol-chloroform extraction and isopropanol precipitation. Purified PAR was fractionated according to chain length by anion exchange high-pressure liquid chromatography (HPLC) protocol. Iso-ADP-ribose was generated and purified in vitro according to the procedure by Wang et al. (2012). Briefly, the purified PAR was digested by 50 U of snake venom phosphodiesterase (Worthington) with 15 mM MgCl2 overnight at room temperature. The product of the phosphodiesterase digestion, iso-ADPR, was further purified by ion exchange chromatography and Superdex 75 on fast protein liquid chromatography (FPLC). Purified iso-ADP-ribose was dried in air, dissolved by ddH2O to 50 mM, stored at −20°C.
Dot blot
Recombinant proteins (10 pmol) were conjugated to the glutathione beads and incubated with PAR (100 pmol, calculated as the ADP-ribose unit) for 2 h at 4°C. The beads were washed four times with NETN-100 buffer. GST fusion proteins were eluted from beads by glutathione and spotted onto a nitrocellulose membrane. The membrane was blocked with TBST buffer (0.15 M NaCl, 0.01 M Tris-HCl at pH 7.4, 0.1% Tween 20) supplemented with 5% milk and extensively washed with TBST. After drying in the air, the membrane was examined by anti-PAR antibody.
GST fusion protein expression and pull-down assay
GST fusion proteins were expressed in Escherichia coli or using the Bac-to-Bac baculovirus expression system (for recombinant CHFR and NBS1) (Invitrogen) and purified under standard procedures. Purified GST fusion proteins (1 pmol) were incubated with biotin-labeled PAR (5 pmol) and streptavidin beads for 2 h at 4°C. After washing with NETN-100 buffer four times, the samples were boiled in the SDS sample buffer. The elutes were analyzed by Western blot with anti-GST antibody.
ITC
ITC was carried out at 16°C with an ITC 200 Microcalorimeter (GE Healthcare). Proteins were dialyzed extensively into the buffer containing 10 mM Na2HPO4 (pH 7.5), and 100 mM NaCl at the final concentration of 20∼60 μM. Ligands (PAR, ADP-ribose, or iso-ADP-ribose) in the injection syringe were also diluted by the same buffer at the final concentration of 150∼750 μM (the concentration of PAR was calculated as the ADP-ribose unit). A typical titration consisted of 19 consecutive 2 μL injections of ligands following a preinjection of 0.4 μL of ligands into the protein solution at time intervals of 120 sec while stirring at 1000 rpm. Binding isotherms were integrated and analyzed using the software Origin 7.0 (OriginLab) provided by the manufacturer.
Molecular modeling
For the FHA domain proteins, crystal structures of CHK2/peptide (HFD-pT-YLIR; Protein Data Bank [PDB] ID: 1GXC) (Li et al. 2002), RNF8/peptide (ELK-pT-ERY; PDB ID: 2PIE) (Huen et al. 2007), PNKP/peptide (YAG-pS–pT-DEN; PDB ID:2W3O) (Ali et al. 2009), and APTX (PDB ID:3KT9) (Becherel et al. 2010) were used. To construct the APTX peptide (D-pS–D-pT-DA), APTX was aligned with the PNKP/XRCC1 peptide, and the XRCC1 peptide was mutated into the D-pS–D-pT-DA peptide followed by a structural minimization using the MOE program (Chemical Computing Group). For the comparison of the binding affinity between iso-ADP-ribose, peptides, and the FHA domain, five amino acids neighboring the pSer or pThr (capped with ACE and NME at their N and C termini) from the peptides comparable with the size of the iso-ADP-ribose were used in the binding free energy calculations. For the BRCT domain proteins, crystal structures of MDC1/peptide (pS-QEY; PDBID: 3K05) (Campbell et al. 2010), BRCA1/peptide (ISRST-pS-PTFNK; PDB ID: 1T29) (Shiozaki et al. 2004), NMR (nuclear magnetic resonance) structure of Ligase4 (PDB ID: 2E2W), and XRCC1 (PDB ID: 2D8M) were used.
For docking simulations, all of the protein structures were processed, and the protons were added according to the pH 7.0 using the MOE program (Chemical Computing Group) before they were used in the docking simulations. Two docking programs were used to evaluate and select the best binding poses. They were the GOLD program (version 4.0.1) (Jones et al. 1997) and the Glide module from Schrodinger program suite (Friesner et al. 2006). For the FHA domains of CHK2, RNF8, PNKP, and APTX, iso-ADP-ribose was docked into the binding site, whereas ADP-ribose was docked into the binding sites of MDC1, BRCA1, XRCC1, and the Ligase4 BRCT domain. In the docking simulation using the GOLD program, the centers of the binding sites for the proteins were selected at the residues mutated in the experiments and showed to be important for binding to peptides, iso-ADP-ribose, and ADP-ribose experimentally. The radius of the binding site was defined as 13 Å, large enough to cover the binding pockets. For each genetic algorithm (GA) run, a maximum of 200,000 operations were performed on a population of five islands of 100 individuals. Operator weights for crossover, mutation, and migration were set to 95, 95, and 10, respectively. The docking simulations were terminated after 20 runs for each ligand. GoldScore implemented in Gold 4.0.1 was used as the fitness function to evaluate the docked conformations. In the docking simulation using Glide, the center of the box was selected at the amino acids mutated in each protein, which were found important for binding, and the XP mode was used in docking. All of the top-ranked binding poses of iso-ADP-ribose and ADP-ribose with the proteins from two docking programs were inspected, and the poses with the phosphate groups similar to the phospho-peptides in the crystal structures and compatible with PAR were selected and are shown. The electrostatic potential surfaces of the proteins were calculated using the APBS (Baker et al. 2001) module in the PyMOL program (http://www.pymol.org) based on parameters generated from the PDB2PQR server (Dolinsky et al. 2004).
The selected binding models were then subjected to the MD simulation using the Amberprogram suite (version 12) , and the binding free energy was calculated using the Amber program suite (version 10). The force field parameters for ADP-ribose and iso-ADP-ribose were derived by using the Antechamber module in Amber. The point charge parameters of both ligands were derived from the minimized geometry at the RHF level using a 6-31G* basis set with Gaussian09 and followed by the RESP fitting of the electrostatic field potential generated from the point charges at each atom site to those calculated from Gaussian09.
The topology and coordinate files for each protein–ligand complex were prepared by first adding counter-ions to neutralize the charges of the system before it was solvated in a 12 Å cubic box of the TIP3P (Jorgensen et al. 1983) water. The system was initially minimized by a 1000-step steepest decent and a 2000-step conjugate gradient minimization procedures to the solvents. Then, a 2-psec simulation was performed to raise the temperature of the system to 150K, followed by another 18 psec of simulation to increase the temperature further to 298K, where the protein ligands were fixed using 10 kcal/mol force constants in reference to the initial structure. A second 60-nsec equilibration of the system at 298K was performed by constraining the backbone atoms of the system with a 2 kcal/mol force constant. The production run was 2 nsec. Conformations were saved from the trajectory at intervals of 1 psec. Conformations collected from 0.5–2 nsec were used for the binding affinity prediction calculations. The MD simulations were performed using the GPU accelerated version of the PMEMD program (Götz et al. 2012) in the isothermal isobaric (NTP, T = 298K and P = 1 atm) ensemble. The SHAKE (Ryckaert et al. 1977) algorithm was used to fix bonds involving hydrogen. The PME method (Darden et al. 1993) was used, and the nonbonded cutoff distance was set at 10 Å. The time step was 2 fsec, and the neighboring pairs list was updated every 20 steps.
The MM-PBSA method was used for binding free energy calculations. In the MM-PBSA calculation, the 31 conformations corresponding to 50-psec intervals in the trajectory were used for the molecular mechanics calculations. Eight conformations (taken at intervals of 200 psec) from the 1.5-nsec trajectory were chosen for the normal mode calculations for entropic contribution to the binding free energy. In the normal mode calculations, a distance-dependent dielectric constant of −4r was used, the maximum cycle was set to 60,000, and the convergence tolerance was 0.0002 kcal mol−1Å−. For the solvent-accessible surface area calculation, the default value of 0.0072 kcal/mol × Å2 for the surface tension coefficient was used.
For Ligase4 and XRCC1, we analyzed all 20 conformations (or models) of the NMR structures and performed docking simulations followed by MD simulations to evaluate the stability of binding models. For Ligase4, we found that the first and 18th conformations of the NMR structures are suitable for binding model determination. The binding model of ADP-ribose with the 18th conformation from the NMR structure yielded stable structures in the 4-nsec MD simulations. For XRCC1, we performed molecular dynamics simulations of the protein ligand-binding model based on four different XRCC1 conformations. We found that the second, third, sixth, and 17th conformations of the NMR structures gave well-defined and open binding sites and are suitable for generating the binding models with ADP-ribose for further MD simulations. Only the binding model between ADP-ribose and the 17th conformation of XRCC1 gave stable structures in a 4-nsec MD simulation.
Laser microirradiation and live-cell imaging
U2OS cells and MEFs were plated on glass-bottomed culture dishes (Mat Tek Corporation). Laser microirradiation was performed using an IX 71 microscope (Olympus) coupled with the MicoPoint laser illumination and ablation system (Photonic Instruments, Inc.). A 337.1-nm laser diode (3.4 mW) transmitted through a specific dye cell and then yielded a 365-nm wavelength laser beam that was focused through 60× UPlanSApo/1.35 oil objective to yield a spot size of 0.5–1 μm. The time of cell exposure to the laser beam was ∼3.5 nsec. The pulse energy was 170 μJ at 10 Hz. Images were taken by the same microscope with the CellSens software (Olympus). GFP fluorescence at the laser line was converted into a numerical value using Axiovision software (version 4.5). Normalized fluorescent curves from 50 cells from three independent experiments were averaged. The error bars represent the standard deviation.
Comet assays
Single-cell gel electrophoretic comet assays were performed under neutral conditions according to a previous study (Olive and Banath 2006). Briefly, U2OS cells were treated with or without 4 Gy of IR and recovered in normal culture medium for the indicated time at 37°C. Cells were collected and rinsed twice with ice-cold PBS; 2 × 104 cells per milliliter were combined with 1% LMAgarose at 40°C at the ratio of 1:3 (v/v) and immediately pipetted onto slides. For cellular lysis, the slides were immersed in the neutral lysis solution (2% sarkosyl, 0.5 M Na2EDTA, 0.5 mg/mL proteinase K at pH 8.0) overnight at 37°C in the dark followed by washing in the rinse buffer (90 mM Tris buffer, 90 mM boric acid, 2 mM Na2EDTA at pH 8.5) for 30 min with two repeats. Next, the slides were subjected to electrophoresis at 20 V (0.6 V/cm) for 25 min and stained in 2.5 μg/mL propidium iodide for 20 min. All images were taken with a fluorescence microscope and analyzed by Comet Assay IV software.
IR treatment and colony formation assay
Cells were irradiated with a 137Cs source at a dose of 10 Gy (or at the indicated doses). After irradiation, cells were lysed at the indicated time points for immunoprecipitation or Western blot. For colony formation assay, 500 wild-type or H2AX−/− MEFs were seeded into six-well plates and then treated by various doses of IR with or without olaparib. After a 7-d culture, the viable cells were fixed and stained with crystal violet. The number of colonies (>50 cells for each colony) was calculated.
G2/M checkpoints assay
Cells expressing the wild-type NBS1 or NBS1 K160A pretreated with or without olaparib were treated with or without 2 Gy of IR. After 1 h of recovery, cells were fixed with 70% (v/v) ethanol, stained with rabbit antibody to phospho-histone H3 (pSer10), and then incubated with FITC-conjugated goat secondary antibody to rabbit. The stained cells were treated with RNase A and then incubated with propidium iodide. Samples were analyzed by flow cytometry.
Drug treatment
For live-cell imaging, immunoprecipitation, or Western blot, 100 nM olaparib, 10 μM GLTN, or 10 μM wortmannin was added into the cell culture medium 1 h before laser microirradiation or cell lysis. For colony formation assay, 100 nM olaparib was added into the medium during the culture.
Statistical analyses
All experiments were performed in triplicates unless indicated otherwise. Means and standard deviations were plotted. Student's t-test was used for statistical analyses.
Acknowledgments
We thank Drs. Ming Lei and Feng Zhang for technical support. This work was supported by the National Institute of Health (CA132755 and CA130899 to X.Y.). X.Y. is a recipient of the Era of Hope Scholar Award from the Department of Defense.
Supplemental material is available for this article.
Article is online at http://www.genesdev.org/cgi/doi/10.1101/gad.226357.113.
The majority of neural stem cells (NSCs) in the adult brain are quiescent, and this fraction increases with aging. The transcriptional mechanisms that promote NSC quiescence are largely unknown. Here, Martynoga et al. establish the first cell culture model of NSC quiescence and identify nuclear factor one (NFI) transcription factor family member NFIX as an essential regulator of the quiescent state. Interestingly, analyses of the hippocampus of Nfix mutant mice suggest that NFIX controls quiescence by regulating NSC interactions with their microenvironment.
The majority of neural stem cells (NSCs) in the adult brain are quiescent, and this fraction increases with aging. Although signaling pathways that promote NSC quiescence have been identified, the transcriptional mechanisms involved are mostly unknown, largely due to lack of a cell culture model. In this study, we first demonstrate that NSC cultures (NS cells) exposed to BMP4 acquire cellular and transcriptional characteristics of quiescent cells. We then use epigenomic profiling to identify enhancers associated with the quiescent NS cell state. Motif enrichment analysis of these enhancers predicts a major role for the nuclear factor one (NFI) family in the gene regulatory network controlling NS cell quiescence. Interestingly, we found that the family member NFIX is robustly induced when NS cells enter quiescence. Using genome-wide location analysis and overexpression and silencing experiments, we demonstrate that NFIX has a major role in the induction of quiescence in cultured NSCs. Transcript profiling of NS cells overexpressing or silenced for Nfix and the phenotypic analysis of the hippocampus of Nfix mutant mice suggest that NFIX controls the quiescent state by regulating the interactions of NSCs with their microenvironment.
Cellular quiescence is a reversible state of growth and proliferation arrest that can be adopted by many types of cells, from bacteria and yeast to cultured mammalian fibroblasts and adult tissue stem cells (Coller et al. 2006; Valcourt et al. 2012). It is an active state that involves important changes in cell physiology, including energy metabolism and cell adhesion (Venezia et al. 2004; Coller et al. 2006; Fukada et al. 2007; Pallafacchina et al. 2010; Lien et al. 2011; Brohl et al. 2012; Valcourt et al. 2012).
Quiescence is essential to prevent the premature exhaustion of long-lived self-renewing stem cell populations (Orford and Scadden 2008). The balance between neural stem cell (NSC) proliferation and quiescence in the adult brain is regulated by diverse physiological stimuli, and disruption of this balance is thought to contribute to the cognitive decline of old age (Lee et al. 2011; Faigle and Song 2013). However, the cell-intrinsic mechanisms that mediate these effects and control NSC quiescence and activity remain poorly understood.
Stem cells are present in two regions of the postnatal and adult brain: the subependymal zone (SEZ) adjacent to the lateral ventricles and the dentate gyrus (DG) of the hippocampus, where they continuously generate new neurons that integrate into neuronal circuits of the olfactory bulb and hippocampus, respectively (Temple 2001; Fuentealba et al. 2012). In contrast to the highly proliferative stem cells of the embryonic neural tube, NSCs in the postnatal and adult brain are relatively quiescent (Temple 2001; Niu et al. 2011; Fuentealba et al. 2012). Adult NSCs are stimulated to divide by diverse physiological stimuli, including physical exercise and cognitive stimulation, while conversely, stress, anxiety, and old age suppress their divisions (Fabel and Kempermann 2008; Ma et al. 2009; Lucassen et al. 2010). Seizures stimulate NSC divisions in aged mice, suggesting that this cell cycle arrest is reversible (Lugert et al. 2010). Adult stem cells inhabit specialized niches that produce signals controlling their lifelong self-renewal and production of differentiated progeny (Fuchs et al. 2004; Riquelme et al. 2008; Fuentealba et al. 2012; Faigle and Song 2013). In particular, Notch and BMP signaling, activated by ligands presented by differentiating neural precursors in the neurogenic niches, provide negative feedback signals that maintain the quiescent state of SEZ and hippocampal stem cells (Bonaguidi et al. 2008; Ables et al. 2010; Ehm et al. 2010; Imayoshi et al. 2010; Mira et al. 2010).
Although progress has been made in identifying transcription factors (TFs) that regulate different steps of neurogenesis in the postnatal and adult brain (Hsieh 2012), the nature of the factors that mediate the activity of extrinsic signals and determine cell-intrinsically the quiescent or proliferating state of NSCs is still largely unknown. Elucidating these transcriptional mechanisms is essential to understand how the physiology and behavior of NSCs is regulated and, in the longer term, develop therapeutic interventions; e.g., to counteract the decline of neurogenesis in the aging brain. The FoxO proteins are currently the best-studied TFs promoting quiescence in NSCs. In mice mutant for FoxO1, FoxO3, and FoxO4 or for FoxO3 alone, an initial excess of NSC proliferation is followed by a depletion of the NSC pool and a decline in neurogenesis (Paik et al. 2009; Renault et al. 2009). Whether other TFs act downstream from quiescence-promoting signals and regulate common or distinct aspects of the physiology of quiescent NSCs is not known.
The main obstacles to studying NSC quiescence are the difficulty of isolating these cells in sufficient numbers from highly complex adult neurogenic niches (Beckervordersandforth et al. 2010; Fuentealba et al. 2012) and the lack of a well-characterized cell culture model. NSCs are routinely maintained in culture (NS cell cultures) in the presence of high concentrations of mitogens and are highly proliferative (Pastrana et al. 2009; Ehm et al. 2010; Mira et al. 2010; Sun et al. 2011). However, BMP ligands have recently been shown to promote cell cycle arrest in adherent cultures of mouse and rat NS cells (Mira et al. 2010; Sun et al. 2011). In this study, we examined in detail BMP-treated, embryonic stem cell-derived NS cells and demonstrated that they have characteristic features of quiescent cells. We also showed that entry into quiescence involves major changes in the transcriptional profile of these cells and particularly in their expression of cell adhesion and extracellular matrix (ECM) molecules.
We used this NS cell quiescence model to identify TFs that participate in the gene regulatory network (GRN) that governs the quiescent state in NSCs. For this, we characterized regulatory elements that are active in quiescent NS cells by genome-wide mapping of the enhancer-associated histone mark H3 Lys 27 acetylation (H3K27ac) and coactivator p300 (Heintzman et al. 2009; Creyghton et al. 2010; Rada-Iglesias et al. 2011; Rada-Iglesias et al. 2012). We found that proteins of the nuclear factor one (NFI) family bind to a very large fraction of these enhancers and that family member NFIX is required for the establishment of a significant portion of the gene expression program of quiescent NS cells and the suppression of a significant part of the gene expression program of proliferating NS cells. Finally, we show that mutation of the Nfix gene results in loss of quiescence in a significant fraction of hippocampal NSCs in vivo.
Together, this study shows that establishing a cell culture model of NSC quiescence has allowed us to characterize fundamental aspects of the biology of NSCs and identify a key TF that plays an essential role in implementing the quiescent NSC gene expression program.
ResultsBMP4-treated NS cells are quiescent
To model NSC quiescence in culture, we replaced the mitogen EGF with BMP4 in the culture medium of NS cells, which also contains FGF2 (Conti et al. 2005; Mira et al. 2010; Sun et al. 2011). We monitored cell proliferation by staining for the proliferation marker Ki67 and measuring incorporation of the thymidine analog EdU. We observed that NS cells had stopped proliferating 24 h after addition of BMP and remained cell cycle-arrested when maintained in the presence of BMP for 3 d and up to 28 d (Fig. 1A–E; data not shown). The cell cycle arrest was due to exposure to BMP, since removing EGF from the culture medium without adding BMP4 did not block proliferation (Supplemental Fig. S1A), and adding the BMP signaling inhibitor Noggin to the BMP4-containing medium prevented NS cells from exiting the cell cycle or caused cell cycle re-entry when cells had previously been exposed to BMP4 for 3 d (Supplemental Fig. S1A). Flow cytometry analysis revealed that BMP-treated cells were arrested with a 2N DNA content; i.e., in the G1 or G0 phase of the cell cycle (Supplemental Fig. S1B). Antibody staining confirmed that the cell cycle-arrested cells maintained expression of the NSC markers Sox2, Nestin, and BLBP and did not express the astrocyte marker S100β or the neuronal marker βIII-tubulin, while expression of the NSC/astrocyte marker GFAP was increased and expression of EGFR, a marker of activated NSCs (Pastrana et al. 2009), was suppressed by the BMP treatment (Supplemental Fig. S1C).
Characterization of cell cycle-arrested NS cell cultures. (A) Time course of experimental treatments. E cultures were maintained continuously in EGF-containing medium. EB and EBE cultures were first transferred into BMP4-containing medium for 3 d and then replated into either BMP4-containing (EB) or EGF-containing (EBE) medium for 1–6 d. (B,C) Analysis of proliferation by Ki67 immunostaining (B, red) and EdU detection (C, red) after 4-h exposure of NS cells in E, EB, or EBE cultures as indicated. Cells were counterstained with DAPI (blue). (D) Percentages of EdU-positive NS cells in E, EB, and EBE cultures. EB and EBE cultures were replated in BMP4 or EGF, respectively, for 1–6 d, as indicated. Cell cycle-arrested cells promptly resume proliferation when EGF replaces BMP4. Error bars represent the standard deviation (n = 3 biological replicates). (E) Percentages of EdU-positive cells in cultures exposed to BMP4 for 28 d and to EGF for another 3 d (right) or maintained in EGF for the same period (left). BMP-induced cell cycle arrest remains fully reversible even after a prolonged exposure. (F) Expression of the neuronal-specific gene βIII-tubulin in E and EBE cultures (exposed to BMP4 for 3 d) switched to a neuronal differentiation medium for 10 d. (G,H) Percentages of βIII-tubulin+ cells in E and EBE cultures maintained in BMP medium for 3 d (G) or 28 d (H) before being switched back to EGF medium for 6 d and neuronal differentiation medium for 10 d. Prolonged exposure to BMP4 does not affect the differentiation potential of NS cells. (I) Hierarchical clustering of normalized expression values from gene microarray analysis of NS cells in E, EB, and EBE cultures. Cycling NSCs (E cultures) cluster separately from cell cycle-arrested NS cells (EB) but together with reactivated NS cells (EBE). Three independent samples were hybridized to microarrays for each condition (Rep1–Rep3). (J) Comparison of transcript levels in cycling and cell cycle-arrested NS cells by RNA-seq. Genes are ranked along the X-axis according to the statistical significance (log10P-value) of difference in normalized expression levels (FPKM [fragment per kilobase transcriptome per million mapped reads]) between proliferating NS cells (E cultures) and arrested NS cells (EB cultures). Transcripts down-regulated in arrested NS cells are given a negative value. The horizontal dotted line represents the P = 0.05 significance threshold. The corresponding heat map is shown at the bottom. (K, L) GO analysis of genes down-regulated (K) and up-regulated (L) in cell cycle-arrested NS cells. The X-axis values correspond to DAVID P-values. All terms reported have a false discovery rate (FDR) < 5%. The number of genes belonging to each category is shown in brackets. See also Supplemental Figure S1 and Supplemental Table S1.
To determine whether the proliferation arrest of BMP-treated NS cells is reversible, a defining property of quiescent cells, we removed the BMP4-containing medium after 3 or 28 d and returned the cells to EGF-containing medium (Fig. 1A). After 3 d in proliferation medium, NS cells had resumed proliferation at a rate similar to that of control NS cells (Fig. 1B–E; Supplemental Fig. S1D) and had also retained their neuronal differentiation potential (Fig. 1F–H). To confirm that the effect of BMP4 is fully reversible, we used expression microarrays to compare the transcript profile of NS cells cultured sequentially in EGF medium, BMP medium for 3 d, and EGF for 6 d (called “EBE cultures” below and in Fig. 1) with the transcriptome of NS cells cultured continuously in EGF medium (“E cultures”) and with that of NS cells cultured in EGF and then BMP for 3 d (“EB cultures”) (Fig. 1A). Only 49 genes were significantly deregulated in EBE cultures compared with E cultures (17 down-regulated and 32 up-regulated more than twofold; P < 0.05). Moreover, cluster analysis of the microarray data showed that EBE cultures clustered together with E cultures and separately from EB cultures, thus suggesting that they had reverted to a transcriptional state indistinguishable from that of cells that had proliferated continuously (Fig. 1I). We thus conclude that exposure of NS cells to BMP4 for 3–28 d induces a state of cell cycle arrest that is entirely reversible.
To further examine the changes in gene expression associated with BMP4-induced cell cycle arrest, transcripts from cell cycle-arrested and proliferating NS cells were compared by RNA sequencing (RNA-seq). We found that 2475 genes were up-regulated and 1980 genes were down-regulated in arrested NS cells compared with proliferating NS cells (P < 0.05) (Fig. 1J). The quality of this data set was assessed by quantitative PCR (qPCR) analysis, which confirmed the regulation of a selection of up-regulated and down-regulated genes in BMP4-treated cells (Supplemental Fig. S1F). Gene ontology (GO) analysis using DAVID (Database for Annotation, Visualization, and Integrated Discovery; http://david.abcc.ncifcrf.gov) showed that down-regulated mRNAs were mostly involved in the cell cycle (e.g., GO terms: “cell cycle” and “chromosome”) and DNA and RNA metabolism (“DNA metabolic process” and “RNA processing”), as expected for a cell cycle-arrested cell population (Fig. 1K). Other down-regulated genes were associated with protein translation (“ribonucleotide complex” and “ribosome biogenesis”), which is reminiscent of the reduction in protein synthesis associated with quiescence in many mammalian cells as well as yeast and bacteria (Valcourt et al. 2012).
Conversely, up-regulated genes included the cyclin-dependent kinase inhibitor Cdkn2b/p15/INK4B (fold change = 17.5; P = 6.56 × 10−7) as well as many cell cycle inhibitors induced in other types of quiescent cells (Venezia et al. 2004; Coller et al. 2006; Fukada et al. 2007; Lien et al. 2011). However, the most significantly enriched up-regulated gene categories in cell cycle-arrested NS cells were associated with the ECM (“extracellular matrix” and “polysaccharide binding”) and cell–cell adhesion (“adherens junction”) (Fig. 1L), including a large number of ECM genes (15 collagens, three laminins, and one spondin), receptors for ECM proteins (nine integrins), and cell adhesion molecules (four cadherins, two protocadherins, six cell adhesion molecules [CAMs], and four claudins) (Supplemental Table S1). All of these classes of gene are known to control the interaction of stem cells with their niche and signaling environments (Chen et al. 2013).
We then used gene set enrichment analysis (GSEA) (Subramanian et al. 2005) to directly compare the genes up-regulated and down-regulated in arrested NSCs with genes induced in published microarray profiling studies of different types of quiescent cells, including hematopoietic stem cells (Venezia et al. 2004), skeletal muscle stem cells (Fukada et al. 2007), hair follicle stem cells (Lien et al. 2011), and fibroblasts (Coller et al. 2006). All gene sets expressed in these quiescent cell populations were highly enriched in transcripts up-regulated in arrested NS cells (Fig. 2A–D), and several of the GO terms associated with quiescence-enriched gene sets in other cell types were also associated with cell cycle-arrested NS cell genes (Fig. 2F,G; Supplemental Fig. S2A,B; Beckervordersandforth et al. 2010). It is noteworthy that although many genes up-regulated in arrested NS cells were enriched in one or two other quiescent cell types, there was no common gene signature shared by all of the quiescent cells analyzed (Fig. 2E). GSEA also showed strong enrichment among the genes induced in arrested NSCs; in genes expressed in adult SEZ NSCs, which are mostly in a quiescent state (Beckervordersandforth et al. 2010); and in genes induced in neurosphere cultures by the quiescence-promoting factor FoxO3 (Supplemental Fig. S2C,E; Renault et al. 2009). Conversely, genes expressed by non-NSC astrocytes (Beckervordersandforth et al. 2010) were not significantly enriched, suggesting that BMP-treated NS cells in culture are more similar to SEZ NSCs and other quiescent adult stem cell populations than to differentiated parenchymal astrocytes (Supplemental Fig. S2D). Collectively, these results indicate that BMP4 induces in cultured NS cells a state of reversible cell cycle arrest and a transcriptome profile that are characteristic of quiescent cells.
Cell cycle-arrested NS cells share transcriptomic features with various types of quiescent cells. (A–D) GSEA analysis shows that genes that are up-regulated in cell cycle-arrested NS cells (ranked by expression along the X-axis) are highly enriched in gene sets that are up-regulated in quiescent muscle stem cells (A), hematopoetic stem cells (B), hair follicle stem cells (C), and quiescent fibroblasts (D). (NES) Normalized enrichment score. (E) Hierarchical clustering of the 455 genes up-regulated in cell cycle-arrested NSCs, quiescent muscle stem cells (MSCs), hematopoetic stem cells (HSCs), hair follicle stem cells (HFSCs), and quiescent fibroblasts (Fib). (F,G) GO analysis of genes expressed by quiescent muscle stem cells (F) and fibroblasts (G). GO terms also enriched in cell cycle-arrested NSCs (see Fig. 1L) are indicated by darker blue bars. See also Supplemental Figure S2.
Identification of active enhancers in quiescent and proliferating NS cells
To identify components of the GRN that control the quiescent state in NS cells, we characterized the enhancer elements that recruit TFs in these cells. To identify putative active enhancers in quiescent NS cells, we performed chromatin immunoprecipitation (ChIP) coupled to high-throughput DNA sequencing (ChIP-seq) to locate the histone acetlytransferase p300 and the histone modification H3K27ac in the genome of these cells. We defined an active enhancer as a genomic region located >2 kb from a gene transcription start site (TSS) where a ChIP-seq peak for p300 occurred within an island of H3K27ac, in agreement with recent reports (Rada-Iglesias and Wysocka 2011; Rada-Iglesias et al. 2012). Using this definition, we identified 16,810 active enhancers in the genome of quiescent NS cells (Fig. 3A; Supplemental Table S2). The large majority of these enhancers was located <20 kb (38%) or between 20 and 100 kb (42%) of the nearest genes (Supplemental. Fig. S3A).
Identification of active enhancers in quiescent NS cells. (A) Heat map representation of the density of ChIP-seq reads for H3K27ac and p300 ±2 kb relative to the midpoint of enriched regions at 16,246 active enhancers in NS cells. This panel represents the merger of data obtained in proliferating and quiescent NS cells. A large fraction of the regions displayed presents active enhancer features only in proliferating NS cells or only in quiescent NS cells, and a smaller fraction presents these features in both cellular states. Intensity of color represents the normalized statistical significance of the signal versus input control sequences. (B,C) H3K27ac and p300 ChIP-seq signal and RNA expression level (FPKM) in quiescent (blue) and proliferating (green) NS cells in the vicinity of Id4 and Vash1, two representative genes that are up-regulated in quiescent and proliferating NS cells, respectively. Regions defined as quiescent and proliferating NS cell-specific enhancers are indicated by blue and green rectangles, respectively. ChIP-seq peak height corresponds to SICER P-value for H3K27ac and MACS Q-value for p300. (D) Average ChIP-seq signal profile for H3K27ac in quiescent (blue line) and proliferating (green line) NS cells and several other epigenetic marks in proliferating NS cells at regions defined as quiescent (left) and proliferating (right) NS cell-specific enhancers. Plots are centered on the p300 summit. Quiescent NS cell-specific enhancers show strong signals for the enhancer-associated H3K4me1 mark and weak signals for the open chromatin-associated H3K4me2 and H3K27ac marks in proliferating NS cells, consistent with these regions being marked as enhancers but minimally active in proliferating NS cells. Proliferating NS cell-specific enhancers have strong signals for H3K27ac, H3K4me1, and H3K4me2 but not the other nonenhancer-associated epigenetic modifications. Note the dip in the enrichment profile for H3K27ac, indicative of a localized depletion of nucleosomes characteristic of enhancers (Heintzman et al. 2007; Bonn et al. 2012). (E) Box plots of normalized transcript counts (FPKM) for all genes expressed in quiescent NS cells (left) and genes associated with quiescent NS cell-specific enhancers (right). The latter are expressed at higher levels than the transcriptomic average (Wilcoxon test, P < 2.2 × 10−16). (F) Fraction of genes up-regulated in quiescent NS cells whose closest enhancer is quiescent NS cell-specific (left), pan-NS cell (middle), or proliferating NS cell-specific (right). Asterisk denotes significant P-value (Wilcoxon test). See also Supplemental Figure S3.
We next asked whether enhancers identified in quiescent NS cells were present in all NS cells regardless of their cell cycle status or whether they were specific for the quiescent state. We examined the location of p300 and the H3K27ac mark in the genome of proliferating (EGF-treated) NSCs by ChIP-seq and identified 10,270 active enhancers in these cells using the same definition as above (Fig. 3A; Supplemental Fig. S3B; Supplemental Table S2). As expected from the large differences in transcript profiles between quiescent and proliferating NSCs (Fig. 1J), the majority of genomic regions with active enhancer features in quiescent NS cells did not have these features in proliferating cells (9157 quiescence-specific enhancers), while a smaller proportion of enhancers active in proliferating NS cells was not found in quiescent cells (3098 proliferation-specific enhancers), and the remaining enhancers were equally active in either both quiescent and proliferating NS cells (3991 pan-NS cell enhancers) or an intermediate activity state (Fig. 3A–C; Supplemental Table S2).
To validate the enhancers that we identified in NS cells, we compared our p300 and H3K27ac profiles with published data sets on the distribution of histone marks in proliferating neural precursor cells (Mikkelsen et al. 2007; Meissner et al. 2008). We found that proliferating NS cell-specific enhancers were strongly enriched for the histone modifications H3K4me1 and H3K4me2, previously associated with enhancers in many cell types (Fig. 3D; Heintzman et al. 2007; Ernst et al. 2011). Quiescent NS cell-specific enhancers were also enriched in H3K4me1 and H3K4me2 in proliferating neural precursors but to a much lower degree. There was also some residual signal for H3K27ac, although it lacked the “valley” shape that is characteristic of active enhancers (Fig. 3D; Heintzman et al. 2007; Bonn et al. 2012). These observations suggest that these quiescence-specific enhancers exist in a primed but less active state in proliferating NS cells (Creyghton et al. 2010; Zentner et al. 2011; Bogdanovic et al. 2012), although they lack the H3K27me3 mark that some studies have found to be characteristic of “poised” enhancers (Rada-Iglesias et al. 2011). Furthermore, both proliferation-specific and quiescence-specific enhancers showed little or no enrichment for histone modifications associated with promoters (H3K4me3), gene transcription (H3K36me3), or repression (H3K27me3 and H3K9me3) (Fig. 3D).
We also examined whether epigenomically defined regulatory elements displayed enhancer activity in a luciferase reporter assay. Six out of the seven quiescence-specific enhancers analyzed drove significant reporter gene activity in quiescent NS cells, and all of these regions were silent in proliferating cells, as expected (Supplemental Fig. S3C,D). We also obtained a good validation rate for proliferation-specific enhancers, with five out of seven enhancers analyzed showing activity in proliferating NS cells. Other recent studies have shown similar, or slightly lower, validation rates for epigenomically defined enhancers in luciferase assays (Zentner et al. 2011; Ostuni et al. 2013). Unexpectedly, we observed that all of the proliferation-specific enhancers were active in quiescent NS cells, suggesting that epigenetic factors that are absent from the transfected reporter constructs are required to silence proliferation-specific enhancers in NS cell quiescence (Supplemental Fig. S3C,D).
Altogether, these results confirm that the genomic elements identified by the coincidence of p300 and H3K27ac signals have an overall epigenetic signature of active enhancers (Heintzman et al. 2007, 2009; Ernst et al. 2011; Rada-Iglesias and Wysocka 2011; Rada-Iglesias et al. 2012).
Quiescent NS cell enhancers are associated with highly expressed genes
When we assigned enhancers to their nearest gene, we found a strong positive correlation between quiescence-specific enhancers and genes highly expressed in quiescent NS cells (Wilcoxon test, P < 2.2 × 10−16) (Fig. 3E). To investigate the functions of these enhancer-associated genes, we used the Genomic Regions Enrichment of Annotations Tool (GREAT) (McLean et al. 2010). We found that genes associated with quiescence-specific enhancers are highly overrepresented for generic stem cell terms (“stem cell development”) and terms associated with quiescent NS cell-enriched genes (“cell junction assembly”) (Fig. 1G; Supplemental Fig. 3E).
We then examined the reciprocal association of quiescent NSC-enriched genes with the different types of NS cell enhancers and found, as predicted, that genes up-regulated in quiescent NS cells are more likely to have a quiescence-specific element than a proliferation-specific or pan-NS cell element as their nearest enhancer (Wilcoxon test, P < 2.2. × 10−16) (Fig. 3F). The strong association between quiescent NS cell-enriched genes and quiescence-specific enhancers is also observed if one considers not just the nearest element, but an aggregate score of all enhancers present in the intervals between the genes and their two neighbors (see the Materials and Methods). Genes with a higher aggregate enhancer score in quiescent NS cells than in proliferating NS cells were significantly more likely to be induced in quiescence than expected by chance (P = 1.24 × 10−6). Altogether, our definition of enhancers succeeds in identifying genomic regions that have the epigenetic characteristics of active enhancers and are associated with genes that are highly expressed in quiescent NS cells and have functions relevant to the quiescent NS cell state.
Widespread binding of NFI TFs to quiescent NS cell enhancers
We hypothesized that TFs that play major roles in the GRN of quiescent NS cells should bind to a large fraction of enhancers in these cells. Quiescence-specific enhancers should therefore be enriched in the DNA-binding motifs of these TFs. De novo motif searches using the algorithms GADEM (Li 2009), MEME-chip (Machanick and Bailey 2011), and RSAT (Thomas-Chollier et al. 2012) consistently recovered three distinct motifs that were significantly enriched, specifically around the summit of p300 binding, in all three categories of NS cell enhancers (Fig. 4A–G; data not shown). These motifs closely resemble consensus binding sites for NFI (Fig. 4A), Sox factors (Fig. 4B), and basic helix–loop–helix (bHLH) factors (Fig. 4C). When taking into account the frequency of random occurrence of these motifs in the genome, the NFI motif was by far the most strongly enriched in quiescence-specific enhancers (Fig. 4D,E) and was also more prevalent in pan-NS cell enhancers (Fig. 4D,F), while the bHLH motif (E-box) was the most abundant in proliferation-specific enhancers (Fig. 4D,G). These results suggest that members of the NFI TF family bind to a large fraction of enhancers in quiescent NS cells and may therefore contribute significantly to the GRN that operates in these cells.
The NFI motif is most strongly overrepresented in quiescent NS cell enhancers. (A–C) DNA sequence motifs matching consensus binding sites for NFI (A), Sox (B), and bHLH (C, E-box) TFs are found overrepresented in quiescence-specific, pan-NS cell, and activity-specific enhancers by de novo motif searches. (D) Enrichment values (E-values) of NFI, Sox, and E-box motifs in quiescence-specific, pan-NS cell-specific, and activity-specific enhancers as reported by DREME. (E–G) Observed frequency of motif occurrence around the summit of p300 binding in quiescence-specific, pan-NS cell, and activity-specific enhancers. The NFI motif is the most overrepresented in quiescence-specific and pan-NS cell enhancers. (H) RNA-seq shows that all four NFI genes are transcribed in NS cells, with transcript levels of Nfix increasing sharply in quiescent cells, while those of Nfia, Nfib, and Nfic decrease or remain unchanged. (I,J) Immunocytochemistry shows that NFIX protein is strongly induced in quiescent NS cells, while NFIA expression is reduced. See also Supplemental Figure S4.
To address whether motif enrichment predicts TF binding and examine the function of enhancer-bound TFs, we chose to focus on the NFI family, since the NFI motif was the most prevalent in quiescent NSC enhancers, and these factors were not previously known to regulate NSC biology, whereas the functions of Sox and bHLH factors have already been extensively studied in these cells (Bylund et al. 2003; Ligon et al. 2007; Scott et al. 2010; Castro et al. 2011). Our RNA-seq, immunocytochemistry, and Western blot data showed that the four members of the NFI family (NFIA, NFIB, NFIC, and NFIX) are expressed in both proliferating and quiescent NSCs, but NFIX is up-regulated when NS cells become quiescent (the nuclear protein ratio in quiescent cells/proliferating cells is 226%), while NFIA, NFIB, and NFIC are down-regulated or unchanged (64%, 51%, and 87%, respectively) (Fig. 4H–J; Supplemental Fig. S4A–C). To examine NFI protein binding to the NS cell genome, we performed a ChIP-seq analysis with an antibody that specifically recognizes the four NFI factors, which are closely related in sequence (Mason et al. 2009; Pjanic et al. 2011). NFI factors bound to 25,807 high-confidence sites in quiescent NS cells (Fig. 5A,B). As expected, de novo analysis identified the consensus NFI-binding motif as the most overrepresented in NFI-bound regions (Supplemental Fig. S4D,E). NFI-binding events were found in a very large fraction of the enhancers in quiescent NS cells (12,222; 73%), thus confirming that the strong enrichment of NFI motifs is a useful predictor of the widespread binding of these factors in quiescent NS cell enhancers (Fig. 5A,B). Conversely, an unusually high fraction of the NFI-binding sites (12,323; 48%) mapped within an active enhancer, while a further 2675 binding sites (10.3%) mapped in promoter regions (Fig. 5B; Supplemental Fig. S4F). Furthermore, within quiescent NS cell enhancers, the significance of the NFI-binding peaks correlated strongly with that of p300 peaks (Fig. 5C), and the peak summits mapped closely to each other (Fig. 5D), suggesting that NFI factors have a central role in enhancer activity in quiescent NS cells.
NFI TFs bind to the majority of quiescent NS cell enhancers. (A) Heat map representation of all enhancers active in quiescent NS cells sorted into quiescent-specific and pan-NS cell enhancers showing ChIP-seq signal for NFI TFs, H3K27ac, and p300. (B) Venn diagram showing the large overlap of enhancers in quiescent NS cells with regions of significant NFI TF binding. (C,D) Strong correlation of the strength of ChIP-seq signals for p300 and NFI in enhancers (C; correlation coefficient = 0.67) and close proximity of their summits (D; median intersummit distance = 35 base pairs [bp]), consistent with p300 recruitment by NFI TFs. (E,F) Functional annotation of quiescence-specific (E) and pan-NS cell (F) enhancers bound by NFI TFs by GREAT according to GO biological process. Enhancers bound by a NFI factor (purple) and those that are not significantly bound (green) were examined separately. The X-axis values represent the binomial FDR Q-values; the numbers in parentheses are the number of binomial region hits.
Functional annotation of NFI-bound enhancers using GREAT also supported an important role for these enhancers in regulating the quiescent NS cell state. Genes linked to quiescent-specific enhancers that are bound by NFI factors are particularly enriched for the processes of cell junction organization and assembly, while genes associated with pan-NS cell enhancers bound by NFI are mostly involved in carbohydrate metabolism (Fig. 5E,F). These activities are specifically associated with the quiescent NS cell state (Fig. 1L), which argues for a central role of NFI proteins in the GRN operating in quiescent NS cells.
NFIX is both required and sufficient to induce aspects of quiescence in NS cells
To directly address the role of NFI factors in quiescent NS cells, we focused on NFIX, as it is the only family member whose expression is up-regulated when cells enter quiescence. NS cells were transduced with a lentivirus encoding an shRNA for Nfix that reduced Nfix mRNA and protein levels to 37%–30% and 51%–30% of controls, respectively (Fig. 6A; Supplemental Fig. S5A–D; Messina et al. 2010). EdU incorporation and flow cytometry analysis showed that in proliferation conditions, Nfix shRNA-expressing and control cells proliferated at the same rate, suggesting that Nfix does not play a major role in NS cell proliferation (Supplemental Fig. S5E). In contrast, the entry into quiescence of Nfix shRNA-expressing cells was delayed, and a significant fraction of the cells remained proliferative even after 3 d of exposure to BMP (Fig. 6B,C; Supplemental Fig. S5E). Conversely, overexpressing NFIX in proliferating NS cells resulted in a rapid cell cycle arrest of electroporated cells without induction of markers of astrocytic (S100β) or neuronal (βIII-tubulin) differentiation (Fig. 6D,E; Supplemental Fig. S5F), suggesting that NFIX is sufficient and to some extent required for NS cells to exit the cell cycle.
NFIX is both required and sufficient to induce aspects of quiescence in NS cell cultures. (A) Efficiency of Nfix silencing in NS cells exposed to BMP to induce quiescence at the time of shRNA electroporation analyzed by qPCR 1, 2, and 3 d after shRNA transfection and BMP exposure. A scrambled shRNA was used in the control experiment, and expression of the gene ActB is analyzed for comparison. Note that Nfix transcript levels increase progressively between days 1 and 3 as cells enter quiescence in both control and Nfix knockdown experiments. (B) Analysis of proliferation by EdU immunostaining after 4 h of exposure in NS cell cultures following 1, 2, and 3 d of BMP exposure as indicated. Cells are counterstained with DAPI (blue). (C) Percentages of EdU-positive NS cells in Nfix shRNA transfected and control cultures. The BMP-induced cell cycle arrest is delayed by Nfix silencing. The progressive reduction in cell proliferation of Nfix shRNA-treated cultures between days 1 and 3 might be due to the progressive increase in Nfix expression during this period (shown in A). Error bars represent the standard deviation (n = 3 biological replicates). (D) Analysis of proliferation by EdU immunostaining in NS cell cultures transfected 18 h earlier with a Nfix expression construct and GFP or with GFP alone. (E) Percentages of EdU-positive cells in NS cell cultures transfected with GFP or GFP and Nfix. Nfix efficiently promotes cell cycle arrest. (F) Venn diagram showing the large fraction of genes regulated in quiescent NS cells that are also regulated by Nfix. GO analysis of Nfix-activated genes that are also induced in quiescent NS cells (G) and Nfix-repressed genes that are up-regulated in proliferating NS cells (H). (I) Representative examples of putative NFI direct target genes (associated with a NFI-bound enhancer/promoter and activated by Nfix) induced in quiescent NS cells and belonging to functionally important GO categories. (J) ChIP-seq signal for H3K27ac, p300, and NFI and RNA-seq signal (FPKM) for Svep1 and Bgn, two representative NFI direct target genes up-regulated in quiescent NSCs. Significant NFI binding within enhancer regions is indicated by pale blue rectangles. Peak height corresponds to SICER P-value for H3K27ac and MACs Q-value for p300 and NFI. See also Supplemental Figure S5 and Supplemental Table S2.
To further examine the cellular states induced by Nfix silencing and overexpression, we analyzed the transcriptome of NS cells expressing Nfix, Nfix shRNA, or control vectors with microarrays. Silencing Nfix in quiescent NS cells resulted in the up-regulation or down-regulation of 1677 genes, while its overexpression in proliferating cells resulted in the regulation of 2565 genes, and 628 genes were regulated in both experiments (i.e., 37% of the genes regulated by Nfix loss of function and 24% of the genes regulated by Nfix gain of function) (Supplemental Fig. S5G–I). Remarkably, 69% of the 3634 genes regulated by Nfix in either of the two conditions were part of the set of genes that is regulated when BMP-treated NS cells enter quiescence, and Nfix-regulated genes represented 48% of all of the quiescence-regulated genes (Figs. 1J, 6F). Among all Nfix-regulated genes, 1713 genes were activated by Nfix (i.e., down-regulated by Nfix knockdown and/or up-regulated by Nfix overexpression), and 44% of these were part of the gene expression program induced in quiescent NS cells (hypergeometric test, P < 2.2 × 1016), including genes involved in vasculature development, morphogenesis, cell adhesion, and ECM that were also overrepresented in the quiescence program (Figs. 1J,L, 6G; Supplemental Fig. S5H). Two-thousand-three-hundred -forty-nine genes were repressed by Nfix (i.e., up-regulated by Nfix knockdown and/or down-regulated by Nfix overexpression), and 38% of these were part of the program activated in proliferating NS cells (hypergeometric test, P < 2 × 1016), including, predictably, genes involved in the cell cycle, DNA metabolism, and protein translation (Fig. 6H; Supplemental Fig. S5I). Genes shown to be regulated by microarray were validated by qPCR (Supplemental Fig. S5J–M). A majority of Nfix-activated quiescence-specific genes (71%) were associated with NFI-binding events (Supplemental Fig. S5 H), suggesting that NFIX directly activates an important fraction of the genes in the quiescence program (534 genes corresponding to 20% of quiescence up-regulated genes). In contrast, NFI factors were only bound to a minority of Nfix-repressed proliferation-specific genes (43%), suggesting that its suppression of the proliferation program is more indirect (Supplemental Fig. S5 I).
To examine with an independent method the role of Nfix in activation of the quiescence gene expression program, we used a dominant-negative construct that interferes with the activity of all NFI family members (NFI-EnR or DN-NFI) (Supplemental Fig. S6A–C; Bachurski et al. 2003) Microarray analysis of NS cells electroporated with DN-NFI or a control vector and placed in BMP4-containing medium for 18 h, showed that 58% of 638 genes down-regulated by DN-NFI were part of the quiescence program induced by BMP4 in NSCs (hypergeometric test, P < 2.2 × 10−16). These 368 NFI-induced genes represented 19% of the quiescence program and were mostly involved in cell adhesion and the ECM. Conversely, 43% of 518 genes up-regulated by DN-NFI were part of the proliferation-specific program and were predominantly involved in protein translation (Supplemental. Fig. S6D–G).
Together, this functional analysis demonstrates that NFIX is both required and sufficient to activate a very significant portion of the gene expression program of NS cell quiescence and suppress an important part of the program of NS cell proliferation.
NFIX is required for NSC quiescence in the postnatal brain
We next asked whether our finding that NFIX regulates part of the quiescence program in cultured NS cells was predictive of a role for this factor in NSCs in vivo. We examined the expression of NFIX in the adult brain and found that it was expressed by NSCs in the two adult neurogenic regions (Fig. 7; data not shown). In the DG of the hippocampus, NFIX was expressed by Nestin+, GFAP+ radial NSCs, including both quiescent (MCM2-negative, 49%) and proliferating cells (MCM2-positive, 66.6%) (Fig. 7A–E). Thus, NFIX is expressed by a subset of quiescent NSCs in vivo and might play a similar role in regulating their gene expression programs, as we observed in cultured quiescent NS cells.
NFIX is required for NSC quiescence in the postnatal brain. (A–D) NFIX expression in NSCs in the postnatal day 90 (P90) mouse DG subgranular zone (SGZ). NSCs express GFAP and have a distinctive radial morphology. Proliferative NSCs express MCM2. (E) NFIX is expressed in 49% of quiescent and 66% of proliferative NSCs. Increased proliferation, reduced quiescence, and abnormal morphology and position of NSCs in the P20 Nfix−/− DG (H,I,K,N,O) compared with wild-type (WT) littermates' DG (F,G,J,L,M). (Q) Quantification of the number of NSCs in the wild-type and Nfix−/− DG reveals no difference in the density of NSCs. An increased proportion of Nfix−/− NSCs exhibit abnormal morphology (P) and position within the DG (T). (GL) Granule cell layer. Examples are indicated with orange arrows in H, I, K, and N. An increased number of Nfix−/− NSCs are in a proliferative state compared with wild-type littermates, as measured by expression of Ki67+ (R) and incorporation of BrdU (S). In A–D, yellow arrows highlight examples of proliferating NSCs, and blue arrowheads highlight examples of quiescent NSCs. In J, yellow arrowheads demonstrate the typical radial orientation of wild-type DG NSC processes, while K shows an Nfix−/− NSC with an abnormally oriented main process. Values in P–T are the mean plus standard deviation from counts in three mice. Asterisks in P–T indicate statistical significance of difference between wild type and Nfix mutants (t-test, P < 0.05; n = 3 independent mice). See also Supplemental Figure S6.
Nfix-null mutant mice present severe morphological defects in both the SEZ and DG at postnatal stages as well as defects in progenitor cell differentiation in the DG at birth (Campbell et al. 2008; Heng et al. 2012b). They die at 3 wk of age, which precludes an analysis of hippocampal NSCs by label retention, but postnatal hippocampal NSCs can also be identified by their radial morphology and the coexpression of Nestin and GFAP. Analysis of the DG of Nfix mutant mice soon before they die revealed a marked reduction in the number of GFAP+, Nestin+ NSCs with a typical radial morphology compared with wild-type control mice as well as a concomitant increase in the number of GFAP+, Nestin+ NSCs with a polarized but abnormal morphology, presumably due to a deleterious effect of loss of NFIX on NSC integrity (Fig. 7P). When grouping together GFAP+, Nestin+ NSCs with radial and abnormal morphologies, their number per area unit was not significantly different in Nfix mutant and wild-type control mice, indicating that NFIX is not essential for the maintenance of NSCs (Fig. 7Q). However, for both radial and abnormal GFAP+, Nestin+ NSCs, the fraction of cells that divide, as marked by Ki67 expression, was dramatically increased in Nfix mutants (28.6% ± 6%) compared with control mice (5.8% ± 0.6%; t-test, P = 0.02; n = 3) (Fig. 7F–J,R). We obtained similar results by quantifying proliferating NSCs with BrdU after a 2-h BrdU incorporation (12.0% ± 0.6% BrdU+ NSCs in Nfix mutant mice and 4.2% ± 1.3% in control mice; t-test, P = 0.006; n = 3) (Fig. 7L–O,S). Thus, Nfix is essential to maintain hippocampal NSCs in a quiescent state. Moreover, the distribution of proliferating NSCs in the Nfix mutant DG was highly abnormal, with a large fraction of these cells ectopically located in the granular layer of the DG (Fig. 7T). This suggests that Nfix mutant NSCs are not properly anchored to the subgranular layer of the DG, and therefore, similar to NFI factors in cultured NSCs, NFIX regulates the cell adhesion properties of hippocampal NSCs, which in turn might directly influence the quiescent status of NSCs.
Discussion
Our study illustrates how modeling stem cell physiology in culture can be used in combination with epigenomic profiling to identify major TFs regulating stem cell states. Here, we characterized and validated a cell culture model of NSC quiescence. We used this model to demonstrate that proteins of the NFI family bind to a large fraction of enhancer elements active in quiescent NS cells and that NFIX regulates a gene expression program controlling multiple aspects of the quiescent NSC phenotype and in particular their cell adhesion properties. We discuss below our findings on the quiescent state in NSCs and the role and mode of action of NFIX in these cells.
An in vitro model of NSC quiescence
In contrast to embryonic NSCs that are highly proliferative, NSCs in neurogenic regions of the postnatal and adult brain are relatively quiescent, with the vast majority resting in G0, and only a small fraction actively progressing through the cell cycle (Temple 2001; Fuentealba et al. 2012). To understand how adult NSCs select between proliferative and quiescent states and how this selection is biased by a variety of physiological and pathological stimuli, it is necessary to characterize the transcriptional mechanisms that control the quiescent state in these cells. However, NSCs are intermingled with other cell types in the neurogenic niches of the brain and are difficult to purify in significant numbers, thus precluding a systematic study of the mechanisms regulating gene expression in NSCs in vivo (Pastrana et al. 2009; Beckervordersandforth et al. 2010). Homogenous NS cells have been established from embryonic stem cells and embryonic and adult brain tissue (Conti and Cattaneo 2010). NS cells are highly proliferative and therefore have been used to investigate mechanisms controlling cell proliferation, fate specification, and differentiation (Conti and Cattaneo 2010; Castro et al. 2011). In contrast, the quiescent state of NS cells and the GRN inducing and maintaining this cellular state have been poorly studied.
Our extensive characterization of embryonic stem cell-derived NS cells cultured in the presence of BMP4 and FGF2 has shown that these cells have hallmarks of quiescent stem cells. Different types of quiescent cells share few characteristic properties beyond reversible cell cycle arrest, but gene expression profiles can be used as a representation of the unique physiology of cellular quiescence (Coller et al. 2006). Analysis of the gene expression profile of cell cycle-arrested NS cells showed that these cells are very significantly enriched for genes induced by fibroblasts and different types of adult tissue stem cells when they enter quiescence, thus suggesting that different cell types, including NS cells, employ overlapping quiescence gene expression programs. In contrast, these cells were not significantly similar to parenchymal astrocytes, in agreement with a recent report showing that while BMP signaling alone promotes terminal astrocytic differentiation, exposure to both BMP and FGF2 maintains the stem cell character of NS cells (Sun et al. 2011).
Beside the down-regulation of genes related to the cell cycle and protein translation, entry of NS cells into quiescence is accompanied by the up-regulation of many genes encoding ECM proteins, receptors for ECM molecules, and cell–cell adhesion molecules. This suggests that quiescence involves a profound change in the attachment of NSCs to the ECM and neighboring cells, as previously suggested for NSCs in the adult SEZ (Kazanis et al. 2010; Kokovay et al. 2010, 2012) and for other types of adult stem cells (Venezia et al. 2004; Fukada et al. 2007; Pallafacchina et al. 2010; Brohl et al. 2012). For example, entry of hematopoietic stem cells into quiescence involves homing to their cellular niche, which is mediated by integrins and the transmembrane glycoprotein endoglin (which are significantly induced in quiescent NS cells) (Supplemental Table S1; Venezia et al. 2004). Similarly, it has been proposed recently that NSCs in the adult SEZ move from an ependymal niche to a vascular niche as they become activated (Kokovay et al. 2010). Interestingly, α6 integrin, a factor that we found expressed by proliferating NSCs and down-regulated when cells enter quiescence (Supplemental Table S1), is required for the binding of NSCs to endothelial cells in the SEZ (Shen et al. 2008). Expression of different repertoires of adhesion molecules, ECM proteins, and ECM receptors by quiescent and activated NSCs is therefore likely to promote or facilitate their interactions with different niche cells and hence play an important role in their exposure to different signaling environments as well as influence how these cells respond to such signals (Kerever et al. 2007; Riquelme et al. 2008; Hynes 2009).
BMP signaling and quiescence
BMP signaling promotes quiescence in not only NS cell cultures (Mira et al. 2010; Sun et al. 2011; this study), but also the adult hippocampus, where perturbation of BMP signaling results in excessive proliferation and eventual depletion of hippocampal stem cells (Mira et al. 2010). BMPs have also been implicated in the quiescent state of other types of adult stem cells, including hair follicle, intestinal, and hematopoietic stem cells (Kobielak et al. 2007; Li and Clevers 2010; Lien et al. 2011). Surprisingly, Smads, the main transcriptional effectors of BMPs, do not seem to play a major role in regulating gene expression in BMP-treated quiescent NS cells, since the consensus Smad-binding motif is not significantly overrepresented in enhancers active in quiescent NS cells (data not shown). Genes of the inhibitor of differentiation (Id) family are major targets of BMP signaling in many tissues, including the embryonic nervous system (Nakashima et al. 2001; Samanta and Kessler 2004; Vinals et al. 2004). The four family members, and particularly Id1 and Id4, are highly up-regulated in quiescent NS cells and may therefore contribute significantly to the quiescence-inducing activity of BMP. Id proteins inhibit the activity of bHLH TFs by disrupting dimerization with their E protein partners and preventing their binding to DNA (Massari and Murre 2000). Id proteins may contribute to the quiescent state by antagonizing bHLH factors that promote the proliferation of neural progenitors, including Olig2 (Ligon et al. 2007) and Ascl1 (Castro et al. 2011). This idea is supported by the finding that consensus binding sites for bHLH proteins are much more prevalent in enhancers active in proliferating NSCs than in enhancers active in quiescent NSCs. However, induction of Id genes is unlikely to be the main mechanism by which BMP4 drives NS cell quiescence, since overexpression of Id1 in NS cells is sufficient to arrest their divisions but not to induce the broad changes in gene expression observed in BMP-treated quiescent NS cells (B Martynoga, unpubl.). Hes1 is another gene induced by BMP signaling in both the embryonic brain and cultured NS cells (Nakashima et al. 2001; this study) and inhibits progenitor cell proliferation (Baek et al. 2006; Yu et al. 2006). Moreover, Hes1 is essential for the maintenance of fibroblast quiescence (Sang et al. 2008). However, Hes1-binding motifs are not strongly overrepresented in NS cell enhancers (data not shown), arguing against an important role of Hes factors in these cells. Our analysis shows instead that NFIX has a major role in the GRN operating in quiescent NS cells, as discussed below. How BMP signaling regulates NFI proteins is currently not known, and further investigations are required to address this important question.
Widespread enhancer binding of NFI factors in quiescent NS cells
To identify TFs with important roles in the regulation of NS cell quiescence, we first annotated enhancer elements that are active in these cells, defined as genomic regions that recruit the coactivator p300 and harbor H3K27ac (Heintzman et al. 2009; Creyghton et al. 2010; Rada-Iglesias and Wysocka 2011; Rada-Iglesias et al. 2012). We found a larger number of enhancers in quiescent NSCs (16,810) than in proliferating NSCs (10,270), in keeping with the greater number of genes that are up-regulated in quiescent NS cells (2475) than in proliferating NS cells (1980) and with the fact that quiescence is an active state that involves both large-scale induction and suppression of gene expression (Coller et al. 2006).
We interrogated quiescent NS cell enhancers for enriched DNA sequence motifs in order to predict TFs that regulate the quiescent state (Rada-Iglesias et al. 2012). Finding the NFI-binding motif as the most overrepresented in these enhancers was unexpected, since the NFI gene family previously had no known function in NSC biology. Location analysis demonstrated that NFI factors are indeed bound to a remarkable 73% of enhancers in the quiescent NSC genome. Reciprocally, half of NFI-binding sites are located within epigenomically defined enhancers. In contrast, a recent study showed that despite strong enrichment of its motif, only 10% of TFAP2A TF ChIP-seq peaks mapped within enhancers in cultured human neural crest cells, and just 30% of all active enhancers in these cells were TFAP2A-bound (Rada-Iglesias et al. 2012). This, together with the strong correlation observed between NFI- and p300-binding strengths, argues for a central role for NFI factors in regulating gene expression in quiescent NS cells.
NFIX targets cell adhesion and ECM genes to promote NS cell quiescence
Three of the four NFI genes—Nfia, Nfib, and Nfix—are widely expressed in the developing nervous system, including in progenitors, post-mitotic neurons, and glial cells (Mason et al. 2009; Heng et al. 2012a). These genes have been implicated in multiple aspects of neural development, including the specification, differentiation, and migration of both astrocytes and neurons (Shu et al. 2003; Deneen et al. 2006; Campbell et al. 2008). The four family members are expressed in cultured NS cells; however, NFIX is the only member whose expression is sharply up-regulated at the transition from proliferation to quiescence and is the most abundant in quiescent NS cells. We therefore focused our functional analysis on Nfix by overexpressing or silencing the gene in NS cells and examining Nfix mutant brains. Our results show that Nfix does indeed play an essential role in the regulation of the quiescent state of NS cells.
A prominent change in the transcriptome of NS cells when Nfix was induced or silenced was the regulation of a large number of ECM and cell adhesion molecules. The mislocalization of NSCs in Nfix mutant hippocampus suggests that Nfix controls the cell adhesion properties of NSCs in vivo. As already discussed, ECM and cell adhesion molecules constitute a large fraction of the gene expression programs of different types of quiescent cells, and adhesion of stem cells to their niche is thought to be important for maintenance of the quiescent state, particularly for hematopoietic stem cells (Hurley et al. 1995; Scott et al. 2003; Venezia et al. 2004). Since our results show that Nfix directly regulates only a few cell cycle arrest genes, the primary cause of the loss of quiescence and excess proliferation of NSCs in Nfix mutant hippocampus might be the loss of cell adhesion and disruption of interactions with their niche.
Previous analyses of Nfi mutant mice have shown that Nfi family members have divergent roles in brain development (Mason et al. 2009). Nfix mutants present an overexpansion of the embryonic brain and a delay of hippocampal progenitor differentiation that are not seen in other Nfi mutants (Driller et al. 2007; Campbell et al. 2008; Heng et al. 2012b). Moreover, manipulation of Nfia expression alongside that of Nfix in NS cells shows that the two genes have different activities in the cells, and Nfia does not have a prominent role in the regulation of quiescence (B Martynoga, unpubl.). Given the close structural similarity between NFI factors, it will be interesting to elucidate how these factors exert their divergent functions. Different NFI proteins might regulate different target genes by recognizing subtly different DNA motifs or may target the same genes but regulate them differently. Nfi genes also have distinct expression patterns, and it will be important to characterize the pathways that regulate their expression; e.g., the up-regulation of Nfix and the down-regulation of Nfia and Nfib in quiescent NS cells.
Together, our study establishes a platform to understand how the signaling environment of the niche influences NSC physiology and decipher the regulatory networks that control the different NSC states.
Materials and methodsNS cells
NS5 cells were cultured according to standard methods (Conti et al. 2005) with the following minor modification: Cells were plated onto uncoated tissue culture plastic with the addition of 2 μg/mL laminin (Sigma) to the medium. To induce quiescence, 35,000–65,000 cells per square centimeter were plated into normal proliferation medium (EGF and FGF2, both at 10 ng/mL; Peprotech), and, after 16 h, fresh NSC medium was added without EGF and with 50 ng/mL BMP4 (R&D Systems) and 20 ng/mL FGF2. For reactivation, after at least 3 d in BMP4-containing medium, cells were passaged with Accutase (Sigma) and plated into proliferation medium at a density of 35,000–65,000 cells per square centimeter. Details of immunostaining, cell cycle analysis by FACS, qPCR, and induction of neurogenesis are described in the Supplemental Material.
shRNA knockdown of Nfix
Lentiviral particles encoding a control or Nfix-specific shRNA construct and a puromycin resistance cassette (Messina et al. 2010) were generated in 293T cells according to standard procedures. Proliferating NS cells were transduced with the lentiviruses, and, 24 h later, 2 μg/mL puromycin was added to select for shRNA-expressing cells. After a further 48 h, NS cells were assessed for proliferation or plated into quiescence medium for proliferation, RNA, and FACS analysis.
Transfection and cell sorting
Cells were electroporated with NFIX (isoform X2), NFIA (isoform A1), NFIC (isoform C2), or DN-NFI (NFI-enR) (Bachurski et al. 2003) cloned upstream of an IRES and NLS-tagged GFP under the control of the CAGGs promoter or GFP control, both in the pCAGGS expression vector with AAD-1011 nucleofector (Amaxa). NFIX transfected cells were plated into proliferation medium, trypsinized, and FACS-sorted 18 h later. DN-NFI transfected cells were plated into prewarmed proliferation medium for 8 h, and then medium was replaced with quiescence medium. After 24 h, cells were trypsinized and FACS-sorted for GFP expression directly into Trizol LS (Invitrogen) for RNA extraction and downstream analysis.
ChIP
NS cells were fixed sequentially with di(N-succimidyl) glutarate and 1% formaldehyde in phosphate-buffered saline and then lysed, sonicated, and immunoprecipitated as described previously (Castro et al. 2011) using material from ∼5 × 106 cells per sample. Immunoprecipitations were with rabbit anti-H3K27ac (4 μg per ChIP sample; Abcam, ab4729), rabbit anti-p300 (3 μg per ChIP sample; Santa Cruz Biotechnology, sc-585), or goat anti-NFI (6 μg per ChIP sample; Santa Cruz Biotechnology, sc-30918).
ChIP-seq data generation and processing
DNA libraries were prepared from 10 ng of immunoprecipitated DNA according to the standard Illumina ChIP-seq protocol for quiescent NS cell H3K27ac ChIP, quiescent NS cell p300 ChIP, quiescent NS cell NFI ChIP, quiescent NS cell input DNA, proliferating NS cell H3K27ac ChIP, proliferating NS cell p300 ChIP, and proliferating NS cell input DNA. Libraries were sequenced with the Genome Analyzer IIx (Illumina). The raw reads for p300, H3K27ac, and NFI in quiescent NS cells and p300 and H3K27ac in proliferating NS cells were mapped to the mouse genome (mm9, including random chromosomes) with Bowtie version 0.12.5 (Langmead 2010). For each cell condition, an input chromatin sample was mapped in the same way. The number of uniquely mapped reads in quiescent NS cells was 28.0 million for p300, 38.0 million for H3K27ac, 24.9 million for NFI, and 27.3 million for input. In proliferating NS cells, 18.8 million p300, 33.5 million H3K27ac, and 31.0 million input unique reads were mapped.
H3K27ac data sets were processed further with SICER version 1.1 (Zang et al. 2009) to define islands of enrichment, and we used MACS version 2.0.9 (Zhang et al. 2008) to define bound regions for p300 and NFI. Further details of data processing are described in the Supplemental Material.
ChIP-seq data generated in this study have been deposited in the European Nucleotide Archive's Sequence Read Archive under accession number ERP002084 and are also available via ArrayExpress under accession number E-MTAB-1423.
The ChIP-seq data sets for H3K4me1, H3K4me2, H3K4me3, H3K27me3, H3K9me3, and H3K36me3 in neural progenitors were retrieved from Gene Expression Omnibus repository with accession numbers GSE8024 and GSE11172.
Definition of active enhancers
In order to define active enhancer regions in quiescent and proliferating NS cells, we used the p300 peaks as reference, selecting peaks whose summit is included within a H3K27ac island. We removed promoter-proximal peaks whose summit is closer than 2 kb to the TSS of any gene in ENSEMBL version 61 annotated as known protein-coding. We used the Q-value reported by MACS for each p300 peak as the enhancer score. We divided these enhancer sets into quiescence-specific, proliferation-specific, or pan-NS cells by considering the H3K27ac and p300 signal in both cell states. For H3K27ac, we used SICER to call differentially enriched regions. Quiescence-specific enhancers are defined as p300 peaks that are only called in quiescent NS cells or that fall within an H3K27ac island that is specifically enriched in quiescence and vice versa for proliferation-specific enhancers. Pan-NS cell enhancers are defined by the consistent presence of p300 peaks in the two cell states that fall within a nondifferentially enriched H3K27ac island.
Motif analysis
To identify motifs overrepresented in the active enhancer regions, we used three tools that are based on different approaches: MEME-ChIP (Machanick and Bailey 2011), GADEM (Li 2009), and RSAT peak motifs (Thomas-Chollier et al. 2012). Parameters and further motif analysis are described in the Supplemental Material.
Generation and analysis of microarray and RNA-seq data
For microarray analysis, RNA from three biological replicates per conditions was prepared and hybridized to Illumina Mouseref-8 version 2.0 bead chips according to the manufacturer's specifications. Normalization and statistical analysis were carried out with GeneSpring software (Agilent). Probes were considered deregulated if there was ≥1.5-fold differential expression with a Benjamini-Hochberg-corrected P-value < 0.05 (t-test). Generation of RNA-seq data will be described elsewhere (S Hadjur and D Georgopoulou, in prep.). We obtained a total of 48.3 million pairs of 75-base-pair (bp) paired-end reads for quiescent NSCs and 98.8 million pairs for proliferating NSCs and processed with TopHat and the Cufflinks package (Trapnell et al. 2012).
Functional classification of gene lists and TF-bound genomic regions
GO analysis was conducted with DAVID using functional annotation clustering. Representative terms from the top-ranking clusters of GO terms are reported, all with P < 0.05 and false discovery rate (FDR) < 5%. GSEA was carried out with 1000 permutations. Functional classification of genes associated with enhancers was conducted with GREAT using default settings.
Mice and immunohistochemistry
To characterize expression patterns of NFI factors in DG NSCs, we used wild-type MF1 mice. To test the function of Nfix in postnatal neurogenesis, we used mice carrying an allele of Nfix that lacks exon 2 (Campbell et al. 2008). We demonstrated previously that these mice do not produce NFIX protein (Campbell et al. 2008). For BrdU analysis, BrdU was administered intraperitoneally 2 h prior to sacrifice. Postnatal pups were transcardially perfused with 0.9% saline followed by 4% paraformaldehyde (PFA) and then post-fixed in 4% PFA at 4°C, and 40-μm vibratome sections were cut for immunohistochemistry. Immunohistochemistry was performed on free-floating sections in 10% normal donkey serum and 0.1% Triton-X100 according to standard protocols. Antibodies used are described in the Supplemental Material. All work with laboratory mice was conducted according to the relevant national and international guidelines and regulations.
Quantification of NSCs in postnatal DG
For each image counted, seven to 10 1-μm confocal Z-stacks were merged for quantification. For all counts, images from at least three sections from two or three independent mice were quantified. For quantification of the number of NSCs in the wild-type DG expressing NFI factors and/or MCM2 or BrdU, only NSCs with a clear radial GFAP+ process that could be confidently linked to a nucleus within the subgranular zone (SGZ) were considered. To quantify NSCs in the Nfix−/− DG, counts were made in the proximal blade of the DG, where gross morphology was more similar to wild-type DG. To estimate total density of NSCs, the number of GFAP+, Nestin+ processes was counted. Cells were deemed to reside in the granule cell layer (GL) if their main process terminated more than two nucleus widths from the bottom of the SGZ. Morphology of NSCs was considered abnormal if the angle of their primary process deviated by >30° from perpendicular to the SGZ surface or they exhibited more than one main process. For assessment of proliferation, only NSC processes that could be clearly associated with a DAPI-positive nucleus were considered.
Acknowledgments
We are grateful to Abdul Sesay and Harsha Jani for expert technical assistance with the high-throughput sequencing and DNA microarray experiments, respectively. We thank members of the National Institute for Medical Research (NIMR) FACS facility for their expertise in cell sorting, and Graziella Messina for providing shRNA constructs. We thank members of the Guillemot laboratory for suggestions and comments on the manuscript. This work was supported by a Small Collaborative Project Grant from the 7th Framework Programme of the European Commission (FP7-223210 to L.E., J.W., and F.G.) and a Grant-in-Aid from the Medical Research Council (U117570528 to F.G.).
Supplemental material is available for this article.
Article is online at http://www.genesdev.org/cgi/doi/10.1101/gad.216804.113.
Senescence is a stable proliferation arrest implicated in tumor suppression and aging. While recent studies have implicated changes in chromatin and lamin B1 regulation in senescence, functional interactions between them are poorly understood. Here, genome-wide mapping of histone modifications in proliferating and senescent primary human lung fibroblasts reveal profound changes in the senescent chromatin landscape. The data suggest that lamin B1 down-regulation in senescence is a key trigger of chromatin changes that impact gene expression, aging, and cancer.
Senescence is a stable proliferation arrest, associated with an altered secretory pathway, thought to promote tumor suppression and tissue aging. While chromatin regulation and lamin B1 down-regulation have been implicated as senescence effectors, functional interactions between them are poorly understood. We compared genome-wide Lys4 trimethylation on histone H3 (H3K4me3) and H3K27me3 distributions between proliferating and senescent human cells and found dramatic differences in senescence, including large-scale domains of H3K4me3- and H3K27me3-enriched “mesas” and H3K27me3-depleted “canyons.” Mesas form at lamin B1-associated domains (LADs) in replicative senescence and oncogene-induced senescence and overlap DNA hypomethylation regions in cancer, suggesting that pre-malignant senescent chromatin changes foreshadow epigenetic cancer changes. Hutchinson-Gilford progeria syndrome fibroblasts (mutant lamin A) also show evidence of H3K4me3 mesas, suggesting a link between premature chromatin changes and accelerated cell senescence. Canyons mostly form between LADs and are enriched in genes and enhancers. H3K27me3 loss is correlated with up-regulation of key senescence genes, indicating a link between global chromatin changes and local gene expression regulation. Lamin B1 reduction in proliferating cells triggers senescence and formation of mesas and canyons. Our data illustrate profound chromatin reorganization during senescence and suggest that lamin B1 down-regulation in senescence is a key trigger of global and local chromatin changes that impact gene expression, aging, and cancer.
senescencelamin B1chromatingene expression
Cellular senescence is a stable proliferation arrest implicated in aging and tumor suppression and can be induced by telomere attrition, activation of oncogenes, and other cellular stresses (Campisi 2005; Adams 2009). Senescent cells are further characterized by altered physical morphology and transcription, including up-regulation of genes involved in cellular defense and the inflammatory response, many of which are secreted and are a part of the senescence-associated secretory phenotype (SASP) (Acosta et al. 2008; Coppe et al. 2008; Kuilman et al. 2008). The SASP is thought to promote immune clearance of pre-malignant senescent cells to facilitate tumor suppression (Xue et al. 2007; Krizhanovsky et al. 2008; Kang et al. 2011; Sagiv et al. 2013). However, a chronic SASP is also suggested to promote tissue aging, and, accordingly, clearing of senescent cells in mice delays the onset of age-related pathology (Krtolica et al. 2001; Baker et al. 2011). Recent studies have also described changes in chromatin during cell senescence. Many mammalian cell types develop regions of condensed chromatin, called senescence-associated heterochromatin foci (SAHFs), suggestive of large-scale changes in senescent chromatin (Funayama et al. 2006; Narita et al. 2006; Zhang et al. 2007). Also, a deficiency in the Polycomb-repressive protein EZH2 and subsequent decrease of repression-associated Lys27 trimethylation on histone H3 (H3K27me3) leads to rapid senescence in primary human cells, in part through up-regulation of p16INK4a (hereafter “p16”) (Bracken et al. 2007). In addition, histone loss leads to premature aging of yeast, and overexpression of histones results in life span extension (Dang et al. 2009; Feser et al. 2010), suggesting that the impact of chromatin on cell senescence extends to an impact on cell and organismal aging.
Longevity is also associated with alterations in specific histone post-translational modifications. In yeast, Lys16 acetylation on histone H4 (H4K16ac) increases with age at heterochromatic regions, including telomeres, due to reduction of histone deacetylase Sir2 protein levels (Dang et al. 2009), providing a potential mechanism for the observation that Sir2 overexpression extends yeast life span (Kaeberlein et al. 1999). Likewise, in mice, the histone deacetylase SIRT6 regulates senescence and longevity through Lys9 deacetylation on histone H3 (H3K9ac) (Michishita et al. 2008; Kawahara et al. 2009), and premature aging is linked to altered chromatin at telomeres in both yeast (Dang et al. 2009) and mice (Michishita et al. 2008; Dang et al. 2009). Observations in Caenorhabditis elegans link the loss of the Trithorax-mediated active transcription histone modification H3K4me3 and gain of repressed transcription modification H3K27me3 to extended longevity through an effect that may be inherited transgenerationally (Greer et al. 2010, 2011). Alterations in heterochromatin factors have also been described in prematurely aging cells from Hutchinson-Gilford progeria syndrome (HGPS) patients; namely, decreased levels of heterochromatin protein 1 (HP1), H3K9me3, and H3K27me3 and increased levels of H4K20me3 (Scaffidi and Misteli 2005; Shumaker et al. 2006; Taimen et al. 2009; McCord et al. 2013).
Results
These studies highlight a relationship between chromatin regulation in cell senescence, cancer, and aging; however, there is limited understanding of specific chromatin changes that occur on a genome-wide scale. Here we report genome-wide chromatin changes during senescence in IMR90 primary human lung fibroblasts. The cells were serially passaged in culture at physiological oxygen (3%) until replicative senescence and maintained in culture in a senescent state for 2 wk prior to analysis (Supplemental Fig. 1A). As expected, the early passage cells (population doubling [PD] 24; hereafter “proliferating cells”) exhibit hallmarks of proliferation, including few senescence-associated β-galactosidase (SA-β-gal)-positive cells and low levels of p16 (Supplemental Fig. 1B–D); comparatively, late passage senescent cells (PD87; hereafter “senescent cells”) show nearly 100% SA-β-gal-positive cells, up-regulated p16 levels (Supplemental Fig. 1B–D), and shortened telomeres (data not shown).
To survey chromatin changes that occur during senescence, we performed chromatin immunoprecipitation (ChIP) followed by genome-wide parallel sequencing (ChIP-seq) for total histone H3 and two H3 modifications—H3K4me3 and H3K27me3—in proliferating cells and senescent cells. Trithorax-mediated H3K4me3 is canonically associated with promoters of transcriptionally active genes (Barski et al. 2007; Guenther et al. 2007; Shilatifard 2012), whereas Polycomb-mediated H3K27me3 is associated with facultative heterochromatin (Lee et al. 2006a; Schwartz et al. 2006; Barski et al. 2007; Schuettengruber et al. 2009). We also performed a transcriptome analysis using microarrays, assessing RNA levels at 33,288 RefSeq transcripts from the same cell samples used for ChIP (Supplemental Text 1; Supplemental Fig. 2; Supplemental Tables 1, 2). Our microarray data largely agree with other previously published data sets (Shelton et al. 1999; Zhang et al. 2003) and were further validated by quantitative RT–PCR (qRT–PCR) of >50 randomly selected genes that show altered expression, including known down-regulated cell cycle genes and up-regulated SASP genes (e.g., Supplemental Fig. 2B,C). Hence, by several independent assays, the proliferating and senescent cells show expected patterns of physiology and gene expression.
We mapped ChIP-seq data for the histone modifications to the human genome, quantified binding enrichment by normalization to total histone H3, and subsequently assessed each resulting enrichment map for regions of significant binding. We validated these maps by performing qPCR across >100 genomic loci; indeed, qPCR strongly correlated with ChIP-seq results (R = 0.83) (e.g., Supplemental Fig. 3). It is important to note that while total histone H3 decreases significantly during senescence as measured by Western blot (Supplemental Fig. 4A, lysates normalized by cell number; O'Sullivan et al. 2010), the relative levels of H3K4me3 and H3K27me3 (normalized to histone H3) do not significantly change between proliferating and senescent cells (senescent sample concentrated ∼13-fold for equivalent loading of H3 level) (Supplemental Fig. 4B). Furthermore, ChIP-seq and ChIP-qPCR data were normalized to total histone H3 ChIP, which accounted for any regional differences in histone occupancy that could affect modification levels, thereby providing a platform to specifically identify regions of differential histone modifications (see Supplemental Fig. 4C for track views of total H3 and the modifications).
Both modifications (normalized to total H3) show altered patterns genome-wide in senescence (Supplemental Fig. 5). By visual inspection, both histone modifications appear to be changed in large domains in senescence. Notably, H3K4me3 is surprisingly enriched across the genome in extremely large domains, often hundreds of kilobases. We developed a new algorithm to identify large, differentially enriched H3K4me3 regions in senescent cells (H3K4me3-enriched mesas; hereafter “K4me3 mesas”) (Fig. 1A, H3K4me3 shown in the top track, proliferating tracks in orange and senescent tracks in blue; see the Supplemental Material for detailed analysis description). We identified 648 mesas spanning 50 kb (minimal size) to several hundred kilobases in length, occupying ∼17% of the senescent genome (Fig. 1A, green bars representing computationally defined H3K4me3 mesas). Unlike the canonical, sharp, highly enriched peaks of H3K4me3 observed over gene promoters (Roh et al. 2006), H3K4me3 enrichment in mesas is broad, covering the entire region with few, if any, peaks of enrichment that stand out above the general elevated signal.
Large-scale chromatin changes occur in the senescence genome for both histone modifications. (A) Sample tracks of H3K4me3 and H3K27me3 ChIP-seq data over a 3-Mb region of chromosome 1 (Chr1: 38,016,703–41,116,920) show major regions of gains and losses for both histone modifications. Proliferating tracks are shown in orange, and senescence tracks are shown in blue. K4me3 mesas are shown in green blocks, the top 1% gain regions and K27me3 mesas are shown in blue blocks, and the top 1% loss regions and K27me3 canyons are shown in red blocks. (B) Example of overlapped K4me3 and K27me3 mesas over a 750-kb region of chromosome 7 (Chr7: 35,060,526–35,814,473). Proliferating tracks are shown in orange, and senescence tracks are shown in blue. (C,D) ChIP-qPCR validation of the K4me3 (C) and K27me3 (D) mesas shown in B. Proliferating data are shown in orange, and senescence data are shown in blue. ChIP-qPCR primers are tiled across the entire region of the mesa and include 3′ and 5′ flanking primers. ChIP-qPCR data are shown as ratios of modification to total histone H3. ChIP-qPCR data are the average of three biological replicates, and error bars represent standard deviation from the mean. (E) Example of K27me3 canyon over a 1.3-Mb region of chromosome 6 (Chr6: 36,239,850–37,570,928). Proliferating tracks are shown in orange, and senescence tracks are shown in blue. (F) ChIP-qPCR validation of the K27me3 canyon shown in E. Proliferating data are shown in orange, and senescence data are shown in blue. ChIP-qPCR primers are tiled across the entire region of the canyon and include 3′ and 5′ flanking primers. ChIP-qPCR data are shown as ratios of modification to total histone H3. ChIP-qPCR data are the average of three biological replicates, and error bars represent the standard deviation from the mean.
The ChIP-seq analysis also showed large-scale changes in the senescence-associated H3K27me3 profile, notably both regions of gain and loss, with varying sizes. H3-normalized H3K27me3 sequence reads were first analyzed in a sliding window analysis to define regions of significant gain or loss. The top 1% gain and loss regions were then “stitched” together to define contiguous regions of gain or loss across the genome for H3K27me3 (see the Supplemental Material for detailed analysis description). Using this method, we computationally defined 1440 H3K27me3-enriched mesas (hereafter “K27me3 mesas”) across the senescence genome spanning 4.3 kb to 27.5 Mb (average size of 394 kb), which showed striking overlap with the previously defined K4me3 mesas (Fig. 1A, blue bars representing computationally defined K27me3 mesas).
Analysis also revealed large domain losses of H3K27me3 (H3K27me3-depleted canyons; hereafter “K27me3 canyons”) (Fig. 1A, red bars representing computationally defined canyons). We defined 1374 K27me3 canyons across the senescent genome spanning 7 kb to 19 Mb (average size of 730.9 kb). Interestingly, the canyon phenomenon was limited to H3K27me3, as H3K4me3 did not show any significant loss across broad domains.
To validate the discovery methods, we performed ChIP-qPCR in triplicate (using three independently senesced sets of IMR90 cells) at varying intervals across and outside (upstream and downstream) multiple mesas and canyons from different chromosomes (Fig. 1B [example of overlapped K4me3 mesa and K27me3 mesa is validated by ChIP-qPCR in C,D], E [example K27me3 canyon is validated by ChIP-qPCR in F]). We also discovered mesa and canyon trends by ChIP-qPCR analysis in a different fibroblast strain (senescent MRC5), suggesting that these large-scale chromatin changes occur in other senescent cells (Supplemental Fig. 6A–C). We further confirmed K27me3 canyons by using a second, distinct antibody against H3K27me3 (Supplemental Fig. 6D). Moreover, we validated the K27me3 mesas and canyons by generating additional H3 and H3K27me3 ChIP-seq maps using three biological replicates of proliferating and senescent IMR90s (Supplemental Fig. 7). These repeated ChIP-seq data confirmed our observation that K27me3 mesas and canyons form across the senescent genome (Supplemental Fig. 7). Finally, we also observed K27me3 mesa and canyon trends in another published data set (Chandra et al. 2012), analyzing H3K27me3 enrichment or loss in that data set using our defined mesa and canyon regions, albeit to a lesser degree than in our study (Supplemental Fig. 7, far right boxes).
Thus, we defined three major domains of chromatin change in senescence: K4me3 mesas, K27me3 mesas, and K27me3 canyons. Interestingly, both types of mesas often overlap, whereas K27me3 canyons typically occupy different regions of the genome (Fig. 2A, K4me3 mesas in green, K27me3 mesas in blue, and K27me3 canyons in red). By nucleotide measure, ∼38% of K27me3 mesas overlap with K4me3 mesas, and 42% of K4me3 mesas overlap with K27me3 mesas, showing a high degree of overlap (Fig. 2B, blue and green circles). In contrast, by nucleotide measure, only 15% of K4me3 mesas overlap with K27me3 canyons, and only 7% of K27me3 canyons overlap with K4me3 mesas, highlighting the separation of canyons from the mesas (Fig. 2B, green and red circles). The observation of overlap between mesas, but not with canyons, is also reflected when measuring the overlap between features at an 80% cutoff parameter (Supplemental Fig. 8). Thus, the computational measure underscores the visual overlaps and differences observed in the track views (Fig. 2A), highlighting the greater concordance of the two types of mesas compared with mesas and canyons, which appear to occupy different regions of the genome.
Mesas are overlapped features that are enriched for LADs; canyons are outside of LADs and enriched for enhancers. (A, top) Representation of all K4me3 (green blocks) and K27me3 (blue blocks) mesas and K27me3 canyons (red blocks) across the whole of chromosome 9. LADs are shown in gold blocks, enhancers are shown in black blocks, and genes are shown in blue blocks at the bottom of the image. Whole-chromosome views show the enrichment for mesas within LADs and canyon formation adjacent to LADs or in general LAD-poor regions. Moreover, canyons appear to be enriched for enhancers, which are also located outside of LADs and in gene-rich regions. (Bottom) Closer view of a section of a 9.2-Mb region of chromosome 9 (Chr9: 118,000,000–127,200,000) highlights the relationship of mesas with LADs and canyons with enhancer regions. (B) Venn diagram representation of the nucleotide overlap between K4me3 mesas, K27me3 mesas, and K27me3 canyons shows a high degree of overlap between the two mesas but little overlap between canyons and mesas, highlighting the difference between the large-scale changes that are visually depicted in A. (C) Pie chart representations of the percent of nucleotides in K4me3 mesas (top left panel in green), K27me3 mesas (top right panel in blue), and K27me3 canyons (bottom panel in red) that overlap with LADs. Most nucleotides in mesas overlap with LADs (90% for K4me3 mesas and 92% for K27me3 mesas), but only 41% of canyon nucleotides overlap with LADs, highlighting the difference between the large-scale changes that are visually depicted in A. (D) Column plot representation of the overlap between K4me3 mesas, K27me3 mesas, and K27me3 canyons with two categories of enhancers (H3K4me1 shown in red and H3K4me1+H3K27ac shown in blue) highlights the enrichment of enhancers in K27me3 canyons over background measure. Mesas, as expected, are not enriched for enhancers. (E) Example track view of the MMP cluster on chromosome 11 that shows H3K27me3 loss over genes and enhancers. The bottom view is a closeup of the MMP3 and MMP12 genes (up-regulated in senescence; Chr11: 102,017,147–102,489,087) with a cluster of enhancers (circled in red) nearby that are also contained within the K27me3 canyon.
The large-scale and distinct nature of mesas and canyons strongly indicates a profound change in global chromatin organization during senescence. DNA–lamin interactions at the nuclear membrane are a foundation for chromatin organization, and DNA at the nuclear periphery is strongly associated with heterochromatin and transcriptional silencing (Kourmouli et al. 2000; Mattout-Drubezki and Gruenbaum 2003; Shaklai et al. 2007; Finlan et al. 2008; Reddy et al. 2008; Towbin et al. 2010). Given the key role of lamins in controlling chromatin organization (Peric-Hupkes et al. 2010) and down-regulation of lamins in senescent cells (Shimi et al. 2011; Freund et al. 2012), we hypothesized that disruption of nuclear lamin interaction in senescent cells may contribute to the large-scale chromatin changes.
To investigate this possibility, we first investigated the proximity of canyons and mesas to lamin-associated domains (LADs). We performed ChIP-seq of lamin B1 in proliferating cells and defined 2054 LADs across the genome (see the Supplemental Material), which largely confirms a previous LAD DamID ChIP–chip data set in proliferating IMR90 cells (Supplemental Fig. 9; Guelen et al. 2008). The median size of the LADs is 450 kb, and by nucleotide measurement, LADs cover ∼61% of the genome in proliferating cells. Interestingly, by visual inspection, K4me3 and K27me3 mesas appear to be strongly associated with LADs, whereas K27me3 canyons appear to form in between LADs or over “gaps” in the LAD profile (Fig. 2A, LADs shown in gold).
We computationally defined the mesa–LAD and canyon–LAD relationship by assessing overlapping nucleotides (Fig. 2C). Approximately 60% of LAD nucleotides have overlap with mesas or canyons (Supplemental Fig. 10A). Indeed, 90% of K4me3 mesa nucleotides and 92% of K27me3 nucleotides overlap with LADs (Fig. 2C, green and blue pie charts at the top), underscoring the occurrence of mesas within LADs. In contrast, only 41% of K27me3 canyon nucleotides overlap with LADs (Fig. 2C, red pie chart at the bottom), highlighting the visual observation of canyon formation adjacent to and outside of LADs. The high overlap of mesas with LADs, but not with canyons, is also reflected when measuring the overlap between features at an 80% cutoff parameter (Supplemental Fig. 10B). We further analyzed the mesa–LAD and canyon–LAD relationship by assessing the size of the LADs that overlap these features (Supplemental Fig. 10C); LADs that overlap K27me3 canyons are smaller than the LADs overlapping K27me3 mesas (Supplemental Fig. 10C, cf. red vs. gray LAD size distribution), again emphasizing the incidence of canyons outside of the large LADs that overlap mesas. Together, this analysis highlights a nonrandom nature of mesa and canyon formation in the genome.
LADs are generally gene-poor relative to their genomic occupancy (Pickersgill et al. 2006; Guelen et al. 2008; Peric-Hupkes et al. 2010; van Bemmel et al. 2010), suggesting that K27me3 canyons, often outside of LADs, may be enriched for genes and their regulatory regions. Strikingly, we found both inactive enhancers (as defined by regions enriched for H3K4me1) and active enhancers (as defined by regions enriched for both H3K4me1 and H3K27ac) strongly overlapping with K27me3 canyons (Fig. 2D). Specifically, 51% of inactive enhancers and 49% of active enhancers are within canyons, and this association is nonrandom and significant (P = 0.001 for enrichment over background), as shown by permutation analysis (Fig. 2D). Thus, K27me3 canyons, mostly outside of LADs, are strongly enriched for enhancers. Furthermore, enhancers near key senescence genes are often contained within K27me3 canyons, where the H3K27me3 loss is also associated with up-regulated gene expression (65.1% of SASP genes are in canyons), as seen in the MMP cluster on chromosome 11 and specifically for the MMP3 gene (Fig. 2E). Hence, H3K27me3 loss at enhancer regions may play an important role in regulating transcriptional changes in senescence.
Similar profound alterations of chromatin have also been reported in genome-wide analyses of DNA methylation in cancer cells (Schuster-Bockler and Lehner 2012). Since senescence is a tumor suppression mechanism, we examined the intersection of senescent canyons and mesas with DNA differentially methylated regions (DMRs) in cancer. We found that 60% of K4me3 mesas overlap at least 50% with cancer-specific DNA hypomethylated DMRs (P < 2.2 × 10−16) in human primary colon adenocarcinoma (Berman et al. 2012; data not shown). Comparison with colorectal cancer reveals that 76% of K4me3 mesas overlap at least 50% with large (≥100 kb) DNA hypomethylated DMRs (example in Fig. 3A; Hansen et al. 2011). The data indicate similarities between chromatin regulation in senescent cells and cancer cells.
H3K4me3 mesas overlap hypomethylated regions in cancer; mesas form in an OIS model and HGPS. (A) Example track view of a cluster of H3K4me3 mesas on chromosome 1 (Chr1: 30,500,000–40,000,000) shows a strong degree of overlap with cancer hypomethylated regions (black blocks). Proliferating tracks are shown in orange, and senescence tracks are shown in blue. (B) Example track view of the cluster of H3K4me3 mesas between replicative senescence and OIS shows a strong degree of overlap between the mesas of the two different senescence models. Proliferating tracks are shown in orange, and senescence tracks are shown in blue. Replicative senescence mesas are shown in green blocks, OIS mesas are shown in teal blocks, and LADs are shown in gold blocks. (C) Venn diagram representing the degree of overlap between the OIS and replicative senescence mesas shows a high degree of overlap between the two sets. Note that the OIS model has almost 200 more mesas than the replicative senescence. (D) Box plot analysis of the K4me3 mesas in replicative senescence and OIS (blue boxes) compared with nonmesa control regions (gray) shows an even higher enrichment for H3K4me3 gain in OIS mesas than replicative senescence mesas. (E) Histogram analysis of the OIS and replicative senescence mesas shows a bimodal pattern for the OIS mesas (green dotted line) compared with the single mode of replicative senescence mesas (green solid line). This suggests that a subset of OIS mesas is particularly enriched for H3K4me3, even more than the replicative senescence mesas. Control nonmesa regions are shown in black solid (replicative senescence) and dotted (OIS) lines. (F,G) Western blot analysis of progerin expression in proliferating and senescent IMR90s (F) and parental control and HGPS cell strains (G) shows progerin expression only in HGPS cells, not in IMR90 or parental control cells. (H) Western blot analysis of lamin B1 expression in parental control and HGPS cell strains shows similar levels of lamin B1 in both cell populations at the time of experimentation, underscoring the proliferating state of the HGPS cells at this point. (I,J) ChIP-qPCR evidence for K4me3 mesa formation in HGPS. Replicative senescence qPCR mesa validation shown in I (proliferating shown in orange and senescence shown in blue). The same K4me3 mesa analysis shown in J (parent control shown in purple and HGPS shown in green) indicates that K4me3 mesas may be a shared feature of HGPS. ChIP-qPCR primers are tiled across the region of the mesas and include 3′ and 5′ flanking primers. ChIP-qPCR data are shown as ratios of modification to total histone H3. ChIP-qPCR data are the average of three biological replicates, and error bars represent the standard deviation from the mean. (K,L) ChIP-qPCR test for K27me3 mesa formation in HGPS suggests that these features may not be forming in proliferating HGPS. Replicative senescence qPCR mesa validation shown in K (proliferating shown in orange and senescence shown in blue). The same K4me3 mesa analysis shown in L (parent control shown in purple and HGPS shown in green) indicates that K27me3 mesas may not be a shared feature of HGPS. Note that the K27me3 signal in HGPS is generally lower than the parental control, even at flanking regions. ChIP-qPCR primers are tiled across the region of the mesas and include 3′ and 5′ flanking primers. ChIP-qPCR data are shown as ratios of modification to total histone H3. ChIP-qPCR data are the average of three biological replicates, and error bars represent the standard deviation from the mean. (M,N) ChIP-qPCR test for K27me3 canyon formation in HGPS suggests that these features may not be forming in proliferating HGPS. Replicative senescence qPCR canyon validation shown in M (proliferating shown in orange and senescence shown in blue). The same K4me3 canyon analysis shown in N (parent control shown in purple and HGPS shown in green) indicates that K27me3 canyons are maybe not forming in still-proliferating HGPS cells. Note that the K27me3 signal in HGPS is generally lower than the parental control, even at flanking regions. ChIP-qPCR primers are tiled across the region of the canyons and include 3′ and 5′ flanking primers. ChIP-qPCR data are shown as ratios of modification to total histone H3. ChIP-qPCR data are the average of three biological replicates, and error bars represent the standard deviation from the mean.
To further investigate this connection, we examined K4me3 mesa formation in the cancer-related senescence model of oncogene-induced senescence (OIS) (Cosme-Blanco et al. 2007; Feldser and Greider 2007). IMR90 cells harboring a tamoxifen-inducible RAS oncogene undergo rapid senescence (Young et al. 2009). H3 and H3K4me3 ChIP-seq was performed for control and OIS cells, leading to identification of 806 K4me3 mesas, more than observed in replicative senescence (648 mesas). There is a striking overlap of OIS mesas with the original K4me3 mesas in replicative senescence (Fig. 3B, replicative senescence mesas shown in green and OIS mesas shown in teal): Sixty percent of replicative senescence K4me3 mesas overlap with OIS mesas (Fig. 3C), highlighting the obvious visual overlap in the tracks. Interestingly, the H3K4me3 gains in mesas in OIS seem more pronounced than in replicative senescence (Fig. 3D), and further analysis revealed a subset that is particularly enriched for H3K4me3 (Fig. 3E, cf. green dotted line for OIS mesas and the solid green line for replicative senescence mesas).
We further examined mesa and canyon formation in primary skin fibroblasts from HGPS patients, a segmental premature aging syndrome associated with accelerated cell senescence. HGPS is most commonly associated with a lamin A splicing mutation (to generate progerin), and we reasoned that lamin A mutation might phenocopy the major chromatin changes associated with lamin B1 down-regulation in senescence. We performed ChIP-qPCR on patient-derived proliferating HGPS cells across regions corresponding to the K4me3 mesas, K27me3 mesas, and K27me3 canyons of senescent IMR90 cells. Importantly, the patient-derived HGPS cell strain was progerin-positive, compared with progerin-negative IMR90 and parental control cells (Fig. 3F,G). Moreover, the HGPS cells were proliferating and not senescent at the time of the experimentation (at passage doubling 14; both cyclin A-positive and p16-negative measured by qRT–PCR) (data not shown), and the parental and HGPS lamin B1 levels were equivalent at the time cells were harvested for experiments (Fig. 3H).
We performed ChIP-qPCR across regions corresponding to K4me3 and K27me3 mesas and to K27me3 canyons of senescent IMR90 cells. As a control, we used passage-matched cells from an unaffected progerin-negative parent. We did not detect either K27me3 mesas or K27me3 canyons in HGPS across the regions tested (Fig. 3K–N); we note that, in general, our results show that H3K27me3 is broadly reduced in HGPS compared with the parental control, as previously observed (Shumaker et al. 2006; McCord et al. 2013). Hence, it appears that lowered H3K27me3 in HGPS may extend across the genome prior to senescence. Interestingly, in contrast, K4me3 mesas arise in the HGPS cell line compared with parent control across two different K4me3 mesa regions (Fig. 3I,J [IMR90 proliferating and senescent control shown in I, and HGPS and parent control shown in J]). Taken together, the data indicate that large-scale chromatin changes, particularly K4me3 mesas, may be a shared feature in prematurely aging HGPS cells and senescent cells. In particular, premature formation of H3K4me3 mesas might contribute to premature senescence of the HGPS cells.
The analysis above focused on large-scale chromatin changes that occur in senescence. Given the overlap of canyons with enhancer and gene regions, we next compared chromatin alteration with gene expression. First, we quantified H3K4me3 and H3K27me3 enrichment at the top 500 most up-regulated and down-regulated genes in senescence (Fig. 4A, gene expression fold changes between −40-fold and 20-fold as shown in the right panel). H3K4me3 is enriched at up-regulated genes and is notably depleted at down-regulated genes compared with randomly selected genes with no transcriptional change (Fig. 4A, green boxes). H3K27me3 shows the opposite pattern, where the modification is depleted at up-regulated genes and enriched at down-regulated genes. To assess specific genes, we performed a scatter plot analysis, comparing gene expression with changes in H3K4me3 (Fig. 4B) and H3K27me3 (Fig. 4C). We found that H3K4me3 depletion is correlated to cell cycle and other proliferation genes that are transcriptionally down-regulated in senescence (Fig. 4B, cell cycle genes marked in green diamonds, expanded in the insert; Supplemental Fig. 11A shows the gene ontology [GO] analysis for down-regulated genes that lose H3K4me3; Chicas et al. 2012). Consistently, the OIS cells also display H3K4me3 loss at down-regulated cell cycle genes (Supplemental Fig. 11B).
Changes in H3K4me3 and H3K27me3 are correlated to gene expression changes in senescence. (A) Box plot representation of H3K4me3 (green) and H3K27me3 (pink) at the top 500 up-regulated and down-regulated genes in senescence shows loss of H3K4me3 at down-regulated genes and loss of H3K27me3 at up-regulated genes. Control genes (no change) are shown in gray boxes. Box plot on the right depicts the range of expression changes at the top 500 up-regulated and down-regulated genes. Enrichment is reported as a ratio of senescence/proliferating, and all ChIP-seq signal is normalized to total histone H3. (B) Scatter plot analysis of H3K4me3-mediated gene regulation in senescence; fold change H3K4me3 is shown on the Y-axis, and fold change gene expression is shown on the X-axis. Cell cycle genes are marked in green. The bottom left quadrant contains down-regulated genes that also lose H3K4me3. The boxed region in the bottom left quadrant is shown in the closeup on the right to highlight the enrichment of cell cycle genes for H3K4me3 loss. (C) Scatter plot analysis of H3K27me3-mediated gene regulation in senescence; the fold change H3K27me3 is shown on the Y-axis, and the fold change gene expression is shown on the X-axis. SASP genes are marked in red. The bottom right quadrant contains up-regulated genes that also lose H3K27me3. The boxed region in the bottom right quadrant is shown in the closeup on top to highlight the enrichment of SASP genes for H3K27me3 loss. (D) GO analysis of up-regulated genes that lose H3K27me3 shows a strong enrichment for genes that are involved in the senescence pathway, including senescence genes, anti-proliferation genes, and cell death/stress genes. Each GO category color indicates the level of significance (yellow to orange, as shown on the key in the bottom left). (E) Sample tracks of H3K4me3 and H3K27me3 ChIP-seq data over a down-regulated cell cycle gene, CCNA2 (Chr4: 122,955,000–122,966,500), show pronounced loss of H3K4me3 at the promoter and TSS of the gene. Proliferating tracks are shown in orange, senescence tracks are shown in blue, and H3K27me3 change is shown in gray. The gene location is shown on the bottom track. (F) Sample tracks of H3K4me3 and H3K27me3 ChIP-seq data over the up-regulated SASP gene TNFRSF10c (Chr8: 23,012,616–23,034,551) show pronounced loss of H3K27me3 across the gene. Proliferating tracks are shown in orange, senescence tracks are shown in blue, and H3K27me3 change is shown in gray. The gene location is shown on the bottom track.
Strikingly, scatter plot analysis revealed that H3K27me3 loss is strongly correlated to a number of genes involved in the senescence response, including key SASP genes up-regulated in senescence (Fig. 4C, SASP genes marked in red diamonds, expanded in the insert; SASP and other senescence-related genes listed in Supplemental Table 3). GO analysis of the lower right quadrant of the H3K27me3 scatter plot (H3K27me3 loss and increased gene expression) shows significant enrichment for senescence-related categories, including senescence, anti-proliferation, and stress response (Fig. 4D; Supplemental Table 4 for full gene list of all GO terms and genes and statistics for H3K4me3 and H3K27me3; Supplemental Text 2 for GO and scatter plot analysis). It is important to note that, while some of the genes enriched in this quadrant also gain H3K4me3 in the senescent state, the key defining chromatin change correlated to this category of genes is the loss of H3K27me3.
As predicted by the scatter plot analysis, overlaid proliferating and senescence ChIP-seq tracks clearly show H3K4me3 loss at the promoter and transcriptional start site (TSS) of the senescence-down-regulated CCNA (cyclin A) gene and little change of H3K27me3 (Fig. 4E, proliferating tracks in orange, senescent tracks in blue, and H3K27me3 difference track in gray). Tumor necrosis factor receptor 10c (TNFRSF10c) is an up-regulated SASP gene (18-fold up-regulated) and is enriched in the scatter plot analysis for H3K27me3 loss. ChIP-seq tracks show H3K27me3 loss across the entire TNFRSF10c locus as well as H3K4me3 increase at the promoter and TSS (Fig. 4F). Together, these analyses highlight a second type of chromatin change in senescent cells, where localized chromatin changes are correlated to the transcriptional change in senescence. Remarkably, these local changes are typically encompassed within large-scale features; namely, the K27me3 canyons.
Our observations above show large-scale chromatin changes, including mesas that overlap almost entirely with LADs as well as canyons, typically located in between LADs and associated with localized chromatin changes correlated to gene expression. In light of this and knowing that lamin B1 is down-regulated in senescent cells (Shimi et al. 2011; Freund et al. 2012), we investigated whether down-regulation of lamin B1 is a trigger of chromatin changes in senescent cells. As previously shown, we detected a significant reduction in lamin B1 transcript and protein levels to nearly undetectable levels in senescent cells (Fig. 5A), whereas lamins A and C do not change (data not shown). Interestingly, we also detected reduced protein levels of other lamin interaction and nuclear organizational components, including the cohesin component SA1, the boundary element CTCF, and the lamin-interacting deacetylase Lap2 (lamin-associated deacetylase) (Fig. 5B). Importantly, the loss of lamin B1 was not due to the nonproliferative state in senescent cells, as quiescent cells achieved by serum starvation did not exhibit lamin B1 loss (Supplemental Fig. 12A; Shimi et al. 2011), and quiescent cells caused by contact inhibition did not exhibit canyon and mesa formation (Supplemental Fig. 12B,C).
Lamin B1 is significantly decreased in senescence, and LMNB1 knockdown in proliferating cells causes formation of canyons and mesas. (A, left panel) Western blot analysis of total lamin B1 in proliferating and senescent cells shows lamin B1 loss in senescent cells. Lysates were normalized by total cell count; GAPDH was used as a loading control. (Right panel) CTCF (boundary element), SA1 (cohesin component), and Lap2 are decreased in senescent lysates compared with proliferating cells. Lysates were normalized by total cell count with GAPDH as a loading control. (B) qRT–PCR measure of lamin B1 mRNA expression in knockdown compared with control. LMNB1 expression is reduced by >60% by two different shRNA constructs (shRNA 1 and shRNA 2) compared with scrambled hairpin and vector-only control infected cells. (C) Western blot analysis of total protein in LMNB1 knockdown compared with control. Total lamin B1 is significantly reduced at the protein level in two knockdowns (shRNA1 and shRNA 2) compared with scrambled hairpin and vector-only control infected cells. Lysates were normalized by total cell count with GAPDH used as a loading control. (D) Life span analysis of LMNB1 knockdown shows rapid senescence within two PDs following cell infection compared with controls. Life span is visualized by plotting PDs (Y-axis) to days of growth in culture (X-axis). Two life spans of wild-type (WT), uninfected IMR90 show senescence after 78 (red diamonds) and 81 (blue diamonds) PDs. EZH2 knockdown causes senescence after only two cell passages following infection (green and purple triangles). LMNB1 knockdown by two different shRNA constructs shows the same kinetics of rapid senescence as EZH2 (turquoise and orange circles). The rapid senescence kinetics are specific to EZH2 and LMNB1 knockdown, as empty vector-only and scrambled hairpin infections (blue and pink squares) have a nearly normal life span, although not as long as wild-type controls (red and blue diamonds). The flattening of the growth curves indicates the length of time that cells were kept in a senescent state prior to harvesting for further experimentation. (E) LMNB1 knockdown results in up-regulation of p16. qRT–PCR measure of p16 expression in LMNB1 knockdown cells compared with control shows expected senescence up-regulation of p16 following LMNB1 knockdown. Relative p16 expression is up-regulated in two knockdowns (shRNA1 and shRNA 2) compared with scrambled hairpin and vector-only control infected cells; GAPDH was used as a control. (F) LMNB1 and EZH2 knockdown causes cellular senescence, not cell death. EZH2 and LMNB1 knockdown does not cause significant cell death, as measured by annexin staining compared with empty vector control. EZH2 and LMNB1 knockdown cells undergo rapid senescence, within two PDs, as measured by SA-β-gal staining, compared with empty vector control. For both types of knockdown, stably infected cells were maintained in culture for 2–3 d to ensure growth cessation prior to experimentation. (G,H) Western blot analysis of lamin B1 expression (G) and histone H3 expression (H) in a time course of IMR90 cells approaching senescence. The PD60, PD70, and PD78 time points are all proliferating states compared with senescence at PD80. Lamin B1 levels appear to be decreasing in proliferating cells prior to achievement of senescence compared with GAPDH control (G), whereas histone H3 decrease does not appear to occur until cells are in a senescent state compared with GAPDH control (H). (I–K) ChIP-qPCR evidence for K4me3 mesa (Chr7: 61,558,937–64,197,991), K27me3 mesa (Chr6: 167,743,380–169,126,883), and K27me3 canyon (Chr2: 85,956,542–86,782,134) formation in LMNB1 knockdown (KD; regions defined in replicative senescence). Control knockdown data are shown in orange, and LMNB1 knockdown is shown in blue. ChIP-qPCR primers are tiled across the entire region of the mesas and canyon and include 3′ and 5′ flanking primers. ChIP-qPCR data are shown as ratios of modification to total histone H3. ChIP-qPCR data are the average of three biological replicates, and error bars represent standard deviation from the mean.
To assess the functional link between lamin B1 loss and senescence, expression of the lamin B1 gene LMNB1 was reduced using lentiviral expression of targeted shRNA in proliferating cells. Two different shRNAs lowered lamin B1 RNA levels >50% (Fig. 5B), and protein levels significantly declined (Fig. 5C). Compared with scrambled shRNA or vector controls, the LMNB1 knockdown cells exhibited rapid cessation of cell division within two passages following infection (Fig. 5D, blue and orange circles) and became senescent, exhibiting up-regulated p16 expression (Fig. 5E) and high SA-β-gal-positive cells (senescence marker) compared with low annexin-positive cells (apoptosis marker) (Fig. 5F). The premature senescence phenotype was specific to the LMNB1 knockdown cells, as vector control cells showed almost normal life span (Fig. 5D, blue and pink squares) and little cell death or senescence following infection (Fig. 5E,F). We further note that lamin B1 loss appears to occur in nearly senescent cells across a time course of cells approaching senescence (Fig. 5G). Compared with midlife PD60 cells, lamin B1 appears to be slightly reduced at PD70 and largely reduced at PD78, which is two PDs before senescence at PD80. In contrast, the major loss of total histone associated with senescence does not appear until after lamin B1 is reduced, with slight reduction of histone H3 at PD78 and significant reduction at PD80, when the cells reach senescence (Fig. 5H).
The rapidity of premature senescence after LMNB1 knockdown was similar to the senescence following knockdown of the H3K27me3 methylase EZH2 (Fig. 5D, green and purple triangles) as previously described (Bracken et al. 2007). This and the observation that lamin B1 loss appears to be occurring in nearly senescent cells prior to senescence suggested that a functional link may exist between lamin disruption and the large-scale chromatin changes that we detected in senescent cells. To address this possibility, we performed ChIP-qPCR for H3K4me3 and H3K27me3 in the proliferating cells following lamin B1 knockdown (Fig. 5I–K). Strikingly, LMNB1 knockdown led to development of specific K4me3 mesas, K27me3 mesas, and K27me3 canyons across the genome (canyon and mesa regions were selected at random from the originally defined set and were remarkably similar to these features in senescent cells) (cf. Figs. 5I–K and 1C,D,F). The observation of mesa and canyon formation in LMNB1 knockdown strongly suggests that disruption of nuclear lamina-mediated organization may facilitate global chromatin changes during senescence.
Discussion
Here we show profound reorganization in the human epigenome upon senescence, with acquisition of large, contiguous stretches of altered chromatin—K4me3 mesas, K27me3 mesas, and K27me3 canyons—focused in and around the location of LADs in proliferating cells (Figs. 1A, 2A). We note that correlated K4me3 and K27me3 mesas over LADs may represent unusually large regions of “bivalent” chromatin (that is, marked by the presence of both active and repressive chromatin modifications) (Bernstein et al. 2006); whether the bivalency represents chromatin instability remains to be determined. Importantly, several lines of evidence indicate that mesa and canyon chromatin changes in senescence are closely linked to altered structure of LADs in senescence: First, lamin B1 declines to undetectable levels during senescence (Fig. 5A; Shimi et al. 2011; Freund et al. 2012), and the initial lamin B1 decline is prior to reduction of histone levels (Fig. 5G,H). Second, forced reduction of lamin B1, which leads to premature senescence, concomitantly results in mesas and canyons (Fig. 5I–K). Third, K4me3 mesas are detected in patient-derived HGPS cells (Fig. 3I,J), which harbor a lamin A/progerin mutation and undergo premature senescence.
HGPS is a segmental progeroid syndrome; thus, our results indirectly link formation of H3K4me3 mesas to accelerated tissue and organismal aging, perhaps due to acquisition of premature senescence chromatin changes. We also connect the senescence-associated chromatin features to alterations in cancer genomes. Specifically, mesas strongly correlate to the broad-scale, differentially DNA hypomethylated regions in colorectal cancer (Fig. 3A). Additionally, we detected H3K4me3 mesas in OIS (Fig. 3B,C). Together, these findings implicate the large-scale changes in senescent chromatin to both cancer and aging.
The functional consequence of the large-scale alterations in the senescence epigenome is not yet known; however, several lines of evidence are consistent with the notion that these large-scale chromatin changes are effectors of senescence. First, knockdown of lamin B1 triggers formation of mesas, overlapping with LADs, and accelerated senescence (Fig. 5D,I–K). Second, mesas form prematurely in proliferating HGPS cells (Fig. 3H–J) and thus are not simply a consequence of premature senescence in these cells. Third, localized H3K27me3 loss in canyons is strongly connected to up-regulation of key senescence and anti-proliferation genes, including the canonical SASP genes (Fig. 4C). Many of these genes are located within and near K27me3 canyons. Moreover, K27me3 canyons are enriched for inactive and active enhancers (Fig. 2D), strongly suggesting that canyons may contribute to the senescent state by altering gene expression via regulation of enhancer activity.
Interestingly, the K4me3 mesa and K27me3 canyon alterations that we observed are consistent with acquisition of a more open chromatin state, which supports previous findings that, in general terms, histone modifications characteristic of open chromatin tend to oppose longevity (Michishita et al. 2008; Dang et al. 2009; Kawahara et al. 2009; Greer et al. 2010, 2011). These results are in apparent contrast to the description of SAHFs in some senescent cell types (Funayama et al. 2006; Narita et al. 2006; Zhang et al. 2007). In a recent study, the higher-order heterochromatin organization reflected in SAHFs in oncogene-induced senescent cells was reported to be independent of regional changes in chromatin modifications (Chandra et al. 2012); however, since the genomic locations of SAHFs are not yet defined, the relationship between mesas, canyons, and SAHFs remains to be determined.
Our analysis provides a framework for understanding how down-regulation of lamin B1 in senescent cells initiates global chromatin changes, including reduction of histone levels, K4me3 and K27me3 mesas, and K27me3 canyons. Importantly, some of these global chromatin changes (K27me3 canyons) encompass local changes at gene enhancers that are likely to drive expression of SASP genes, key effectors of the senescence program. Indeed, lamin B1 is reduced in many normal cell types undergoing senescence, and overexpression delays senescence (Shimi et al. 2011; Freund et al. 2012). We note that certain cell types and disease states exhibit opposite effects of lamin B1 alteration (Barascu et al. 2012; Dreesen et al. 2013). In these diseased cells types, there may be other chromatin changes that alter the nuclear lamin organization and/or affect the cell cycle, perhaps leading to the opposite response to lamin B1 fluctuations compared with the nondiseased IMR90 model system used in our study. This is particularly interesting given our observation that proliferating but prematurely aging HGPS cells exhibit a subset of the global chromatin changes that we observed in senescent normal fibroblast cells. Moreover, we also link global chromatin changes in normal senescent cells to epigenetic alterations in cancer. In sum, we identified a lamin B1-regulated program of global chromatin reconfiguration in senescence that extends to local gene regulatory changes and to both aging and cancer.
Materials and methodsAntibodies
All antibodies used for this work are listed in Supplemental Table 5.
Cell culture
IMR90 cells (obtained from Coriell Institute for Medical Research, Camden, NJ) were grown in standard tissue culture medium (DMEM with 10% FBS and 1% penicillin/streptomycin) at 3% oxygen. For senescence studies, cells were replicatively passaged until growth ceased. Senescent cells were maintained in dishes for 2 wk to ensure growth termination. For knockdown experiments, midlife IMR90 cells were infected with lentiviral hairpin constructs (TRC collection) designed against lamin B1 in medium containing 8 μg/mL polybrene and 5% FBS for 24 h. Cells were then maintained in standard tissue culture medium with addition of 0.5 μg/mL puromycin to select a stably infected population. For lamin B1 knockdown studies, stably infected cells were established within two PDs, soon after which replication ceased. Senescence cells were maintained in dishes for 2–3 d to ensure growth termination prior to downstream experiments. For all cells, senescence was determined by monitoring p16 up-regulation and down-regulation of cyclin genes and by SA-β-gal staining. HGPS cells (AGO3199 and AG11498) and the parent control (AGO3257) (obtained from Coriell Institute for Medical Research, Camden, NJ) were grown in Eagle's MEM with Earle's salts, nonessential amino acids, 15% FBS, and 1% penicillin/streptomycin at 3% oxygen. For the experiment with HGPS and control, both cell lines were harvested for experiments at passage 14.
Whole-cell extract preparation and Western blot analysis
Medium from cells growing in 10-cm2 dishes was removed, and cells were washed in 1× PBS. Cells were scraped into 5 mL of 1× PBS, collected by centrifugation at 1200 rpm for 3 min at 4°C, and resuspended in 500 μL to 1 mL of standard SDS-PAGE loading buffer. Volumes were adjusted based on cell count, resulting in cell number-normalized lysate. Samples were boiled for 10 min, vortexed, and spun down to remove debris. Extracts were resolved on standard SDS-PAGE gels, transferred to nitrocellulose, and blotted for proteins of interest. All blots were blocked in 5% milk in TBST (1× TBS + 0.1% Tween). Antibodies were diluted in the same blocking solution. The antibody dilutions are listed in Supplemental Table 4. Washes were done in TBST. Visualization of HRP-conjugated secondary antibodies was achieved by standard chemiluminescence. Images of Western blots were taken with a Fujifilm LAS-4000 imager.
RNA preparation and qRT–PCR
RNA was purified using Trizol extraction. RT was performed using the Applied Biosystems High-Capacity RNA-to-cDNA kit (4387406). cDNA was quantified by qRT–PCR using standard procedures on a 7900HT Fast-Real-Time PCR (ABI). Primer sequences are available on request.
ChIP-qPCR or ChIP-seq
Cells in 10-cm2 dishes were fixed in 1% formaldehyde for 10 min, and fixation was quenched with addition of glycine to 125 mM for an additional 5 min. Cells were harvested by scraping from plates and washed twice in 1× PBS before storage at −80°C. ChIP was performed as described in the Young laboratory protocol (Lee et al. 2006b), except that extracts were sonicated twice for 9 min each round (30 sec of sonication with intermediate incubation of 30 sec per round) using a Bioruptor (Diagenode). All ChIPs were performed using 500 μg of extract and 2 μg of antibody per sample. Thirty microliters of Protein G Dynabeads (Invitrogen, 100.02D) was used per ChIP. Controls with IgG and no antibody controls were routinely performed, and all antibodies were tested by titration to be functioning within the linear range of the protocol. Following elution, ChIP DNA was analyzed by standard qPCR methods on a 7900HT Fast-Real-Time PCR (ABI). Primer sequences are available on request. For sequencing, 10 ng of ChIP DNA was used to make sequencing libraries using standard Illumina library single-end construction procedures. Sequencing was performed on either Illumina GAIIx (36 base pairs [bp], single-end reads) or Hi-Seq (100 bp, paired-end or single-end reads) platforms.
Computational methods
ChIP-seq analysis and computational methods used to characterize canyons and mesas are in the Supplemental Material.
Acknowledgments
We thank the NIH Roadmap Epigenomics Initiative (http://nihroadmap.nih.gov/epigenomics) and ENCODE for the use of their H3K4me1 and H3K27ac ChIP-seq data sets in proliferating IMR90 cells, the IDOM Functional Genomics Core Sequencing Facility and the Penn Genome Frontiers Institute for ChIP-seq, the Beatson Institute microarray core facility for microarray hybridization, and the Protein Expression Core (Epigenetics of Aging P01) for various support. This work was supported by an NIA P01 grant (P01AG031862) awarded to P.D.A. and S.L.B., and a CRUK program grant (A10250) awarded to P.D.A.
Supplemental material is available for this article.
Article published online ahead of print. Article and publication date are online at http://www.genesdev.org/cgi/doi/10.1101/gad.223834.113.
ReferencesAcostaJC, O'LoghlenA, BanitoA, GuijarroMV, AugertA, RaguzS, FumagalliM, Da CostaM, BrownC, PopovN, 2008Chemokine signaling via the CXCR2 receptor reinforces senescence.
133: 1006–101818555777AdamsPD2009Healing and hurting: Molecular mechanisms, functions, and pathologies of cellular senescence.
36: 2–1419818705BakerDJ, WijshakeT, TchkoniaT, LeBrasseurNK, ChildsBG, van de SluisB, KirklandJL, van DeursenJM2011Clearance of p16Ink4a-positive senescent cells delays ageing-associated disorders.
479: 232–23622048312BarascuA, Le ChalonyC, PennarunG, GenetD, ImamN, LopezB, BertrandP2012Oxidative stress induces an ATM-independent senescence pathway through p38 MAPK-mediated lamin B1 accumulation.
31: 1080–109422246186BarskiA, CuddapahS, CuiK, RohTY, SchonesDE, WangZ, WeiG, ChepelevI, ZhaoK2007High-resolution profiling of histone methylations in the human genome.
129: 823–83717512414BermanBP, WeisenbergerDJ, AmanJF, HinoueT, RamjanZ, LiuY, NoushmehrH, LangeCP, van DijkCM, TollenaarRA, 2012Regions of focal DNA hypermethylation and long-range hypomethylation in colorectal cancer coincide with nuclear lamina-associated domains.
44: 40–4622120008BernsteinBE, MikkelsenTS, XieX, KamalM, HuebertDJ, CuffJ, FryB, MeissnerA, WernigM, PlathK, 2006A bivalent chromatin structure marks key developmental genes in embryonic stem cells.
125: 315–32616630819BrackenAP, Kleine-KohlbrecherD, DietrichN, PasiniD, GargiuloG, BeekmanC, Theilgaard-MonchK, MinucciS, PorseBT, MarineJC, 2007The Polycomb group proteins bind throughout the INK4A-ARF locus and are disassociated in senescent cells.
21: 525–53017344414CampisiJ2005Senescent cells, tumor suppression, and organismal aging: Good citizens, bad neighbors.
120: 513–52215734683ChandraT, KirschnerK, ThuretJY, PopeBD, RybaT, NewmanS, AhmedK, SamarajiwaSA, SalamaR, CarrollT, 2012Independence of repressive histone marks and chromatin compaction during senescent heterochromatic layer formation.
47: 203–21422795131ChicasA, KapoorA, WangX, AksoyO, EverttsAG, ZhangMQ, GarciaBA, BernsteinE, LoweSW2012H3K4 demethylation by Jarid1a and Jarid1b contributes to retinoblastoma-mediated gene silencing during cellular senescence.
109: 8971–897622615382CoppeJP, PatilCK, RodierF, SunY, MunozDP, GoldsteinJ, NelsonPS, DesprezPY, CampisiJ2008Senescence-associated secretory phenotypes reveal cell-nonautonomous functions of oncogenic RAS and the p53 tumor suppressor.
6: 2853–286819053174Cosme-BlancoW, ShenMF, LazarAJ, PathakS, LozanoG, MultaniAS, ChangS2007Telomere dysfunction suppresses spontaneous tumorigenesis in vivo by initiating p53-dependent cellular senescence.
8: 497–50317396137DangW, SteffenKK, PerryR, DorseyJA, JohnsonFB, ShilatifardA, KaeberleinM, KennedyBK, BergerSL2009Histone H4 lysine 16 acetylation regulates cellular lifespan.
459: 802–80719516333DreesenO, ChojnowskiA, OngPF, ZhaoTY, CommonJE, LunnyD, LaneEB, LeeSJ, VardyLA, StewartCL, 2013Lamin B1 fluctuations have differential effects on cellular proliferation and senescence.
200: 605–61723439683FeldserDM, GreiderCW2007Short telomeres limit tumor progression in vivo by inducing senescence.
11: 461–46917433785FeserJ, TruongD, DasC, CarsonJJ, KieftJ, HarknessT, TylerJK2010Elevated histone expression promotes life span extension.
39: 724–73520832724FinlanLE, SproulD, ThomsonI, BoyleS, KerrE, PerryP, YlstraB, ChubbJR, BickmoreWA2008Recruitment to the nuclear periphery can alter expression of genes in human cells.
4: e100003918369458FreundA, LabergeRM, DemariaM, CampisiJ2012Lamin B1 loss is a senescence-associated biomarker.
23: 2066–207522496421FunayamaR, SaitoM, TanobeH, IshikawaF2006Loss of linker histone H1 in cellular senescence.
175: 869–88017158953GreerEL, MauresTJ, HauswirthAG, GreenEM, LeemanDS, MaroGS, HanS, BankoMR, GozaniO, BrunetA2010Members of the H3K4 trimethylation complex regulate lifespan in a germline-dependent manner in C. elegans.
466: 383–38720555324GreerEL, MauresTJ, UcarD, HauswirthAG, ManciniE, LimJP, BenayounBA, ShiY, BrunetA2011Transgenerational epigenetic inheritance of longevity in Caenorhabditis elegans.
479: 365–37122012258GuelenL, PagieL, BrassetE, MeulemanW, FazaMB, TalhoutW, EussenBH, de KleinA, WesselsL, de LaatW, 2008Domain organization of human chromosomes revealed by mapping of nuclear lamina interactions.
453: 948–95118463634GuentherMG, LevineSS, BoyerLA, JaenischR, YoungRA2007A chromatin landmark and transcription initiation at most promoters in human cells.
130: 77–8817632057HansenKD, TimpW, BravoHC, SabunciyanS, LangmeadB, McDonaldOG, WenB, WuH, LiuY, DiepD, 2011Increased methylation variation in epigenetic domains across cancer types.
43: 768–77521706001KaeberleinM, McVeyM, GuarenteL1999The SIR2/3/4 complex and SIR2 alone promote longevity in Saccharomyces cerevisiae by two different mechanisms.
13: 2570–258010521401KangTW, YevsaT, WollerN, HoenickeL, WuestefeldT, DauchD, HohmeyerA, GerekeM, RudalskaR, PotapovaA, 2011Senescence surveillance of pre-malignant hepatocytes limits liver cancer development.
479: 547–55122080947KawaharaTL, MichishitaE, AdlerAS, DamianM, BerberE, LinM, McCordRA, OngaiguiKC, BoxerLD, ChangHY, 2009SIRT6 links histone H3 lysine 9 deacetylation to NF-κB-dependent gene expression and organismal life span.
136: 62–7419135889KourmouliN, TheodoropoulosPA, DialynasG, BakouA, PolitouAS, CowellIG, SinghPB, GeorgatosSD2000Dynamic associations of heterochromatin protein 1 with the nuclear envelope.
19: 6558–656811101528KrizhanovskyV, YonM, DickinsRA, HearnS, SimonJ, MiethingC, YeeH, ZenderL, LoweSW2008Senescence of activated stellate cells limits liver fibrosis.
134: 657–66718724938KrtolicaA, ParrinelloS, LockettS, DesprezPY, CampisiJ2001Senescent fibroblasts promote epithelial cell growth and tumorigenesis: A link between cancer and aging.
98: 12072–1207711593017KuilmanT, MichaloglouC, VredeveldLC, DoumaS, van DoornR, DesmetCJ, AardenLA, MooiWJ, PeeperDS2008Oncogene-induced senescence relayed by an interleukin-dependent inflammatory network.
133: 1019–103118555778LeeTI, JennerRG, BoyerLA, GuentherMG, LevineSS, KumarRM, ChevalierB, JohnstoneSE, ColeMF, IsonoK, 2006aControl of developmental regulators by Polycomb in human embryonic stem cells.
125: 301–31316630818LeeTI, JohnstoneSE, YoungRA2006bChromatin immunoprecipitation and microarray-based analysis of protein location.
1: 729–74817406303Mattout-DrubezkiA, GruenbaumY2003Dynamic interactions of nuclear lamina proteins with chromatin and transcriptional machinery.
60: 2053–206314618255McCordRP, Nazario-TooleA, ZhangH, ChinesPS, ZhanY, ErdosMR, CollinsFS, DekkerJ, CaoK2013Correlated alterations in genome organization, histone methylation, and DNA-lamin A/C interactions in Hutchinson-Gilford progeria syndrome.
23: 260–26923152449MichishitaE, McCordRA, BerberE, KioiM, Padilla-NashH, DamianM, CheungP, KusumotoR, KawaharaTL, BarrettJC, 2008SIRT6 is a histone H3 lysine 9 deacetylase that modulates telomeric chromatin.
452: 492–49618337721NaritaM, KrizhanovskyV, NunezS, ChicasA, HearnSA, MyersMP, LoweSW2006A novel role for high-mobility group a proteins in cellular senescence and heterochromatin formation.
126: 503–51416901784O'SullivanRJ, KubicekS, SchreiberSL, KarlsederJ2010Reduced histone biosynthesis and chromatin changes arising from a damage signal at telomeres.
17: 1218–122520890289Peric-HupkesD, MeulemanW, PagieL, BruggemanSW, SoloveiI, BrugmanW, GrafS, FlicekP, KerkhovenRM, van LohuizenM, 2010Molecular maps of the reorganization of genome–nuclear lamina interactions during differentiation.
38: 603–61320513434PickersgillH, KalverdaB, de WitE, TalhoutW, FornerodM, van SteenselB2006Characterization of the Drosophila melanogaster genome at the nuclear lamina.
38: 1005–101416878134ReddyKL, ZulloJM, BertolinoE, SinghH2008Transcriptional repression mediated by repositioning of genes to the nuclear lamina.
452: 243–24718272965RohTY, CuddapahS, CuiK, ZhaoK2006The genomic landscape of histone modifications in human T cells.
103: 15782–1578717043231SagivA, BiranA, YonM, SimonJ, LoweSW, KrizhanovskyV2013Granule exocytosis mediates immune surveillance of senescent cells.
32: 1971–197722751116ScaffidiP, MisteliT2005Reversal of the cellular phenotype in the premature aging disease Hutchinson-Gilford progeria syndrome.
11: 440–44515750600SchuettengruberB, GanapathiM, LeblancB, PortosoM, JaschekR, TolhuisB, van LohuizenM, TanayA, CavalliG2009Functional anatomy of polycomb and trithorax chromatin landscapes in Drosophila embryos.
7: e1319143474Schuster-BocklerB, LehnerB2012Chromatin organization is a major influence on regional mutation rates in human cancer cells.
488: 504–50722820252SchwartzYB, KahnTG, NixDA, LiXY, BourgonR, BigginM, PirrottaV2006Genome-wide analysis of Polycomb targets in Drosophila melanogaster.
38: 700–70516732288ShaklaiS, AmariglioN, RechaviG, SimonAJ2007Gene silencing at the nuclear periphery.
274: 1383–139217489096SheltonDN, ChangE, WhittierPS, ChoiD, FunkWD1999Microarray analysis of replicative senescence.
9: 939–94510508581ShilatifardA2012The COMPASS family of histone H3K4 methylases: Mechanisms of regulation in development and disease pathogenesis.
81: 65–9522663077ShimiT, Butin-IsraeliV, AdamSA, HamanakaRB, GoldmanAE, LucasCA, ShumakerDK, KosakST, ChandelNS, GoldmanRD2011The role of nuclear lamin B1 in cell proliferation and senescence.
25: 2579–259322155925ShumakerDK, DechatT, KohlmaierA, AdamSA, BozovskyMR, ErdosMR, ErikssonM, GoldmanAE, KhuonS, CollinsFS, 2006Mutant nuclear lamin A leads to progressive alterations of epigenetic control in premature aging.
103: 8703–870816738054TaimenP, PfleghaarK, ShimiT, MollerD, Ben-HarushK, ErdosMR, AdamSA, HerrmannH, MedaliaO, CollinsFS, 2009A progeria mutation reveals functions for lamin A in nuclear assembly, architecture, and chromosome organization.
106: 20788–2079319926845TowbinBD, MeisterP, PikeBL, GasserSM2010Repetitive transgenes in C. elegans accumulate heterochromatic marks and are sequestered at the nuclear envelope in a copy-number- and lamin-dependent manner.
75: 555–56521467137van BemmelJG, PagieL, BraunschweigU, BrugmanW, MeulemanW, KerkhovenRM, van SteenselB2010The insulator protein SU(HW) fine-tunes nuclear lamina interactions of the Drosophila genome.
5: e1501321124834XueW, ZenderL, MiethingC, DickinsRA, HernandoE, KrizhanovskyV, Cordon-CardoC, LoweSW2007Senescence and tumour clearance is triggered by p53 restoration in murine liver carcinomas.
445: 656–66017251933YoungAR, NaritaM, FerreiraM, KirschnerK, SadaieM, DarotJF, TavareS, ArakawaS, ShimizuS, WattFM2009Autophagy mediates the mitotic senescence transition.
23: 798–80319279323ZhangH, PanKH, CohenSN2003Senescence-specific gene expression fingerprints reveal cell-type-dependent physical clustering of up-regulated chromosomal loci.
100: 3251–325612626749ZhangR, ChenW, AdamsPD2007Molecular dissection of formation of senescence-associated heterochromatin foci.
27: 2343–235817242207oai:pubmedcentral.nih.gov:37596962013-09-04genesdevpmc-openGenes DevGenes DevGADGenes & Development0890-93691549-5477Cold Spring Harbor Laboratory PressPMC3759696PMC375969637596962396409423964094871166010.1101/gad.217281.113Research PaperRedistribution of the Lamin B1 genomic binding profile affects rearrangement of heterochromatic domains and SAHF formation during senescenceSadaie et al.LMNB1 redistribution during senescenceSadaieMahito167SalamaRafik16CarrollThomas1TomimatsuKosuke12ChandraTamir18YoungAndrew R.J.1NaritaMasako1Pérez-ManceraPedro A.1BennettDorothy C.3ChongHeung4KimuraHiroshi5NaritaMasashi19Cancer Research UK Cambridge Institute, Li Ka Shing Centre, University of Cambridge, Cambridge CB2 0RE, United Kingdom;Research Centre for Bioscience and Technology, Tottori University, Tottori 683-8503, Japan;Biomedical Sciences Research Centre, St. George's, University of London, London SW17 0RE, United Kingdom;Cellular Pathology, Division of Biomedical Sciences, St. George's, University of London, London SW17 0RE, United Kingdom;Graduate School of Frontier Biosciences, Osaka University, Osaka 565-0871, Japan
These authors contributed equally to this work.
Present addresses: 7Graduate School of Biostudies, Kyoto University, Kyoto 606-8501, Japan;
Laboratory of Developmental Genetics and Imprinting, The Babraham Institute, Cambridge CB22 3AT, UK.
The nuclear lamina component Lamin B1 is specifically down-regulated during senescence, but the functional consequences are unclear. Narita and colleagues present evidence that Lamin B1 binding varies across the genome. Major reductions in Lamin B1 binding occur in regions associated with the repressive histone mark H3K9me3. Surprisingly, however, increased LMNB1 binding in some genomic regions is associated with elevated H3K27me3 marks and gene repression. These findings indicate that Lamin B1 redistribution may be part of a global reorganization of chromatin and gene expression in senescent cells.
Senescence is a stress-responsive form of stable cell cycle exit. Senescent cells have a distinct gene expression profile, which is often accompanied by the spatial redistribution of heterochromatin into senescence-associated heterochromatic foci (SAHFs). Studying a key component of the nuclear lamina lamin B1 (LMNB1), we report dynamic alterations in its genomic profile and their implications for SAHF formation and gene regulation during senescence. Genome-wide mapping reveals that LMNB1 is depleted during senescence, preferentially from the central regions of lamina-associated domains (LADs), which are enriched for Lys9 trimethylation on histone H3 (H3K9me3). LMNB1 knockdown facilitates the spatial relocalization of perinuclear H3K9me3-positive heterochromatin, thus promoting SAHF formation, which could be inhibited by ectopic LMNB1 expression. Furthermore, despite the global reduction in LMNB1 protein levels, LMNB1 binding increases during senescence in a small subset of gene-rich regions where H3K27me3 also increases and gene expression becomes repressed. These results suggest that LMNB1 may contribute to senescence in at least two ways due to its uneven genome-wide redistribution: first, through the spatial reorganization of chromatin and, second, through gene repression.
Lamin B1senescenceepigenetics
Cellular senescence is a state of stable cell cycle exit triggered by cellular stress, including genotoxic, oxidative, and oncogenic stress. Senescence is characterized by the activation of tumor suppressor proteins, including p53 and Rb, and limits cellular immortalization in culture. These tumor-suppressive hallmarks are reinforced by the paradoxical induction of senescence observed upon the activation or loss of certain oncogenes or tumor suppressors in both cultured cells and premalignant tumors (Serrano et al. 1997; Collado and Serrano 2006). Senescence has also been implicated in aging: Cells derived from premature aging syndrome patients become senescent more readily in culture compared with young or healthy individuals (McClintock et al. 2006). Age-dependent accumulation of senescent cells in progenitor compartments has also been observed in various tissues, limiting the regenerative capacity of those organs during aging (Sharpless and DePinho 2007).
Senescent cells exhibit dynamic gene expression changes, including some specific features such as the up-regulation of p16INK4A and components of senescence-associated secretory phenotype (SASP) as well as the stable repression of proliferation-related genes (Shelton et al. 1999). In addition, senescence is often accompanied by changes in chromatin structure, forming senescence-associated heterochromatic foci (SAHFs) (Narita et al. 2003). Through the study of senescence induced by the ectopic expression of oncogenic Ras in human fibroblasts, a number of functional and physical components of the process of SAHF formation have been identified (Chan et al. 2005; Zhang et al. 2005; Funayama et al. 2006; Narita et al. 2006; Ye et al. 2007). SAHFs are highly organized structures, where Lys9 trimethylation on histone H3 (H3K9me3; a constitutive heterochromatin marker) forms the core, which is surrounded by a layer of H3K27me3 (a facultative heterochromatin marker). These repressive layers are clearly separated from the outer transcriptionally active layer, supporting the idea that SAHF formation may contribute to gene expression profile stability for both active and repressive genes, although the direct relationship between SAHFs and gene regulation is still elusive (Chandra and Narita 2013). Despite the striking structural alteration in chromatin, the global landscapes of the repressive histone marks are highly static during SAHF formation, with only localized alterations in some genic regions, thus suggesting that SAHFs are formed through a spatial repositioning of repressively marked chromatin (Chandra et al. 2012).
The nuclear lamina is a filamentous structure, forming a scaffold underneath the inner nuclear membrane. In addition to its role in maintaining nuclear structural integrity, it has been implicated in the nuclear positioning of chromatin and transcription regulation (Dechat et al. 2010; Kind and Van Steensel 2010; Peric-Hupkes et al. 2010; Dittmer and Misteli 2011; Burke and Stewart 2012). The major structural components of the lamina in mammals are the intermediate filament proteins Lamin A/C (LMNA/C), LMNB1, and LMNB2. Genome-wide mapping of LMNB1 identified large lamina-associated domains (LADs), which are devoid of active histone marks and enriched for repressive marks (Pickersgill et al. 2006; Guelen et al. 2008). LADs are relatively gene-poor, and those genes included in LADs are generally silenced in both humans and Drosophila. During mouse embryonic stem cell (ESC) differentiation, the overall LAD pattern remains largely static. However, more localized changes in lamina–chromatin association are also observed, and their correlation with gene expression reinforces the idea that LADs provide a transcriptionally repressive environment (Peric-Hupkes et al. 2010).
Recent reports have shown that the global LMNB1 level is altered during senescence (Shimi et al. 2011; Barascu et al. 2012; Freund et al. 2012; Dreesen et al. 2013). However, it is unclear how the LMNB1 alterations affect the dynamic alterations seen in chromatin structure and the gene expression profile during senescence. Here we show that the alterations in LMNB1 in senescent human diploid fibroblasts (HDFs) are far from uniform along the genome. Despite the global down-regulation in LMNB1, LMNB1 binding is reduced mainly in H3K9me3-enriched regions, and this reduction is correlated with the spatial repositioning of H3K9me3-enriched chromatin but not with gene expression changes during senescence. Furthermore, we also found de novo gains in LMNB1 binding in small sections of the genome, which are correlated with increased H3K27me3 and gene repression. Our data suggest that the alterations in LMNB1–genome binding have different roles between regions with different chromatin states.
ResultsLMNB1 genomic binding is redistributed during senescence
We first examined the senescence-associated alteration in the LMNB1 level using IMR90 HDFs. Consistent with previous reports (Shimi et al. 2011; Freund et al. 2012; Dreesen et al. 2013), total LMNB1 protein levels were down-regulated during oncogenic H-Ras-induced senescence (RIS) and replicative exhaustion senescence but not in quiescent cells (induced by serum starvation) or cells immortalized through adenoviral oncoprotein E1A expression (Fig. 1A; Supplemental Fig. S1A,B). The chromatin-bound LMNB1 level was also reduced in RIS cells (Supplemental Fig. S1C).
Uneven alterations in LMNB1 genomic profile during RIS. (A) Western blotting with the indicated antibodies for whole-cell lysates collected over 10 consecutive days after estrogen receptor (ER)-tagged H-RasG12V (ER:Ras) induction by adding 100 nM 4-hydroxytamoxifen (4-OHT) in IMR90 cells. (B) A smoothhist2D plot showing the global reduction of LMNB1 during RIS. Normalized ChIP-seq counts of LMNB1 in 100,000-segmented regions in growing versus RIS cells are plotted (see the Materials and Methods for normalization). Only fragments that are LMNB1-positive in either condition were included. (C) A representative whole-chromosome browser snapshot overlap of normalized ChIP-seq reads from growing and RIS IMR90 conditions. (D) Pie chart describing six nonoverlapping classes of genomic regions based on differential binding events of LMNB1 between growing and RIS. Numbers represent percentage of the genome in each class. (“RIS-only” or “Grow-only”) LMNB1 is positive only in one condition; (“RIS-up” or “RIS-down”) LMNB1 is positive in both conditions, and its levels are significantly up-regulated or down-regulated, respectively, in RIS cells; (“Unchanged”) LMNB1 is positive in both conditions with no significant differences. (E) Genomic features of each class as defined in D. The features are represented as a percentage of total base pairs in each class. (Gene) Protein-coding gene bodies; (LINE) long interspersed nuclear element; (SINE) short interspersed nuclear element; (CpG) CpG islands; (Promoter) −1.5 to +0.5 kb of transcription start sites.
To analyze how the genome-wide LMNB1 DNA binding profile changes during senescence, we performed three replicates of LMNB1 chromatin immunoprecipitation (ChIP) followed by genome-wide parallel sequencing (ChIP-seq; mean Spearman correlation coefficient rs = 0.9) in growing and RIS IMR90 cells. Since the global level of LMNB1 protein in RIS was substantially reduced, the LMNB1 ChIP-seq libraries generated from RIS cells suffered from lower complexity, which was taken into account through intercondition and intracondition normalization for all read-based analyses (Supplemental Material; Supplemental Fig. S2). The similarity between our ChIP-seq results (growing cells) and the LADs identified in different HDFs by the DamID (DNA adenine methyltransferase identification) technique using a Dam-LMNB1 fusion protein (Guelen et al. 2008) was high (rs = 0.7), and >80% of DamID-defined LADs were detected by ChIP-seq (Supplemental Table S1; Supplemental Fig. S3, see the legends for details). Consistent with the global down-regulation in the LMNB1 protein level, LMNB1-binding events were reduced overall during RIS (Fig. 1B,C). Nevertheless, we identified a substantial number of LADs in RIS cells (Supplemental Table S1). Indeed, in some regions, LMNB1 binding was even increased. The increase of LMNB1 binding was confirmed for several regions by independent ChIP-qPCR experiments, thereby supporting our normalization method for the LMNB1 ChIP-seq data across the two conditions (Supplemental Fig. S4). Thus, despite the global reduction in LMNB1 protein levels, the alterations in the LMNB1 binding profile during RIS were not uniform across the genome.
LMNB1 is preferentially reduced from H3K9me3 regions
To investigate the changes in the LMNB1 binding profile further, we divided the genome into six classes according to the differential LMNB1 binding between growing and RIS (Fig. 1D, see the legend for details). LMNB1-decreased regions (“Grow-only” + “RIS-down”) contained a relatively low frequency of genes (Grow-only in particular), CpG islands, and promoters (Fig. 1E; Supplemental Fig. S5 shows three representative examples of genome browser shots of LMNB1 ChIP-seq from “RIS-only” regions that overlap with genes). Consistently, chromosome-wide analyses revealed that the extent of LMNB1 reduction varied between chromosomes with the more prominent reductions in gene-poor chromosomes (Supplemental Fig. S6A).
Next, to examine the genome-wide dynamic association between LMNB1 and repressive histone marks, we conducted principal component analyses (PCAs) on the ChIP-seq reads for LMNB1, H3K9me3, and H3K27me3 in growing and RIS cells. The profiles of H3K9me3 and H3K27me3 were highly distinct, and, consistent with results from our previous study (Chandra et al. 2012), the profile of each mark was globally unaltered during RIS. The DNA binding profile of LMNB1 was altered during RIS, and its orientation in the PCA was shifted away from the projection of H3K9me3 to that of H3K27me3, suggesting that the nature of the differential LMNB1 binding may be distinct between H3K9me3 and H3K27me3 regions (Supplemental Fig. S6B). To confirm this, we performed PCAs for both the H3K9me3 and H3K27me3 reads in growing and RIS conditions and colored them differentially based on the nature of the LMNB1 alteration. The LMNB1-decreased and LMNB1-increased/unchanged regions were biased toward the H3K9me3 and H3K27me3 PCA projections, respectively (Fig. 2A). Consistently, H3K9me3 occupancy was relatively high in LMNB1-decreased regions (Grow > RIS), whereas LMNB1-increased regions (Grow < RIS) tended to be more H3K27me3-enriched (Supplemental Fig. S6C). These data indicate that LMNB1 binding is primarily reduced from H3K9me3-positive regions during RIS and that LMNB1 gain is more likely to occur in H3K27me3-positive regions.
Preferential reduction of LMNB1 from H3K9me3 regions. (A) PCA of ChIP-seq data for H3K9me3 and H3K27me3 using a 500-kb window size in growing and RIS conditions. LMNB1-positive segments in either condition are colored based on the classes defined in Figure 1D. (B) Profiles of the indicated chromatin features around LADs in growing cells. To align LADs, all LADs were segmented using a moving window of 1% of the domain size. LMNB1 and other ChIP-seq profiles were overlaid on the segmented windows using median reads per kilobase per million mapped (RPKM). (C, top) Genome browser representations of enrichment profiles of LMNB1 and indicated histone marks. The gray shading represents the LAD. (Middle) A standardized score (Z-score) is used to show the signature of each histone mark overlapping the LAD. (Bottom) The log ratio of LMNB1 read counts between RIS and growing conditions around scaled LADs is shown. To avoid any contamination of the outside of scaled LADs with neighboring LADs, only short (10-kb) leading and tailing regions were added. (D) Profiles of histone marks overlapping aligned LMNB1-increased (Grow < RIS in Fig. 1D) regions (the gray shading) in the RIS condition. Domain alignment was performed such that all domains were orientated 5′ to 3′; a distance equal to the domain width was added upstream of and downstream from the region, and the whole region was cut into equal 100 percentiles, where the LMNB1-enriched regions are now the mid-tertile of the data.
We next examined the alteration in LMNB1 binding in the H3K9me3 and H3K27me3-enriched regions in the context of LADs. It has been shown that LADs are enriched for H3K27me3 at their edges, whereas H3K9me2 occupies the whole LAD (Guelen et al. 2008). Our data agree with this and, in addition, we found a predominant enrichment of H3K9me3 in the central region of LADs in growing conditions (Fig. 2B). Consistent with the PCA, the average reduction in LMNB1 was more prominent in the H3K9me3-enriched central regions than it was in the flanking H3K27me3 regions (Fig. 2C). A focused analysis of the LMNB1-increased regions (“RIS-up” + “RIS-only”) revealed that gains in LMNB1 were closely associated with H3K27me3 regions (Fig. 2D). Thus, there is effectively an element of redistribution in LMNB1 together with a preferential reduction in LMNB1 from H3K9me3 regions despite the global reduction of LMNB1 during RIS.
Perinuclear heterochromatin is reduced during senescence
In human diploid cells, an array of H3K9me3 foci are often observed in perinuclear regions, whereas H3K27me3 is localized to the interior of nuclei (Chandra et al. 2008, 2012). In addition, we showed recently that repressively marked chromatin is spatially repositioned during RIS to form SAHFs (Chandra et al. 2012). We found that the thickness of perinuclear electron-dense heterochromatin (by electron microscopy) and the number of H3K9me3 foci (by confocal microscopy) adjacent to the nuclear envelope were significantly diminished in RIS cells (Fig. 3A,B). Indeed, although some SAHFs contacted the nuclear envelop, the major part of the H3K9me3 core of SAHFs was located in the interior of the nucleus (Supplemental Movies S1, S2). The perinuclear heterochromatin reduction was also observed in replicative senescent IMR90 cells and RIS BJ cells (Supplemental Fig. S7A–D). In addition, oncogenic BRAF-induced senescent human melanocytes also exhibited SAHF formation as well as a reduction in perinuclear H3K9me3 foci and LMNB1 levels, although the LMNB1 reduction in senescent melanocytes was less pronounced than in HDFs (Supplemental Fig. S7E–I). We hypothesized that the reduction of LMNB1, particularly from H3K9me3 regions, during RIS facilitates the spatial redistribution of heterochromatin, providing a “pro-SAHF” nuclear environment. Although the reduction in perinuclear H3K9me3 foci was observed in both SAHF-positive and SAHF-negative RIS cells, it was more prominent in SAHF-positive cells (Fig. 3B). In addition, in single-cell analyses using laser-scanning cytometry (LSC) on IMR90 cells, we mapped the majority of SAHF-positive nuclei to the LMNB1-low RIS population (Fig. 3C). These data indicate that the reduction of both LMNB1 and perinuclear heterochromatin is closely correlated with SAHF formation.
Reduction of perinuclear heterochromatin during senescence. (A) Representative electron micrographs of growing (G) and RIS IMR90 cells. Areas indicated by the rectangles are magnified. (Right) Thickness of high-electron density perinuclear areas was measured from the inner nuclear membrane using a superimposed grid and ImageJ. The data represent the average thickness of five to 17 points per nucleus. Black bars represent mean. Representative plot of two independent experiments is shown. (***) P < 0.001. (B) Confocal images of the equatorial section for the indicated antibodies in growing and RIS cells with (+) or without (−) obvious SAHFs. The number of H3K9me3 foci adjacent to the peripheral LMNA/C signals per nucleus (n ≥ 25) was counted. Black bars indicate mean. Representative plot of three independent experiments is shown. (***) P < 0.001. (C) LMNB1 intensity per nucleus was quantified with a laser-scanning cytometer (n > 1000). (Top) Averaged histogram of three independent experiments is shown. (Bottom) RIS cells with low (RIS-1) and high (RIS-2) LMNB1 levels were gated, and SAHFs were assessed in each gated population.
To confirm that specific genomic regions do relocalize upon LMNB1 reduction during RIS, we used DNA fluorescence in situ hybridization (FISH). We chose seven bacterial artificial chromosome (BAC) probes mapped to genomic regions where LMNB1 became reduced to various degrees during RIS (Fig. 4A). These were mostly H3K9me3-enriched regions. Consistent with previous reports (Guelen et al. 2008; Haferkamp et al. 2009), we detected a correlation between the LMNB1 density associated with a particular DNA region and its perinuclear localization in combined samples from growing and RIS conditions. In addition, each probe tended to show a more perinuclear pattern in growing conditions relative to RIS (Fig. 4A). These results suggest that the global reduction in LMNB1 levels, particularly in H3K9me3 regions, during RIS is correlated with the release of heterochromatin from the nuclear perimeter.
Correlation between LMNB1 reduction and intranuclear location of genomic regions. (A) Projections of confocal FISH images. (Left) FISH signals of the BAC clone 784D7 are shown. (Right) Mean percentage of cells that displayed at least one FISH signal at the nuclear periphery (n = 3) is plotted against mean RPKM of LMNB1 ChIP-seq within regions covered by each probe in each condition (n = 3). (B) Western blot analysis for the indicated antibodies in IMR90 cells expressing three different sh-LMNB1. (C) Confocal images of LMNB1-depleted cells stained with the indicated antibodies. (Left) Inset numbers represent percentage of cells with SAHFs (mean ± SEM). (Right) The number of perinuclear H3K9me3 foci was assessed as in Figure 3B. Two independent experiments for three independent shRNAs are shown. (D) Combination of LMNB1 depletion and ectopic expression of HMGA promotes SAHF formation and senescence. sh-LMNB1-1 or the control miR30 vector (−) and either HA-HMGA1 (A1), HA-HMGA2 (A2), or the control vector (−) were coexpressed in IMR90 cells. Cells were assessed for protein expression, SAHFs, BrdU incorporation, and SA-β-galactosidase activity. Values represent mean ± SEM from six experiments. (E) Ectopic expression of LMNB1 inhibits SAHF formation. LMNB1 or control vector was retrovirally transduced to IMR90 cells expressing ER:Ras. Cells were assessed for protein expression, SAHFs, and BrdU incorporation at the indicated time points after ER:Ras induction. (*) P < 0.05; (**) P < 0.01; (***) P < 0.001.
Modulation of the global LMNB1 level affects SAHF formation
To examine the direct consequences of LMNB1 reduction in normal HDFs, we generated shRNAs targeting LMNB1 using a miR30-based design (Fig. 4B; Silva et al. 2005). LMNB1 depletion in IMR90 and BJ cells resulted in a significant reduction in the number of perinuclear H3K9me3 foci, although it induced only a modest senescence phenotype (including SAHFs) (Fig. 4C; Supplemental Fig. S8A). We showed previously that the HMGA proteins are required for SAHF formation as architectural components (Narita et al. 2006). Thus, we hypothesized that the spatial removal of perinuclear heterochromatin together with the accumulation of these architectural components of SAHFs might cooperatively contribute to SAHF formation. Indeed, LMNB1 and HMGA2 protein levels were inversely correlated during the time course of RIS (Fig. 1A). To test this, we stably expressed sh-LMNB1 and/or HA-tagged HMGA proteins in IMR90 cells. Consistent with results from our previous study, overexpression of the HMGA proteins alone using a weak promoter showed a minimal effect, similar in extent to that of sh-LMNB1 alone (Fig. 4D; Narita et al. 2006). However, the coexpression of both sh-LMNB1 and HA-HMGA1 (or HA-HMGA2) more efficiently induced SAHFs (Fig. 4D; Supplemental Fig. S8B). Similar results were also obtained using BJ cells (Supplemental Fig. S8C). We next examined whether ectopic expression of LMNB1 affects SAHF formation. The overall effect of LMNB1 overexpression on senescence has been controversial: LMNB1 overexpression was originally shown to increase the proliferation rate and delay the onset of senescence in HDFs (Shimi et al. 2011), but more recent studies suggested that it actually induces senescence (Barascu et al. 2012; Dreesen et al. 2013). When we ectopically expressed LMNB1 using the weak promoter, we managed to express it at a level comparable with endogenous LMNB1 in growing cells (Fig. 4E). In these conditions, while ectopic LMNB1 induced an irregular nuclear shape in both growing and RIS conditions, the impact on cell proliferation was marginal (Fig. 4E). Nevertheless, SAHF formation was partially, yet significantly, reduced by the ectopic expression of LMNB1 (Fig. 4E; Supplemental Fig. S8D). These data suggest that LMNB1 reduction during senescence induces the spatial repositioning of perinuclear heterochromatin and thus, together with architectural proteins, facilitates SAHF formation.
De novo LMNB1 binding is correlated with gene repression
It has been suggested that the association of genes with the nuclear lamina is involved in their silencing, but this repressive effect appears to depend on the genes and their chromatin context (Kind and Van Steensel 2010). Thus, we next examined whether there was any relationship between differential LMNB1 binding on genic regions and the differential expression of those genes during RIS. A genome-wide analysis failed to show any global correlation (Fig. 5A). This is consistent with a recent study showing that gene expression changes due to the double knockout of Lmnb1 and Lmnb2 in mouse trophectoderm and ESCs are not correlated with the binding of Lmnb1 to the promoters of those genes (Kim et al. 2011). Thus, our data suggest that expression of only a small subset of genes, if any, is correlated with the alteration of regional LMNB1 occupancy during RIS. To identify such genes, we took advantage of gene sets included in the regions defined in Figure 1D based on the differential binding of LMNB1 and performed a gene set enrichment analysis (GSEA) (Subramanian et al. 2005). This technique allowed us to test for a nonrandom distribution of an a priori defined subset of genes within a ranked expression data set. Genes within the RIS-only class were significantly down-regulated during RIS (Fig. 5B; Supplemental Table 2), while genes in all other classes failed to show significant association with the phenotype (Supplemental Fig. S9). High-scoring “leading-edge” genes (Subramanian et al. 2007) from the RIS-only class as well as other classes were subjected to a gene ontology (GO) analysis, revealing a distinct GO feature in the RIS-only class with prominent enrichment in cell cycle GO categories (Fig. 5C; see also Supplemental Fig. S5, where the examples overlap with leading-edge genes). In contrast, other classes mostly share a ‘”membrane” signature, which we previously showed is strongly associated with RIS, suggesting that this correlation is likely to be nonspecific (Narita et al. 2011). Furthermore, the H3K27me3 ratio between RIS and growing cells (log2 RIS/Grow) was elevated within the gene bodies of the leading-edge genes, whereas the same ratio for H3K36me3 dropped (Fig. 5D; Supplemental Fig. S10). These data suggest that the potential role of LMNB1 in gene repression is limited to a specific subset of genes representing a small fraction of the genome, wherein LMNB1 binding is paradoxically increased during RIS.
Gene repression in de novo LMNB1 regions in RIS cells. (A) Scatter plot comparing log2 fold change of gene expression and LMNB1 ChIP-seq reads on the gene body during RIS. (B) GSEA profile of genes within the classes defined in Figure 1D. Genes are ranked into an ordered list based on differential expression between growing and RIS cells. Genes that are repressed during RIS are ranked toward the right (blue). The rectangle indicates the “leading-edge” subset of genes. (C) Heat map showing GO analysis in genes in the leading-edge subsets from GSEA analyses of genes included in the indicated classes. (D) Log2 ratios of normalized read counts for the indicated ChIP-seq data between RIS and growing conditions around scaled gene bodies associated with the RIS-only region are shown for the indicated gene sets. Gene bodies were aligned 5′ to 3′ according to transcription direction. (B–D) Note that genes are included when at least 50% of gene bodies overlap with RIS-only regions. (E) Localization of selected leading-edge genes by FISH in growing and RIS cells. Mean percentage of cells that displayed at least one FISH signal at the nuclear periphery (n = 3) is plotted against mean RPKM of LMNB1 ChIP-seq within regions covered by each probe in each condition (n = 3). See also Supplemental Figure S11.
Finally, we performed additional DNA FISH experiments using probes that overlapped with the RIS-only region leading-edge genes, where H3K27me3 was more prominent than H3K9me3 (Supplemental Fig. S11). Similar to the H3K27me3 immunolabeling pattern (Chandra et al. 2012), these regions were mostly located in the interior of the nucleus in both growing and RIS cells (Fig. 5E). In contrast to the first probe set, however, most RIS-only probes exhibited a shift toward the nuclear periphery, although the dynamic range was small (Fig. 5E; see Supplemental Fig. S11 for a better comparison between the two probe sets). These data further suggest that the regional gains in LMNB1 binding in part facilitate a perinuclear relocalization and thus may contribute to gene silencing.
Discussion
Increasing evidence has shown that LMNB1 global levels are altered during senescence, but the directionality of the change and its impact on the senescence phenotype are unclear (Shimi et al. 2011; Barascu et al. 2012; Freund et al. 2012; Dreesen et al. 2013). In this study, we confirmed that LMNB1 expression levels are down-regulated during senescence using human fibroblasts and melanocytes. At the same time, our data revealed that the alterations in LMNB1 binding to chromatin are uneven throughout the genome. This uneven redistribution is not random, since the major reductions in LMNB1 occur mostly in H3K9me3-enriched regions. In addition, in a small but substantial portion of the genome, particularly in H3K27me3-enriched regions, gains in LMNB1 binding occur during senescence. The global reduction and regional gains in LMNB1 binding are correlated with SAHF formation and gene repression, respectively; thus, it is possible that these two parallel processes collectively contribute to the senescence phenotype.
High-order chromatin structure alteration is a complex and perhaps multimodular process. The “pro-SAHF” nuclear environment achieved through the preferential reduction of LMNB1 from H3K9me3 regions appears to be independent of the architectural aspect of SAHF formation. Our findings therefore provide an additional element in the machinery of SAHF formation. Considering the modest effects of both the depletion and overexpression of LMNB1 on the process of SAHF formation, it is clear that other factors are involved in this process. It also remains to be elucidated how such an organized redistribution of LMNB1 during senescence is regulated. Interestingly, it has recently been shown that nuclear envelope proteins and their associated factors are involved in heterochromatin tethering to the nuclear periphery. For example, Lap2β, a nuclear envelope protein, forms a complex with the transcription repressor cKrox (ThPOK) and the histone deacetylase HDAC3 and tethers LADs to the nuclear lamina in a sequence-dependent manner (Zullo et al. 2012). In addition, Lamin B receptor (LBR), another nuclear envelope protein, has been shown to contribute to tethering peripheral heterochromatin: The absence of both LBR and LMNA/C leads to the loss of peripheral heterochromatin (Solovei et al. 2013). LBR preferentially binds to B-type lamins as well as heterochromatin protein 1 (HP1), a H3K9me3-binding protein (Ye and Worman 1996). Although the dynamic status of these factors during senescence is not known, it is possible that, together with the global regulation of the LMNB1 level itself, modulation of more specific mechanisms involving genomic sequence- or histone mark-dependent chromatin tethering to the nuclear lamina contribute to the complex and uneven redistribution of LADs. Regardless of the nature of the mechanism, our results imply that the differential affinity of LMNB1 for regions with different genomic and chromatin features may itself be a mechanism that allows for the fine-tuning of LMNB1 functions. It has also been shown that LMNB1 defects affect interphase chromosomal positioning and gene expression (Malhas et al. 2007) and that the nuclear location of genes can change in a disease- and loci-specific manner during tumorigenesis (Meaburn and Misteli 2008; Meaburn et al. 2009). Thus, the regulation of the balance between alterations in LMNB1 on both global and regional levels might have wider implications for the development of disease.
Materials and methodsAntibodies
Antibodies used for Western blotting were as follows: LMNA/C (sc-7292, Santa Cruz Biotechnology), LMNB1 (ab16048, Abcam), Cyclin A (C4710, Sigma), p16 (sc-759, Santa Cruz Biotechnology), HMGA2 (sc-30223, Santa Cruz Biotechnology), H-Ras (OP23, Calbiochem), β-actin (A5441, Sigma), Rb (9309, Cell Signaling), HA (3724, Cell Signaling), and BRAF (F7, sc-5258, Santa Cruz Biotechnolgy).
Cell culture and gene transfer
IMR90 (female, embryonic lung-derived) and BJ (male, neonatal foreskin-derived) fibroblasts (American Type Culture Collection) were cultured in phenol-red-free DMEM with 10% FBS under 5% oxygen. The retroviral and lentiviral gene transfer was carried out as described (Narita et al. 2006) using the vectors detailed in the Supplemental Material. RIS was induced by retrovirus-mediated expression of either H-RasG12V or H-RasG12V fused to the estrogen receptor (ER) ligand-binding domain (ER:Ras) (Young et al. 2009). Primary human melanocytes (Tissue Bank, Yale SPORE in Skin Cancer, Yale University) were grown in 254 medium, including a melanocyte growth supplement (Cascade Biologics/Life Technologies), under 5% oxygen. Senescence was induced by lentivirus-mediated expression of BRAFV600E (Michaloglou et al. 2005).
Immunofluorescence, LSC, electron microscopy, and senescence assays
Immunofluorescence and senescence-associated β-galactosidase were performed as described (Narita et al. 2003). The primary antibodies used were H3K9me3 (07-523, Millipore), H3K27me3 (CMA323) (Hayashi-Takanaka et al. 2011), LMNA/C (sc-7292, Santa Cruz Biotechnology), LMNB1 (ab16048, Abcam; sc-6216, Santa Cruz Biotechnology), HA (H9658, Sigma), and BrdU (555627, Becton Dickinson). For quantitative imaging, LSC (Compucyte iCys) was used. Electron microscopy was performed as reported (Motskin et al. 2011).
ChIP-seq
ChIP was performed as described (Chandra et al. 2012). For LMNB1 and control IgG ChIP-seq, three independent biological replicates were performed with matched inputs for each condition using a LMNB1 antibody (ab16048, Abcam) and then anti-rabbit IgG-conjugated Dynabeads (112.04D, Invitrogen). All ChIP-seq data from this study have been submitted to the Gene Expression Omnibus database (GSE49341). All other ChIP-seq data came from our previous study and were obtained using the same conditions (Chandra et al. 2012), except for H3K79me1 (Hawkins et al. 2010). CTCF-binding regions were obtained from published data (Kim et al. 2007). For detailed ChIP-seq data analyses, see the Supplemental Material.
Expression microarray
The gene expression data sets on HumanHT-12 version 3.0 Expression BeadChip (Illumina) arrays were those published previously (Chandra et al. 2012). For detailed data analyses, see the Supplemental Material.
DNA-FISH
DNA-FISH was performed as described elsewhere (Chandra et al. 2012).
Acknowledgments
We thank members of the Narita laboratory for helpful discussions; L. Blackburn for editing; CI core Genomics, Bioinformatics, and Microscopy (H. Zecchini and A. Schreiner) facilities for technical support; J. Skepper for electron microscopy; and D. Peeper, P. Adams, and M. Soengas for reagents and technical advice. This work was supported by the University of Cambridge; Cancer Research UK; Hutchison Whampoa; the Human Frontier Science Program (M.N. and R.S.); Cancer Research UK consortium grant, Senectus Therapeutics (M.N., D.C.B., and M.S.); and fellowships from the Uehara Memorial Foundation (M.S.) and the Japan Society for the Promotion of Science (K.T.).
Supplemental material is available for this article.
Article is online at http://www.genesdev.org/cgi/doi/10.1101/gad.217281.113.
Here, Mohr and colleagues establish a role for the poly(A)-binding protein (PABP) repressor Paip2 in viral infection. The investigators find that human cytomegalovirus (HCMV) infection causes the up-regulation of Paip2 as well as PABP. The data indicate that Paip2 accumulation represents an innate host response to counteract the virus-induced increase in PABP abundance, limit the assembly of translation initiation factor complexes, and restrict viral growth. Paip2 thus plays a significant role in an innate defense mechanism to restrict viral protein synthesis and replication.
The capacity of polyadenylate-binding protein PABPC1 (PABP1) to stimulate translation is regulated by its repressor, Paip2. Paradoxically, while PABP accumulation promotes human cytomegalovirus (HCMV) protein synthesis, we show that this is accompanied by an analogous increase in the abundance of Paip2 and EDD1, an E3 ubiquitin ligase that destabilizes Paip2. Coordinate control of PABP1, Paip2, and EDD1 required the virus-encoded UL38 mTORC1 activator and resulted in augmented Paip2 synthesis, stability, and association with PABP1. Paip2 synthesis also increased following serum stimulation of uninfected normal fibroblasts, suggesting that this coregulation may play a role in how uninfected cells respond to stress. Significantly, Paip2 accumulation was dependent on PABP accrual, as preventing PABP1 accumulation suppressed viral replication and inhibited the corresponding Paip2 increase. Furthermore, depleting Paip2 restored the ability of infected cells to assemble the translation initiation factor eIF4F, promoting viral protein synthesis and replication without increasing PABP1. This establishes a new role for the cellular PABP1 inhibitor Paip2 as an innate defense that restricts viral protein synthesis and replication. Moreover, it illustrates how a stress-induced rise in PABP1 triggered by virus infection can counter and surpass a corresponding increase in Paip2 abundance and stability.
The control of gene expression at the level of mRNA translation facilitates swift responses to an ever-changing environment and plays a fundamental role in the response to stress and viral infection (Mohr and Sonenberg 2012). To replicate productively, viruses must seize control of the cellular protein synthesis machinery and the complex networks that regulate its activity. At the same time, host defenses use translational control mechanisms to respond to microbial infection and mobilize innate host defenses (Walsh and Mohr 2011). While some of these responses can affect specific mRNAs, others can globally impact the translational capacity of the cell (Mohr and Sonenberg 2012; Walsh et al. 2013). As such, the steady-state level of translation initiation factors and repressors are poised to play a powerful role in controlling translation in infected and uninfected cells.
Unlike many viruses that impair host mRNA translation or inactivate cellular translation factors, host mRNA translation proceeds in human cytomegalovirus (HCMV)-infected cells (Stinski 1977). To ensure that viral mRNAs (which contain 7-methyl guanosine [m7G]-capped 5′ and 3′ polyadenylated termini) effectively compete for limiting translation factors, HCMV infection triggers an increase in the intracellular concentration of key initiation factors, including the cellular polyadenylate-binding protein PABPC1 (PABP1) and the multisubunit cap-binding complex eIF4F (Kudchodkar et al. 2004; Isler et al. 2005; Walsh et al. 2005; Perez et al. 2011). Comprised of the cap-binding protein eIF4E and the RNA helicase eIF4A bound to the large molecular scaffold eIF4G, eIF4F complex assembly is regulated on the m7GTP cap structure, facilitating 40S ribosome subunit loading onto the mRNA 5′ end (Jackson et al. 2010). PABP1 bound to the mRNA 3′ poly(A) tail stimulates translation by binding to eIF4G and in effect juxtaposes the mRNA 3′ end with the eIF4F complex assembled on the 5′ end, forming a closed loop or circular topology (Kahvejian et al. 2001). While the level of mRNAs encoding the individual eIF4F subunits increases upon infection, PABP1 abundance is controlled at the level of translation and is dependent on the virus-encoded mTORC1 activator encoded by the UL38 gene. Importantly, preventing the HCMV-induced rise in PABP1 levels interferes with eIF4F assembly, viral protein production, and viral replication (McKinney et al. 2012). Thus, successfully increasing the concentration of the host PABP1 represents an important aspect of the productive viral life cycle.
Multiple distinct regulatory pathways maintain PABP1 homeostasis in cells. An adenine-rich autoregulatory region in the PABP1 mRNA 5′ untranslated region (UTR) reportedly represses its synthesis (Wu and Bag 1998; Patel et al. 2005). As a member of an mRNA family that contains a 5′-terminal oligopyrimidine (TOP) region in the 5′ UTR, PABP1 synthesis is coordinately regulated at the level of translation, together with other host ribosomal protein genes and translation factors, and is stimulated in response to mitogenic, growth, and nutritional stimuli (Hornstein et al. 1999a). TOP mRNA translation is also stimulated by HCMV infection (McKinney et al. 2012). PABP1 activity, however, is controlled through its association with interacting protein partners (Derry et al. 2006). Besides associating with eIF4G, PABP1 interacts with Paip1 (an eIF4G-like molecule that stimulates translation) and Paip2 (a potent translation repressor) (Craig et al. 1998; Khaleghpour et al. 2001a; Roy et al. 2002; Martineau et al. 2008).
Paip2 inhibits translation by displacing PABP1 from the poly(A) tail at the mRNA 3′ end and preventing PABP1 from binding eIF4G (Khaleghpour et al. 2001b; Karim et al. 2006). To ensure proper control of gene expression, PABP1 and its repressor, Paip2, are tightly coregulated, as PABP1 depletion by RNAi triggers a corresponding reduction in Paip2. This requires binding of Paip2 to the E3 ubiquitin ligase EDD1, as opposed to PABP1, and illustrates how differential binding to EDD1 and PABP1 control Paip2 levels (Yoshida et al. 2006). A role for the converse scenario—namely, how Paip2 might respond to elevated PABP1 levels—has not been investigated and remains unknown, in part due to the difficulty in overexpressing PABP (Wormington et al. 1996; Wu and Bag 1998; Hornstein et al. 1999b; Ma et al. 2006). Here we show that EDD1 and newly synthesized Paip2 levels surprisingly increased in response to PABP1 accumulation induced upon infection with HCMV. This required expression of the virus-encoded UL38 mTORC1 activator. Paip2 stability also increased in HCMV-infected cells, and Paip2 was detected bound to PABP1 even though EDD1 levels were elevated. Significantly, besides suppressing eIF4F assembly, viral protein production, and replication, preventing the HCMV-induced increase in PABP1 abundance inhibited Paip2 accumulation. Moreover, depleting Paip2 restored virus protein production and replication in cells where the PABP increase was blocked by RNAi. Thus, Paip2 depletion obviates the need to increase PABP1 abundance in HCMV-infected cells. Taken together, this demonstrates that while increasing PABP and Paip2 levels are both dependent on mTORC1 signaling, Paip2 abundance specifically increases in response to PABP accumulation. This not only suggests that the coordinate increase in Paip2 synthesis and stability naturally limits PABP1 activity, but establishes that it can function as a previously unrecognized innate host defense with the capacity to restrict viral protein synthesis and replication. Sufficiently increasing cellular PABP1 levels allows HCMV to overcome this host restriction and effectively counter the potential negative impact of the Paip2 repressor. This is the first example illustrating how a virus can manipulate the PABP1/Paip2 homeostatic switch by increasing PABP1 levels to facilitate productive virus growth.
ResultsPABP1 accumulation induced by HCMV infection is accompanied by a corresponding increase in Paip2 and EDD1 abundance
To investigate how PABP1 function is influenced by its endogenous cellular inhibitor, Paip2, in HCMV-infected cells, we first compared the overall abundance of Paip2 mRNA and protein in mock-infected versus HCMV-infected normal human diploid fibroblasts (NHDFs). Given that Paip2 inhibits PABP1 (Fig. 1A) and that the virus-induced increase in PABP1 levels stimulates HCMV replication, it was surprising that Paip2 and EDD1 both accumulated along with PABP1 between 30 and 48 h post-infection (hpi) (Fig. 1B). By 72 hpi, overall levels of Paip2 and EDD1 rose between fourfold and fivefold, and PABP abundance increased between sevenfold and eightfold (Supplemental Fig. S1). The HCMV-induced increase in EDD1 protein abundance was not accompanied by a detectable corresponding increase in steady-state mRNA levels, whereas the increase in Paip2 protein was only accompanied by a modest increase in mRNA (Fig. 1C), suggesting an underlying post-transcriptional control strategy similar to what we previously defined for PABP1 (Perez et al. 2011). Importantly, while reducing PABP1 abundance has been shown to trigger a corresponding decrease in Paip2 and EDD1 levels (Yoshida et al. 2006), this is the first demonstration that these factors are coordinately up-regulated. Since PABP1 function is controlled by the relative abundance of Paip2 and PABP1, the mechanisms underlying the coordinate regulation of PABP1, Paip2, and EDD1 in HCMV-infected cells were investigated.
Accumulation of the cellular PABP repressor Paip2; its negative regulator, EDD1; and PABP1 in HCMV-infected cells. (A) Cartoon illustrating control of PABP homeostasis in response to serum stimulation or HCMV infection. PABP is encoded by an mRNA containing a TOP element and is repressed in growth-arrested cells (Hornstein et al. 1999a). Serum stimulation or HCMV infection activates mTORC1 and stimulates new PABP synthesis, which in turn promotes eIF4F assembly and 40S ribosome recruitment to the mRNA 5′ end (Perez et al. 2011; McKinney et al. 2012). The PABP repressor Paip2 regulates PABP1 function by (1) preventing its association with the mRNA poly(A) tail and (2) binding to PABP1 and preventing its association with eIF4G, both of which repress translation (Derry et al. 2006). Excess Paip2 that is not bound to PABP interacts with the E3 ubiquitin ligase EDD1, which targets Paip2 for proteolysis (Yoshida et al. 2006). While PABP depletion stimulates Paip2 degradation, how Paip2 responds to PABP accumulation is unknown and is depicted as a question mark. (B) Asynchronous NHDFs were mock-infected (0 hpi) or infected with HCMV at a multiplicity of 3. At the indicated times post-infection, total protein was isolated, fractionated by SDS-PAGE, and analyzed by immunoblotting with the indicated antisera. (C) As in B except at the indicated times post-infection, total RNA was isolated through Trizol extraction and subjected to RT-qPCR using the indicated primer sets. Each reaction product was normalized to the signal obtained using 18S rRNA, and the fold induction upon infection was calculated as described (Perez et al. 2011).
Paip2 is regulated by the HCMV-encoded UL38 mTORC1 activator
UL38 is a multifunctional HCMV-encoded early gene product that may serve different roles in discrete compartments of the cell (Tenney and Colberg-Poley 1991; Varnum et al. 2004; Qian et al. 2011). One characterized role for UL38 in positively regulating translation is binding to TSC2, inhibiting the tuberous sclerosis complex, constitutively activating mTORC1, and subsequently disabling the 4E-BP1 translation initiation inhibitor (Moorman et al. 2008). We showed that UL38 not only regulates PABP1 accumulation in an mTORC1-dependent manner that requires 4E-BP1 inactivation, but also translationally activates the TOP motif in the 5′ UTR of the mRNA (McKinney et al. 2012). While the TOP motif confers translational repression during nutrient and growth factor starvation and consists of a 5′ terminal cytosine residue followed by a 4- to 13-nucleotide (nt) sequence of pyrimidine residues (Meyuhas 2000), the Paip2 mRNA does not contain a canonical TOP motif in its 5′ UTR (Fig. 2A). To determine whether Paip2 accumulation in HCMV-infected cells was, like PABP1, dependent on UL38, NHDFs were mock-infected or infected with wild-type HCMV, a UL38-deficient HCMV, or a virus in which the UL38 mutation was repaired to wild type, and Paip2, PABP1, and EDD1 abundance was evaluated by immunoblotting. Although Paip2, PABP1, and EDD1 all accumulated in cells infected with wild-type HCMV, their accumulation was impaired in cells infected with a UL38-deficient virus (Fig. 2B). Infection with a revertant virus where the UL38 gene was reintroduced into the UL38-deficient (ΔUL38) HCMV genome restored robust accumulation of PABP1, Paip2, and EDD1 to near wild-type levels, proving that the coordinate accumulation of PABP, Paip2, and EDD1 was indeed UL38-dependent (Fig. 2B).
Paip2 and EDD1 mRNAs do not contain a canonical TOP element but encode proteins that accumulate in a UL38-dependent manner together with PABP1. (A) Comparison of the Paip2 mRNA 5′ terminus with those of canonical TOP-bearing mRNAs, which contain a 5′-terminal cytosine residue followed by a 4- to 14-nt pyrimidine tract and a GC-rich region. (B) Asynchronous NHDFs were either mock-infected or infected (MOI = 3) with wild-type (WT) HCMV, a UL38-deficient mutant virus (ΔUL38), or a revertant virus where the UL38 deficiency was repaired by reintroducing a wild-type UL38 gene (Rev). At 48 hpi, total protein was isolated, fractionated by SDS-PAGE, and analyzed by immunoblotting with the indicated antisera.
UL38 is a multifunctional protein that not only activates mTORC1, but also regulates apoptosis and the unfolded protein response (Terhune et al. 2007; Qian et al. 2011). In addition, the control of Paip2 abundance is incompletely understood. To determine whether mTOR activity in HCMV-infected cells regulates Paip2 steady-state levels, mock-infected and HCMV-infected NHDFs were treated with the mTORC1-selective inhibitor rapamycin or the mTOR active site inhibitor PP242 (Feldman et al. 2009). The virus-induced increase in PABP1 and Paip2 was subsequently evaluated by immunoblotting. Figure 3A demonstrates that Paip2 accumulation in HCMV-infected cells was reduced by rapamycin treatment and nearly abrogated by PP242. Thus, Paip2 accumulation in response to HCMV infection, similar to PABP accumulation, requires mTOR signaling. The greater reduction in PABP1 and Paip2 achieved with PP242 versus rapamycin could indicate that the levels of both proteins are controlled by 4E-BP1 phosphorylation, which is differentially sensitive to PP242 compared with rapamycin (Feldman et al. 2009). Indeed, 4E-BP1 phosphorylation controls the virus-induced PABP1 increase (Perez et al. 2011). Alternatively, as HCMV infection becomes resistant to rapamycin after 36 hpi, it is possible that mTORC2 could potentially be involved (Kudchodkar et al. 2004, 2006).
Paip2 abundance in response to HCMV infection, serum stimulation, or UL38 induction is regulated by mTORC1. (A) Asynchronous NHDFs were either mock-infected or HCMV-infected in the presence of DMSO, the mTORC1-selective inhibitor rapamycin (100 nM), or the active site mTOR inhibitor PP242 (2.5 μM). At 48 hpi, total protein was collected, fractionated by SDS-PAGE, and analyzed by immunoblotting with the indicated antisera. (B, lanes 1–4) NHDFs that express UL38 upon induction with dox cells were growth-arrested by serum deprivation; treated for 1 h with DMSO, rapamycin (100 nM), or PP242 (2.5 μM); and subsequently untreated or serum-stimulated with 20% FBS for 20 min in the presence or absence of drug. (Lanes 5–7) Alternatively, cells were dox-treated for 72 h. Subsequently, cultures were metabolically pulse-labeled for 1 h with [35S]Met-Cys, and total protein was collected. A sample of the lysate was immunoprecipitated using anti-Paip2 antisera. Lysates (input panel) and isolated immune complexes (Paip2 panel) were fractionated by SDS-PAGE, and radiolabeled polypeptides were directly visualized by exposing the fixed, dried gel to X-ray film. The migration of molecular weight standards (in kilodaltons) is shown to the left of the top panel.
To determine whether ectopic expression of UL38 is sufficient to induce Paip2 accumulation in uninfected cells, primary NHDFs were transduced with a lentivirus expressing UL38 under the control of a doxycycline (dox)-inducible promoter. Following growth arrest by serum deprivation, the cells were either exposed to serum in the absence of dox or treated with dox in the absence of serum. After metabolic labeling of newly synthesized proteins with 35S amino acids for 1 h, total protein was harvested, fractionated by SDS-PAGE, and visualized by autoradiography (Fig. 3B). In addition, Paip2 synthesis was evaluated by immunoprecipitation. Both serum exposure in the absence of dox and dox treatment in the absence of serum were sufficient to stimulate total cellular protein synthesis and new Paip2 synthesis (Fig. 3B, cf. lanes 2–5). Importantly, UL38 induction was more effective than serum stimulation at stimulating 35S amino acid incorporation into total protein and newly synthesized Paip2. Thus, ectopic UL38 expression in uninfected cells in the absence of any other HCMV-encoded proteins is necessary and sufficient to increase Paip2 abundance.
To determine whether Paip2 synthesis in response to UL38 expression in uninfected NHDFs requires mTOR activity, serum-deprived NHDFs that express inducible UL38 were treated with serum or dox in the presence of rapamyin or PP242. Figure 3B shows that either serum or UL38 induced Paip2 accumulation in a rapamycin or PP242-sensitive manner. Thus, new Paip2 synthesis and accumulation in uninfected NHDFs is induced in response to UL38 expression in a manner dependent on mTORC1 activity. Notably, unlike PABP mRNA, whose translation is also stimulated by mTORC1, Paip2 mRNA does not contain a canonical TOP element, raising the possibility that Paip2 synthesis and accumulation are intrinsically responsive to PABP abundance.
Paip2 stability increases during infection through increased binding to PABP1
By associating with PABP1 in a 2:1 stoichiometry via two conserved PABP-interacting motifs (PAM1 and PAM2), Paip2 inhibits PABP1 (Karim et al. 2006). When PABP1 is limiting, PAM2 binds to the PABC domain of PABP1 or EDD1 (Yoshida et al. 2006). To understand how the virus might benefit from increasing PABP1 while simultaneously stimulating accumulation of Paip2 and EDD1, the possibility that the association of PABP1 with Paip2 during infection might be impaired was investigated. Nonionic detergent lysates prepared from mock-infected or HCMV-infected cultures were immunoprecipitated using anti-Paip2 or anti-PABP1 sera, and the amount of Paip2 and PABP1 present in immune complexes was evaluated by immunoblotting. While the total levels of both proteins increased in HCMV-infected cells (see Input in Fig. 4A), it was surprising that (1) PABP1 and Paip2 remained associated, and (2) the overall amount of associated Paip2 and PABP increased (Fig. 4A). However, it remained possible that the ratio of free PABP1 to free Paip2 could be altered during infection in a manner dependent on their rates of synthesis and turnover. To determine the impact of HCMV infection on the stability of PABP and Paip2, primary NHDFs either mock-infected or infected with HCMV were treated with cycloheximide (CHX) to attenuate new protein synthesis, and total EDD1, Paip2, and PABP1 levels were measured over a 9-h time period by immunoblotting. The efficacy of CHX treatment was verified by monitoring the inhibition of 35S amino acid incorporation into acid-insoluble material (data not shown). While PABP1 and EDD1 levels remained fairly consistent throughout the CHX treatment in mock-infected cells, decreased Paip2 levels were readily detected shortly after treatment (Fig. 4B). In contrast, Paip2 abundance decreased more slowly in HCMV-infected cells than in mock-infected cells. Quantification of this decay revealed that Paip2 half-life nearly doubled in HCMV-infected cells, as Paip2 levels decreased by 50% after 5 h of CHX treatment in uninfected cells versus 9 h in infected cells (Fig. 4B). Since Paip2 homeostasis is dependent on its ability to bind PABP1 (Yoshida et al. 2006), variation in Paip2 stability likely reflects changes in the balance of free PABP1 to free Paip2. In this case, increasing PABP1 concentration in HCMV-infected cells would serve as a sink to stabilize the pool of newly synthesized Paip2.
Paip2 associates with PABP and is stabilized in HCMV-infected cells. (A) Soluble cell-free extracts were prepared from asynchronous NHDFs that were either mock-infected (−) or HCMV-infected (+) at 48 hpi (MOI = 3). Samples were immunoprecipitated (IP) using anti-Paip2 or anti-PABP1 antisera. Isolated immune complexes and a sample of the input extract were fractionated by SDS-PAGE and analyzed by Western blotting (WB) using antisera specific for Paip2 or PABP1. (B) NHDFs that were mock-infected or infected with HCMV (MOI = 3) were treated with 100 μg/mL CHX at 48 hpi. At the indicated times (hours) following CHX addition, total protein was isolated, fractionated by SDS-PAGE, and analyzed by immunoblotting using the indicated primary antibodies and a secondary antibody covalently linked to an infrared fluorophore. The membrane was scanned, and the fraction of Paip2 remaining at each time point was quantified using an Odyssey infrared imager. Each band was measured for raw intensity value, and each time point was normalized to the amount of Paip2 present at time 0. (C) Paip2 was immunodepleted from soluble cell-free extracts prepared from asynchronous NHDFs that were either mock-infected (Mock) or HCMV-infected (wild type vs. ΔUL38) at 48 hpi (MOI = 3). Samples of the input extracts and the fractions not bound to anti-Paip2 antibody (flow-through) were fractionated by SDS-PAGE and analyzed by immunoblotting using the indicated antisera.
A key prediction of this model is that the virus-induced increase in PABP1 concentration is sufficient to maintain a pool of free PABP not bound to Paip2. To test this hypothesis, the amounts of free PABP not bound to Paip2 in cell-free lysates prepared from NHDFs infected with wild-type HCMV or a UL38-deficient virus (ΔUL38) that is unable to increase PABP abundance were compared. Initial levels of PABP1, Paip2, and UL38 were first evaluated in input lysates prepared from mock-infected versus HCMV-infected (wild type vs. ΔUL38) by immunoblotting. As expected, PABP1 and Paip2 abundance both increased in a UL38-dependent manner (Fig. 4C). However, following immunodepletion of Paip2 from cell-free lysates, more PABP remained in the unbound fraction isolated from wild-type HCMV-infected cells compared with cells infected with ΔUL38 or uninfected cells (Fig. 4C). Thus, the HCMV-induced PABP1 increase is sufficient to increase the concentration of free PABP1 not bound to its inhibitor, Paip2, in a UL38-dependent manner.
Paip2 accumulation is triggered by the HCMV-induced PABP1 increase and regulates eIF4F assembly in infected cells
Having shown previously that the virus-induced PABP1 increase promoted virus replication, it was puzzling to observe a concomitant rise in the level and stability of its cognate repressor, Paip2, along with its negative regulator, EDD1. Thus, we next considered the hypothesis that HCMV protein accumulation and replication were in fact controlled by the dynamic interplay between PABP1 and Paip2 levels. By reducing the abundance of the E3 ubiquitin ligase EDD1, binding of Paip2 to PABP1 increases, stabilizing Paip2 and inhibiting PABP1 function (Yoshida et al. 2006). Indeed, Paip2 abundance, but not PABP1 abundance, increased upon EDD1 depletion in HCMV-infected cells. (Fig. 5A). Moreover, Paip2 accrual induced by EDD1 depletion coincided with a reduction in overall levels of representative viral proteins (UL44 and ICP8) and suggests that further increasing Paip2 concentration in HCMV-infected cells limits viral protein accumulation (Fig. 5A). Preventing the virus-induced PABP1 increase using RNAi in HCMV-infected cells markedly reduced accumulation of viral gene products representing different temporally expressed classes of viral genes (Fig. 5B, cf. lane 1 vs. 2) in agreement with our earlier work (McKinney et al. 2012). Moreover, we now demonstrate that it also resulted in a corresponding reduction in Paip2 levels. Notably, no differences in PABP1 or viral protein accumulation were detected when Paip2 was depleted with siRNA (Fig. 5B, cf. lane 1 vs. 3). This raised the possibility that PABP1 is produced in vast excess in order to saturate the inhibitory effects of Paip2.
Increased Paip2 abundance in response to virus-induced PABP accumulation regulates eIF4F assembly and HCMV protein production. (A) NHDFs transfected with the indicated siRNA (control nonsilencing siRNA [ctrl] or EDD1) were infected with HCMV at MOI = 0.1. After 5 d, total protein was collected, fractionated by SDS-PAGE, and analyzed by immunoblotting with the indicated antisera. (*) A nonspecific, cross-reacting band. (B) NHDFs transfected with a control nonsilencing siRNA (ctrl) or the indicated pairs of siRNAs were infected with HCMV at MOI = 0.1. After 5 d, total protein was collected and analyzed as in A. (C) As in B, except that total RNA was isolated using Trizol and subjected to RT-qPCR using primers specific for the HCMV IE2 or pp28 genes. Each reaction product was normalized to the signal obtained using primers specific for 18S rRNA and expressed as the fold change relative to HCMV-infected cells treated with control nonsilencing siRNA. (D) As in B, except that siRNA-treated NHDFs were mock-infected or HCMV-infected (MOI = 3). After 5 d, cell-free extracts prepared using a nonionic detergent were subject to batch chromatography on m7GTP Sepharose. A sample of input extract or m7GTP-bound proteins was fractionated by SDS-PAGE and analyzed by immunoblotting with the indicated antisera.
To test the hypothesis that raising PABP1 abundance in HCMV-infected cells is required to counteract the inhibitory impact of Paip2, the remaining pool of Paip2 was depleted in cells where the HCMV-induced PABP increase was blocked using RNAi. Remarkably, Paip2 depletion in HCMV-infected cells that were unable to increase PABP1 effectively restored viral protein production to levels observed in cultures treated with control, nonsilencing siRNA, or Paip2 siRNA alone (Fig. 5B, cf. lanes 4 and 3 or 1). Thus, similar levels of viral protein accumulation can be achieved by depleting Paip2, effectively obviating the need to increase intracellular PABP concentration. This suggests that rising Paip2 levels function to restrict viral replication, and this is effectively counteracted by raising PABP1 concentration. It further implies that the increase in Paip2 abundance occurs in response to rising PABP levels.
To understand the underlying mechanism whereby Paip2 depletion allows HCMV protein accumulation to proceed without raising intracellular PABP1 levels, the varied ways by which PABP1 post-transcriptionally controls gene expression were examined. PABP1 regulates mRNA stability by protecting mRNAs from deadenylation-mediated decay, and this could account for the observed changes in viral protein accumulation (for review, see Mangus et al. 2003). However, the steady-state abundance of HCMV pp28 and IE2 mRNAs did not detectably change when infected cells were treated with siRNAs targeting PABP, Paip2, or both PABP and Paip2 (Fig. 5C). PABP is also a translation initiation factor (Kahvejian et al. 2005). While HCMV infection, unlike many viruses, does not normally impair ongoing cellular protein synthesis, altering the balance between PABP and Paip2 may impact the efficiency with which viral versus host mRNAs access the translation initiation machinery. Although preventing the HCMV-induced PABP increase modestly reduced global infected cell protein synthesis, the overall profile of protein synthesis in infected cells was not grossly altered by Paip2 depletion in the presence or absence of PABP siRNA treatment (Supplemental Fig. S2). Thus, translation of viral mRNAs was not detectably enriched at the expense of cellular mRNAs.
Besides stimulating protein accumulation, the HCMV-induced PABP increase effectively stimulated assembly of eIF4E, eIF4G, and eIF4A into the heterotrimeric eIF4F translation initiation factor complex (Kudchodkar et al. 2004; Walsh et al. 2005; McKinney et al. 2012). To evaluate the impact of perturbing PABP/Paip2 homeostasis on eIF4F assembly, cell-free lysates prepared from NHDFs treated with siRNAs and subsequently mock-infected or infected with HCMV were subjected to batch chromatography using m7GTP Sepharose. After washing the resin to remove unbound and nonspecifically bound components, bound proteins were fractionated by SDS-PAGE and analyzed by immunoblotting to evaluate binding of eIF4G and eIF4A to the cap-binding protein eIF4E (Fig. 5D). While preventing the HCMV-induced PABP1 increase using RNAi substantially reduced the association of eIF4G and eIF4A with eIF4E bound to the m7GTP cap affinity resin (Fig. 5D, cf. lanes 2 and 3), Paip2 depletion modestly increased eIF4G binding to eIF4E compared with lysates prepared from cultures treated with control, nontargeting siRNA (Fig. 5D, lane 2 vs. 4). Remarkably, Paip2-depletion restored eIF4G and eIF4A binding to eIF4E in cells where the HCMV-induced PABP increase was inhibited by PABP siRNA. This was specific for HCMV-infected cells and not observed in uninfected cells (Supplemental Fig. S3). Thus, the defect in translation initiation factor eIF4F assembly resulting from failure to increase PABP1 concentration in HCMV-infected cells can be effectively corrected by Paip2 depletion. Furthermore, it implies that existing supplies of Paip2 effectively restrict eIF4F assembly, and this can be counteracted by a commensurate virus-induced rise in PABP1 concentration.
Paip2 is a host antiviral restriction factor antagonized by PABP accumulation in HCMV-infected cells
To determine whether Paip2 acts as a restriction factor that limits viral productive growth and spread in cells where the HCMV-induced PABP increase is blocked, NHDFs treated with siRNA (control, PABP1, Paip2, or both PABP and Paip2) were infected with HCMV, and viral replication was evaluated. Under these conditions (multiplicity of infection [MOI] = 0.1), preventing the virus-induced PABP increase using PABP1 siRNA reduced viral replication, whereas Paip2 depletion had a very slight but not significant impact. Significantly, depleting Paip2 in cells where the HCMV-induced PABP increase was blocked resulted in a 40-fold increase of viral replication, restoring viral growth in NHDFs to levels observed in cultures treated with control, nonsilencing siRNA or Paip2 siRNA alone (Fig. 6A,B). Thus, the HCMV-induced increase in PABP1 levels controls viral growth by antagonizing the corresponding increase in the host Paip2 inhibitor. This establishes that the cellular Paip2 can function as a host restriction factor to antagonize viral replication. Furthermore, it provides the first example of a virus that manipulates the PABP/Paip2 axis to specifically counteract Paip2 and foster viral protein accumulation and replication.
Endogenous Paip2 functions to restrict productive HCMV growth when the virus-induced PABP accumulation is blocked. (A) NHDFs transfected with a control nonsilencing siRNA (ctrl) or the indicated pairs of siRNAs were infected with HCMV (MOI = 0.1) and imaged for EGFP expression 5 d post-infection. (B) Supernatants collected from three independent experiments detailed in A were assayed for viral particle production based on their TCID50 (Heider et al. 2002). Error bars indicate standard error of the mean. (*) P < 0.022 for control versus PABP1 siRNA; (**) P < 0.01 for PABP1 versus PABP1/Paip2 siRNA. Control versus Paip2 was not statistically significant (P > 0.1).
Discussion
To ensure that viral mRNAs effectively compete with cellular mRNAs for limiting translation factors, HCMV uses an unusual strategy and forces its host to increase the overall abundance of several critical initiation factors, one of which is PABP1 (Isler et al. 2005; Walsh et al. 2005). The resulting accumulation of PABP1 in HCMV-infected cells is required to stimulate eIF4F assembly, viral protein production, and replication (McKinney et al. 2012). Here, we show that the PABP1 repressor Paip2 and EDD1, a ubiquitin E3 ligase that regulates Paip2 stability, unexpectedly increase together with PABP upon infection. The coordinate regulation of PABP1, Paip2, and EDD1 is dependent on the HCMV mTORC1 activator encoded by the UL38 gene. Preventing the HCMV-induced increase in PABP1 abundance not only impaired viral protein production and replication, but also inhibited the rise in Paip2 levels. Surprisingly, depleting Paip2 in cells where the HCMV-induced PABP increase was blocked restored their ability to support viral protein production and replication without a commensurate increase in PABP1 levels. This establishes a role for the host PABP inhibitor Paip2 in infection biology, where it can function as an innate defense that restricts viral protein synthesis and replication. Moreover, it provides the first example of how a pathogen can manipulate the cellular PABP1–Paip2 homeostatic axis by raising PABP1 levels to overwhelm the functional capacity of its inhibitor, Paip2.
While previous studies showed that depleting PABP1 coordinately reduced Paip2 stability in a manner requiring the ubiquitin E3 ligase EDD1 (Yoshida et al. 2006), the effect of increasing PABP1 concentration on Paip2 levels was never addressed, in part due to difficulties associated with overexpressing PABP (Wormington et al. 1996; Wu and Bag 1998; Hornstein et al. 1999b; Ma et al. 2006). Indeed, the biological benefit of destabilizing Paip2 when PABP1 levels fall intuitively preserves a pool of functional PABP1, which would otherwise be inhibited by Paip2 under these circumstances (Yoshida et al. 2006). The rise in Paip2 and EDD1 that accompanies the PABP1 increase, however, is more perplexing at first glance, as it has the potential to neutralize the investment involved in making more PABP1. Perhaps the coordinate increase in activator and inhibitor allows for a ready-made supply of inhibitor that can be rapidly enlisted to upset the delicate PABP1–Paip2 balance and thereby antagonize the newly set, elevated PABP1 levels. As PABP1 binds to mRNA in both noncooperative and cooperative modes, the latter involving homophillic protein–protein interactions that promote PABP multimerization on the mRNA poly(A) tail (Kühn and Pieler 1996; Melo et al. 2003; Lin et al. 2012), relatively small changes in the free PABP pool could have a significant impact on binding of the initial subunits to poly(A) mRNA termini. Thus, besides expanding the available pool of free PABP not bound to Paip2, the virus-induced PABP1 increase likely ensures that the intracellular PABP concentration is sufficient for PABP binding to and subsequent multimerization on mRNA. Alternatively, there may also be ways to increase PABP1 without elevating Paip2 or EDD1, and the corresponding rise in Paip2 could be viewed as a host response to infection or stress. In support of this hypothesis, preventing the HCMV-induced PABP increase also suppressed Paip2 accumulation, suggesting that increasing Paip2 abundance is a host response to elevated PABP concentration.
While cellular translation repressors that regulate eIF4F assembly from its component subunits (eIF4E, eIF4G, and eIF4A) are inactivated in herpesvirus-infected cells (Walsh and Mohr 2011), this is the first demonstration of a role for the PABP1–Paip2 axis. Assembly of eIF4F is controlled in part by the availability of its eIF4E subunit, which is regulated by the abundance and phosphorylation status of its repressor protein, 4E-BP1 (Gingras et al. 1999). Remarkably, the levels of the repressors 4E-BP1 and Paip2 are controlled by a homeostatic mechanism that responds to the abundance of their respective cognate initiation factors, eIF4E and PABP1 (Yoshida et al. 2006; Yanagiya et al. 2010). Furthermore, both repressors appear to be regulated by mTORC1 signaling, which is constitutively activated in HCMV-infected cells by the TSC2-binding protein UL38. Notably, antagonizing translation repressors like 4E-BP1 and now Paip2 by virus-encoded factors in infected cells is consistent with the notion that host translation control pathways function as innate defenses with the capacity to restrict productive viral replication. Indeed, the observation that Paip2 depletion restores viral replication in infected cultures where the virus-induced PABP1 increase is blocked implies that Paip2 restricts viral replication by limiting PABP1 availability. Furthermore, restoration of replication to levels that exceed wild type by depleting Paip2 in cells where the HCMV-induced PABP increase was blocked suggests that the scope of PABP1 function can be maintained with far less PABP1 than is induced by HCMV. It also implies that translationally up-regulating PABP1 is meant to overcome the antiviral response of increasing Paip2. Additional virus-encoded functions, including the UL69 RNA-binding protein that associates with PABP and eIF4A (Aoyagi et al. 2010), may also help preserve PABP function by limiting its availability to interact with Paip2.
Although Paip2 clearly is an important inhibitor of PABP1 function, spermatogenesis in mice, growth control in Drosophila, synaptic plasticity, and memory in mice are the only biological process with a documented role for Paip2 (Roy et al. 2002; Yanagiya et al. 2010; Khoutorsky et al. 2013). Our study establishes that Paip2 can play a significant role in infection biology by altering the balance of PABP1 with its cognate repressor. Importantly, infection is a potent stress, and translational control is a major means whereby stress-induced gene expression is controlled. This raises the possibility that Paip2 might play a more global role in controlling PABP1 levels in response to stress in uninfected cells. Future studies in other model systems are required to address this possibility and uncover any parallels with changes to Paip2, EDD1, and PABP1 in HCMV-infected cells.
Even though Paip2 protein synthesis increases in an mTORC1-dependent manner, Paip2 mRNA does not contain a TOP sequence element. While TOP mRNA translation is stimulated in HCMV-infected cells (McKinney et al. 2012), Paip2 represents the first host mRNA whose translation is specifically controlled in infected cells and is not a TOP family member. Thus, while mTORC1 activation controls translation of canonical TOP mRNAs in HCMV-infected cells, non-TOP-containing mRNAs like Paip2 are poised to respond to other translational regulators, like PABP. By not suppressing host mRNA translation, HCMV provides a unique model to evaluate how differential host mRNA translation might impact infection. Translationally regulated host factors might be required, like PABP1, to stimulate viral replication or contribute to host defenses and thereby restrict viral replication like Paip2. Although the extent to which host mRNAs are translationally controlled in HCMV-infected cells and their contribution to virus biology remain to be elucidated, they could reveal a powerful means to control viral replication and allow viruses to access a wealth of functions encoded by host genomes. Likewise, it may also provide infected hosts with a valuable opportunity to attempt to restrict viral replication and identify new innate defense effectors whose synthesis is transitionally regulated. Finally, the HCMV model system may provide valuable insight into how different mRNA populations are either recruited or excluded from polyribosomes.
Materials and methodsCell culture and viruses
NHDFs (purchased from Clonetics) were routinely subcultured by a 1:3 split and maintained until passage 20 in DMEM supplemented with 5% FBS, 1% L-glutamine, and 1% penicillin/streptomycin (v/v). GFP-expressing HCMV (AD169 strain) wild-type, UL38-null (ΔUL38), or UL38 revertant (REV) viruses were propagated as described (Terhune et al. 2007). Akt inhibitor VIII was purchased from Sigma.
Quantitative PCR (qPCR), antibodies, and Western blotting
RNA was isolated from cell-free lysates and processed into cDNA for qPCR as described (McKinney et al. 2012). The following primers were used for qPCR analysis: PABP1 (fw, CCCAGCTGCTCCTAGACC; and rev, GAGTAGCTGCAGCGGCT), Paip2 (purchased from Origene), EDD1 (UBR5; purchased from Origene), 18s (fw, AGGAATTGACGGAAGGGCACCA; and rev, TTATCGGAATTAACCAGACAAATCG), pp28 (fw, TGCTCTGGGTCGCCAGGTGT; and rev, CAGCCACTACCGCAGAGCC), and IE2 (fw, CGGGTGGATGTGTCACGGGC; and rev, ACGCACCCGCTCTCCCAGA).
Protein lysates were collected and subjected to SDS-PAGE and Western blotting as described (Walsh and Mohr 2004). Anti-PABP1 rabbit polyclonal was a gift from Simon Morley (University Sussex, UK), and anti-UL38 mouse monoclonal was a gift from Tom Shenk (Princeton University). The following primary antibodies were purchased commercially: anti-Paip2 (Sigma, no. P0087), anti-actin (Calbiochem, no. CP01), anti-EDD1 (Bethyl Laboratories, no. A300-573A-2), anti-Akt (Cell Signaling, no. 9272); anti-pp28 (Abcam, no. ab6502), anti-IE1/2 (Millipore, no. MAB810), and anti-UL44 (Virusys, no. CA006). For quantitative immunoblotting, a secondary antibody covalently linked to an infrared fluorophore was used (Li-Cor Biosciences, no. 827-08365) and the membrane was scanned using an Odyssey infrared imager (Li-Cor Biosciences).
Protein half-life analysis, immunoprecipitation, and Paip2 immunodepletion.
Mock-infected or HCMV-infected cells were treated with 100 μg/mL CHX (Research Products International Corp.) for up to 9 h. Western blots were quantified (n = 3) using an Odyssey infrared imager (Li-Cor Biosciences). To determine the efficacy of CHX treatment, cells were pulse-labeled with 35S-labeled amino acids at the indicated times, and the amount of acid-insoluble radioactivity present in cell-free lysates was quantified as described (McKinney et al. 2012).
Paip2 was immunoprecipitated from cell-free lysates prepared from mock-infected or HCMV-infected NHDFs (1.3 × 106 cells per sample) harvested at 48 hpi. After washing with cold PBS, cells were suspended in NP40 lysis buffer (50 mM HEPES at pH 7.4, 100 mM NaCl, 1.5 mM MgCl2, 2 mM EDTA, 0.25% NP40, 1× Roche Phos Stop, protease inhibitor tablets). Lysates were gently rocked for 30 min at 4°C, and, subsequently, lysates were clarified by centrifugation (12,000g) for 10 min at 4°C. CaCl2 (1 mM final concentration) was added to soluble supernatants (0.5 mL), which were subsequently treated with a nuclease cocktail (4 U of micrococcal nuclease [New England Biolabs, no. M0247S], 150 U of RNase T1, 3.75 U of RNase A [Ambion, no. AM2286]) for 20 min at room temperature. Nuclease-treated lysates were precleared by adding 5 μg of purified normal rabbit serum (Invitrogen, no. 17-0780-01) for 1 h at 4°C, and nonspecific binding proteins were subsequently collected by incubation with 10 μL of packed bed volume of protein A Sepharose CL-4B (GE Healthcare, no. 17-780-01) for 1 h at 4°C. An aliquot of soluble, precleared extract (10%) was removed (input fraction), diluted 1:2 in SDS sample buffer, and reserved for later analysis. Anti-Paip2 antibody (5 μg) was added to the remainder of the precleared lysate, and the mixture used to resuspend 10 μL of packed bed volume of protein A Sepharose. After overnight incubation at 4°C, the beads and supernatant were partitioned by brief centrifugation in a microfuge, and the soluble flow-through fraction was removed for analysis by immunoblotting (together with the input fraction). For analysis of Paip2 and Paip-2-associated proteins, the beads were next washed three times (0.5 mL per wash) with NP40 lysis buffer, suspended in SDS-containing sample buffer, and boiled for 3 min. Immune complexes were fractionated by SDS-PAGE and analyzed by immunoblotting. For analysis of PABP and PABP-associated proteins, a similar immunoprecipitation protocol was performed using anti-PABP antibody.
Generating UL38-expressing cells and 35S metabolic labeling
To generate a dox-inducible, UL38-expressing cell line, the 990-base-pair (bp) UL38 gene was PCR-amplified from a pRetro-EBNA-UL38 vector (Moorman et al. 2008) with fw (GACAGGAACTAGTATACCACGCATAGCACT) and rev (ATACGGCCTCGAGCTGACCACCATCTGTAC) primers bearing SpeI and XhoI overhangs, respectively. The product was then digested, purified, and cloned into a pSLIK tet-inducible vector as described (Kobayashi et al. 2012).
Lentiviruses were produced as described (Kobayashi et al. 2012), and a population of transduced cells was selected using 50 μg/mL hygromycin B (Life Technologies).
RNAi, microscopy, and viral replication assay
PABP1, Paip2, and EDD1 siRNA smartpools were purchased from Dharmacon (catalog nos. M019598, M015376, and M007189, respectively). Transfections of siRNAs, low MOI HCMV infections, and viral replication assays were conducted as described (McKinney et al. 2012). For depleting EDD1, 20 nM siRNA was transfected. For depleting two target genes, 20 nM siRNA was transfected for each target (40 nM siRNA total siRNA). EGFP expression was used to measure HCMV spread and was visualized under 5× magnification using a Zeiss Axiovert 200 fluorescent microscope. A paired two-tailed t-test was used to calculate P-values.
Acknowledgments
We thank S. Morley, R. Schneider, and T. Shenk for helpful reagents; D. Walsh and H. Burgess for their critical review of the manuscript; and members of the Mohr laboratory, R. Schneider, and A. Wilson for many helpful discussions. This work was supported by National Institutes of Health (NIH) grants AI073898 and GM056927 to I.M., and CA120768 to D.Y. C.M. was supported by NIH training grant T32 AI007647.
Supplemental material is available for this article.
Article is online at http://www.genesdev.org/cgi/doi/10.1101/gad.221341.113.
ReferencesAoyagiM, GasparM, ShenkT2010Human cytomeglovirus UL69 protein facilitates translation by associating with the mRNA cap-binding complex and excluding 4E-BP1.
107: 2640–264520133758CraigAW, HaghighatA, YuAT, SonenbergN1998Interaction of polyadenylate-binding protein with the eIF4G homologue PAIP1 enhances translation.
392: 520–5239548260DerryMC, YanagiyaA, MartineauY, SonenbergN2006Regulation of poly(A)-binding protein through PABP-interacting proteins.
71: 537–54317381337FeldmanME, ApselB, UotilaA, LoewithR, KnightZA, RuggeroD, ShokatKM2009Active-site inhibitors of mTOR target rapamycin-resistant outputs of mTORC1 and mTORC2.
7: 371–383GingrasAC, GygiSP, RaughtB, PolakiewiczRD, AbrahamRT, HoekstraMF, AebersoldR, SonenbergN1999Regulation of 4E-BP1 phosphorylation: A novel two-step mechanism.
13: 1422–143710364159HeiderJA, YuY, ShenkT, AlwineJC2002Characterization of a human cytomegalovirus with phosphorylation site mutations in the immediate-early 2 protein.
76: 928–93211752183HornsteinE, GitA, BraunsteinI, AvniD, MeyuhasO1999aThe expression of the poly(A)-binding protein gene is translationally regulated in a growth-dependent fashion through a 5′terminal oligopyrimidine tract motif.
274: 1708–17149880551HornsteinE, HarelH, LevyG, MeyuhasO1999bOverexpression of poly(A)-binding protein down-regulates the translation or the abundance of its own mRNA.
457: 209–21310471780IslerJA, SkaletAH, AlwineJC2005Human cytomegalovirus infection activates and regulates the unfolded protein response.
79: 6890–689915890928JacksonRJ, HellenCUT, PestovaTV2010The mechanisms of eukaryotic translation initiation and principles of its regulation.
10: 113–12720094052KahvejianA, RoyG, SonenbergN2001The mRNA closed-loop model: The function of PABP and PABP-interacting proteins in mRNA translation.
66: 293–30012762031KahvejianA, SvitkinYV, SukariehR, M'BoutchouM, SonenbergN2005Mammalian poly(A)-binding protein is a eukaryotic translation initiation factor, which acts via multiple mechanisms.
19: 104–11315630022KarimMM, SvitkinYV, KahvejianA, De CerscenzoG, Costa-MattioloM, SonenbergN2006A mechanism of translational repression by competition of Paip2 with eIF4G for poly(A) binding protein (PABP) binding.
103: 9494–949916772376KhaleghpourK, SvitkinYV, CraigAW, DeMariaCT, DeoRC, BurleySK, SonenbergN2001aTranslational repression by a novel partner of human poly(A) binding protein, Paip2.
7: 205–21611172725KhaleghpourK, KahvejianA, De CrescenzoG, RoyG, SvitkinY, ImatakaH, O'Connor-McCourtM, SonenbergN2001bDual interactions of the translational repressor Paip2 with poly(A) binding protein.
21: 5200–521311438674KhoutorskyA, YanagiyaA, GkogkasCG, FabianMR, Prager-KhoutorskyM, CaoR, GamacheK, BouthietteF, ParsyanA, SorgeRE, 2013Control of synaptic plasticity and memory via suppression of poly(A)-binding protein.
78: 298–31123622065KobayashiM, WilsonAC, ChaoMV, MohrI2012Control of viral latency in neurons by axonal mTOR signaling and the 4E-BP translation repressor.
14: 1527–153222802527KudchodkarSB, YuY, MaguireTG, AlwineJC2004Human cytomegalovirus infection induces rapamycin-insensitive phosphorylation of downstream effectors of mTOR kinase.
78: 11030–1103915452223KudchodkarSB, YuY, MaguireTG, AlwineJC2006Human cytomegalovirus infection alters the substrate specificities and rapamycin sensitivities of raptor- and rictor-containing complexes.
103: 14182–1418716959881KühnU, PielerT1996Xenopus poly(A) binding protein: Functional domains in RNA binding and protein–protein interaction.
256: 20–308609610LinJ, FabianM, SonenbergN, MellerA2012Nanopore detachment kinetics of poly(A) binding proteins from RNA molecules reveals the critical role of C-terminus interactions.
102: 1427–143422455926MaS, MusaT, BagJ2006Reduced stability of mitogen-activated protein kinase-2 mRNA and phosphorylation of poly(A)-binding protein (PABP) in cells overexpressing PABP.
281: 3145–315616332685MangusDA, EvansMC, JacobsonA2003Poly(A)-binding proteins: Multifunctional scaffolds for the post-transcriptional control of gene expression.
4: 22312844354MartineauY, DerryMC, WangX, YanagiyaA, BerlangaJJ, ShyuAB, ImatakaH, GehringK, SonenbergN2008Poly(A)-binding protein-interacting protein 1 binds to eukaryotic translation initiation factor 3 to stimulate translation.
28: 6658–666718725400McKinneyC, PerezC, MohrI2012Poly(A) binding protein abundance regulates eukaryotic translation initiation factor 4F assembly in human cytomegalovirus-infected cells.
109: 5627–563222431630MeloEO, DhaliaR, Martins de SaC, StandartN, de Melo NetoOP2003Identification of a C-terminal poly(A)-binding protein (PABP)–PABP interaction domain: Role in cooperative binding to poly (A) and efficient cap distal translational repression.
278: 46357–4636812952955MeyuhasO2000Synthesis of the translational apparatus is regulated at the translational level.
267: 6321–633011029573MohrI, SonenbergN2012Host translation at the nexus of infection and immunity.
12: 470–48323084916MoormanNJ, CristeaIM, TerhuneS, RoutMP, ChaitBT, ShenkT2008Human cytomegalovirus protein UL38 inhibits host cell stress responses by antagonizing the tuberous sclerosis protein complex.
3: 253–26218407068PatelGP, MaS, BagJ2005The autoregulatory translational control element of poly(A)-binding protein mRNA forms a heteromeric ribonucleoprotein complex.
33: 7074–708916356927PerezC, McKinneyC, ChuluunbaatarU, MohrI2011Translational control of cytoplasmic poly(A) binding protein in human cytomegalovirus-infected cells.
85: 156–16420980505QianZ, XuanB, GualbertoN, YuD2011The human cytomegalovirus protein pUL38 supresses endoplasmic reticulum stress-mediated cell death independently of its ability to induce mTORC1 activation.
85: 9103–911321715486RoyG, De CrescenzoG, KhaleghpourK, KahvejianA, O'Connor-McCourtM, SonenbergN2002Paip1 interacts with poly(A) binding protein through two independent binding motifs.
22: 3769–378211997512StinskiM1977Synthesis of proteins and glycoproteins in cells infected with human cytomegalovirus.
23: 751–767197270TenneyDJ, Colberg-PoleyAM1991Expression of the human cytomegalovirus UL36-38 immediate early region during permissive infection.
182: 199–2101850901TerhuneS, TorigoiE, MoormanN, SilvaM, QianZ, ShenkT, YuD2007Human cytomegalovirus UL38 protein blocks apoptosis.
81: 3109–312317202209VarnumSM, StreblowDN, MonroeME, SmithP, AuberryKJ, Pasa-TolicL, WangD, CampDG2nd, RodlandK, WileyS, 2004Identification of proteins in human cytomegalovirus (HCMV) particles: The HCMV proteome.
78: 10960–1096615452216WalshD, MohrI2004Phosphorylation of eIF4E by Mnk-1 enhances HSV-1 translation and replication in quiescent cells.
18: 660–67215075293WalshD, MohrI2011Viral subversion of the host protein synthesis machinery.
9: 860–87322002165WalshD, PerezC, NotaryJ, MohrI2005Regulation of the translation initiation factor eIF4F by multiple mechanisms in human cytomegalovirus-infected cells.
79: 8057–806415956551WalshD, MathewsMB, MohrI2013Tinkering with translation: Protein synthesis in virus-infected cells.
5: a01235123209131WormingtonM, SearfossAM, HurneyCA1996Overexpression of poly(A) binding protein prevents maturation-specific deadenylation and translational inactivation in Xenopus oocytes.
15: 900–9098631310WuJ, BagJ1998Negative control of the poly(A)-binding protein mRNA translation is mediated by the adenine-rich region of its 5′-untranslated region.
273: 34535–345429852123YanagiyaA, DelbesG, SvitkinYV, RobaireB, SonenbergN2010The poly(A)-binding protein partner Paip2a controls translation during late spermatogenesis in mice.
120: 3389–340020739757YoshidaM, YoshidaK, KozlovG, LimNS, BerlangaJJ, KahvejianA, GehringK, WingSS, SonenbergN2006Poly(A) binding protein (PABP) homeostasis is mediated by the stability of its inhibitor, Paip2.
25: 1934–194416601676oai:pubmedcentral.nih.gov:37596982014-02-15genesdevpmc-openGenes DevGenes DevGADGenes & Development0890-93691549-5477Cold Spring Harbor Laboratory PressPMC3759698PMC375969837596982396409623964096871166010.1101/gad.210211.112Research PaperE3-ubiquitin ligase Nedd4 determines the fate of AID-associated RNA polymerase II in B cellsSun et al.Fate of AID-associated RNA polII in B cellsSunJianbo14KeimCelia D.14WangJiguang12KazadiDavid1OliverPaula M.3RabadanRaul2BasuUttiya15Department of Microbiology and Immunology,Department of Biomedical Informatics, College of Physicians and Surgeons, Columbia University, New York, New York 10032, USA;Department of Pathology and Laboratory Medicine, University of Pennsylvania School of Medicine, Philadelphia, Pennsylvania 19104, USA
During B-lymphocyte class switch recombination, mutagenesis of the immunoglobulin (Ig) locus requires RNA polII-dependent targeting of the DNA mutator AID, yet the mechanism is unknown. Sun et al. now show that E3-ubiquitin ligase Nedd4 destabilizes AID-associated RNA polII at Ig switch regions. Loss of Nedd4 activity leads to RNA exosome substrate accumulation at AID target genes. This study links noncoding RNA processing following RNA polII pausing with AID regulation during class switch recombination.
Programmed mutagenesis of the immunoglobulin locus of B lymphocytes during class switch recombination (CSR) and somatic hypermutation requires RNA polymerase II (polII) transcription complex-dependent targeting of the DNA mutator activation-induced cytidine deaminase (AID). AID deaminates cytidine residues on substrate sequences in the immunoglobulin (Ig) locus via a transcription-dependent mechanism, and this activity is stimulated by the RNA polII stalling cofactor Spt5 and the 11-subunit cellular noncoding RNA 3′–5′ exonucleolytic processing complex RNA exosome. The mechanism by which the RNA exosome recognizes immunoglobulin locus RNA substrates to stimulate AID DNA deamination activity on its in vivo substrate sequences is an important question. Here we report that E3-ubiquitin ligase Nedd4 destabilizes AID-associated RNA polII by a ubiquitination event, leading to generation of 3′ end free RNA exosome RNA substrates at the Ig locus and other AID target sequences genome-wide. We found that lack of Nedd4 activity in B cells leads to accumulation of RNA exosome substrates at AID target genes and defective CSR. Taken together, our study links noncoding RNA processing following RNA polII pausing with regulation of the mutator AID protein. Our study also identifies Nedd4 as a regulator of noncoding RNAs that are generated by stalled RNA polII genome-wide.
activation-induced deaminaseimmunoglobulin locus transcriptionNedd4RNA polymerase II stallingRNA polymerase II ubiquitinationnoncoding RNA
Prior to the discovery of noncoding RNA (ncRNA) as a major subclass of eukaryotic genome regulators (Ebert and Sharp 2012; Rinn and Chang 2012), the presence of noncoding germline transcripts in the immunoglobulin (Ig) locus had attracted the attention of many molecular biologists and immunologists (Alt et al. 1982). Accumulating over the last four decades, ample evidence has unequivocably established that the synthesis of long noncoding germline transcripts in the Ig locus plays a pivotal role in recruiting B-cell-specific DNA mutator factors recombination activation genes (RAG-1 and RAG-2) and activation-induced cytidine deaminase (AID) to their target DNA sequences (for review, see Schatz et al. 1992; Keim et al. 2013). AID is a ssDNA cytidine deaminase; AID's activity depends on transcription, cofactors, and transcription-driven secondary DNA structures to identify substrate DNA, which it subsequently mutates to promote class switch recombination (CSR) and somatic hypermutation (SHM) (Chaudhuri et al. 2007; Keim et al. 2013). CSR is an AID-dependent chromosomal deletion–recombination event that alters the IgH locus in such a way that the host B cell is now capable of expressing antibodies that have an isotype different from IgM. One important, unanswered question relates to how a genome-wide process like transcription regulates AID in such a fashion that specific DNA single-strand mutations and DNA double-strand breaks at variable (V) genes and switch (S) sequences are generated in a controlled manner in the Ig locus. Recent advances in the understanding of RNA polymerase II (polII) regulation during transcription initiation, elongation, and termination at various DNA sequences provided insights that have helped to elucidate RNA polII's role in regulating AID targeting and mutagenic activity (Besmer et al. 2006; Wang et al. 2006; Rajagopal et al. 2009; Pavri et al. 2010; Basu et al. 2011). Work in multiple laboratories has focused on the state of the eukaryotic transcription complex with which AID is associated. Following transcription initiation at transcription start sites (TSSs), RNA polII undertakes “promoter escape,” a process that is regulated stringently by many regulatory mechanisms. These regulatory mechanisms include the action of various DNA helicases that catalyze melting of supercoiled promoters (a TFIIH-dependent mechanism) and recruitment of various RNA polII-associated cofactors signaled by RNA polII C-terminal phosphorylation events at the Ser-5 (S5) residue. Following promoter escape, another step that regulates the entry of RNA polII into elongation mode is “RNA polII pausing,” also referred to as “promoter-proximal transcription pausing (PPTP).” Paused RNA polII molecules are poised to undergo rapid entry into transcription elongation mode if provided with adequate signaling cues. The paused RNA polII complex is associated with additional cofactors NELF and DSIF (containing the proteins Spt4 and Spt5). Following phosphorylation of NELF, DSIF, and the C-terminal tail of RNA polII at Ser-2 (S2) by the kinase P-TEFb, NELF is released from the RNA polII stalled complex, and this event signals that the RNA polII can now enter the elongation phase. Capping of the nascent transcript associated with the paused RNA polII promotes RNA polII entry into the elongation phase (Cramer et al. 2008; Cheung and Cramer 2012). It is believed that AID associates with one or many states of the RNA polII following “promoter escape” (Pavri and Nussenzweig 2011; Kenter 2012; Keim et al. 2013).
Recent studies using a combination of genome sequencing of AID-expressing B cells devoid of repair pathways (thus a lack of repair of AID-generated DNA lesions genome-wide) and AID-DNA chromatin immunoprecipitation (ChIP) studies have revealed that AID can mutate various parts of the B-cell genome (Liu et al. 2008; Pavri et al. 2010; Yamane et al. 2011). Based on these studies, it has been proposed that AID identifies its potential target sequences genome-wide by binding to the Spt5-containing transcriptional “pausing” complex and promoting DNA deamination due to the stimulatory role of its cofactor, the ncRNA processing complex RNA exosome (Pavri et al. 2010; Basu et al. 2011; Stavnezer 2011; Kenter 2012). In a previous study, we proposed that the stalled RNA polII-bound AID complex is first identified by the 3′–5′ RNA exonuclease RNA exosome (Liu et al. 2006; Lykke-Andersen et al. 2009) that then displaces the RNA from the DNA/RNA hybrid formed in the transcription complex bubble. The displacement provides AID access to the ssDNA on template and nontemplate strands of the RNA polII-associated transcription bubble at AID target sequences in the Ig locus (Basu et al. 2011). We reported that in in vitro reactions, AID can deaminate cytosines on template and nontemplate strands of transcribed dsDNA in the presence of the purified RNA exosome complex (Basu et al. 2011). How the 3′–5′ RNA exonucleolytic RNA exosome identifies a transcriptionally stalled RNA polymerase complex and displaces the nascent RNA from the transcription complex to generate ssDNA substrates on which AID can act is an important question. Paused RNA polII can undergo two fates. First, it can backtrack and realign with the nascent transcript to continue transcription with the help of its cofactor TFIIS (which cleaves the 3′ end of the nascent transcript to effectively realign the transcript with the template DNA), allowing transcription elongation (Fig. 1A). Alternatively, the RNA polII can be destabilized by its ubiquitination to effect transcription termination and thereby prevent accumulation of stalled RNA polII on the DNA (Anindya et al. 2007, 2010; Svejstrup 2010). It has been reported that degradation-inducing polyubiquitination of RNA polII is initiated by a monoubiquitination event signaled by the HECT domain-containing E3 ligase Nedd4 (Fig. 1A; Rotin and Kumar 2009). Following Nedd4-mediated K63 monoubiquitination, RNA polII is polyubiquitinated by another enzyme at a K48 site to promote proteasome-mediated degradation (Harreman et al. 2009). Here we investigate the state of RNA polII that is complexed with AID and its target sequences at the Ig locus and in the remainder of the genome. We observed that Nedd4 ubiquitinates the AID-associated transcription complex to regulate its CSR-catalyzing activity at the Ig locus. We provide evidence that Nedd4 activity is required for generation of RNA exosome target sequences at potential oncogenic “off-targets” of AID, such as c-myc. Taken together, these observations generate compelling evidence for the mechanism by which AID bound with stalled RNA polII uses RNA exosome to mutate its DNA substrates without allowing the initiation of catastrophic genomic instability.
AID-associated transcription pausing complex in B cells contains E3-ubiquitin ligase Nedd4. (A) Schematic of possible “transition states” of transcriptionally paused Spt5-associated RNA polII complex in transcribed Ig switch (IgS) sequences in B cells. Following TFIIH-driven promoter escape, transcriptionally stalled RNA polII may enter the transcription elongation phase through a combination of backtracking and reinitiation (a TFIIS-dependent process) or undergo pausing and destabilization followed by ubiquitination-mediated degradation (a Nedd4-dependent process). (B) 5′IgSμ bound by AID, ExoSc3, Nedd4, TFIIS, or TFIIH in mouse primary B cells stimulated for CSR to IgG1, determined with ChIP assay. IgSα is a switch sequence that is not transcriptionally activated and acts as a negative control. (C) Interaction of AID with Nedd4, TFIIS, ExoSc3, and Spt5 observed by AID coimmunoprecipitation followed by Western blotting in CSR-stimulated B cells. (D) Two rounds of immunoprecipitation with extracts of CH12-F3 cells stimulated for IgA CSR and treated with MG132. The first immunoprecipitates with AID antibody (lane 2) were boiled and then subjected to immunoprecipitation with K48 or K63 linkage-specific poly-Ub antibodies (lanes 3,4) along with IgG control (lane 5).
ResultsNedd4 binds the AID complex in B cells
Following transcriptional pausing, RNA polII can either (1) backtrack and reinitiate its transcriptional attempts, (2) terminate its transcriptional attempts, or (3) overide the whole pausing process and enter the elongation phase. Possibilities 1 and 2 are outlined in Figure 1A. To determine which of these states is attained by the RNA polII during transcription of switch sequence IgSμ, we used ChIP to monitor enrichment of IgSμ sequences from B cells stimulated for IgG1 CSR with all of the following markers: TFIIH (a marker for the RNA polII complex in cells that have undergone “promoter escape”), TFIIS (RNA polII backtracking and nascent RNA cleavage), Nedd4 (RNA polII ubiquitination), and known IgSμ-binding proteins Spt5, AID, and Exosc3. In these experiments, we isolated naïve B cells from the spleen and stimulated them to undergo CSR to IgG1 using a cocktail of LPS and IL4. After 2 d of CSR stimulation, we performed ChIP reactions with the above-mentioned proteins and identified the binding efficiencies to switch sequence IgSμ with the help of quantitative PCR (qPCR) (for details, see the Materials and Methods). We found that Nedd4, TFIIS, and TFIIH bind to the nonrepetitive assayable DNA regions of IgSμ (5′IgSμ), indicating that RNA polII (as identified by TFIIH) is present on 5′IgSμ poised to undergo backtracking and elongation (as marked by the presence of TFIIS) and/or to undergo ubiquitination-mediated destabilization (as marked by the binding of Nedd4) (Fig. 1B). However, we realize that not all 5′IgSμ-bound RNA polII isolated from a heterogeneous population of CSR-stimulated B cells will be in the same configuration simultaneously. Some 5′IgSμ bound by RNA polII will be in the elongation phase, some will be in the backtracking phase, and a small subset will be undergoing ubiquitination-mediated degradation. Next, we wanted to know which of these RNA polII complexes is associated with AID (schematized in Fig. 1A). To determine the state of AID-bound RNA polII in B cells, we immunoprecipitated the AID complex from nuclear extracts obtained from AID−/− and AID+/+ mouse splenic B cells that were stimulated for CSR (to IgG1) and evaluated the presence of various components of the transcription complex (for details, see the Materials and Methods). We observed that AID immunoprecipitates RNA polII stalling factor Spt5, RNA exosome subunit Exosc3, and RNA polII-destabilizing E3-ubiquitin ligase Nedd4 (Fig. 1C, lane 4). We did not find the marker for backtracking RNA polII complex (TFIIS) in our AID immunoprecipitates (Fig. 1C). In support of some of these observations, interaction-mapping analysis based on published literature provides evidence that AID family member protein APOBEC3G (Conticello et al. 2005) forms a functionally relevant complex with RNA polII and Nedd4 (Supplemental Fig. S1A; Dussart et al. 2005). In Supplemental Figure S1A, we show the interaction network of APOBEC3G with Nedd4 and RNA polII. We also performed immunoprecipitation reactions in mouse B cells with TFIIS and detected the presence of TFIIS and Spt5 in the immunoprecipitate but not AID (Supplemental Fig. S1B). We conclude that AID interacts with stalled RNA polII component Spt5, RNA exosome, and Nedd4. The Spt5-associated, AID-associated, and RNA exosome-associated RNA polII is enriched with Nedd4 but does not contain detectable levels of TFIIS (Fig. 1C).
To determine whether AID-associated RNA polII is ubiquitinated in B cells and probe the nature of this ubiquitination linkage, we immunoprecipitated AID-associated RNA polII (first immunoprecipitation) from a B-cell line (CH12F3) that can undergo CSR to IgA following stimulation with LPS, IL-4, and TGFβ cytokines. We boiled this sample to remove protein–protein interactions and then immunoprecipitated RNA polII with anti-K48Ub (Fig. 1D, lane 3) or anti-K63Ub (Fig. 1D, lane 4) antibodies to determine the nature of the specific ubiquitination linkage. We probed the K48-modified or K63-modified RNA polII immunoprecipitate with specific antibodies against Ser-5 phosphorylated RNA polII or Ser-2 phosphorylated RNA polII. We observed in repeated experiments that AID is bound to Ser-5 phosphorylated RNA polII (higher exposures reveal the Ser-5 phosphorylated RNA polII signal on the Western blot) (see Supplemental Fig. S1C) and also to Ser-2 phosphorylated RNA polII (Fig. 1D; Supplemental Fig. S1C). Moreover, we observed that Ser-2 phosphorylated and ubiquitinated RNA polII immunoprecipitates with AID and is enriched at a higher molecular weight due to ubiquitination (Fig. 1D). We note that in addition to K48 linkage, RNA polII may be modestly modified by a K63-Ub linkage (Fig. 1D). Taken together, these experiments show that in B cells, AID associates with RNA polII ubiquitinated with a K48 linkage; this RNA polII has undergone promoter escape and is poised to enter elongation, since it is marked by RNA polII C-terminal domain (CTD) Ser-2 phosphorylation. Consistent with reports using yeast as a model system, we found that Ser-2 phosphorylated RNA polII is a target of ubiquitination in B cells (Harreman et al. 2009).
Nedd4 promotes AID activity in B cells
To check whether Nedd4 is involved in AID function, we generated Nedd4 knockdown CH12F3 cells by lentiviral transduction of shRNA specifically targeting Nedd4 mRNA (we refer to these cells as shNedd4). A nonmammalian shRNA control (SHC002) was prepared as well. CH12F3 cells can be stimulated for IgA CSR in ex vivo conditions following incubation with LPS, IL4, and TGFβ. We observed a clear diminution in CSR efficiency to IgA in shNedd4 cells after 72 h of stimulation (Fig. 2A, left panel). The reduction in CSR in shNedd4 cells compared with SHC002 control cells was confirmed in five separate experiments (Fig. 2A, right panel; three individual experiments shown in Supplemental Fig. S2A). To check whether the CSR deficit in shNedd4 cells results from the change of expression level of important proteins for CSR such as AID and RNA exosome, we examined the protein levels of shNedd4 and SHC002 cells by Western blot. Using actin as a loading control, it was immediately evident that shNedd4 cells express reduced levels of Nedd4 protein, while the expression of the other proposed components of the AID/RNA polII complex in these cells, such as phosphorylated RNA polII (Rpb1: Ser-2 and Ser-5 phosphorylated), Spt5, AID, or Exosc3, did not show any appreciable decrease in their expression levels (Fig. 2B). Normal cell proliferation is required for optimal CSR. To exclude the possibility that the CSR deficit in shNedd4 cells is due to defective proliferation of shNedd4 cells, we examined cell proliferation of shNedd4 and SHC002 cells via cell number counting and observed no variation in proliferation between shNedd4 cells and control SHC002 cells after 72 h in culture (Supplemental Fig. S2B). We confirmed that observation by using the VPD450 dye dilution technique (Supplemental Fig. S2C). Another requirement during CSR is the robust transcription of switch sequences. We found that shNedd4 cells do not have a deficiency in switch sequence transcripts (Iμ or Iα) (discussed later; see Fig. 4; Supplemental Fig. S5). Since Nedd4 shares some protein homology with another potential E3-ligase, Nedd4L (Nedd4-2) (Kamadurai et al. 2009), we also generated a knockdown line of Nedd4L protein in CH12F3 cells. We confirmed knockdown of Nedd4 and Nedd4L transcripts by qPCR and found that the transcript levels of both genes were approximately half of those expressed in CH12F3 cells (Supplemental Fig. S3A). We found that shNedd4 cells are defective in CSR, but the shNedd4L cells undergo IgA CSR similar to that of the control SHC002 (wild-type) cells (Supplemental Fig. S3B).
Nedd4-deficient B cells have impaired CSR. (A) Representation of CSR efficiency of shNedd4 and SHC002 CH12F3 cells stimulated for IgA CSR (left) from five independently performed experiments (right). (—) The mean. (B) Protein expression in shNedd4 and SHC002 cells illustrated with Western blotting. IgG3 CSR (C), IgG1 CSR (D), and protein expression (E) of B cells obtained from Nedd4−/− and Nedd4+/+ mice. AID−/− is a negative control for CSR; different chimeric mice were used in each figure for the data of each Nedd4 genotype.
Since the cell line CH12F3 might override certain epigenetic programs by virtue of its transformed status, we evaluated the role of Nedd4 in stimulating CSR in primary B cells. For this purpose, we obtained fetal liver chimeric mice that express a C-terminally deleted form of Nedd4 (consistent with previous published literature, we refer to these B cells as Nedd4−/− [Yang et al. 2008]) (for details of the generation of these chimeric mice, see the Materials and Methods). As a control, we also obtained chimeric mice that expressed wild-type Nedd4 (Nedd4+/+). Using LPS, we stimulated splenic B cells from these mice to undergo CSR to IgG3 and with anti-CD40 antibody together with IL4 for CSR to IgG1. We found that CSR to IgG3 (Fig. 2C) and IgG1 (Fig. 2D) is reduced in Nedd4−/− B cells in comparison with Nedd4+/+ controls. We show the defect in IgG3 CSR levels using a FACS plot, since efficiency of IgG3 CSR in ex vivo cultures is low, and FACS plots are widely accepted as the proper representation of the actual phenotype. The efficiency of IgG1 CSR is high, and thus we represent the data using a quantitative plot. We used these IgG1CSR-stimulated B cells to prepare protein extracts from the Nedd4+/+ and Nedd4−/− B cells and observed the loss of full-length Nedd4 expression in the Nedd4−/− B cells (Fig. 2E, top panel). However, Nedd4 deficiency does not affect AID expression levels in these cells (Fig. 2E, middle panel). We performed growth curves using these B-cell cultures and concluded that Nedd4−/− and Nedd4+/+ B cells proliferated similarly during the 72-h assay (Supplemental Fig. S2D). To determine whether AID activity is indeed decreased during IgG1 CSR, we evaluated the level of mutation at the 5′ end of the IgSμ switch regions (donor switch sequence and direct target of AID activity) and found reduced AID-induced mutations in Nedd4−/− B cells (∼50% reduction of overall mutation frequency in Nedd4−/− B cells) (Supplemental Table SI). We would have liked to measure the mutation frequency at the core IgSμ, a region where we expect complexation of AID/Spt5/RNA exosome/Nedd4 (discussed in our proposed model in Fig. 6, below), but these regions are difficult to clone due to their G-richness and other DNA sequence properties.
Next, we examined whether Nedd4 deficiency may induce genomic instability in B cells and lead to inhibition of CSR. For that purpose, we assayed for genomic stability in Nedd4+/+ and Nedd4−/− B cells using telomeric FISH (T-FISH) assays (Franco et al. 2006). In these T-FISH assays, chromosome instability can be detected easily by observing loss of labeled telomeres. We did not find any observable genomic instability in Nedd4−/− B cells in comparison with Nedd4+/+ cells (Supplemental Fig. S4). Taken together, we conclude that Nedd4 functions to promote CSR for both IgG3 and IgG1 isotypes in primary B cells (Fig. 2C,D). Nedd4 deficiency also leads to decreased CSR to IgA in CH12F3 cells (Fig. 2A). Thus, Nedd4 is an important component of the CSR machinery in B cells at multiple isotypes in the IgH locus. Based on these observations, we proceeded to determine the mechanism by which Nedd4's E3 ubiquitination activity promotes AID function during CSR.
Nedd4 promotes AID interaction with its cofactors, Spt5 and RNA exosome, on transcribed Ig switch sequences in B cells
As Nedd4 binds the AID complex and promotes AID function in B cells, we wanted to evaluate whether the level of RNA exosome subunit Exosc3 and Spt5, two marker proteins of the active AID/RNA polII complex, associates with AID in a Nedd4-dependent fashion. We noted that Nedd4 deficiency does not affect the overall expression of AID, Spt5, or Exosc3, as the total content of these proteins in the input samples is comparable in SHC002 and shNedd4 cells (Fig. 3A). We then immunoprecipitated AID from SHC002 and shNedd4 cells and found that in the shNedd4 cells, the steady-state level of the AID/RNApolII/RNA exosome complex is significantly reduced, as observed by the lack of Spt5 and Exosc3 in AID immunoprecipitates obtained from shNedd4 cells (Fig. 3A). However, at this moment, it is not possible to conclude whether Nedd4 influences the order of recuitment of RNA exosome and AID to the stalled RNA polII complex.
Nedd4 promotes AID interaction with its cofactors, Spt5 and RNA exosome, on transcribed Ig switch (IgS) sequences while decreasing steady-state levels of RNA polII bound to IgS. (A) Coimmunoprecipitation assays to evaluate interaction of AID with Spt5 and Exosc3 in SHC002 or shNedd4 cells. (B) Association with IgSμ switch region DNA by AID and Exosc3 in SHC002 and shNedd4 cells, determined with ChIP. Ubiquitination of RNA polII (C) and steady-state binding of RNA polII at IgSμ (D,E) in SHC002 and shNedd4 cells treated with MG132. (St) Stimulated for CSR; (Un) without CSR stimulation. The ChIP products were either assayed by DNA gel electrophoresis (D) or quantitated by qPCR (E).
To determine the effect of Nedd4 deficiency on AID and RNA exosome binding to IgSμ, we used ChIP assays in SHC002-CH12F3 or shNedd4-CH12F3 cells. We found that AID and Exosc3 bind to IgSμ following CSR stimulation in CH12F3 (SHC002) cells, but this interaction is reduced in Nedd4-deficient (shNedd4) cells (Fig. 3B). Thus, we conclude that in B cells, Nedd4 facilitates the complexation of AID and its cofactors, RNA exosome and Spt5, on its physiological DNA substrates. We proceeded to seek a mechanistic interpretation of Nedd4-dependent AID interaction with the RNA polII-associated Spt5 and RNA exosome complex during CSR.
Nedd4 controls the steady-state level of IgS-associated RNA polII by its ubiquitination activity
We focused on determining the regulation of IgSμ-bound RNA polII by Nedd4. From previously published literature, we were aware that monoubiquitination of RNA polII by Nedd4 at sites of DNA damage marks the RNA polII complex for polyubiquitination-mediated degradation (Anindya et al. 2007; Harreman et al. 2009). Given that Nedd4 is a component of the AID-bound transcription complex and binds to transcription-activated IgSμ (Fig. 1B,C), we investigated whether RNA polII ubiquitination is Nedd4-dependent and determined its accumulation levels in transcribed IgSμ. We assayed for RNA polII binding to IgSμ in CH12F3 cells proficient and deficient in Nedd4. We chose to analyze only IgSμ, since it is the most robustly transcribed switch sequence and thus provides the best opportunity to perform biochemical assays that require protein complex detection. We performed ubiquitinated protein immunoprecipitation experiments with IgA CSR-stimulated cells and observed that unlike in the SHC002 cells, shNedd4 cells have reduced ubiquitinated RNA polII (Fig. 3C, cf. lanes 1–3 and 7–9). In contrast, we did not see a significant decrease in the levels of ubiquitinated PCNA in these cells (a known monoubiquitinated protein in B cells) (Langerak et al. 2007) and saw only a slight change of AID ubiquitination (Fig. 3C). A fraction of AID is ubiquitinated in B cells, as has been reported previously by other groups (Aoufouchi et al. 2008; Delker et al. 2013). These results demonstrate that in CSR-stimulated B cells, a portion of cellular RNA polII is ubiquitinated by Nedd4.
We subsequently wanted to determine whether the steady-state level of IgSμ-bound RNA polII is dependent on the Nedd4 activity in these B cells. We chromatin-immunoprecipitated RNA polII from SHC002 and shNedd4 cells and assayed for RNA polII association with IgSμ by conventional PCR (Fig. 3D) and qPCR (Fig. 3E). By both methods, we observed that there is increased accumulation of RNA polII on IgSμ in shNedd4 cells in comparison with SHC002 controls. Taken together, these experiments demonstrate a role of Nedd4 in the turnover of RNA polII resident at IgS sequences during CSR in activated B cells, potentially by a polyubiquitination-mediated degradation event.
Because there is more RNA polII associated with IgS in Nedd4-deficient B cells, we checked whether there are more IgS transcripts in Nedd4-deficient B cells. We observed that in shNedd4 cells, IgSμ and IgSα germline transcripts levels are stabilized at levels above those seen in control SHC002 cells (Supplemental Fig. S5A,B). Consistently, we also found that in Nedd4−/− primary B cells, the steady-state level of IgSγ1 switch sequence transcript is higher than that seen in Nedd4+/+ B cells upon CSR stimulation (Supplemental Fig. S5C). The stabilizing effect of germline transcripts in Nedd4-depleted CH12F3 cells (Supplemental Fig. S5A,B) is much higher than that seen in Nedd4 mutant primary B cells (Supplemental Fig. S5C). This may be due to the fact that CH12F3 cell lines survive better than primary B cells in culture and accumulate more RNA, thus providing robust means to evaluate increased accumulation of germline transcripts. Based on these observations, we conclude that Nedd4 determines CSR efficiency in B cells, potentially by cotranscriptionally regulating the steady-state levels of RNA polII associated with IgS switch regions.
Nedd4 regulates ncRNA biogenesis at AID target sequences
If Nedd4 mediates destabilization of RNA polII at AID target sequences and activates RNA exosome-mediated degradation of nascent ncRNAs, we would expect that there would be an accumulation of RNA exosome substrate transcripts at AID target loci in Nedd4−/− B cells. To determine whether this is indeed the case, we decided to perform RNA sequencing of the whole genome of Nedd4+/+ and Nedd4−/− B cells. We isolated ribosomal RNA (rRNA)-depleted total RNA from Nedd4+/+ and Nedd4−/− B cells following CSR activation to IgG1. We performed high-throughput RNA sequencing and analyzed the levels of various coding RNAs and ncRNAs in these B cells (for details of total mapped reads, see Supplemental Fig. S6A). We quantitated the genome-wide exome expression level in Nedd4+/+ and Nedd4−/− cells and found slight but not statistically significant differences in mRNA levels genome-wide over the total genome expression. We listed the read counts per million bases analyzed of the Nedd4+/+ and Nedd4−/− coding gene transcriptomes (Supplemental Table S2) and identified pathways that could vary between these two genotypes (see Supplemental Fig. S6B,C for pathways that could be up-regulated or down-regulated in the Nedd4−/− B cells, respectively). We did not observe significant changes in the subset of genes that are expressed in the Nedd4+/+ and Nedd4−/− samples, although there are a few genes that are up-regulated or down-regulated as shown in Figure 4A. As determined using STRING pathway analysis software (Supplemental Fig. S6B,C), we did not observe any changes in pathways that directly affect transcription, cell cycle progression, or DNA repair pathways that indirectly could have affected CSR. We then assayed for various types of RNAs that are noncoding in nature to determine whether transcription of the noncoding genome by RNA polII is affected in Nedd4−/− B cells. A family of ncRNAs that is expressed in B cells is large intergenic ncRNAs (lincRNAs). We are interested in this subgroup, since IgS germline transcripts share a number of properties with lincRNAs, including the observations that germline transcripts do not code for proteins, are spliced, and are polyadenylated, all three properties having also been observed in recently characterized lincRNAs (Rinn and Chang 2012). Indeed, we found that lincRNA levels are increased in Nedd4−/− B cells in comparison with those present in Nedd4+/+ cells (Fig. 4B). (A detail of the TSS and transcription end site of each of the lincRNAs that we analyzed is presented in Supplemental Table S3.) Based on this finding, we specifically inspected the steady-state expression level of germline transcript IgSγ1 in Nedd4−/− B cells. We chose IgSγ1 transcript levels due to the fact that they are expressed in primary B cells only following CSR stimulation, are known to form a secondary RNA structure, and are polyadenylated and thus represent closely the canonical description of lincRNAs. We observed that the percentage of mapped reads (normalized to the remainder of the RNA genome) at the IgSγ1 region is increased in Nedd4−/− B cells over that seen in Nedd4+/+ cells (Fig. 4C) in two separate RNA sequencing experiments. The increase in the IgSγ1 mapped reads per million reads of the total Nedd4−/− transcriptome is ∼25%–40% compared with that found in the Nedd4+/+ control.
Nedd4-deficient B cells have increased levels of long ncRNA genome-wide. Whole-genome RNA sequencing from Nedd4−/− and Nedd4+/+ B cells grown under conditions of IgG1 CSR for 3 d. (A) Gene expression of coding genes in Nedd4+/+ and Nedd4−/− B cells was evaluated from RNA sequencing data sets for genes that are diffentially expressed in the two sets. Fold change in gene expression of individual genes for Nedd4−/− B cells over that of Nedd4+/+ B cells is presented. Coding genes expressed at the same levels in Nedd4−/− and Nedd+/+ B cells are represented by blue dots, genes expressed higher in Nedd4−/− B cells are represented by red dots (log2 sum reads per kilobase per million mapped reads [RPKM] > 0; log2 fold change Nedd4−/−/Nedd4+/+ > 1), and genes that have decreased expression in Nedd4−/− B cells are represented by green dots (log2 sum RPKM > 0; log2 fold change Nedd4−/−/Nedd4+/+ < −1). (B) Large ncRNA subtype lincRNA expression levels in Nedd4+/+ and Nedd4−/− B cells. lincRNAs identified by Guttman et al. (2009) were assayed for expression levels in this plot. (C) Summation of total transcript reads of the IgSγ1 switch sequence in Nedd4+/+ and Nedd4−/− B cells from two separate analyses performed on RNA isolated from two different mice of each genotype.
Another type of ncRNA that could be a target of the RNA exosome complex is TSS-associated ncRNA (TSS-ncRNA). TSS-ncRNAs have been implicated in recruitment and regulation of RNA polII during the transcription initiation phase, although the exact role in this process is not understood. We evaluated the expression levels of TSS-RNAs genome-wide for various loci and found that there is no significant alteration in the spectrum of genes that express TSS-ncRNAs in Nedd4+/+ and Nedd4−/− B cells (Fig. 5A). We evaluated and found no differences in the start site positioning or length of TSS-ncRNAs in the Nedd4−/− and Nedd4+/+ B cells compared with TSSs genome-wide (Fig. 5B). These findings indicate that Nedd4 activity deficiency does not alter the distribution or location of TSS-ncRNAs genome-wide. Since we found that Nedd4 interacts with RNA polII associated with AID (Fig. 1C), we decided to investigate the levels of TSS-associated transcripts in Nedd4−/− cells specifically at genes that have been previously reported to be mutated by AID or bound with AID in B cells based on data obtained from AID ChIP assays (Fig. 5C). Recently published literature indicates that on a genome-wide basis, AID binds to many genes where RNA polII stalling occurs. We found that AID-bound genes have higher amounts of TSS-RNAs in Nedd4−/− B cells compared with those observed in Nedd4+/+ B cells (Fig. 5C). We wanted to directly visualize and evaluate the presence of TSS-RNA at AID “off-target” genes. We are aware that c-myc has been reported in various studies to be a robust “off-target” of AID (Ramiro et al. 2006; Liu et al. 2008; Pasqualucci et al. 2008). More importantly, AID-mediated mutations in c-myc and p53 have now been established to initiate Burkitt's lymphoma (Ramiro et al. 2007) and gastric epithelial cancer (Matsumoto et al. 2007; Shimizu et al. 2012), respectively. As seen in Figure 5, D and E, the level of upstream TSS-associated transcripts at c-myc is increased in Nedd4−/− B cells. In quantitative terms, the level of upstream transcripts at the c-myc TSS is approximately fourfold higher in Nedd4-deficient B cells based on the level found in Nedd4+/+ B cells (P-value = 3.2700 × 10−7) when normalized to downstream reads initiating from the TSS of c-Myc (Fig. 5D). Similarly, when we assayed for the presence of TSS-associated upstream transcripts at the Trp53 locus, we found increased stabilization of these transcripts in conditions of Nedd4 deficiency (Fig. 5D,E). These experiments provide further evidence that Nedd4 regulates nascent noncoding transcript levels at certain loci that are mutatable by AID in B cells by initiating their processing and/or degradation by the AID cofactor complex RNA exosome.
Nedd4 is required for processing of TSS transcripts at AID target sequences in B cells. TSS-ncRNA in Nedd4+/+ and Nedd4−/− B cells. (A) Comparison of expressed TSS-ncRNAs in Nedd4+/+ and Nedd4−/− B cells depicted by their RPKM levels. (B) Evaluation of the TSS position and expression level of TSS-ncRNA transcripts and their cognate coding transcripts genome-wide in Nedd4+/+ or Nedd4−/− B cells. (C) Evaluation of TSS-ncRNA levels (evaluated for the 1000 base pairs [bp] upstream of the genic TSS) at nontargets and targets of AID (Yamane et al. 2011) in Nedd4−/− B cells relative to that seen in Nedd4+/+ B cells. (D) A quantitative estimate of the level of TSS-associated upstream transcripts (≤1 kb in length) at the cMyc locus (top panel) and the p53 locus (bottom panel) in Nedd4+/+ and Nedd4−/− B cells. (E) The expression profile of TSS-RNA at the c-myc locus (top panel) and the p53 locus (bottom panel) in Nedd4+/+ and Nedd4−/− B cells. The region of TSS-RNA is shown with an arrow; the RNA sequencing data plots were derived using Investigative Genomics Viewer software.
Discussion
In this study, we evaluated the fate of the RNA polII complex that stimulates AID's ability to deaminate DNA target sequences genome-wide. We propose two fates of RNA polII that it could encounter following promoter escape (a TFIIH-dependent mechanism) and after attaining transcriptional pausing: (1) cleavage of the nascent RNA in the transcription elongation complex followed by reinitiation of the polII (a TFIIS-dependent mechanism) or (2) destabilization of RNA polII via the ubiquitination pathway that leads to exposure of the nascent RNA in the collapsing transcription bubble (a Nedd4-dependent mechanism) (Anindya et al. 2007; Cheung and Cramer 2012). We found that Nedd4 and TFIIH complex with AID; these observations suggest that a combination of RNA polII backtracking and RNA polII destabilization functions as a possible mechanism that supports AID mutagenesis activity. Based on these observations and published work from other laboratories, we updated our model of how AID identifies its target sequences in the IgH locus and genome-wide and incorporates mutations. Following transcription initiation at various IgH- and non-IgH-localized DNA sequences, RNA polII undergoes promoter-proximal stalling or may undergo stalling in its elongation phase (Fig. 6B). The conditions that promote promoter-proximal stalling are in the process of being unraveled, although it is quite possible that environmental cues and promoter-proximal DNA sequences may promote RNA polII stalling (Saunders et al. 2006). RNA polII stalling at regions significantly downstream from TSSs could be caused by pretermination events of RNA polII induced by various conditions, including the presence of secondary structures on the template DNA (Li and Manley 2006; Richard and Manley 2009). In either case, the stalled RNA polII will be required to relinquish the associated nascent transcripts as an RNA exosome substrate in order to resolve the paused state. Indeed, AID-mediated mutations can occur within the first 100–500 base pairs (bp) from the TSS (associated with the promoter-proximal stalled complex), as seen during SHM in variable genes or certain other AID “off-target” genes. Mutations also occur at >2 kb downstream from TSSs (associated with the pretermination RNA polII complex), as seen in switch sequences. For the RNA exosome to be able to identify and degrade stalled RNA polII-associated transcripts at DNA sequences where AID mutates its targets, it has to be able to find a 3′ end free transcript. In this study, we provide evidence that Nedd4 induces RNA polII ubiquitination, which consequently promotes the generation of the 3′ end free nascent RNA transcript by displacing the associated RNA polII transcription complex (Fig. 6B,C). We propose that this RNA polII ubiquitination event occurs in a complex that contains Spt5, RNA exosome, AID, and Nedd4 (Fig. 6B,C); the absence of any of the components destabilizes the complex (Fig. 3A; Basu et al. 2011). We propose that once the RNA exosome degrades the nascent germline transcript associated with the AID-associated transcription complex, the template and nontemplate strands of the IgS sequence are exposed for AID-mediated mutagenesis (Fig. 6D). This model is in line with the known activity of RNA exosome. Transcriptional complex-associated ncRNAs are substrates of the RNA exosome complex. For example, in Saccharomyces pombe, RNA exosome-mediated cotranscriptional degradation of centromeric RNA has been implicated in maintenance of transcriptional silencing and genomic integrity (Buhler et al. 2007; Reyes-Turcu and Grewal 2012; Yamanaka et al. 2013). RNA exosome substrate TSS-associated ncRNA promoter upstream transcripts (PROMPTs) have also been identified in mammalian cells (Preker et al. 2008; Seila et al. 2008). In addition, other RNAs—such as viral RNAs (Zhu et al. 2011), intergenic cryptic unstable transcripts (Wyers et al. 2005), rRNA, small nucleolar RNAs (snoRNAs) (Allmang et al. 1999), etc.—are reported to be RNA exosome substrates in vivo (Houseley et al. 2006). Understanding of the transcriptional properties (length, TSS information, and directionality) of these substrate RNAs and the mechanism of their recognition by the RNA exosome complex will provide further insight into how RNA exosome identifies and processes transcripts in the IgH locus, as proposed in Figure 6.
A schematic representation of AID regulation via the transcription complex in the Ig switch sequence. (A) A simplified representation of an Ig switch region structure that contains a G-rich core sequence (blue oval) and the promoter preceding Iγ1. (B) Transcription at switch sequences can cause secondary DNA structures (R-loops) that impede and stall RNA polII (RNAP II). Moreover, stalled RNA polII (bound with Spt5 and RNA exosome) recruits AID and Nedd4. (C) Nedd4 present in the RNA polII/RNA exosome/AID complex promotes the ubiquitination of RNA polII and disengages it from the nascent RNA transcript. (D) The RNA exosome is now able to degrade the transcript in the transcription bubble to allow AID access to the template and nontemplate strands of its target sequence. In this model (adapted from Sun et al. 2013), multiple sets of the stalled AID–RNA polII–RNA exosome–Nedd4 complex are present in the R-loop ssDNA structure in the switch sequence; for the purpose of simplicity, we show two representative complexes.
We do not exclude the possibility that other mechanisms can also provide the RNA exosome access to the AID/RNA polII-associated nascent RNA. Residual CSR in Nedd4−/− B cells indicates that there could be other pathways that provide AID access to transcribed switch sequences. Interaction of AID with the replication protein complex RPA due to phosphorylation of AID at residue Ser-38 has been implicated in providing AID the ability to mutate nontemplate strand switch sequences. A high density of nontemplate strand DNA breaks can be sufficient to induce some DNA double-strand breaks and cause inefficient CSR. It is also possible that “scrunching” of RNA polII may displace the template DNA strand from the RNA and override the role of the RNA exosome and/or Nedd4 in providing AID with a suitable substrate. However, until now, “scrunching” has only been demonstrated with bacterial RNA polII (Revyakin et al. 2006), and thus its feasibility as a mechanism for generation of ssDNA with mammalian RNA polII requires further work. Negative supercoiling preceding the transcribing RNA polII can also generate ssDNA that is the target of AID protein (Shen and Storb 2004; Longerich et al. 2006). Finally, we do realize that transcription-coupled DNA mutagenesis by the nucleotide excision repair (NER) pathway may also provide AID a mechanism to initiate and spread mutations in the genome. NER components are associated with factors that promote “promoter escape.” However, we note that patients with NER pathway mutations do not manifest any defects in SHM, the physiological target of AID activity (Kim et al. 1997).
Proper distribution and rapid resolution of AID-induced mutations at switch sequences and variable region genes is important, since they can otherwise be intermediates for deleterious chromosomal translocations. On the other hand, it is also important to remove residual stalled RNA polII resident genome-wide, which may have been generated during the G1 phase of the cell cycle and prior to the onset of DNA replication during S phase. Indeed, a low level of AID-promoted mutations may induce RNA polII stalling at various regions of the B-cell genome during the G1 phase of the cell cycle. In addition, stalled RNA polII may also induce AID-independent mutations in the B-cell genome (Unniraman et al. 2004; Barlow et al. 2013). In either case, failure to remove stalled RNA polII may induce collision of these transcription complexes with components of the replication machinery during S phase and generate DNA double-strand breaks. Thus, the role of Nedd4 in destabilizing stalled RNA polII molecules that are stalled genome-wide in an AID-dependent or AID-independent manner may have significant implications in preventing genomic instability. HECT domain E3 ligases have been implicated in the prevention of various oncogenic events by catalyzing proteasomal degradation of various oncogenes like PTEN, p53, Notch-1, etc.; thus, understanding its mechanism of function in B cells is important (Bernassola et al. 2008). Our work points toward a novel role of Nedd4 as a checkpoint of oncogenesis by prevention of aberrant mutagenesis of the B-cell genome.
Materials and methodsAntibodies and plasmids
AID antibodies were generated as described (Chaudhuri et al. 2003). Anti-exosome subunit antibodies were purchased from GenWay Biotech, Inc., or Abcam, Plc. Details of additional antibodies are as follows: Actin was purchased from Sigma-Aldrich; Nedd4 was purchased from R&D Systems; rabbit IgG and RNA polII (4H8) were purchased from Abcam; Spt5, TFIIS, and TFIIH p52 were purchased from Santa Cruz Biotechnology; and IgG1, IgA, and B220 were purchased from Becton Dickinson. All secondary HRP-conjugated antibodies were purchased from Sigma. shNedd4 (TRCN0000092433), shNedd4L (TRCN0000086869), nonmammalian shRNA control (SHC002), and MISSION Lentiviral Packaging Mix (SHP001) were all purchased from Sigma.
Cell culture, transfection, infection, and selection
Splenic B cells were prepared with CD43-negative selection and cultured in RPMI1640 medium containing 15% FBS plus 20 μg/mL LPS and/or 20 ng/mL IL-4. CH12F3 cells were maintained in RPMI1640 medium containing 10% FBS or were stimulated for IgA CSR with the addition of 20 μg/mL LPS, 20 ng/mL IL-4, and 1 ng/mL TGFβ1. HEK293T cells were cultured in DMEM with 10% FBS. HEK293T cells were transfected with specific plasmids for protein production. All cells were incubated at 37°C in a 5% CO2 humidified incubator. HEK293T cells were cotransfected with shRNA plasmid and MISSION Lentiviral Packaging mix for lentivirus production. (For details of lentiviral knockdown assays, see the Supplemental Material.)
RNA extraction, protein preparation, immunoprecipitation, and ChIP
For RNA extraction, protein preparation, immunoprecipitation, and ChIP, see the Supplemental Material.
Real-time qPCR and primers
Real-time PCR was performed with SYBR Green ROX (Roche Applied Science) using Eppendorf Realplex2. For quantification, a standard curve was established with a serial dilution of samples with each primer pair. Assays for germline S region transcript levels were performed according to published protocols (Muramatsu et al. 2000). Primer pairs for germline transcripts were as follows: Iμ, 5′-CTCTGGCCCTGCTTATTGTTG-3′ and 5′-GAAGACATTTGGGAAGGACTGACT-3′; Iα, 5′-CCTGGCTGTTCCCCTATGAA-3′ and 5′-GAGCTGGTGGGAGTGTCAGTG-3′; and Iγ1, 5′TATGATGGAAAGAGGGTAGCATT-3′ and 5′-CTGGGCTGGTCTGTCAACTCCTT-3′.
Other q-PCR primers were GAPDH, 5′-TGGCCTTCCGTGTTCCTAC-3′ and 5′-GAGTTGCTGTTGAAGTCGCA-3′; Nedd4, 5′-GTGGGAAGAGAGGCAGGATGTC-3′ and 5′-GCGAATTCACAGGAAGTGTAGGC-3′; and Nedd4L, 5′-GAGGCTCCAGTTCATGTGGG-3′ and 5′-GGATACGGGATTCTCCCTGTC-3′. Primer pairs for switch region ChIP were 5′Sμ, 5′-TAGTAAGCGAGGCTCTAAAAAGCAT-3′ and 5′-AGAACAGTCCAGTGTAGGCAGTAGA-3′; and Sα, 5′-TGAAAAGACTTTGGATGAAATGTGAACCAA-3′ and 5′-GATACTAGGTTGCATGGCTCCATTCACACA-3′.
High-throughput sequencing
Splenic B cells of Nedd4−/− and Nedd4+/+ were cultured with LPS and IL-4 for 3 d, and rRNA-depleted total RNA was prepared from these cells as described in the Supplemental Material. Libraries were prepared with Illumina's TruSeq RNA prep kit and then sequenced with 50 million to 60 million 2 × 100-bp paired raw passing filter reads on an Illumina HiSeq 2000 V3 instrument at the Columbia Genome Center. We mapped the pass filter reads to the genome (human, NCBI build37; mouse, UCSC mm9) using TopHat version 2.0.4 (Trapnell et al. 2009). TopHat infers novel exon–exon junctions ab initio and combines them with junctions from known mRNA sequences (refgenes) as the reference annotation. For each read, we allowed up to two mismatches and 10 multiple hits during the mapping. The gene expression level was calculated using FPKM (fragment per kilobase transcriptome per million mapped reads) by Cufflinks version 2.0.2. The RNA sequencing data from Nedd4−/− and Nedd4+/+ cells can be obtained from Gene Expression Omnibus (accession no. GSE49027).
Nedd4-deficient mouse B cells
The generation of Nedd4+/+ and Nedd4−/− fetal liver chimeras was described previously (Yang et al. 2008). Heterozygous mice encoding a gene trap inserted between the first two WW domains of Nedd4 (Bay Genomics, XA398) were crossed in timed matings, and fetal livers were harvested at day 16 of fetal development. Fetal liver cell suspensions were frozen in medium containing 90% FCS and 10% DMSO, while DNA samples from the fetuses were genotyped. Fetal liver suspensions from Nedd4+/+ and Nedd4−/− embryos were transferred by tail vein injection into lethally irradiated 6- to 10-wk-old Rag1−/− recipients that had received a “split dose” of 800 and 400 rads separated by 2–4 h. Each fetal liver was used to reconstitute five lethally irradiated recipients. Recipient mice were analyzed between 7 and 9 wk post reconstitution.
Acknowledgments
We thank Drs. Lorraine Symington, Kang Liu, and Gerson Rothschild for critically reading the manuscript, and other members of the Basu laboratory for reagents and discussions. We thank Dr. Vundavalli Murty (Columbia University) for help in the metaphase analysis of Nedd4-deficient B cells, and Dr. Olivier Couronne (Columbia University Genome Center) for RNA sequencing. This work was supported by grants from the NIH (1DP2OD008651-01) and National Institute of Allergy and Infectious Diseases (NIAID) (1R01AI099195-01A1), Trustees of Columbia University faculty startup funds, and the Irma Hirschl Charitable Trust (to U.B.). D.K. is supported by a grant from the NIAID (F31AI098411-01A1).
Supplemental material is available for this article.
Article is online at http://www.genesdev.org/cgi/doi/10.1101/gad.210211.112.
ReferencesAllmangC, KufelJ, ChanfreauG, MitchellP, PetfalskiE, TollerveyD1999Functions of the exosome in rRNA, snoRNA and snRNA synthesis.
18: 5399–541010508172AltFW, RosenbergN, EneaV, SidenE, BaltimoreD1982Multiple immunoglobulin heavy-chain gene transcripts in Abelson murine leukemia virus-transformed lymphoid cell lines.
2: 386–4006810096AnindyaR, AygunO, SvejstrupJQ2007Damage-induced ubiquitylation of human RNA polymerase II by the ubiquitin ligase Nedd4, but not Cockayne syndrome proteins or BRCA1.
28: 386–39717996703AnindyaR, MariPO, KristensenU, KoolH, Giglia-MariG, MullendersLH, FousteriM, VermeulenW, EglyJM, SvejstrupJQ2010A ubiquitin-binding domain in Cockayne syndrome B required for transcription-coupled nucleotide excision repair.
38: 637–64820541997AoufouchiS, FailiA, ZoberC, D'OrlandoO, WellerS, WeillJC, ReynaudCA2008Proteasomal degradation restricts the nuclear lifespan of AID.
205: 1357–136818474627BarlowJH, FaryabiRB, CallenE, WongN, MalhowskiA, ChenHT, Gutierrez-CruzG, SunHW, McKinnonP, WrightG, 2013Identification of early replicating fragile sites that contribute to genome instability.
152: 620–63223352430BasuU, MengFL, KeimC, GrinsteinV, PefanisE, EcclestonJ, ZhangT, MyersD, WassermanCR, WesemannDR, 2011The RNA exosome targets the AID cytidine deaminase to both strands of transcribed duplex DNA substrates.
144: 353–36321255825BernassolaF, KarinM, CiechanoverA, MelinoG2008The HECT family of E3 ubiquitin ligases: Multiple players in cancer development.
14: 10–2118598940BesmerE, MarketE, PapavasiliouFN2006The transcription elongation complex directs activation-induced cytidine deaminase-mediated DNA deamination.
26: 4378–438516705187BuhlerM, HaasW, GygiSP, MoazedD2007RNAi-dependent and -independent RNA turnover mechanisms contribute to heterochromatic gene silencing.
129: 707–72117512405ChaudhuriJ, TianM, KhuongC, ChuaK, PinaudE, AltFW2003Transcription-targeted DNA deamination by the AID antibody diversification enzyme.
422: 726–73012692563ChaudhuriJ, BasuU, ZarrinA, YanC, FrancoS, PerlotT, VuongB, WangJ, PhanRT, DattaA, 2007Evolution of the immunoglobulin heavy chain class switch recombination mechanism.
94: 157–21417560275CheungAC, CramerP2012A movie of RNA polymerase II transcription.
149: 1431–143722726432ConticelloSG, ThomasCJ, Petersen-MahrtSK, NeubergerMS2005Evolution of the AID/APOBEC family of polynucleotide (deoxy)cytidine deaminases.
22: 367–37715496550CramerP, ArmacheKJ, BaumliS, BenkertS, BruecknerF, BuchenC, DamsmaGE, DenglS, GeigerSR, JasiakAJ, 2008Structure of eukaryotic RNA polymerases.
37: 337–35218573085DelkerRK, ZhouY, StrikoudisA, StebbinsCE, PapavasiliouFN2013Solubility-based genetic screen identifies RING finger protein 126 as an E3 ligase for activation-induced cytidine deaminase.
110: 1029–103423277564DussartS, DouaisiM, CourcoulM, BessouG, VigneR, DecrolyE2005APOBEC3G ubiquitination by Nedd4-1 favors its packaging into HIV-1 particles.
345: 547–55815581898EbertMS, SharpPA2012Roles for microRNAs in conferring robustness to biological processes.
149: 515–52422541426FrancoS, GostissaM, ZhaS, LombardDB, MurphyMM, ZarrinAA, YanC, TepsupornS, MoralesJC, AdamsMM, 2006H2AX prevents DNA breaks from progressing to chromosome breaks and translocations.
21: 201–21416427010GuttmanM, AmitI, GarberM, FrenchC, LinMF, FeldserD, HuarteM, ZukO, CareyBW, CassadyJP, 2009Chromatin signature reveals over a thousand highly conserved large non-coding RNAs in mammals.
458: 223–22719182780HarremanM, TaschnerM, SigurdssonS, AnindyaR, ReidJ, SomeshB, KongSE, BanksCA, ConawayRC, ConawayJW, 2009Distinct ubiquitin ligases act sequentially for RNA polymerase II polyubiquitylation.
106: 20705–2071019920177HouseleyJ, LaCavaJ, TollerveyD2006RNA-quality control by the exosome.
7: 529–53916829983KamaduraiHB, SouphronJ, ScottDC, DudaDM, MillerDJ, StringerD, PiperRC, SchulmanBA2009Insights into ubiquitin transfer cascades from a structure of a UbcH5B approximately ubiquitin-HECT(NEDD4L) complex.
36: 1095–110220064473KeimC, KazadiD, RothschildG, BasuU2013Regulation of AID, the B-cell genome mutator.
27: 1–1723307864KenterAL2012AID targeting is dependent on RNA polymerase II pausing.
24: 281–28622784681KimN, KageK, MatsudaF, LefrancMP, StorbU1997B lymphocytes of xeroderma pigmentosum or Cockayne syndrome patients with inherited defects in nucleotide excision repair are fully capable of somatic hypermutation of immunoglobulin genes.
186: 413–4199236193LangerakP, NygrenAO, KrijgerPH, van den BerkPC, JacobsH2007A/T mutagenesis in hypermutated immunoglobulin genes strongly depends on PCNAK164 modification.
204: 1989–199817664295LiX, ManleyJL2006Cotranscriptional processes and their influence on genome stability.
20: 1838–184716847344LiuQ, GreimannJC, LimaCD2006Reconstitution, activities, and structure of the eukaryotic RNA exosome.
127: 1223–123717174896LiuM, DukeJL, RichterDJ, VinuesaCG, GoodnowCC, KleinsteinSH, SchatzDG2008Two levels of protection for the B cell genome during somatic hypermutation.
451: 841–84518273020LongerichS, BasuU, AltF, StorbU2006AID in somatic hypermutation and class switch recombination.
18: 164–17416464563Lykke-AndersenS, BrodersenDE, JensenTH2009Origins and activities of the eukaryotic exosome.
122: 1487–149419420235MatsumotoY, MarusawaH, KinoshitaK, EndoY, KouT, MorisawaT, AzumaT, OkazakiIM, HonjoT, ChibaT2007Helicobacter pylori infection triggers aberrant expression of activation-induced cytidine deaminase in gastric epithelium.
13: 470–47617401375MuramatsuM, KinoshitaK, FagarasanS, YamadaS, ShinkaiY, HonjoT2000Class switch recombination and hypermutation require activation-induced cytidine deaminase (AID), a potential RNA editing enzyme.
102: 553–56311007474PasqualucciL, BhagatG, JankovicM, CompagnoM, SmithP, MuramatsuM, HonjoT, MorseHC3rd, NussenzweigMC, Dalla-FaveraR2008AID is required for germinal center-derived lymphomagenesis.
40: 108–11218066064PavriR, NussenzweigMC2011AID targeting in antibody diversity.
110: 1–2621762814PavriR, GazumyanA, JankovicM, Di VirgilioM, KleinI, Ansarah-SobrinhoC, ReschW, YamaneA, Reina San-MartinB, BarretoV, 2010Activation-induced cytidine deaminase targets DNA at sites of RNA polymerase II stalling by interaction with Spt5.
143: 122–13320887897PrekerP, NielsenJ, KammlerS, Lykke-AndersenS, ChristensenMS, MapendanoCK, SchierupMH, JensenTH2008RNA exosome depletion reveals transcription upstream of active human promoters.
322: 1851–185419056938RajagopalD, MaulRW, GhoshA, ChakrabortyT, KhamlichiAA, SenR, GearhartPJ2009Immunoglobulin switch μ sequence causes RNA polymerase II accumulation and reduces dA hypermutation.
206: 1237–124419433618RamiroAR, JankovicM, CallenE, DifilippantonioS, ChenHT, McBrideKM, EisenreichTR, ChenJ, DickinsRA, LoweSW, 2006Role of genomic instability and p53 in AID-induced c-myc-Igh translocations.
440: 105–10916400328RamiroA, San-MartinBR, McBrideK, JankovicM, BarretoV, NussenzweigA, NussenzweigMC2007The role of activation-induced deaminase in antibody diversification and chromosome translocations.
94: 75–10717560272RevyakinA, LiuC, EbrightRH, StrickTR2006Abortive initiation and productive initiation by RNA polymerase involve DNA scrunching.
314: 1139–114317110577Reyes-TurcuFE, GrewalSI2012Different means, same end-heterochromatin formation by RNAi and RNAi-independent RNA processing factors in fission yeast.
22: 156–16322243696RichardP, ManleyJL2009Transcription termination by nuclear RNA polymerases.
23: 1247–126919487567RinnJL, ChangHY2012Genome regulation by long noncoding RNAs.
81: 145–16622663078RotinD, KumarS2009Physiological functions of the HECT family of ubiquitin ligases.
10: 398–40919436320SaundersA, CoreLJ, LisJT2006Breaking barriers to transcription elongation.
7: 557–56716936696SchatzDG, OettingerMA, SchlisselMS1992V(D)J recombination: Molecular biology and regulation.
10: 359–3831590991SeilaAC, CalabreseJM, LevineSS, YeoGW, RahlPB, FlynnRA, YoungRA, SharpPA2008Divergent transcription from active promoters.
322: 1849–185119056940ShenHM, StorbU2004Activation-induced cytidine deaminase (AID) can target both DNA strands when the DNA is supercoiled.
101: 12997–1300215328407ShimizuT, MarusawaH, EndoY, ChibaT2012Inflammation-mediated genomic instability: Roles of activation-induced cytidine deaminase in carcinogenesis.
103: 1201–120622469133StavnezerJ2011Complex regulation and function of activation-induced cytidine deaminase.
32: 194–20121493144SunJ, RothschildG, PefanisE, BasuU2013Transcriptional stalling in B-lymphocytes: A mechanism for antibody diversification and maintenance of genomic integrity.
4: 127–135SvejstrupJQ2010The interface between transcription and mechanisms maintaining genome integrity.
35: 333–33820194025TrapnellC, PachterL, SalzbergSL2009TopHat: Discovering splice junctions with RNA-Seq.
25: 1105–111119289445UnniramanS, ZhouS, SchatzDG2004Identification of an AID-independent pathway for chromosomal translocations between the Igh switch region and Myc.
5: 1117–112315489857WangL, WhangN, WuerffelR, KenterAL2006AID-dependent histone acetylation is detected in immunoglobulin S regions.
203: 215–22616418396WyersF, RougemailleM, BadisG, RousselleJC, DufourME, BoulayJ, RegnaultB, DevauxF, NamaneA, SeraphinB, 2005Cryptic pol II transcripts are degraded by a nuclear quality control pathway involving a new poly(A) polymerase.
121: 725–73715935759YamanakaS, MehtaS, Reyes-TurcuFE, ZhuangF, FuchsRT, RongY, RobbGB, GrewalSI2013RNAi triggered by specialized machinery silences developmental genes and retrotransposons.
493: 557–56023151475YamaneA, ReschW, KuoN, KuchenS, LiZ, SunHW, RobbianiDF, McBrideK, NussenzweigMC, CasellasR2011Deep-sequencing identification of the genomic targets of the cytidine deaminase AID and its cofactor RPA in B lymphocytes.
12: 62–6921113164YangB, GayDL, MacLeodMK, CaoX, HalaT, SweezerEM, KapplerJ, MarrackP, OliverPM2008Nedd4 augments the adaptive immune response by promoting ubiquitin-mediated degradation of Cbl-b in activated T cells.
9: 1356–136318931680ZhuY, ChenG, LvF, WangX, JiX, XuY, SunJ, WuL, ZhengYT, GaoG2011Zinc-finger antiviral protein inhibits HIV-1 infection by selectively targeting multiply spliced viral mRNAs for degradation.
108: 15834–1583921876179oai:pubmedcentral.nih.gov:37596992014-02-15genesdevpmc-openGenes DevGenes DevGADGenes & Development0890-93691549-5477Cold Spring Harbor Laboratory PressPMC3759699PMC375969937596992393465723934657871166010.1101/gad.219105.113Resource/MethodologyNovel proteomic approach (PUNCH-P) reveals cell cycle-specific fluctuations in mRNA translationAviner et al.Proteomic monitoring of mRNA translationAvinerRanen1GeigerTamar234Elroy-SteinOrna134Department of Cell Research and Immunology, George S. Wise Faculty of Life Sciences, Tel Aviv University, Tel Aviv 69978, Israel;Department of Human Molecular Genetics and Biochemistry, Sackler Faculty of Medicine, Tel Aviv University, Tel Aviv 69978, Israel
Monitoring protein synthesis is required to understand gene expression regulation. Aviner et al. developed a system-wide proteomic approach for direct monitoring of translation, termed puromycin-associated nascent chain proteomics (PUNCH-P), which is based on incorporation of biotinylated puromycin into newly synthesized proteins followed by streptavidin affinity purification and LC-MS/MS analysis. Using PUNCH-P, cell cycle-specific fluctuations in synthesis for >5000 proteins were measured in mammalian cells. This approach also identified proteins not previously implicated in cell cycle processes and proteins that were not detected using other methods.
Monitoring protein synthesis is essential to our understanding of gene expression regulation, as protein abundance is thought to be predominantly controlled at the level of translation. Mass-spectrometric and RNA sequencing methods have been recently developed for investigating mRNA translation at a global level, but these still involve technical limitations and are not widely applicable. In this study, we describe a novel system-wide proteomic approach for direct monitoring of translation, termed puromycin-associated nascent chain proteomics (PUNCH-P), which is based on incorporation of biotinylated puromycin into newly synthesized proteins under cell-free conditions followed by streptavidin affinity purification and liquid chromatography-tandem mass spectrometry analysis. Using PUNCH-P, we measured cell cycle-specific fluctuations in synthesis for >5000 proteins in mammalian cells, identified proteins not previously implicated in cell cycle processes, and generated the first translational profile of a whole mouse brain. This simple and economical technique is broadly applicable to any cell type and tissue, enabling the identification and quantification of rapid proteome responses under various biological conditions.
mRNA translation is a key step in gene expression that attracts increasing attention at the systems biology level. In past decades, major efforts were invested in studying transcription regulation, while research focusing on post-transcriptional control has lagged behind. Although mRNA levels are commonly used as a proxy of protein amounts, comparative genomic and proteomic analyses of different species and cell types have shown that mRNA and protein levels do not correlate perfectly, thereby stressing the important contribution of translation control and protein stability to gene expression (Vogel and Marcotte 2012). This provides cells with the plasticity needed to rapidly modulate gene expression in response to changes in environmental conditions (e.g., cellular stress) and also for fine-tuning of protein levels during cell cycle progression, proliferation, and differentiation (Calkhoven et al. 2002; Holcik and Sonenberg 2005). The ability to identify and quantify the proteins produced in a population of cells under various conditions is therefore essential to our understanding of the processes underlying gene expression.
Traditionally, translation rates have been monitored by metabolic labeling of cells with radioactive amino acids. For high-resolution identification and quantification of proteins, metabolic labeling was combined with mass spectrometric (MS) analysis, and the radioactive amino acids were replaced by stable isotope-labeled (SILAC [stable isotope labeling by amino acids in cell culture]) amino acids. To distinguish newly synthesized from pre-existing proteins, cells are pulse-labeled with SILAC amino acids, allowing only proteins produced during the pulse to incorporate the label. This method, termed pulsed SILAC (pSILAC), is successfully used to detect long-lasting changes in protein production and degradation but is incompatible for studying rapid fluctuations because accurate quantification of SILAC pairs requires relatively long pulses (Schwanhausser et al. 2009).
Another proteomic approach is based on incorporation of a modified methionine analog called azidohomoalanine (AHA), which is subsequently derivatized to tagged reporter molecules or an affinity purification matrix through click chemistry. This method has been successfully used to visualize mRNA translation in situ; however, it requires predepletion of intracellular endogenous methionine followed by supplementation of the amino acid analog, both of which can result in cellular stress and potential alteration of translation patterns (Vaughan et al. 1971; Kramer et al. 2009). AHA incorporation into tRNA was also shown to occur more slowly than methionine, possibly introducing a measurement bias (Kiick et al. 2002). Attempts to use AHA labeling for MS identification of newly synthesized proteins either alone or in combination with SILAC have shown that this approach is limited to detection of up to several hundreds of proteins (Dieterich et al. 2006; Eichelbaum et al. 2012; Howden et al. 2013).
Protein synthesis has also been indirectly monitored by deep sequencing of ribosome-protected mRNA fragments using a technique called ribosome profiling, otherwise known as ribosome footprinting or ribo-seq (Ingolia et al. 2009). This method involves nuclease digestion of cell extracts to degrade all mRNA molecules that are not protected by ribosomes, leaving out short undigested RNA fragments. These are isolated by a series of purification steps, including PAGE fragment size selection and rRNA subtractive hybridization, followed by sequencing. Ribo-seq is a powerful technique for investigating translation at a subcodon resolution, measuring translation efficiency, and mapping novel ORFs (Guo et al. 2010; Lee et al. 2012). In addition, it does not require in vivo labeling that may induce stress. However, a limitation of ribo-seq is that it only generates a prediction of protein synthesis based on the steady-state amounts of ribosome-bound mRNA molecules. This may be misleading in some cases, as inhibition of protein synthesis is not always accompanied by a decrease in the number of ribosomes associated with the encoding mRNA (Clark et al. 2000; Nottrott et al. 2006; Petersen et al. 2006; Sivan et al. 2007).
In another alternative approach, the naturally occurring antibiotic puromycin is used to label and detect newly synthesized proteins. Puromycin, a tyrosine-tRNA mimetic, is catalytically incorporated by the ribosome into the C terminus of elongating nascent polypeptide chains in a sequence-independent manner. Incorporation leads to translation termination and release of C-terminally truncated peptides bearing a puromycin moiety (Pestka 1971). This so-called “puromycylation” reaction has so far been studied by immunoblotting, immunofluorescence, and flow cytometry using puromycin-specific antibodies and fluorescent or biotinylated puromycin derivatives (Starck et al. 2004; Schmidt et al. 2009; David et al. 2012). In cultured cells, puromycylation followed by Western blotting with an anti-puromycin antibody is used to monitor translation rates in place of radioactive metabolic labeling (Schmidt et al. 2009). Puromycin and its derivatives also enabled in situ visualization of protein synthesis (David et al. 2012; Liu et al. 2012) and isolation of a specific protein produced in rabbit reticulocyte lysate (Starck et al. 2004). However, to the best of our knowledge, attempts to use these antibodies for immunoprecipitation have been unsuccessful.
In this study, we describe a method that combines biotinylated puromycin with MS analysis to globally label newly synthesized proteins and monitor mRNA translation. The method involves isolation of ribosomes by ultracentrifugation followed by cell-free labeling of nascent polypeptide chains with 5′ biotin-dC-puromycin 3′ (Biot-PU), capture on immobilized streptavidin, and analysis by liquid chromatography-tandem MS (LC-MS/MS) (Fig. 1). This work flow leads to the identification of thousands of newly synthesized proteins, generating a snapshot of the cellular translatome. We used this method, termed puromycin-associated nascent chain proteomics (PUNCH-P), to study global cell cycle-dependent variations in translation and identify proteins that are differentially synthesized at specific stages.
Experimental setup of PUNCH-P. Ribosomes are extracted from cultured cells by ultracentrifugation on a sucrose cushion. Ribosomes are then incubated with the labeling reagent Biot-PU, and labeled newly synthesized proteins are isolated by streptavidin affinity purification and analyzed by LC-MS/MS.
ResultsLabeling, capture, and MS analysis of newly synthesized proteins from mammalian cells
The development of a method for proteomic analysis of newly synthesized proteins involved multiple steps of calibration and optimization. As puromycylation can be performed in either cultured cells or isolated polysomes (Blobel and Sabatini 1971; David et al. 2012), we first sought to compare the incorporation of puromycin and Biot-PU under these two experimental conditions. To this end, cultured HeLa cells or polysomes that were isolated by ultracentrifugation on a sucrose cushion were treated with either 1 μM puromycin or 1 μM Biot-PU for 30 min at 37°C. Equal amounts of total protein were resolved on an SDS-PAGE, and the membrane was probed sequentially with a commercial anti-puromycin antibody and a streptavidin-HRP conjugate to detect proteins bearing a terminal puromycin or Biot-PU, respectively. Underivatized puromycin efficiently labeled newly synthesized proteins in both intact cultured cells and isolated polysomes (Supplemental Fig. S1). In contrast, Biot-PU incorporation was only detected in isolated polysomes, possibly due to reduced permeability through the plasma membrane. We reasoned that cell-free puromycylation of isolated polysomes may in fact be preferable, as it does not affect the pattern of protein synthesis or depend on the duration of labeling for obtaining sufficient amounts of puromycylated proteins to allow MS detection. Without a pulse time limitation, Biot-PU can be used to study changes in mRNA translation that occur very rapidly. Furthermore, extraction of ribosomes prior to puromycylation depletes the reaction mixture of endogenously biotinylated proteins that otherwise compete for streptavidin binding and may significantly affect the signal to noise ratio (Robinson et al. 1983; de Boer et al. 2003).
To determine the amount of Biot-PU required for complete labeling of nascent polypeptide chains, we incubated a fixed amount of ribosomes with increasing amounts of Biot-PU (Supplemental Fig. S2). A ratio of 1 pmol of Biot-PU to 1 OD254 ribosomes provided complete labeling and was therefore maintained in all future experiments. We then analyzed the time course of Biot-PU incorporation and found that it proceeds rapidly and reaches saturation within 15 min (Fig. 2A). Incubation on ice or addition of cycloheximide, a competitive inhibitor of the puromycin reaction (Hobden and Cundliffe 1978), each reduced signal intensity, confirming the involvement of ribosome catalysis in the incorporation of Biot-PU (Fig. 2B).
Biochemical characterization of PUNCH-P. (A,B) Newly synthesized proteins were labeled by Biot-PU for the indicated times and under the indicated conditions and detected by Western blotting using streptavidin-HRP. (C) Newly synthesized proteins labeled by Biot-PU were fractionated on a linear 10%–50% sucrose gradient, and fractions corresponding to the entire range of monosomes to heavy polysomes were collected and probed by Western blotting. (Top panel) rRNA absorption at 254 nm, showing the positions of 80S monosome and polysomes. (Bottom panel) Western blotting of gradient fractions using streptavidin-HRP or anti-ribosomal protein L26 antibody. Vertical line traces on the right represent the densitometry of lane 5 (light polysomes) and lane 10 (heavy polysomes).
Pretreatment with cycloheximide is commonly used to arrest elongating ribosomes prior to lysis and fractionation to prevent ribosome runoff during harvesting or ultracentrifugation. To examine whether omitting cycloheximide has a destabilizing effect on ribosomes in our experiments, we compared HeLa cells harvested with or without pretreatment with 100 μg/mL cycloheximide for 5 min and fractionated on a linear 10%–50% sucrose gradient. The data showed that omitting cycloheximide from the harvesting, lysis, and sucrose fractionation steps has a very minor destabilizing effect on polysome size (Supplemental Fig. S3).
To confirm that puromycin labeling correlates with the presence of nascent polypeptide chains, we monitored the cosedimentation of puromycylated peptides with ribosomes by fractionating a saturated Biot-PU-labeling reaction on a similar sucrose gradient under low-salt conditions that prevent ribosome dissociation (Fig. 2C, top panel; Blobel and Sabatini 1971). Western blot analysis of gradient fractions corresponding to the entire range of monosomes through heavy polysomes showed that both the mean molecular weight and the intensity of puromycylated peptides increase with polysome size (Fig. 2C, bottom panel, densitometry). This distribution is consistent with nascent polypeptide chains because heavy faster-sedimenting polysomes translate mRNAs with longer ORFs and produce larger amounts of protein in total.
To examine the applicability of the method to additional cell lines, we isolated ribosomes from HEK293 and RAW264.7 cells and analyzed nascent chain puromycylation as above. Western blot analysis showed that the labeling reaction proceeds efficiently in these cell types, revealing a different repertoire of labeled proteins (Supplemental Fig. S4).
The use of Biot-PU instead of puromycin allowed us to harness the extraordinary affinity between biotin and streptavidin to increase binding specificity and eliminate strong background binders; e.g., ribosomal proteins. We adopted a procedure that consists of overnight incubation with streptavidin beads under stringent denaturing conditions (2% SDS, 8 M urea) and extensive washes prior to elution (Tagwerker et al. 2006). This procedure resulted in very low background binding as compared with the manufacturer's recommended low-stringency protocol while retaining the same levels of the puromycylated peptides (Supplemental Fig. S5).
Using this procedure, we set out to identify and quantify newly synthesized proteins in cycling HeLa cells. Western blotting confirmed that nascent peptides were labeled and captured with high specificity and that components of the translation machinery were removed, as evidenced by the absence of ribosomal protein L26 from the streptavidin eluate (Fig. 3A). Streptavidin beads from similar amounts of puromycylation and control reactions were then subjected to on-bead trypsin digestion followed by single LC-MS/MS runs on the Q-Exactive MS. Data were analyzed with MaxQuant software, and a Student's t-test was used to compare triplicates of puromyclated proteins and control samples. Of 3244 proteins identified in at least two replicates, 3072 were specific to the puromycylated samples relative to nonpuromycylated controls (false discovery rate [FDR] = 0.05, S0 = 0.5) (Fig. 3B; Supplemental Table 1), representing newly synthesized proteins. The technical reproducibility of this experiment was high, with an average Pearson correlation of 0.96 between replicates (Fig. 3C).
PUNCH-P analysis of cycling HeLa cells. (A) Puromycylation and control samples were incubated with streptavidin beads under stringent conditions, and puromycylated newly synthesized proteins were eluted following extensive washing. Equal volumes of starting material (input), flowthrough (FT), wash, and eluate were analyzed by Western blotting. Ribosomal protein L26 is shown as control for removal of ribosomes from the eluate. (B) Volcano plot of proteins identified by MS in puromycylated compared with control samples. (C) Scatter plot showing the high technical reproducibility of PUNCH-P.
To determine whether polysome isolation and labeling procedures influence the identity or relative quantity of the proteins detected, we performed a PUNCH-P analysis of cycling HeLa cells pretreated with emetine, a highly effective irreversible inhibitor of translation elongation that does not interfere with puromycylation (David et al. 2012). We reasoned that any bias introduced by the isolation procedure itself should be eliminated in the presence of emetine, which blocks translation in vivo prior to harvesting and therefore protects from downstream in vitro effects. PUNCH-P profiles for emetine-treated and control cells were strikingly similar in both coverage and relative quantity of proteins detected (r = 0.98), suggesting that no bias is introduced during the in vitro sample preparation procedure (Supplemental Fig. S6; Supplemental Table 1).
Comparative analysis of protein synthesis using PUNCH-P, ribo-seq, and pSILAC
To further evaluate and validate PUNCH-P, we compared it with pSILAC and ribo-seq, two established methods that are used for studying translation at the protein and mRNA levels, respectively. A principle difference between these methods is that PUNCH-P and ribo-seq generate an in vitro snapshot of translation, while pSILAC is based on the in vivo accumulation of labeled proteins during the translation process. In light of these differences, we first compared two pulse durations of pSILAC to determine whether a minimum pulse of 2 h allows reproducible quantification of newly synthesized proteins, as previously reported (Schwanhausser et al. 2009; Eichelbaum et al. 2012). To this end, we pulse-labeled cycling HeLa cells for 2 or 10 h with either heavy or medium-heavy stable isotope amino acids, thereby generating duplicate analyses in single MS runs. While the 10-h pulse yielded reproducible results with a high correlation between heavy and medium-heavy peptides in the same run and a narrow distribution of heavy/medium-heavy ratios between separate runs (ravg = 0.93, median absolute deviation [MAD] of 0.12–0.15), the 2-h pulse showed poor reproducibility (ravg = 0.75, MAD 0.54–0.56) due to insufficient labeling and very large SILAC ratios (Supplemental Fig. S7; Supplemental Table 2). We therefore used the 10-h pSILAC measurements for further comparisons.
Based on observations that steady-state protein levels in mammalian cells are best explained by translation rates (Schwanhausser et al. 2011), we expected the results of each of the above methods to correlate well with overall protein abundance. Therefore, we compared PUNCH-P, 10-h pSILAC, and ribo-seq results (Guo et al. 2010) with publically available whole-proteome analysis of cycling HeLa cells (Nagaraj et al. 2011) and found that the correlation was similar for all three methods (r = 0.41, 0.40, and 0.42, respectively). This suggests that the methods are similarly accurate in quantifying translation products. However, the correlation between PUNCH-P and 10-h pSILAC (ravg = 0.60) was significantly higher than the correlation between ribo-seq and either PUNCH-P or pSILAC (ravg = 0.37 and ravg = 0.43, respectively) (Fig. 4A,B), possibly due to technical differences in detection between MS and deep-sequencing techniques.
Comparison of PUNCH-P, 10-h pSILAC, and ribo-seq. (A) Heat map showing Pearson correlation between the methods for measuring protein synthesis and HeLa whole-cell proteome. (B) Representative scatter plots show the correlation between HeLa whole-cell proteome and PUNCH-P, 10-h pSILAC, and ribo-seq.
In terms of coverage, PUNCH-P identified and quantified thousands of proteins per sample—more than 10-h pSILAC (3072 and 2143, respectively, both analyzed by single 4-h LC-MS/MS runs) but less than ribo-seq (6238 transcripts, sequenced in two lanes per sample). Because ribo-seq coverage depends on the number of lanes used to sequence a single sample, the cost of analysis increases proportionally to the desired depth. PUNCH-P coverage, on the other hand, depends on the amount of starting material (as discussed below) and can also be increased by standard proteomic approaches, including peptide prefractionation and longer overall MS measurement time per sample.
Cell cycle-specific fluctuations in mRNA translation
After establishing that PUNCH-P performs well compared with current alternatives, we applied it to study changes in translation throughout the cell cycle as a test case. Such analysis cannot be accomplished by pSILAC due to the short duration of specific cell cycle phases (e.g., mitosis, which lasts 40–90 min in HeLa cells) (Rao and Engelberg 1968; Sigoillot et al. 2011); it is also challenging for ribo-seq because of the high costs associated with deep-sequencing analysis of a large number of samples.
For cell cycle synchronization, we used double-thymidine block to arrest cultured cells at the G1/S boundary and then released them in fresh medium to allow synchronous progression (Sivan et al. 2011). While the efficiency of synchronization under these conditions is not maximal, it eliminates the necessity of using drugs that arrest progression by inhibiting DNA replication (e.g., mimosine) or microtubule polymerization (e.g., nocodazole) and may confound the results due to nonphysiological effects on translation. We showed previously that arresting cells in mitosis using nocodazole leads to disassembly of polysomes, while a similar effect is not observed in thymidine-synchronized mitotic cells (Sivan et al. 2011).
To generate a global profile of protein synthesis throughout the cell cycle, we harvested thymidine-synchronized HeLa cells at four time points, corresponding to peak S phase, G2/M boundary, mitotic exit, and peak G1 phase, based on fluorescent analysis of DNA content (Fig. 5, bottom). Synchronization was performed in triplicates, and the samples were processed in parallel for PUNCH-P analysis. A total of 5105 proteins were identified in at least two of three samples, of which 4984 were specific to the puromycylated samples relative to nonpuromycylated controls. ANOVA analysis of the differences between cell cycle stages revealed that, while the majority of proteins (4653, or ∼93%) are synthesized at similar levels throughout the cell cycle, a subclass of 339 proteins shows statistically significant variation (FDR = 0.05), representing a unique signature of cell cycle-specific protein expression (Supplemental Table 4). Hierarchical clustering of these fluctuating proteins shows clear segregation into several distinct protein clusters (Fig. 5), which highlight subclasses of functionally interconnected proteins that may play important roles in cell cycle progression. The largest cluster, consisting of proteins that are highly translated during mitosis, includes pivotal mitotic regulators; e.g., CCNB1/2 (cyclin B1/B2) and CDK1 (cyclin-dependent kinase 1). The cluster of proteins highly translated during S phase includes histones and proteins involved in DNA replication and repair. Interestingly, the synthesis of some S-phase proteins is already induced at G1; this cluster includes CCNE2 (G1/S-specific cyclin E2), a classical marker of G1/S transition, as well as several components required for DNA replication (see the Discussion).
Cell cycle-related dynamics in protein synthesis. Hierarchical clustering of proteins was performed on logarithmized intensities after Z-score normalization of the data. The heat map shows proteins with statistically significant differences in synthesis throughout the cell cycle. Selected proteins from each cluster are indicated on the right. Flow cytometry analysis of DNA content using propidium iodide is shown in the bottom panel. Respective cell cycle stages are indicated, with time after release from second thymidine block in parenthesis and percentage of cells in each phase shown below.
To validate some of these results and provide further evidence for the potential of PUNCH-P to measure differences in translation of specific mRNAs, we analyzed the polysome association of selected mRNAs in S and M phase. We chose two proteins whose PUNCH-P expression remained stable throughout the cell cycle and six proteins whose expression fluctuated between S and M phase. Such fluctuations at the protein level can result from differences in total mRNA abundance, translation efficiency, or both. We initially compared the polysome profiles of HeLa cells synchronized to S and M phase and found no differences in polysome size (Fig. 6A, top panel). Next, we extracted total RNA from each of seven polysomal fractions ranging from light to heavy polysomes (Fig. 6A, bottom panel) and measured the abundance of specific mRNAs in each fraction using SYBR fast quantitative PCR (qPCR). As predicted, we found that the relative amount and polysomal distribution of the two transcripts encoding for nonfluctuating proteins (calreticulin [CALR] and cytochrome C [CYCS]) remained constant between the different cell cycle stages (Fig. 6B,C). We then measured the amount of the six transcripts encoding for fluctuating proteins in each polysomal fraction and normalized the results to CALR mRNA to allow a more accurate quantitative comparison of mRNA amounts in the different gradients. As predicted by PUNCH-P, the amounts of polysome-associated FOSL1 (FOS-like antigen 1), CCRN4L (also called nocturnin), and BRIP1 (Fanconi anemia group J protein 1) were significantly higher in S phase (Fig. 6D). Similarly, the amounts of polysome-associated CCNB1 (cyclin B1), GPSM2 (G-protein signaling modulator 2), and PCF11 (pre-mRNA cleavage complex 2 protein) mRNA were higher in M phase, consistent with PUNCH-P results (Fig. 6E). In addition to the differences in absolute amounts, we calculated relative mRNA distribution between heavy (five or more) and light (less than five) polysomes as a percentage of total polysomes. While some mRNAs showed little difference in relative distribution, others changed considerably. Association of CCRN4L and BRIP1 mRNAs with heavy polysomes decreased from 82.4% and 43.3% in S phase to 73.5% and 29.2% in M phase, respectively. Similarly, association of CCNB1 mRNA with heavy polysomes increased from 63.80% in S phase to 84.30% in M phase, as expected for an mRNA that is translationally up-regulated during mitosis (Groisman et al. 2000).
qPCR validation of PUNCH-P results. (A, top panel) Polysome profiles of HeLa cells synchronized to S and M phase by double-thymidine block. (Bottom panel) Total RNA extracted from each of the polysomal fractions visualized by ethidium bromide staining. (B,C) Polysomal association of nonfluctuating mRNAs encoding for CALR (B) and CYCS (C). (D) Polysomal association of mRNAs encoding for proteins that were elevated in S phase according to PUNCH-P. (E) Polysomal association of mRNAs encoding for proteins that were elevated in M phase according to PUNCH-P. Graphs show mean ± SD of qPCR replicates for one of three independent experiments with similar results. C–E represent qPCR results normalized to CALR.
Generating a whole mouse brain translatome
A unique advantage of this method, which analyzes translation based on ex vivo labeling, is its applicability to tissue samples, where in vivo labeling is highly challenging. As a test case, we chose to analyze the translatome of a developing mouse brain. Ribosomes were isolated from brains of three 3-wk-old C57BL mice followed by Biot-PU incorporation. Western blotting confirmed efficient labeling of nascent polypeptide chains (Supplemental Fig. S8A). Streptavidin affinity purification and on-bead digestion followed by LC-MS/MS analysis identified just over 400 proteins specific to the puromycylated samples. This number was increased fivefold (to 2187) when three brain samples were pooled and analyzed together, confirming that starting material is an important determinant of proteome coverage in PUNCH-P analysis (Supplemental Fig. S8B; Supplemental Table 5).
Discussion
In this study, we describe the development of a novel proteomic approach for monitoring translation based on direct measurement of protein amounts without pulse duration limitations. We show that PUNCH-P is equally correlative to steady-state protein levels as pSILAC and ribo-seq, suggesting that it quantitatively detects actively synthesized proteins. Unlike pSILAC, PUNCH-P does not have a minimum pulse time requirement and can therefore be used to detect rapid changes in protein synthesis without influence from protein degradation. While PUNCH-P cannot compete with the single-nucleotide resolution of ribo-seq, it involves simple and rapid experimental setup and data analysis, with turnaround times of ∼2 d from sample preparation to analyzed data. It is also economical, at ∼1% of the cost of ribo-seq, and therefore suitable for the analysis of larger sets of samples. As such, it can also be used to supplement or validate ribo-seq results on a larger scale.
Employing these unique strengths of PUNCH-P, we generated the first atlas of cell cycle-dependent translation. Progression in the cell cycle is known to depend on complex interactions and feedback loops that are based on the temporally precise accumulation and activation of regulatory proteins; e.g., cyclins and CDKs (Lindqvist et al. 2009). While degradation and post-translational modifications have long been implicated in the regulation of cell cycle progression, the relative contribution of translation has yet to be quantified at a global level. Not surprisingly, our results suggest that differential translation may also play an important part in this regulation. We show that synthesis of functionally related proteins is coordinated through the cell cycle and identify novel proteins not previously implicated in cell cycle processes that can serve as a starting point for future research.
The network of proteins with increased synthesis during M phase reveals that mRNA translation is significantly involved in the mitotic program, perhaps more so than other stages of the cell cycle (Supplemental Fig. S9). Key proteins whose synthesis is elevated during M phase include pivotal mitotic regulators; e.g., CCNB1/2 (cyclin B1/B2) and CDK1; PLK1, NEK2, and AURKA kinases, which are involved in functional centrosome maturation and spindle stability; CDC20, which activates the anaphase-promoting complex (APC); and BUB1, which phosphorylates CDC20 as part of mitotic checkpoint complex (MCC) to prevent premature anaphase (highlighted in Supplemental Fig. S9A). Interestingly, of the 10 members of the kinesin motor protein superfamily found to be elevated in M phase, only some have a known role in mitosis (e.g., KIF20B and KIF23) (Miki et al. 2001). Identification of new kinesins that play a part during mitosis or upon entry to G1 may be of clinical significance, as mitotic kinesins are considered potential targets for anti-cancer interventions (Huszar et al. 2009).
PUNCH-P detected increased synthesis of histones in S phase (Supplemental Fig. S9B), consistent with the known temporal relationship between histone expression and DNA synthesis (Marzluff and Duronio 2002). Histone mRNA translation is regulated in part by a conserved stem–loop structure that recruits SLBP (stem–loop-binding protein), which is synthesized in late G1 and degraded upon exit from S phase (Whitfield et al. 2000). Indeed, although SLBP is most abundant in S phase, PUNCH-P identified increased SLBP synthesis in G1 (Supplemental Table 4). The transcription factors ATF3 and NFKB1/B2, which play a role in the intra-S-phase checkpoint and can be induced by DNA damage (Joyce et al. 2001; Fan et al. 2002), are also found in our S-phase list, as well as PARP2 (Supplemental Fig. S9B), which was recently reported to possess transcriptional repression activity by recruiting histone deacetylases and methytransferase to the promoter of cell cycle-related genes (Liang et al. 2013).
Proteins with increased synthesis in both G1 and S include regulators of G1/S transition; e.g., CCNE2, which is known to reach peak nuclear abundance in late G1 (Mumberg et al. 1997). In keeping with this observation, it is tempting to speculate that proteins such as ORC3 (origin recognition complex subunit 3) and DSN1 (kinetochore-associated protein), both enriched in our G1- and S-phase lists, are synthesized late in G1 to allow timely DNA synthesis and recognition of sister chromatids following replication. Interestingly, SMC1A (structural maintenance of chromosomes protein 1A), a protein required for sister chromatid cohesion (Losada and Hirano 2005) was detected by PUNCH-P in S phase, suggesting that its synthesis peaks during DNA replication, consistent with a possible role in DNA repair (Kim et al. 2002).
Of the proteins chosen for mRNA validation (Fig. 6), some have known roles that relate to either S or M phase, while others are implicated as such by the present study. FOSL1 is required for initiation of DNA synthesis (Riabowol et al. 1992), GPSM2 is important for spindle pole orientation (Yasumi et al. 2005), and CCNB1 is a major mitotic regulator whose protein levels peak during G2/M (Pines and Hunter 1989). In contrast, CCRN4L is a circadian deadenylase that turns off the expression of genes in a rhythmic fashion (Garbarino-Pico and Green 2007) but has no known role in the cell cycle; our work shows that CCRN4L is translationally up-regulated and may therefore play a part in S phase. Furthermore, PCF11 has not been implicated in cell cycle-related processes but was shown to interact with Rhn1, which is required for suppression of meiotic mRNAs in mitotically dividing fission yeast (Sugiyama et al. 2012), suggesting a role for PCF11 in mitosis. Additional research is needed to better characterize the involvement of CCRN4L and PCF11 in S and M phase, respectively.
An exciting prospect of PUNCH-P is its unique ability to monitor mRNA translation in whole tissues, which is challenging with ribo-seq and nearly impossible with pSILAC. While tissues with lower translation activity may require pooling to achieve efficient PUNCH-P detection, the test case reported here of a mouse brain analysis confirms that PUNCH-P is indeed applicable to tissues. Based on these results, we estimate that PUNCH-P can be used with any cell type or tissue and may thus represent an important technological advance in studying rapid proteome responses to stress, stimulus, or pharmacological perturbation at a tissue-specific level.
Materials and methodsReagents and antibodies
Biot-PU was custom-synthesized by Dharmacon according to Starck et al. (2004). Streptavidin agarose beads were from Pierce Biotechnology. HRP-conjugated streptavidin for Western blotting was from R&D Systems or Vector Laboratories (Vectastain Elite ABC). Anti-puromycin antibody (12D10) was from Millipore. Control antibodies were rabbit anti-RPL26 (Abcam) and rabbit anti-tubulin (Cell Signaling Technology). HRP-conjugated goat anti-rabbit secondary antibody was from Jackson ImmunoResearch Laboratories. SILAC amino acids were purchased from Cambridge Isotopes Laboratories. All other reagents were from Sigma Aldrich unless otherwise specified.
Cell culture and synchronization
HeLa S3 cells were grown in DMEM (Invitrogen) supplemented with 10% fetal calf serum, 2 mM L-glutamine, and 100 U/mL penicillin/streptomycin (Biological Industries). For double-thymidine block, cells were treated with 2 mM thymidine for 19 h, released from G1/S block in fresh DMEM for 9 h, treated again with 2 mM thymidine for 18 h, released in fresh DMEM, and harvested at different time points, as indicated. Cell cycle distribution was assessed by flow cytometry using the BD Biosciences FACSort instrument after staining with propidium iodide (Sigma).
Ribosome pelleting from cultured cells
For each sample, 3.5 × 107 HeLa cells were washed once, harvested in PBS (Gibco), centrifuged at 1000g for 5 min at 4°C, and then frozen at −80°C for subsequent use. To purify ribosomes, cells were thawed on ice and lysed for 20 min in 500 μL of polysome buffer (18 mM Tris at pH 7.5, 50 mM KCl, 10 mM MgCl, 10 mM NaF, 10 mM α-glycerolphosphate, 1.4 μg/mL pepstatin, 2 μg/mL leupeptin, Complete EDTA-free protease inhibitor cocktail [Roche], 1.25 mM dithiothreitol, 40 U RNase inhibitor [Invitrogen]) supplemented with Triton X-100 and deoxycholate to a final concentration of 1% each. Following centrifugation at 14,000 rpm for 10 min at 4°C, the supernatant was removed and layered on 500 μL of 2 M sucrose in polysome buffer. The sucrose cushion was centrifuged at 37,000 rpm for 4 h at 4°C in a Beckman Coulter TLA120.2 rotor, and the ribosome pellet was resuspended in 100 μL of polysome buffer and processed directly for puromycylation.
Ribosome pelleting from mouse brain tissues
Ribosomes were extracted from mouse brain tissues essentially as described in Darnell et al. (2011), with some modifications. Briefly, 3-wk-old C57BL mice were sacrificed by isoflurane anesthesia and decapitation. Brains were removed and flash-frozen in liquid nitrogen. Prior to PUNCH-P analysis, brains were homogenized in 1 mL of polyribosome buffer with 10 strokes in a dounce homogenizer. NP-40 was added to a final concentration of 1% and incubated for 10 min on ice. The homogenate was spun at 2000g for 10 min at 4°C. The supernatant was respun at 20,000g for 10 min at 4°C, and the resulting supernatant was layered on 500 μL of 2 M sucrose in polysome buffer. The sucrose cushion was centrifuged at 37,000 rpm for 4 h at 4°C in a Beckman Coulter TLA120.2 rotor, and the ribosome pellet was resuspended in 100 μL of polysome buffer and processed directly for puromycylation.
Puromycylation and streptavidin capture
Resuspended ribosomes were incubated for the indicated times at 37°C either with or without Biot-PU at a ratio of 100 pmol per 1 OD254 ribosomes. The reaction was terminated by the addition of Laemmli sample buffer for direct Western blotting or high-stringency wash buffer (100 mM Tris HCl at pH 7.5, 2% SDS, 8 M urea, 150 mM NaCl) for streptavidin capture. The mix was tumbled overnight at room temperature with 50 μL of streptavidin agarose slurry. The beads were then washed four times with 1 mL of high-stringency buffer followed by one 30-min wash in the same buffer at room temperature and then washed for 30 min in 1 mL of high-salt buffer (100 mM Tris HCl at pH 7.5, 1 M NaCl) and five times with ultrapure water. The beads were then incubated for 30 min in 1 mM DTT and then 50 mM iodoacetamide (in the dark) and washed twice with 50 mM ammonium bicarbonate. For MS analysis, beads were resuspended in 50 mM ammonium bicarbonate, and proteins were digested overnight with 0.4 μg of sequencing-grade trypsin (Promega). After overnight incubation, digests were acidified with 0.1% TFA and purified on C18 StageTips (Rappsilber et al. 2007). For Western blot analysis, proteins were released from washed streptavidin beads by boiling in elution buffer (2% SDS, 3 mM biotin, 8 M urea in PBS) for 30 min at 96°C, as previously described (Rybak et al. 2004).
Western blot analysis of puromycylated peptides
For detection of puromycylated peptides, samples were resolved on a 10% or 12% SDS-PAGE and transferred to a nitrocellulose membrane. The membrane was stained with ponceau S, photographed, and blocked for 1 h with 4% BSA in TBST followed by incubation for 1 h with 1:500 streptavidin-HRP in TBST at room temperature or overnight with 1:10,000 anti-puromycin antibody in 4% BSA in TBST. The membrane was washed three times in TBST for 5 min each, and detection was performed using standard ECL technique (GE Healthcare).
Polysome profile analysis
Polysome profile analysis was performed as described previously (Sivan et al. 2007). Briefly, ribosomes were pelleted, labeled with Biot-PU, layered on a 10%–50% linear sucrose gradient in polysome buffer, centrifuged at 37,000 rpm for 100 min at 4°C in an SW41 Beckman Coulter rotor; and fractionated with absorption measured continuously at 254 nm using a Teledyne ISCO UA-6 UV/VIS detector. Fractions were collected and subjected to Western blot analysis as described above.
pSILAC
For pSILAC experiments, cells were cultured in medium deprived of the natural amino acids lysine and arginine and supplemented with light (Lys0 and Arg0), medium (Lys4 and Arg6), or heavy (Lys8 and Arg10) versions of these amino acids. Instead of standard fetal bovine serum, culture medium was supplemented with dialyzed serum. Cells were first cultured in light medium and then pulse-labeled with medium-heavy or heavy medium for 2 h or 10 h. Parallel labeling with medium and heavy medium served as a biological replicate in the same MS run. Each treatment was also conducted in independent duplicates.
Protein digestion LC-MS/MS analysis
PUNCH-P and pSILAC samples were analyzed using single 4-h runs (including loading, gradient, and wash duration) and in replicates as indicated for each experiment. LC-MS/MS analysis was performed on an EASY-nLC1000 ultrahigh-performance LC (UHPLC) (Thermo Scientific) coupled on line to the Q-Exactive mass spectrometer (Thermo Scientific). Peptides were separated on a 50-cm column with 2-μm pepmap beads (Dionex) and connected to the MS through an EASY-spray ionization source. Peptides were loaded onto the column in buffer A (0.5% acetic acid) and separated with a 200-min gradient of 5%–30% buffer B (80% acetonitrile, 0.5% acetic acid) followed by a 10-min wash with 95% buffer B. MS analysis was performed using a data-dependent top 10 method. MS spectra were acquired at 70,000 resolution (at 200 Th) with a target value of 106 ions. MS/MS spectra were acquired at 17,500 resolution with a target value of 105 ions. Dynamic exclusion option was enabled, with exclusion duration of 20 sec.
Data analysis
Raw MS files were analyzed with MaxQuant software (Cox and Mann 2008) and the Andromeda search engine (Cox et al. 2011). MS/MS were searched against the UniProt human database and an additional list of common contaminants, including avidin. Data were filtered with a 1% FDR on the peptide level and the protein level. The “match between runs” option was enabled to transfer identification between runs based on their accurate mass and retention time. Protein abundance was determined as the summed peptide intensities. Bioinformatic analysis was performed using the Perseus program in the MaxQuant environment. t-tests and ANOVA were performed with 5% FDR and S0 = 0.5 (Tusher et al. 2001). Prior to the t-test, data were filtered to have a minimum of two values in at least one of the triplicate noncontrol samples. The missing values were then replaced by a constant value (around the lowest-intensity value). Hierarchical clustering of proteins was performed on logarithmized intensities after Z-score normalization of the data using Euclidean distances.
RNA preparation and qPCR
Equal-volume fractions of sucrose gradients were collected into SDS at a final concentration of 1%. The samples were incubated with 100 μg of proteinase K for 30 min at 37°C, and RNA was extracted with an equal volume of phenol:chloroform:isoamyl (Sigma) followed by centrifugation at 12,000g for 15 min. The aqueous phase was precipitated with 1 vol of isopropanol. Following overnight incubation at −20°C, RNA was pelleted at 12,000g for 30 min at 4°C, washed twice with ice-cold 75% ethanol, and resuspended in water. cDNA was prepared with equal amounts of RNA from each fraction using poly-dT and Verso cDNA enzyme according to the manufacturer's instructions. cDNAs were amplified by qPCR using QuantaBio Fast SYBR mix with 1 μM primers (sequences included below). qPCR was performed using Step-One (Life Technologies) RT–PCR protocol with 60°C as annealing temperature. Relative CALR mRNA amounts were calculated as RQ = 2ΔCt, where ΔCt is the Ct of CALR in fraction × −Ct of CALR in fraction 1. Relative RNA amounts for other mRNAs were calculated as RQ = 2−ΔΔCt, where ΔΔCt is ΔCt for a specific mRNA in fraction × −ΔCt for CALR in the same fraction. Error bars refer to technical replicates of qPCR measurements within single experiments.
O.E.S. acknowledges support from the Israel Science Foundation (grants 131/07 and 1036/12). T.G. is supported by funds from Tel Aviv University.
Supplemental material is available for this article.
Article published online ahead of print. Article and publication date are online at http://www.genesdev.org/cgi/doi/10.1101/gad.219105.113.
ReferencesBlobelG, SabatiniD1971Dissociation of mammalian polyribosomes into subunits by puromycin.
68: 390–3945277091CalkhovenCF, MullerC, LeutzA2002Translational control of gene expression and disease.
8: 577–58312470991ClarkIE, WyckoffD, GavisER2000Synthesis of the posterior determinant Nanos is spatially restricted by a novel cotranslational regulatory mechanism.
10: 1311–131411069116CoxJ, MannM2008MaxQuant enables high peptide identification rates, individualized p.p.b.-range mass accuracies and proteome-wide protein quantification.
26: 1367–137219029910CoxJ, NeuhauserN, MichalskiA, ScheltemaRA, OlsenJV, MannM2011Andromeda: A peptide search engine integrated into the MaxQuant environment.
10: 1794–180521254760DarnellJC, Van DriescheSJ, ZhangC, HungKY, MeleA, FraserCE, StoneEF, ChenC, FakJJ, ChiSW, 2011FMRP stalls ribosomal translocation on mRNAs linked to synaptic function and autism.
146: 247–26121784246DavidA, DolanBP, HickmanHD, KnowltonJJ, ClavarinoG, PierreP, BenninkJR, YewdellJW2012Nuclear translation visualized by ribosome-bound nascent chain puromycylation.
197: 45–5722472439de BoerE, RodriguezP, BonteE, KrijgsveldJ, KatsantoniE, HeckA, GrosveldF, StrouboulisJ2003Efficient biotinylation and single-step purification of tagged transcription factors in mammalian cells and transgenic mice.
100: 7480–748512802011DieterichDC, LinkAJ, GraumannJ, TirrellDA, SchumanEM2006Selective identification of newly synthesized proteins in mammalian cells using bioorthogonal noncanonical amino acid tagging (BONCAT).
103: 9482–948716769897EichelbaumK, WinterM, DiazMB, HerzigS, KrijgsveldJ2012Selective enrichment of newly synthesized proteins for quantitative secretome analysis.
30: 984–99023000932FanF, JinS, AmundsonSA, TongT, FanW, ZhaoH, ZhuX, MazzacuratiL, LiX, PetrikKL, 2002ATF3 induction following DNA damage is regulated by distinct signaling pathways and over-expression of ATF3 protein suppresses cells growth.
21: 7488–749612386811Garbarino-PicoE, GreenCB2007Posttranscriptional regulation of mammalian circadian clock output.
72: 145–15618419272GroismanI, HuangYS, MendezR, CaoQ, TheurkaufW, RichterJD2000CPEB, maskin, and cyclin B1 mRNA at the mitotic apparatus: Implications for local translational control of cell division.
103: 435–44711081630GuoH, IngoliaNT, WeissmanJS, BartelDP2010Mammalian microRNAs predominantly act to decrease target mRNA levels.
466: 835–84020703300HobdenAN, CundliffeE1978The mode of action of α sarcin and a novel assay of the puromycin reaction.
170: 57–61629783HolcikM, SonenbergN2005Translational control in stress and apoptosis.
6: 318–32715803138HowdenAJ, GeogheganV, KatschK, EfstathiouG, BhushanB, BoutureiraO, ThomasB, TrudgianDC, KesslerBM, DieterichDC, 2013QuaNCAT: Quantitating proteome dynamics in primary cells.
10: 343–34623474466HuszarD, TheoclitouME, SkolnikJ, HerbstR2009Kinesin motor proteins as targets for cancer therapy.
28: 197–20819156502IngoliaNT, GhaemmaghamiS, NewmanJR, WeissmanJS2009Genome-wide analysis in vivo of translation with nucleotide resolution using ribosome profiling.
324: 218–22319213877JoyceD, AlbaneseC, SteerJ, FuM, BouzahzahB, PestellRG2001NF-κB and cell-cycle regulation: The cyclin connection.
12: 73–9011312120KiickKL, SaxonE, TirrellDA, BertozziCR2002Incorporation of azides into recombinant proteins for chemoselective modification by the Staudinger ligation.
99: 19–2411752401KimST, XuB, KastanMB2002Involvement of the cohesin protein, Smc1, in Atm-dependent and independent responses to DNA damage.
16: 560–57011877376KramerG, SprengerRR, BackJ, DekkerHL, NessenMA, van MaarseveenJH, de KoningLJ, HellingwerfKJ, de JongL, de KosterCG2009Identification and quantitation of newly synthesized proteins in Escherichia coli by enrichment of azidohomoalanine-labeled peptides with diagonal chromatography.
8: 1599–161119321432LeeS, LiuB, HuangSX, ShenB, QianSB2012Global mapping of translation initiation sites in mammalian cells at single-nucleotide resolution.
109: E2424–E243222927429LiangYC, HsuCY, YaoYL, YangWM2013PARP-2 regulates cell cycle-related genes through histone deacetylation and methylation independently of poly(ADP-ribosyl)ation.
431: 58–6423291187LindqvistA, Rodriguez-BravoV, MedemaRH2009The decision to enter mitosis: Feedback and redundancy in the mitotic entry network.
185: 193–20219364923LiuJ, XuY, StoleruD, SalicA2012Imaging protein synthesis in cells and tissues with an alkyne analog of puromycin.
109: 413–41822160674LosadaA, HiranoT2005Dynamic molecular linkers of the genome: The first decade of SMC proteins.
19: 1269–128715937217MarzluffWF, DuronioRJ2002Histone mRNA expression: Multiple levels of cell cycle regulation and important developmental consequences.
14: 692–69912473341MikiH, SetouM, KaneshiroK, HirokawaN2001All kinesin superfamily protein, KIF, genes in mouse and human.
98: 7004–701111416179MumbergD, WickM, BurgerC, HaasK, FunkM, MullerR1997Cyclin ET, a new splice variant of human cyclin E with a unique expression pattern during cell cycle progression and differentiation.
25: 2098–21059153308NagarajN, WisniewskiJR, GeigerT, CoxJ, KircherM, KelsoJ, PaaboS, MannM2011Deep proteome and transcriptome mapping of a human cancer cell line.
7: 54822068331NottrottS, SimardMJ, RichterJD2006Human let-7a miRNA blocks protein production on actively translating polyribosomes.
13: 1108–111417128272PestkaS1971Inhibitors of ribosome functions.
25: 487–5624949424PetersenCP, BordeleauME, PelletierJ, SharpPA2006Short RNAs repress translation after initiation in mammalian cells.
21: 533–54216483934PinesJ, HunterT1989Isolation of a human cyclin cDNA: Evidence for cyclin mRNA and protein regulation in the cell cycle and for interaction with p34cdc2.
58: 833–8462570636RaoPN, EngelbergJ1968Mitotic duration and its variability in relation to temperature in HeLa cells.
52: 198–2085675557RappsilberJ, MannM, IshihamaY2007Protocol for micro-purification, enrichment, pre-fractionation and storage of peptides for proteomics using StageTips.
2: 1896–190617703201RiabowolK, SchiffJ, GilmanMZ1992Transcription factor AP-1 activity is required for initiation of DNA synthesis and is lost during cellular aging.
89: 157–1611729683RobinsonBH, OeiJ, SaundersM, GravelR1983[3H]biotin-labeled proteins in cultured human skin fibroblasts from patients with pyruvate carboxylase deficiency.
258: 6660–66646406485RybakJN, ScheurerSB, NeriD, EliaG2004Purification of biotinylated proteins on streptavidin resin: A protocol for quantitative elution.
4: 2296–229915274123SchmidtEK, ClavarinoG, CeppiM, PierreP2009SUnSET, a nonradioactive method to monitor protein synthesis.
6: 275–27719305406SchwanhausserB, GossenM, DittmarG, SelbachM2009Global analysis of cellular protein translation by pulsed SILAC.
9: 205–20919053139SchwanhausserB, BusseD, LiN, DittmarG, SchuchhardtJ, WolfJ, ChenW, SelbachM2011Global quantification of mammalian gene expression control.
473: 337–34221593866SigoillotFD, HuckinsJF, LiF, ZhouX, WongST, KingRW2011A time-series method for automated measurement of changes in mitotic and interphase duration from time-lapse movies.
6: e2551121966537SivanG, KedershaN, Elroy-SteinO2007Ribosomal slowdown mediates translational arrest during cellular division.
27: 6639–664617664278SivanG, AvinerR, Elroy-SteinO2011Mitotic modulation of translation elongation factor 1 leads to hindered tRNA delivery to ribosomes.
286: 27927–2793521665947StarckSR, GreenHM, Alberola-IlaJ, RobertsRW2004A general approach to detect protein expression in vivo using fluorescent puromycin conjugates.
11: 999–100815271358SugiyamaT, Sugioka-SugiyamaR, HadaK, NiwaR2012Rhn1, a nuclear protein, is required for suppression of meiotic mRNAs in mitotically dividing fission yeast.
7: e4296222912768TagwerkerC, FlickK, CuiM, GuerreroC, DouY, AuerB, BaldiP, HuangL, KaiserP2006A tandem affinity tag for two-step purification under fully denaturing conditions: Application in ubiquitin profiling and protein complex identification combined with in vivocross-linking.
5: 737–74816432255TusherVG, TibshiraniR, ChuG2001Significance analysis of microarrays applied to the ionizing radiation response.
98: 5116–512111309499VaughanMHJr, PawlowskiPJ, ForchhammerJ1971Regulation of protein synthesis initiation in HeLa cells deprived of single essential amino acids.
68: 2057–20615289364VogelC, MarcotteEM2012Insights into the regulation of protein abundance from proteomic and transcriptomic analyses.
13: 227–23222411467WhitfieldML, ZhengLX, BaldwinA, OhtaT, HurtMM, MarzluffWF2000Stem-loop binding protein, the protein that binds the 3′ end of histone mRNA, is cell cycle regulated by both translational and posttranslational mechanisms.
20: 4188–419810825184YasumiM, SakisakaT, HoshinoT, KimuraT, SakamotoY, YamanakaT, OhnoS, TakaiY2005Direct binding of Lgl2 to LGN during mitosis and its requirement for normal cell division.
280: 6761–676515632202oai:pubmedcentral.nih.gov:37782382013-09-23genesdevpmc-openGenes DevGenes DevGADGenes & Development0890-93691549-5477Cold Spring Harbor Laboratory PressPMC3778238PMC377823837782382401349924013499871166010.1101/gad.221713.113Research CommunicationPhysical clustering of FLC alleles during Polycomb-mediated epigenetic silencing in vernalizationRosa et al.Physical clustering of FLC during vernalizationRosaStefanie15De LuciaFilomena156MylneJoshua S.1257ZhuDanling15OhmidoNobuko3PendleAli1KatoNaohiro4ShawPeter1DeanCaroline18Department of Cell and Developmental Biology, John Innes Centre, Norwich Research Park, Norwich NR4 7UH, United Kingdom;The University of Queensland, Institute for Molecular Bioscience, Brisbane, Queensland 4072, Australia;Graduate School of Human Development and Environment, Kobe University, Kobe 657-8501, Japan;Department of Biological Sciences, Louisiana State University, Baton Rouge, Louisiana 70803, USA
These authors contributed equally to this work.
Present addresses: 6Unité de Biologie des Interactions Cellulaires URA CNRS 2582, Institut Pasteur 25, rue du Docteur Roux 75724 Paris Cedex 15, France.
School of Chemistry and Biochemistry, ARC Centre of Excellence in Plant Energy Biology, The University of Western Australia, Crawley, Perth, Western Australia 6009, Australia.
How nuclear organization influences chromatin changes at individual loci is poorly understood. Vernalization, the promotion of flowering by cold, involves Polycomb-mediated silencing of FLOWERING LOCUS C (FLC). Here, Rosa et al. use live-cell imaging to monitor nuclear organization and FLC dynamics during vernalization. The data reveal that Polycomb-dependent clustering of FLC alleles is a cold-induced step in gene silencing. This study suggests that physical clustering of target genes may be a common feature of Polycomb-mediated epigenetic silencing mechanisms.
Vernalization, the promotion of flowering by cold, involves Polycomb-mediated epigenetic silencing of FLOWERING LOCUS C (FLC). Cold progressively promotes cell-autonomous switching to a silenced state. Here, we used live-cell imaging of FLC-lacO to monitor changes in nuclear organization during vernalization. FLC-lacO alleles physically cluster during the cold and generally remain so after plants are returned to warm. Clustering is dependent on the Polycomb trans-factors necessary for establishment of the FLC silenced state but not on LIKE HETEROCHROMATIN PROTEIN 1, which functions to maintain silencing. These data support the view that physical clustering may be a common feature of Polycomb-mediated epigenetic switching mechanisms.
The ability of plants to sense environmental cues and adapt their growth and development accordingly is key to ensuring their reproductive success. One example of this is vernalization, the process by which flowering is accelerated by prolonged cold. A vernalization requirement aligns reproductive development with the favorable conditions of spring. Vernalization in Arabidopsis thaliana (Arabidopsis) involves the cold-induced epigenetic silencing of FLOWERING LOCUS C (FLC), a MADS box protein that represses flowering (Michaels and Amasino 1999; Sheldon et al. 1999).
Our current understanding of vernalization has been derived from forward genetics combined with chromatin analysis. A well-defined sequence of events has been described (Supplemental Fig. S1) that involves Polycomb-repressive complex 2 (PRC2) and a set of plant homeodomain (PHD) proteins (Gendall et al. 2001; Sung and Amasino 2004; Wood et al. 2006; Greb et al. 2007). An early step in vernalization is disruption of a gene loop (Crevillen et al. 2013), and this is coincident with increased expression of a set of FLC antisense transcripts capable of cold-induced down-regulation of linked coding sequences (Swiezewski et al. 2009). These steps precede and are independent of formation of a modified PHD–PRC2 that accumulates at a specific site in FLC, covering exon 1 and the beginning of the first intron (Finnegan and Dennis 2007; Angel et al. 2011). The PHD–PRC2 is composed of the canonical PRC2 components including VERNALIZATION2 (VRN2), a Su(Z)12 homolog (Gendall et al. 2001), together with PHD proteins VRN5, VEL1, and the cold-induced VIN3 (Sung and Amasino 2004; Wood et al. 2006; De Lucia et al. 2008). It results in progressive accumulation of H3K27me3 at the nucleation site with increasing cold exposure (Angel et al. 2011). A cold-induced noncoding sense RNA, COLDAIR, aids PRC2 recruitment (Heo and Sung 2011). After plants are transferred back to warm conditions, the PHD–PRC2 (now without VIN3) spreads across the whole gene, causing increased H3K27me3 over the whole locus. The levels of the H3K27me3 again quantitatively reflect the length of cold exposure. This quantitative increase has been shown to reflect a population average of an increasing proportion of cells that have switched to the epigenetically silent state (Angel et al. 2011). Maintenance of this silenced state requires LIKE HETEROCHROMATIN PROTEIN 1 (LHP1) (Mylne et al. 2006; Sung et al. 2006).
An emerging theme in gene regulation is the importance of physical position in the nucleus (Fraser and Bickmore 2007; Bickmore and van Steensel 2013; Hubner et al. 2013). In situ hybridization techniques initially revealed the importance of higher-order sequence interactions and nuclear domains in X-chromosome inactivation (Pollex and Heard 2012), heterochromatin formation, and Polycomb silencing (Brown et al. 1997; Lanzuolo et al. 2007; Bantignies and Cavalli 2011). This understanding has been augmented by chromatin conformation capture techniques that have provided high-resolution interaction maps identifying topologically distinct domains within chromosomes (Noordermeer et al. 2011; Naumova et al. 2012; Shen et al. 2012). Association of changes in gene regulation with alteration of these higher-order structures has been shown for Polycomb silenced loci in Drosophila and Hox plus X-linked loci in mammals (Lanzuolo et al. 2007; Noordermeer et al. 2011; Nora et al. 2012; Towbin et al. 2012). Live-cell imaging has also revealed the dynamic nature of this nuclear organization, with short- and long-range motion of Polycomb bodies likely to correspond to movement within and between chromosome territories (Hubner and Spector 2010; Matzke et al. 2010; Cheutin and Cavalli 2012).
How this dynamic nuclear organization interconnects with specific chromatin changes at individual loci is poorly understood. Here, we used live imaging of a transgene containing a FLC-lacO array to monitor nuclear organization and dynamics of FLC during vernalization in Arabidopsis. This revealed a cold-induced physical clustering of FLC-lacO alleles, with the number of cells showing this condition increasing quantitatively with exposure to cold. The clustering is generally stable after cold exposure and is differentially disrupted by the trans-factors required for vernalization. These data suggest that physical clustering of FLC alleles during the cold phase is an important component of FLC epigenetic silencing and support the view that formation of higher-order structures may be an integral component of epigenetic switching mechanisms.
Results and DiscussionVisualization of a functional FLC-lacO transgene in root cells
We wanted to explore the role of nuclear organization during the Polycomb-mediated epigenetic silencing of FLC that occurs during vernalization. The goal was to exploit the well-defined sequence of events established for vernalization in order to associate changes in nuclear organization with functionally important steps in Polycomb silencing. An array of 120 copies of the lacO DNA sequence—each copy separated by a random ∼10-base-pair (bp) sequence to minimize recombinational and replication instability, thus minimizing repeat-induced silencing (Fig. 1A; Lau et al. 2003)—was inserted into the FLC gene either downstream from the poly(A) site (Fig. 1A; Supplemental Fig. S2) or into the first intron (Fig. 2K; Supplemental Fig. S2). A lacO alone was also generated as a control (Supplemental Fig. S2). Only the FLC-lacO construct carrying the lacO array downstream from the poly(A) site was fully functional in planta (Supplemental Fig. S3). Three FLC-lacO transgenic lines (lines 1, 2, and 3) containing single and complete T-DNA insertions of this construct were then selected (Supplemental Figs. S4, S5). A separate Arabidopsis line was generated expressing a LacI-YFP-NLS fusion, and this was crossed with the lacO plants. The LacI-YFP-NLS fusion was expressed from the ethanol-inducible promoter (Roslan et al. 2001). lacO/LacI foci were monitored in root epidermal cells, as these are easily imaged, strongly express FLC, and show cold-induced epigenetic silencing of FLC. Indeed, vernalized root cells maintain FLC silencing even through regeneration into a new plant (Burn et al. 1993). In many experiments, the ethanol-inducible promoter resulted in YFP expression that was too strong to visualize any lacO foci, but in the absence of induction, LacI-YFP expression was at an appropriate level in the differentiation zone above the root meristem (Fig. 1B). In this region, cells undergo chromosome endoreduplication, resulting in larger nuclei with higher ploidy (Galbraith et al. 1991). In nonvernalized plants, multiple (six or more) FLC-lacO foci were detected in this region of the root (Fig. 1C–F). Careful titration of the ethanol exposure enabled meristematic cells to be imaged, and in these cells, only two foci were detected (Fig. 1G,H). These foci did not associate with an obvious location in the nucleus, sometimes being found at one pole in the nucleus of the elongating cells, sometimes in the center.
Monitoring FLC in root nuclei using an FLC-lacO transgene. (A) The lacO array was inserted downstream from the polyadenylation site of FLC. (B) Arabidopsis root showing expression of LacI-YFP in nuclei predominantly in the differentiation zone (bar, 100 μm). (C–F) Representative fluorescence images of nuclei from region 2 showing multiple FLC foci. (G,H) Representative fluorescence images of nuclei from meristematic cells (from region 1) (bars, 5 μm).
Cold-induced clustering of FLC-lacO. (A–D) Representative fluorescence images of Arabidopsis root cells (from region 2) (see Fig. 1B) in plants grown in cold for 2 wk (and imaged immediately afterward) showing clustering of FLC-lacO loci. (E,F) Clustering of FLC-lacO copies in meristematic cells in plants cold-treated for 2 wk. LacI-YFP expression was induced using a 1% ethanol vapor treatment for 1.5 h. (G) Quantification of the FLC-lacO foci (30 cells counted in three different roots) from cells in region 2. (H,I) Representative fluorescence images of nuclei from cells in region 2 expressing a lacO-only transgene. Plants were either nonvernalized (NV) or vernalized for 2 wk (+V). (J) Quantification of the lacO foci (30 cells counted in three different roots) from cells in region 2. (K) FLC-lacO transgene with the lacO array inserted at the BstEII restriction site in FLC intron 1 of the genomic FLC sequence. (L,M) Representative fluorescence images of FLC-lacO-Bst transgene in root cells from region 2 nonvernalized (L) and vernalized for 2 wk (M). (N) Quantification of the FLC-lacO-Bst foci from cells in region 2 (30 cells counted in three different roots) nonvernalized and vernalized for 2 wk (bars, 5 μm).
Clustering of FLC-lacO is induced by cold
After the plants had been exposed for 2 wk to 5°C, a large proportion of nuclei showed FLC-LacO loci that had clustered into one or two foci in endoreduplicated cells (Fig. 2A–D,G) and one in the meristematic cells (Fig. 2E,F). The foci were larger than those in non-cold-treated plants, consistent with a physical clustering (Supplemental Fig. S6Q,R), and again showed no obvious preferential localization in the nucleus. This clustering was found in all of the independent FLC-lacO lines (Supplemental Fig. S6A–P) but was not observed in an Arabidopsis line carrying a single and complete lacO-only transgene (Fig. 2H–J). The noncomplementing FLC-lacO transgene carrying the lacO array in the first intron of FLC (FLC-lacO-Bst) (Fig. 2K–N) also did not show clustering. No RNA was detected from this transgene (Supplemental Fig. S7), suggesting that the clustering process requires expression of the FLC gene.
The clustering of FLC-lacO is impaired in vrn2 and vrn5 but not in lhp1
To investigate factors required for clustering, we crossed the FLC-lacO line 1 containing the LacI-YFP into different mutant backgrounds (vrn2, vrn5, and lhp1). These mutations impair different phases of the epigenetic silencing of FLC; vrn2 disrupts FLC silencing before and during the cold (Gendall et al. 2001), vrn5 disrupts the accumulation of the silencing in the cold (Greb et al. 2007), and lhp1 disrupts the maintenance of silencing after the cold (Mylne et al. 2006; Sung et al. 2006). There was a slightly higher level of preclustering in the vrn5 mutant compared with vrn2, but both abolished the cold-induced clustering of the FLC-lacO foci in the root epidermal cells (Fig. 3A–D,G,H). In contrast, the cold-induced clustering at the end of the cold, or shortly after return to the warm, occurred normally in the lhp1 mutant (Fig. 3E,F). These data suggest that physical clustering of the FLC-lacO alleles in the cold requires the same components as the PHD–PRC2 nucleation event rather than those functioning later in the vernalization process.
The clustering of FLC-lacO is impaired in vrn2 and vrn5 but not in lhp1. (A) Region 2 root cells either nonvernalized (NV) or vernalized for 2 wk (+V). (B) vrn5 nonvernalized (NV) or vernalized for 2 wk (+V). (C) Quantification of FLC-lacO foci in 30 vrn2 cells in region 2. (D) Quantification of FLC-lacO foci in 30 vrn5 cells in region 2. (E) Images of FLC-lacO transgene in lhp1 background nonvernalized (NV) or vernalized for 2 wk (+V). (F) Quantification of FLC-lacO foci in 50 lhp1 cells in region 2. (G,H) Representative fluorescence images of nuclei from meristematic cells harboring the FLC-lacO from plants nonvernalized (NV) or vernalized for 2 wk (+V) in wild type (G) and vrn5 (H) (bars, 5 μm).
The clustering of FLC-lacO increases quantitatively with increasing cold exposure
We next considered how the timing of changes in cold-induced physical clustering compared with other cold-induced processes involved in vernalization. The robust gene loop that is found in many genotypes in warm-grown plants is disrupted in the first few weeks of cold exposure (Crevillen et al. 2013), increased antisense transcription is maximal after 2–3 wk of exposure (Swiezewski et al. 2009), and the increase in H3K27me3 at the nucleation site occurs quantitatively throughout the first 4 wk of cold (Angel et al. 2011). To compare the dynamics of cold-induced clustering with the timing of these other cold-induced steps, the number of cells showing reduced numbers of FLC-lacO foci was measured in plants exposed to different lengths of cold (Fig. 4A; Supplemental Fig. S8). The time-dependent increase in clustering during the cold paralleled the increase in H3K27me3 at the intronic nucleation site (Fig. 4E). Previous work combining modeling and experimental analysis revealed that the quantitative increase in epigenetic silencing, so characteristic of vernalization, is the result of a cell-autonomous switching mechanism, with the probability of switching increasing with increasing exposure to cold (Angel et al. 2011). The parallels observed in this study between clustering and H3K27me3 nucleation suggest that this cell-autonomous switching mechanism might involve the switch of physical position of the FLC locus in the nucleus.
The clustering of FLC-lacO increases quantitatively with increasing cold exposure. (A) Time course of FLC-lacO clustering after different weeks of cold (XT0). The number of foci (one, two, or more than two) was counted in 10 randomly selected nuclei in five different plants. (B) The same analysis but after cold the plants were grown for 7 d (T7) at 20°C before imaging. The number of foci (one, two, or more than two) was counted in 10 randomly selected nuclei in five different plants. (C) Quantification of the proportion of cells in region 2 showing one FLC-lacO cluster after 2 wk of cold and at post-cold time points T7 (7 d post-transfer from cold) and T14. (D) Quantification of cells in the meristematic region showing one FLC-lacO cluster. FLC-lacO clustering in the meristem was quantified in nonvernalized (NV) plants and plants vernalized for different lengths (2–5 wk) as well as at the respective T7 time point. For this analysis, at least three different plants were used for quantification at each time point. (E) Clustering dynamics parallel the cold-induced increases in H3K27me3 at the nucleation site. With increasing cold exposure, H3K27me3 levels increase at the nucleation site (end of exon 1–start intron 1), the timing of which is coincident with the reduction in FLC-lacO foci number. FLC is fully silenced after 6 wk of cold, and this is associated with high H3K27me3 and maintenance of the clustering during subsequent growth at warm temperatures. Gray data points represent the increase of H3K27me3 at the nucleation site normalized to H3 and SHOOT MERISTEMLESS (data from Angel et al. 2011). Black data points represent the frequency of one FLC–lacO cluster.
We then investigated whether the cold-induced clustering was maintained through subsequent cell division after plants were returned to warm conditions. Analysis of endoreduplicated and meristematic cells showed that clustering was generally maintained for 7 d after transfer from the cold (Fig. 4B–D). However, the endoreduplicated cells tended to lose clustering (Fig. 4B,C) slightly more than cells in the meristem (Fig. 4D). We attempted to ask whether the clustering was associated with formation of structures equivalent to Polycomb bodies in mammals (Bantignies and Cavalli 2011) by colocalization of LacI-YFP, FLC-lacO, and VRN2 using immunofluorescence, but the local concentration of the LacI-YFP protein proved too low to detect the FLC-lacO foci using anti-YFP/GFP antibodies. We were also unable to analyze colocalization of FLC-lacO with other Polycomb targets, as clustering was not observed after the extensive fixation and hybridization procedures required for in situ hybridization in plant cells (Supplemental Fig. S9). We interpret this as suggesting that, as in other organisms, the clusters involved in the Polycomb silencing underlying vernalization are highly dynamic structures (Hubner and Spector 2010; Matzke et al. 2010; Cheutin and Cavalli 2012). There is also the possibility that the FLC-lacO repeats stabilize more transient interactions of the endogenous loci.
Our demonstration of cold-induced clustering of FLC-lacO alleles during vernalization suggests that physical repositioning in the nucleus may be a common feature of Polycomb-mediated epigenetic silencing. Robustness in regulatory mechanisms generally involves feedback loops, so we envisage that some aspect of silencing results in clustering, but that clustering then reinforces the silencing. The clustering may be one manifestation of the cell-autonomous switching mechanism proposed for vernalization, and it is an attractive model to generally account for localization of silenced chromosomal domains into Polycomb bodies. Interconnecting the different aspects of these silencing mechanisms will be crucial to fully understand how they reinforce each other, ensuring mitotic inheritance of epigenetic regulation.
Materials and methodsConstructs and transgenic lines
Three constructs were used in this study: a FLC reporting construct containing genomic FLC with a lacO array and kanamycin resistance cassette inserted at either the SwaI site downstream from the FLC poly(A) site (FLC-lacO) or the BstEII site within the first intron (FLC-lacO-Bst) and a control containing only the lacO array and kanamycin resistance cassette (lacO only). To produce the FLC reporting construct (FLC-lacO), a 12-kb SacI fragment of Columbia genomic DNA was cloned in pBS (pFLC15), including 3.7 kb upstream of the FLC start ATG to 2.7 kb downstream from the end of the FLC 3′ untranslated region (UTR). The lacO array was obtained from the vector pLAU41, which contains ∼120 copies of the lacO DNA sequence, each interspersed with random ∼10-mers and a kanamycin resistance (KmR) cassette within a pUC18 backbone similar to the published vector pLAU43. This interspersed repeat array reduced problems with repeat-induced silencing (Lau et al. 2003). To produce the FLC construct with the lacO array in the first intron (FLC-lacO-Bst), an approach similar to FLC-lacO was used, but instead the 5.3-kb lacO-KmR fragment was inserted at the BstEII site. The control contained only the lacO array and kanamycin resistance cassette (lacO only).
The constructs were cloned into Agrobacterium binary vector pSLJ755I5 and transformed into flc-3 FRISF2 containing a small deletion at the 5′ end of FLC and an active FRIGIDA allele from the Arabidopsis accession San-Feliu-2 (Michaels and Amasino 2001). The BstEII lacO-containing FLC constructs did not delay the flowering time of flc-3 FRISF2 at all and therefore were deemed nonfunctional.
Fluorescent in situ hybridization
Arabidopsis roots were fixed for 60 min with 4% (w/v) formaldehyde freshly made from paraformaldehyde (PFA) in PBS buffer. After washing in PBS for 5 min, the roots were digested in a mixture of 1% driselase, 0.5% cellulase, and 0.025% pectolyase for 45–60 min at 37°C. After enzyme treatment, roots were washed in PBS three times for 5 min each and squashed between poly-L lysine slides (Polysine, VWR International) and coverslips in PBS. After the slides were frozen in liquid nitrogen, the coverslips were removed, and the samples were air-dried. Slides were treated with 10 mg/mL RNase for 1 h at 37°C and washed twice in 2× SSC. Probes were labeled with Cy3-dUTP (Sigma) by nick translation. Bacterial artificial chromosome (BAC) clone JAtY79P05, which contains an insert of 60.12 kb, was used as a probe for the native Arabidopsis genome immediately adjacent to the FLC-lacO T-DNA on the bottom arm of chromosome V (“BAC probe”) or the 5.3 kb of the lacO array (“lacO probe”). The hybridization mixture (20 ng/μL labeled DNA, 50% formamide, 10% dextran sulfate, 2× SSC) was applied to the slides, which were then denatured for 7 min at 75°C, 3 min at 55°C, min, and 3 min at 50°C and hybridized overnight at 37°C. After hybridization, slides were washed at 42°C once in 2× SSC, twice in 20% formamide 2× SSC, and twice in 2× SSC and then at room temperature twice in 2× SSC and twice in 4× SSC and 0.1% Tween 20. Nuclei were counterstained with 1 μg/mL DAPI (4′,6-diamidino-2-phenylindole dihydrochloride hydrate; Sigma), and slides were mounted in Vectashield (Vector Laboratories). RNA in situ hybridization was performed on nonvernalized and 4-wk vernalized wild-type and FLC-lacO seedling roots.
Live imaging
Arabidopsis seedlings were placed in a biochamber constructed from a standard coverslip, a well (∼16 × 24 mm) made in Secure Seal (double-sided adhesive sheet; Grace Bio-Labs) filled with MS medium, and a gas-permeable membrane (BioFolie, Viva Science) attached to the Secure Seal as the bottom of the chamber. The chamber with the gas-permeable membrane facing down was then placed over a 1-cm hole drilled into a plastic support slide to allow free gaseous exchange through the gas-permeable membrane, and the edges of the sandwich were sealed with tape.
Imaging was performed using a 60× oil lens on a Nikon Eclipse 600 epifluorescence microscope equipped with a Hamamatsu Orca ER cooled CCD digital camera and a Prior Proscan x–z stage. The following wavelengths were used for fluorescence detection: excitation 340–380 nm and emission 425–475 nm for DAPI, and excitation 490–510 nm and emission 520–550 nm for YFP. For all experiments, series of optical sections with z-steps of 150 or 200 nm were collected using MetaMorph software (Universal Imaging). The images from z sections were projected using the maximum intensity projection algorithm in the ImageJ program.
Acknowledgments
We thank Kim Johnson for help with TAIL-PCR, and Maike Stam, Robert Sablowski, Silvia Costa, and Clive Lloyd for comments on the manuscript. F.D.L. was supported by Biotechnology and Biological Sciences Research Council (BBSRC) grant BB/C517633/1 and European Research Council grant 233039 ENVGENE. S.R. was supported by grant SFRH/BD/23202/2005 from the Portuguese Fundação para a Ciência e a Tecnologia. J.S.M. was supported by European Commission grant QLK5-CT-2001-01412, an ARC QEII Fellowship (DP0879133), and an ARC Future Fellowship (FT120100013). P.S. and C.D. acknowledge support from ISP grant BB/J004588/1 from the BBSRC and the John Innes Foundation.
Supplemental material is available for this article.
Article is online at http://www.genesdev.org/cgi/doi/10.1101/gad.221713.113.
ReferencesAngelA, SongJ, DeanC, HowardM2011A Polycomb-based switch underlying quantitative epigenetic memory.
476: 105–10821785438BantigniesF, CavalliG2011Polycomb group proteins: Repression in 3D.
27: 454–46421794944BickmoreWA, van SteenselB2013Genome architecture: Domain organization of interphase chromosomes.
152: 1270–128423498936BrownKE, GuestSS, SmaleST, HahmK, MerkenschlagerM, FisherAG1997Association of transcriptionally silent genes with Ikaros complexes at centromeric heterochromatin.
91: 845–8549413993BurnJE, BagnallDJ, MetzgerJD, DennisES, PeacockWJ1993DNA methylation, vernalization, and the initiation of flowering.
90: 287–29111607346CheutinT, CavalliG2012Progressive polycomb assembly on H3K27me3 compartments generates polycomb bodies with developmentally regulated motion.
8: e100246522275876CrevillenP, SonmezC, WuZ, DeanC2013A gene loop containing the floral repressor FLC is disrupted in the early phase of vernalization.
32: 140–14823222483De LuciaF, CrevillenP, JonesAM, GrebT, DeanC2008A PHD–polycomb repressive complex 2 triggers the epigenetic silencing of FLC during vernalization.
105: 16831–1683618854416FinneganEJ, DennisES2007Vernalization-induced trimethylation of histone H3 lysine 27 at FLC is not maintained in mitotically quiescent cells.
17: 1978–198317980595FraserP, BickmoreW2007Nuclear organization of the genome and the potential for gene regulation.
447: 413–41717522674GalbraithDW, HarkinsKR, KnappS1991Systemic endopolyploidy in Arabidopsis thaliana.
96: 985–98916668285GendallAR, LevyYY, WilsonA, DeanC2001The VERNALIZATION 2 gene mediates the epigenetic regulation of vernalization in Arabidopsis.
107: 525–53511719192GrebT, MylneJS, CrevillenP, GeraldoN, AnH, GendallAR, DeanC2007The PHD finger protein VRN5 functions in the epigenetic silencing of Arabidopsis FLC.
17: 73–7817174094HeoJB, SungS2011Vernalization-mediated epigenetic silencing by a long intronic noncoding RNA.
331: 76–7921127216HubnerMR, SpectorDL2010Chromatin dynamics.
39: 471–48920462379HubnerMR, Eckersley-MaslinMA, SpectorDL2013Chromatin organization and transcriptional regulation.
23: 89–9523270812LanzuoloC, RoureV, DekkerJ, BantigniesF, OrlandoV2007Polycomb response elements mediate the formation of chromosome higher-order structures in the bithorax complex.
9: 1167–117417828248LauIF, FilipSR, SoballeB, OkstadO, BarreF, SherrattDJ2003Spatial and temporal organization of replicating Escherichia coli chromosomes.
49: 731–74312864855MatzkeAJ, WatanabeK, van der WindenJ, NaumannU, MatzkeM2010High frequency, cell type-specific visualization of fluorescent-tagged genomic sites in interphase and mitotic cells of living Arabidopsis plants.
6: 220148117MichaelsSD, AmasinoRM1999FLOWERING LOCUS C encodes a novel MADS domain protein that acts as a repressor of flowering.
11: 949–95610330478MichaelsSD, AmasinoRM2001Loss of FLOWERING LOCUS C activity eliminates the late-flowering phenotype of FRIGIDA and autonomous pathway mutations but not responsiveness to vernalization.
13: 935–94111283346MylneJS, BarrettL, TessadoriF, MesnageS, JohnsonL, BernatavichuteYV, JacobsenSE, FranszP, DeanC2006LHP1, the Arabidopsis homologue of HETEROCHROMATIN PROTEIN1, is required for epigenetic silencing of FLC.
103: 5012–501716549797NaumovaN, SmithEM, ZhanY, DekkerJ2012Analysis of long-range chromatin interactions using Chromosome Conformation Capture.
58: 192–20322903059NoordermeerD, LeleuM, SplinterE, RougemontJ, De LaatW, DubouleD2011The dynamic architecture of Hox gene clusters.
334: 222–22521998387NoraEP, LajoieBR, SchulzEG, GiorgettiL, OkamotoI, ServantN, PiolotT, van BerkumNL, MeisigJ, SedatJ, 2012Spatial partitioning of the regulatory landscape of the X-inactivation centre.
485: 381–38522495304PollexT, HeardE2012Recent advances in X-chromosome inactivation research.
24: 825–83223142477RoslanHA, SalterMG, WoodCD, WhiteMR, CroftKP, RobsonF, CouplandG, DoonanJ, LaufsP, TomsettAB, 2001Characterization of the ethanol-inducible alc gene-expression system in Arabidopsis thaliana.
28: 225–23511722766SheldonCC, BurnJE, PerezPP, MetzgerJ, EdwardsJA, PeacockWJ, DennisES1999The FLF MADS box gene: A repressor of flowering in Arabidopsis regulated by vernalization and methylation.
11: 445–45810072403ShenY, YueF, McClearyDF, YeZ, EdsallL, KuanS, WagnerU, DixonJ, LeeL, LobanenkovVV, 2012A map of the cis-regulatory sequences in the mouse genome.
488: 116–12022763441SungS, AmasinoRM2004Vernalization in Arabidopsis thaliana is mediated by the PHD finger protein VIN3.
427: 159–16414712276SungS, HeY, EshooTW, TamadaY, JohnsonL, NakahigashiK, GotoK, JacobsenSE, AmasinoRM2006Epigenetic maintenance of the vernalized state in Arabidopsis thaliana requires LIKE HETEROCHROMATIN PROTEIN 1.
38: 706–71016682972SwiezewskiS, LiuF, MagusinA, DeanC2009Cold-induced silencing by long antisense transcripts of an Arabidopsis Polycomb target.
462: 799–80220010688TowbinBD, Gonzalez-AguileraC, SackR, GaidatzisD, KalckV, MeisterP, AskjaerP, GasserSM2012Step-wise methylation of histone H3K9 positions heterochromatin at the nuclear periphery.
150: 934–94722939621WoodCC, RobertsonM, TannerG, PeacockWJ, DennisES, HelliwellCA2006The Arabidopsis thaliana vernalization response requires a polycomb-like protein complex that also includes VERNALIZATION INSENSITIVE 3.
103: 14631–1463616983073oai:pubmedcentral.nih.gov:37782392014-03-01genesdevpmc-openGenes DevGenes DevGADGenes & Development0890-93691549-5477Cold Spring Harbor Laboratory PressPMC3778239PMC377823937782392401350024013500871166010.1101/gad.226019.113Research CommunicationA conserved ncRNA-binding protein recruits silencing factors to heterochromatin through an RNAi-independent mechanismMarina et al.Assembly of heterochromatin by Seb1/Nrd1MarinaDiana B.1ShankarSmita1NatarajanPrashanthiFinnKenneth J.MadhaniHiten D.2Department of Biochemistry and Biophysics, University of California at San Francisco, San Francisco, California 94158, USA
Long noncoding RNAs (lncRNAs) can trigger repressive chromatin, but how they recruit silencing factors remains unclear. Marina et al. find that Seb1, the S. pombe ortholog of the RNA-binding protein Nrd1, is associated with pericentromeric lncRNAs. Mutation of dcr1+ (Dicer) or seb1+ results in equivalent partial reductions of pericentromeric H3K9me levels, while a double mutation eliminates this mark. Seb1 functions independently of RNAi by recruiting the NuRD-related chromatin-modifying complex SHREC.
Long noncoding RNAs (lncRNAs) can trigger repressive chromatin, but how they recruit silencing factors remains unclear. In Schizosaccharomyces pombe, heterochromatin assembly on transcribed noncoding pericentromeric repeats requires both RNAi and RNAi-independent mechanisms. In Saccharomyces cerevisiae, which lacks a repressive chromatin mark (H3K9me [methylated Lys9 on histone H3]), unstable ncRNAs are recognized by the RNA-binding protein Nrd1. We show that the S. pombe ortholog Seb1 is associated with pericentromeric lncRNAs. Individual mutation of dcr1+ (Dicer) or seb1+ results in equivalent partial reductions of pericentromeric H3K9me levels, but a double mutation eliminates this mark. Seb1 functions independently of RNAi by recruiting the NuRD (nucleosome remodeling and deacetylase)-related chromatin-modifying complex SHREC (Snf2–HDAC [histone deacetylase] repressor complex).
Seb1Nrd1RNAiheterochromatinSHRECsilencing
A major unsolved question in chromatin biology is how long intergenic noncoding RNAs (lincRNAs) trigger the formation of repressed chromatin. A large number of mammalian lincRNAs have been identified by systematic studies (Guttman et al. 2009). Many of these ncRNAs associate with chromatin-modifying complexes (Khalil et al. 2009). Multiple models have been proposed for how these ncRNAs are recognized and recruit chromatin-modifying factors, but little is understood mechanistically (Guttman and Rinn 2012).
In Schizosaccharomyces pombe, pericentromeric heterochromatin assembly is promoted by transcription of the dg and dh repeat sequences by RNA polymerase II (Pol II) (Djupedal et al. 2005; Kato et al. 2005). The corresponding long ncRNAs (lncRNAs) are converted into dsRNAs and processed into siRNAs by the combined action of RNA-directed RNA polymerase complex (RDRC) and Dicer (Dcr1) (Verdel et al. 2009; Lejeune and Allshire 2011). siRNAs produced by Dicer are bound by Argonaute (Ago1), a component of the RNA-induced transcriptional silencing (RITS) complex, and together they promote both degradation of pericentromeric ncRNAs and transcriptional silencing via repressive histone methylation (Verdel et al. 2004). These complexes in turn recruit the Clr4 methyltransferase complex (CLRC), which methylates Lys9 on histone H3 (H3K9me) (Nakayama et al. 2001; Zhang et al. 2008). The methyl mark serves as a binding platform for the repressive HP1 proteins Swi6 and Chp2 (Thon and Verhein-Hansen 2000; Bannister et al. 2001; Fischer et al. 2009). Both proteins promote the recruitment of SHREC (Snf2–HDAC [histone deacetylase] repressor complex) to pericentromeric heterochromatin (Sugiyama et al. 2007; Sadaie et al. 2008). Moreover, Chp2 has been found to associate with SHREC to form the SHREC2 complex (SHREC complex associated with Chp2) (Motamedi et al. 2008). The core of SHREC consists of silencing factors Clr1 and Clr2, the HDAC Clr3, and the putative chromatin-remodeling enzyme Mit1 (Sugiyama et al. 2007). SHREC and SHREC2 resemble the mammalian nucleosome remodeling and deacetylase (NuRD) complex (Sugiyama et al. 2007; Motamedi et al. 2008). Previous studies revealed that deletion of clr3+ reduces the levels of pericentromeric H3K9me2 in cells lacking RNAi (Yamada et al. 2005; Reyes-Turcu et al. 2011), indicating that Clr3/SHREC can act independently of RNAi in pericentromeric heterochromatin assembly.
In addition to processing by the RNAi machinery, the pericentromeric lncRNAs are also recognized by several RNA quality control factors such as Mlo3 and Cid14 (Reyes-Turcu et al. 2011). Interestingly, mutations in these factors have been shown to suppress the silencing defect of RNAi mutants at pericentromeric repeats (Reyes-Turcu et al. 2011). This led to the suggestion that the ncRNAs might also act in the RNAi-independent pathway of heterochromatin assembly (Reyes-Turcu et al. 2011). Here we identify a conserved ncRNA-binding protein, Seb1/Nrd1, which binds pericentromeric ncRNAs and is required for H3K9me in cells deficient for RNAi. We demonstrate that this ncRNA-binding protein functions in the RNAi-independent pathway by recruiting the activities of SHREC.
Results and Discussion
To investigate how ncRNAs lead to the formation of repressed chromatin, we identified a candidate ncRNA recognition factor in S. pombe based on prior studies of ncRNAs in Saccharomyces cerevisiae. There, a class of unstable ncRNAs called cryptic unstable transcripts (CUTs) is recognized by an RNA-binding protein called Nrd1 (Arigo et al. 2006; Thiebaut et al. 2006). Nrd1 globally associates with Pol II via a C-terminal domain-interacting domain (CID) and recognizes a specific RNA oligonucleotide through its RNA recognition motif (RRM) domain (Steinmetz and Brow 1996, 1998; Conrad et al. 2000; Carroll et al. 2004; Meinhart and Cramer 2004; Vasiljeva et al. 2008a). Nrd1 also plays a role in recognizing precursors to stable ncRNAs such as nucleolar RNAs (snoRNAs) and small nuclear RNAs (snRNAs) to promote their 3′ end formation (Steinmetz et al. 2001; Kim et al. 2006). Although the S. cerevisiae lineage lost both H3K9me and RNAi during its evolution, we hypothesized that the role of Nrd1 in ncRNA recognition might be conserved. To test this, we tagged the S. pombe Nrd1 ortholog Seb1 (Mitsuzawa et al. 2003) and used cross-linking and RNA immunoprecipitation (RIP) to assess its association with dg and dh ncRNAs as well as the snoRNA snR30. Seb1 displays a strong association with these ncRNAs but not with the act1+ RNA (Fig. 1A). As expected, Hrr1, an RNAi factor, also associates with dg and dh ncRNAs but not with snR30 or act1+ RNA (Supplemental Fig. S1A). The association of Seb1 with dg and dh transcripts is maintained in a clr4Δ mutant that lacks H3K9me, indicating that this conserved heterochromatic methyl mark is not required for the association (Supplemental Fig. S1B). This is in contrast to RNAi factors that require H3K9me to display a RIP signal on the dg and dh transcripts (Rougemaille et al. 2012).
The seb1-1 mutant is defective in heterochromatic silencing at pericentromeric repeats. (A) RIP experiments measuring the enrichment of Seb1-Flag at dg, dh, and act1+ transcripts and snR30 snoRNA in the wild-type strain. (B, top) Schematic of the ectopic heterochromatic silencing reporter construct used in the allele screen. (Bottom) Silencing assays of seb1+ mutations in the ectopic heterochromatic silencing reporter strain background. Cells were plated on nonselective rich YS medium (N/S) and YS medium with 5-FOA (5-FOA). (C, top) Schematic of centromere 1 with the ura4+ reporter gene inserted in the innermost repeat (imr) region. (Bottom) Silencing assay of seb1-1 mutation in the pericentromeric ura4+ reporter strain background. (D,E) RT-qPCR analysis of ura4+ transcript levels (normalized to act1+ transcript levels) (D) and ChIP analysis of H3K9me2 levels at the ura4+ locus (normalized to H3K9me2 levels at the act1+ locus) (E) in the wild-type strain, the seb1-1 mutant, and the strain with three amino acid mutations in seb1+.
We next isolated a mutation in seb1+ that was defective in heterochromatic silencing. Because seb1+, like NRD1, is an essential gene (Mitsuzawa et al. 2003), we mutagenized a gene targeting construct for the endogenous seb1+ locus (Supplemental Fig. S2) and transformed this library into a strain harboring an ectopic heterochromatic silencing reporter system that we had developed for other studies. This system involves the insertion of a 2811-base-pair (bp) fragment of dh repeat downstream from the endogenous ura4+ gene such that the transcription of this fragment is driven by the adh1+ promoter (Supplemental Fig. S3A). This fragment (“fragment 1”) was identified as a highly potent inducer of ura4+ silencing in a systematic study of the activities of dg and dh fragments (data not shown). Silencing by fragment 1 requires functional Clr4 and RNAi as well as the adh1+ promoter; the latter observation indicates that transcription of fragment 1 is required for silencing (Supplemental Fig. S3B). Silencing by fragment 1 also causes a decrease of ura4+ transcript level (Supplemental Fig. S3C) and an increase of H3K9me2 at the ura4+ locus (Supplemental Fig. S3D). Screening of ∼10,000 colonies produced by transformation of the seb1+ mutant library yielded a single mutant that displays a defect in growth on 5-fluoroorotic acid (5-FOA) medium, which selects for strains with a silenced ura4+ gene (Supplemental Fig. S2). This allele, seb1-1, has seven nucleotide substitutions in the seb1+ coding sequence, three of which change the amino acid sequence. Replacement of seb1+ with the seb1-1 allele in the parental strain recapitulates the 5-FOA phenotype (Fig. 1B). This mutation also causes a silencing defect at endogenous heterochromatin: A strain harboring a ura4+ reporter gene inserted into the innermost repeat (imr) region of centromere 1 displays reduced growth on 5-FOA when harboring the seb1-1 allele (Fig. 1C). In the ectopic heterochromatic silencing reporter strain, the seb1-1 mutation causes an accumulation of the ura4+ transcript (Fig. 1D) and a strong defect in H3K9me2 at the ura4+ gene (Fig. 1E), supporting the growth defect observed using the 5-FOA assay.
We constructed all combinations of the three amino acid changes present in the seb1-1 allele (G76S, R442G, and I524V) and used them to replace the wild-type seb1+ sequence in the reporter strain. We found that all three mutations are required to produce a silencing defect on 5-FOA medium (Fig. 1B). However, the mutant with the three amino acid changes (triple mutant) has a milder silencing defect on 5-FOA when compared with the original seb1-1 mutant. The silencing defect of the triple mutant is only obvious at 2 d of growth on 5-FOA, indicating that the silent mutations also contribute to the phenotype (Fig. 1B). Moreover, the triple mutant displays an intermediate defect in H3K9me2 at the ura4+ locus when compared with the seb1-1 allele (Fig. 1E). The triple mutant displays accumulation of ura4+ transcript nearly comparable with that of the seb1-1 mutant (Fig. 1D), likely reflecting distinct sensitivities and thresholds to gene function of the 5-FOA, RNA, and H3K9me2 assays. As each of the four silent mutations (A45G, T132A, T1194C, and T1260A) changes the wild-type codon to a rarer synonymous codon (Forsburg 1994), they could impact protein expression. Indeed, the level of Seb1 protein is lower in the seb1-1 mutant when compared with its level in the wild-type seb1+ strain (Supplemental Fig. S4). Since the seb1-1 allele has a more robust silencing defect compared with the triple mutant, we used the seb1-1 allele in our further analyses.
We examined the effect of seb1-1 mutation on pericentromeric siRNA production using Northern hybridization and discovered that, unlike the clr4Δ mutant, seb1-1 displays normal levels of pericentromeric siRNA accumulation (Fig. 2A). Consistent with the lack of a defect in siRNA production, the seb1-1 mutation does not cause an accumulation of dg transcripts and only causes a slight increase in the level of dh transcripts (Fig. 2B). In contrast, a catalytically dead Dcr1 mutant (dcr1-R1R2) (Colmenares et al. 2007) displays a dramatic increase in dg and dh transcript levels (Fig. 2B). We serendipitously discovered that seb1-1 confers a temperature-sensitive phenotype at 37°C (Supplemental Fig. S5A). As with the silencing phenotype, all seven mutations of seb1+ are required to produce the temperature sensitivity (Supplemental Fig. S5A). Northern hybridization demonstrated no defect in siRNA levels in the seb1-1 mutant, even at the nonpermissive temperature (Supplemental Fig. S5B). Taken together, these data argue that the silencing defect of the seb1-1 allele cannot simply be explained by a defect in RNAi.
The seb1-1 mutation does not affect RNAi. (A) Riboprobe siRNA Northern blot detecting dh siRNAs in the wild-type, clr4Δ, and seb1-1 strains. snoRNA69 was used as a loading control. (B) RT-qPCR analysis of dg and dh transcript levels (normalized to act1+ transcript levels) in wild-type, seb1-1, and dcr1-R1R2 strains. (C) ChIP analysis of H3K9me2 levels at dg and dh repeats (normalized to H3K9me2 levels at the act1+ locus) in a wild-type strain, seb1-1 and dcr1-R1R2 single mutants, the dcr1-R1R2 seb1-1 double mutant, and the clr4Δ mutant. (D) ChIP analysis of H3K9me2 levels at centromere 1 (normalized to H3K9me2 levels at the act1+ locus) in wild-type strain, seb1-1 and dcr1-R1R2 single mutants, and the dcr1-R1R2 seb1-1 double mutant.
To test whether Seb1 functions in the RNAi-independent pathway of heterochromatin formation, we measured H3K9me2 levels at the dg and dh repeats in the single and double combinations of seb1-1 and dcr1-R1R2 alleles. Strikingly, while the levels of H3K9me2 are reduced by only threefold to fivefold in the single mutants, H3K9me2 is virtually abolished in the dcr1-R1R2 seb1-1 double mutant to background levels comparable with those measured in the clr4Δ mutant (Fig. 2C). We observed similar results across the entire pericentromeric region in chromatin immunoprecipitation and quantitative PCR (ChIP-qPCR) experiments using tiled pairs of primers (Fig. 2D; Braun et al. 2011). Moreover, we found that the seb1-1 mutation does not affect Chp1/RITS recruitment to dg and dh repeats (Supplemental Fig. S6). Together, these data demonstrate that Seb1 acts independently of RNAi to promote H3K9me at pericentromeric heterochromatin. To rule out the trivial possibility that the dcr1-R1R2 seb1-1 double mutation reduces the transcription of components of the CLRC, thereby affecting H3K9me2 indirectly, we performed RT-qPCR analyses and found no reduction in them in the mutant (Supplemental Fig. S7).
Although classical heterochromatin mediated by H3K9me was ancestrally lost during the evolution of S. cerevisiae, Nrd1 has been reported to play a role in promoting a different type of silencing that occurs in the recombinant DNA (rDNA) repeats of budding yeast (Vasiljeva et al. 2008b). In this organism, cryptic unstable Pol II transcripts of unknown function are produced from rDNA, which is predominantly transcribed by Pol I. These Pol II transcripts are terminated and targeted for degradation by Nrd1 (Vasiljeva et al. 2008b). In Nrd1 mutants, the defect in termination and turnover leads to dramatic increases in the level of rDNA-derived Pol II transcripts and the accumulation of longer transcripts (Vasiljeva et al. 2008b). This transcriptional readthrough is associated with increased histone acetylation, reduced nucleosome occupancy, activation of inserted Pol II reporter genes, and increased recombination between rDNA repeats (Vasiljeva et al. 2008b). We tested whether Seb1 might act by such a mechanism in S. pombe pericentromeric regions by determining whether the seb1-1 mutation causes an increase in pericentromeric transcript size and levels in cells mutated for Dcr1. Northern hybridization using a riboprobe complementary to the region of the dh repeats encoded by fragment 1 revealed heterogeneously sized transcripts produced from this region that accumulate in the dcr1-R1R2 mutant (Supplemental Fig. S8A). This pattern is consistent with previous reports (Zaratiegui et al. 2011). No obvious increase in transcript size was apparent, and there was no significant increase in transcript level in the seb1-1 dcr1-R1R2 double mutant relative to the dcr1-R1R2 single mutant (Supplemental Fig. S8A). Careful quantification using RT-qPCR also yielded no increase in the abundance of dh transcript produced by the seb1-1 mutation in a dcr1-R1R2 genetic background (Supplemental Fig. S8B). These data indicate that mechanisms observed previously in S. cerevisiae cannot easily explain our observations in S. pombe.
Since it has been previously reported that the clr3Δ mutant, like the seb1-1 mutant, further reduces H3K9me2 in the dcr1Δ mutant (Yamada et al. 2005; Reyes-Turcu et al. 2011), we tested whether Seb1 and SHREC function in the same pathway. We quantitatively measured the levels of H3K9me2 at dg and dh repeats in the dcr1-R1R2 seb1-1 and dcr1-R1R2 clr3Δ double mutants. Consistent with published results (Yamada et al. 2005; Reyes-Turcu et al. 2011), we observed that while H3K9me2 levels at the dg and dh repeats are modestly reduced in the dcr1-R1R2 single mutant, these levels are virtually eliminated in the dcr1-R1R2 clr3Δ double mutant, a phenotype strikingly similar to that of the dcr1-R1R2 seb1-1 double mutant (Fig. 3A). Furthermore, the elimination of H3K9me2 at dg and dh repeats was also observed in strains that have the dcr1-R1R2 mutation combined with a deletion mutation that eliminates the Clr1, Chp2, or Mit1 subunits of SHREC (Fig. 3A). Finally, a catalytically dead point mutation in the Clr3 HDAC or the Mit1 ATPase domain also eliminates H3K9me in dcr1-R1R2 cells (Fig. 3A). These data demonstrate that all known activities of SHREC are required for the RNAi-independent pathway that promotes H3K9me at pericentromeric repeats.
Seb1 functions in the same pathway as SHREC to promote H3K9me. (A) ChIP analysis of H3K9me2 levels at dg and dh repeats (normalized to H3K9me2 levels at the act1+ locus) in wild-type, dcr1-R1R2, dcr1-R1R2 seb1-1, dcr1-R1R2 clr3Δ, dcr1-R1R2 clr3D232N, dcr1-R1R2 mit1Δ, dcr1-R1R2 mit1K587A, dcr1-R1R2 clr1Δ, dcr1-R1R2 chp2Δ, and clr4Δ strains. (B) ChIP analysis of H3K9me2 levels at dg and dh repeats (normalized to H3K9me2 levels at the act1+ locus) in the wild-type strain, the seb1-1 and clr3Δ single mutants, and the seb1-1 clr3Δ double mutant. (C) ChIP analysis of H3K9me2 levels at dg and dh repeats (normalized to H3K9me2 levels at the act1+ locus) in the wild-type strain, the seb1-1 and clr1Δ single mutants, and the seb1-1 clr1Δ double mutant. (D,E) ChIP analyses of Pol II (D) and H3K14ac (E) levels at dg and dh repeats (normalized to the levels at the act1+ locus) in the wild-type strain and seb1-1 and clr3Δ single mutants.
To further test whether Seb1 functions in the same pathway as SHREC in promoting H3K9me, we compared H3K9me2 levels at dg and dh repeats of the seb1-1 clr3Δ double mutant with those of the corresponding single mutants. Significantly, the double and single mutants display very similar levels of H3K9me2, ∼20%–30% of wild-type levels (Fig. 3B). Similar results were obtained when comparing H3K9me2 levels at dg and dh repeats of the seb1-1 clr1Δ double mutant with the levels of the corresponding single mutants (Fig. 3C). These data provide strong genetic support that Seb1 and SHREC proteins function in the same pathway in promoting H3K9me at pericentromeric repeats. Moreover, we found that the seb1-1 mutation increases Pol II occupancy and H3K14 acetylation levels at dg and dh repeats (Figs. 3D,E), similar to that reported previously for the clr3Δ mutation (Sugiyama et al. 2007; Motamedi et al. 2008). To rule out the possibility that seb1-1 mutation indirectly affects heterochromatin assembly by decreasing the levels of mRNAs that encode the subunits of SHREC, we measured their levels in the wild-type and seb1-1 strains. We found that the seb1-1 mutation does not reduce the levels of the mRNAs encoding any of the subunits of SHREC (Supplemental Fig. S9), further arguing for a direct role of Seb1 in heterochromatin formation.
Because Seb1 and SHREC function genetically in the same pathway, we hypothesized that Seb1 physically interacts with SHREC to recruit it to pericentromeric repeats. An observation that supports this hypothesis is that the purification of the Clr2 subunit of SHREC was reported to have yielded two peptides from Seb1 (Supplemental Table S1; Motamedi et al. 2008). However, Seb1 peptide coverage was low and was not obtained in other purifications reported. To test for a physical interaction between Seb1 and SHREC using a more sensitive assay, we performed coimmunoprecipitation (co-IP) immunoblotting experiments. We found that Flag-tagged Seb1 coimmunoprecipitated with Clr3-myc (Fig. 4A) but not with a control protein (Cdc2 detected by anti-PSTAIRE), indicating that these two proteins physically interact in vivo. This interaction is not bridged by RNA, since we could still detect the interaction in the presence of RNase A (Fig. 4B). Seb1 also coimmunoprecipitated with other components of SHREC, including Clr1 and Mit1 (Fig. 4C,D). These data provide strong evidence that Seb1 and SHREC not only function in the same pathway, but also physically interact in vivo.
Seb1 physically associates with SHREC in vivo. (A) Co-IP of Seb1 with Clr3. Strains expressing endogenously tagged Seb1-CBP-2XFlag, Clr3-4myc, or both were subjected to anti-Flag immunoprecipitation. The whole-cell extract (WCE) and immunoprecipitated (IP) samples were detected by anti-myc (top), anti-Flag (middle), and anti-PSTAIRE (bottom) immunoblots; the latter serves as a loading and specificity control. (B) Co-IP of Seb1 with Clr3 in the presence of RNase A. The two bottom panels are images of agarose gels of RT-qPCR experiments to detect dh transcripts in the presence (+RT) or absence (−RT) of reverse transcriptase. (C,D) Co-IP of Seb1 with Clr1 and Mit1. Strains expressing endogenous levels of Seb1-CBP-2XFlag, Clr1-13myc, or both (C) and Seb1-CBP-2XFlag, Mit1-13myc, or both (D) were subjected to anti-Flag immunoprecipitation. The whole-cell extract (WCE) and immunoprecipitated (IP) samples were detected by anti-myc (top) and anti-Flag (bottom) immunoblots.
To gain further insight into how Seb1 functions to promote H3K9me, we tested whether Seb1 acts upstream of SHREC, as might be expected for a recruitment factor. We performed ChIP experiments to measure the enrichment of SHREC in strains harboring an epitope-tagged version of the proteins. We found that every subunit of SHREC tested (Clr3, Clr1, Mit1, and Chp2) is enriched at both dg and dh repeats. However, in the seb1-1 mutant, this enrichment is abolished or strongly reduced (Figs. 5A–D). In contrast, RIP experiments demonstrate that Seb1 still associates with dg and dh transcripts in cells lacking SHREC, indicating that Seb1 recruitment to these ncRNAs is not downstream from SHREC (Fig. 5E). Interestingly, we observed a reproducible increase in Seb1 association with dg and dh ncRNAs in the clr3Δ mutant but not the clr4Δ mutant, suggesting negative regulation of Seb1 recruitment by Clr3 (Fig. 5E; Supplemental Fig. S1B). Together, these data support our hypothesis that Seb1 acts by recruiting SHREC to pericentromeric heterochromatin.
Seb1 recruits SHREC to pericentromeric heterochromatin. (A–D) ChIP analyses of Clr3-myc, Clr1-myc, Mit1-myc, and Chp2-myc levels at dg and dh repeats (normalized to their levels at the act1+ locus) in the wild-type and seb1-1 strains. (E) RIP experiments measuring the enrichment of Seb1-Flag at dg, dh, and act1+ transcripts in the wild-type and clr3Δ strains. (F) Model for the RNAi-independent role of Seb1 in recruiting SHREC to pericentromeric heterochromatin.
In S. pombe, heterochromatin is also found at the silent mating type locus and at subtelomeric regions. These loci contain DNA sequences that are similar to the pericentromeric dg and dh repeats. While Seb1 is also associated with the dg/dh-like transcripts originating from these nonpericentromeric heterochromatic loci (Supplemental Fig. S10), the seb1-1 mutation does not affect SHREC recruitment to these loci (Supplemental Fig. S11). This is not completely unexpected, since there exists additional mechanisms of SHREC recruitment to the silent mating type locus and subtelomeric regions by DNA-binding proteins Atf1/Pcr1 and Taz1, respectively (Yamada et al. 2005; Sugiyama et al. 2007).
Our studies demonstrate an essential role for the ncRNA-binding protein Seb1 in RNAi-independent pericentromeric heterochromatin formation through recruitment of the chromatin-modifying complex SHREC. (Fig. 5F). SHREC may function to promote H3K9me by deacetylating the H3 tail, which is presumably necessary for its methylation by the CLRC. Alternatively, SHREC may have nonhistone substrates whose deacetylation promotes H3K9me. As HP1 proteins also promote the recruitment of SHREC, heterochromatin spread may be promoted by a positive feedback loop involving alternating cycles of histone deacetylation and methylation (Fig. 5F). Consequently, SHREC may be considered both an effector and a trigger of histone methylation.
Candidate recruiters of RNAi to heterochromatic regions in S. pombe include Pol II and the spliceosome (Djupedal et al. 2005; Kato et al. 2005; Bayne et al. 2008), which are essential for viability and operate at a very large number of genomic sites. Presumably, additional factors and signals determine where RNAi-related enzymes are recruited by these multiprotein complexes. Likewise, Seb1 also displays a strong association with nonheterochromatic snR30 snoRNA (Fig. 1A). Interestingly, however, we did not detect recruitment of Clr3 to the snR30 gene (Supplemental Fig. S12), suggesting that Seb1 may play SHREC-independent roles outside of heterochromatin. Our ongoing studies are aimed at defining the molecular cues (including potential RNA ligands) that distinguish Seb1's functions in heterochromatin versus euchromatin.
Materials and methodsStrain construction and growth conditions
Strains were constructed and grown by standard fission yeast methods as described previously (Rougemaille et al. 2012), except that 2 g/L 5-FOA was used in silencing assays.
Immunoprecipitation experiments and RNA analyses
ChIP, RIP, co-IP, and RT-qPCR experiments and Northern analyses were performed using standard molecular biology techniques as described in detail in the Supplemental Material.
Acknowledgments
We thank Danesh Moazed and Karl Ekwall for strains. We are grateful to Ratika Krishnamurty for sharing results of her experiments that were insightful during the revision process. We thank Jeff Corden, Sigurd Braun, Geeta Narlikar, Kristin Patrick, and members of the Madhani laboratory for critical comments on the manuscript. This work was supported by a grant from the National Institutes of Health (R01GM071801). D.B.M. was an American Heart Association predoctoral fellow. S.S. was a post-doctoral fellow of the Leukemia and Lymphoma Society.
Supplemental material is available for this article.
Article is online at http://www.genesdev.org/cgi/doi/10.1101/gad.226019.113.
The p53–Mdm2 feedback loop is thought to be critical for regulating stress-induced p53 activity and levels. Using a novel mouse model, Lozano and colleagues now show that the p53–Mdm2 negative feedback loop is not important in development or longevity but is important in response to DNA damage. Low-dose IR causes a hematopoietic stem cell failure in a Puma-dependent but not a p21-dependent manner. This study suggests that transient disruption of the p53–Mdm2 interaction could be a potential therapeutic strategy for targeting stem cells in hematological malignancies.
The p53–Mdm2 feedback loop is perceived to be critical for regulating stress-induced p53 activity and levels. However, this has never been tested in vivo. Using a genetically engineered mouse with mutated p53 response elements in the Mdm2 P2 promoter, we show that feedback loop-deficient Mdm2P2/P2 mice are viable and aphenotypic and age normally. p53 degradation kinetics after DNA damage in radiosensitive tissues remains similar to wild-type controls. Nonetheless, DNA damage response is elevated in Mdm2P2/P2 mice. Enhanced p53-dependent apoptosis sensitizes hematopoietic stem cells (HSCs), causing drastic myeloablation and lethality. These results suggest that while basal Mdm2 levels are sufficient to regulate p53 in most tissues under homeostatic conditions, the p53–Mdm2 feedback loop is critical for regulating p53 activity and sustaining HSC function after DNA damage. Therefore, transient disruption of p53–Mdm2 interaction could be explored as a potential adjuvant/therapeutic strategy for targeting stem cells in hematological malignancies.
The ubiquitously expressed p53 tumor suppressor is maintained normally in an inactive latent form but functions as the “guardian of the genome” in response to DNA damage (Lane 1992). In response to genotoxic stressors, p53 transactivates target genes involved in cell cycle arrest, senescence, or apoptosis pathways to halt progression of insults into heritable aberrations (Vousden and Lu 2002). A range of inhibitors have been identified that regulate p53 activity under normal and stress conditions. Of these, Mdm2 is the major negative regulator of p53. Genetic deletion of Mdm2 in vivo results in embryonic lethality that is rescued by concomitant deletion of p53 (Jones et al. 1995; Montes de Oca Luna et al. 1995). The prevailing view suggests that Mdm2 inhibits p53 by two different mechanisms. Mdm2 binds and masks the transactivation domain of p53 (Momand et al. 1992; Oliner et al. 1993). Furthermore, Mdm2 is also an E3 ubiquitin ligase that promotes p53 degradation through the 26S proteasome machinery (Haupt et al. 1997; Honda et al. 1997; Kubbutat et al. 1997). Interestingly, Mdm2 itself is a transcriptional target of p53, thus establishing a negative feedback loop. After DNA damage, stabilized/activated p53 binds to the P2 promoter of Mdm2 and promotes its transcription (Barak et al. 1993; Wu et al. 1993). Mdm2 in turn inhibits p53 via one of the two mechanisms described above.
A wealth of correlative evidence suggests that the p53–Mdm2 autoregulatory loop functions as the principal mode of p53 regulation under normal and DNA damage conditions (Haupt et al. 1997; Saucedo et al. 1998; Mendrysa and Perry 2000; Marine et al. 2006). After DNA damage, p53 levels increase, correlating with enhanced p53 binding at the P2-Mdm2 promoter and a subsequent increase in Mdm2 levels (Barak et al. 1993; Wu et al. 1993; Saucedo et al. 1998). This acute response is soon followed by dampening of p53 back to baseline levels. As increased p53 levels are toxic for cell viability, it is generally believed that Mdm2 transactivated by p53 from the P2 promoter is central for down-modulation of p53. Interestingly, this cytoprotective feature of the p53–Mdm2 feedback loop is considered a major impediment in exploiting the potential of p53 reactivation as a therapeutic strategy in tumors with wild-type p53. However, in the absence of an in vivo model, these hypotheses could not be directly evaluated.
To investigate the biological significance of the dual Mdm2 promoters and the p53–Mdm2 autoregulatory loop in vivo, we generated a knock-in mouse model with a defective p53–Mdm2 autoregulatory loop and analyzed the effects of the feedback deficiency during development and under normal and DNA damage conditions.
ResultsGeneration of Mdm2P2/P2 mice
To examine the in vivo significance of the p53–Mdm2 autoregulatory loop, we generated a knock-in mouse by mutating the critical C and G nucleotides in the two p53 response elements of the P2-Mdm2 promoter (Fig. 1A,B). This in vivo approach allowed us to specifically abrogate p53-mediated up-regulation of Mdm2 while maintaining the normal stoichiometry and functionality of other p53 pathway components. The abrogation of P2 promoter function was verified by in vitro luciferase reporter assay prior to cloning of the mutant promoter fragment into the targeting vector (data not shown). The targeting construct (Fig. 1A) with a mutant Mdm2 P2 promoter was electroporated into TC1 mouse embryonic stem (ES) cells. Correctly targeted ES clones were identified by Southern blotting using 5′ and 3′ external probes (Fig. 1A) and injected into C57BL/6 blastocysts to generate Mdm2+/P2 chimeras. Male chimeras (>80%) were backcrossed to C57BL/6 mice to secure germline transmission of the mutant allele. The Neomycin selection cassette was subsequently deleted by crossing with Zp3-Cre deleter mice (Lewandoski et al. 1997). A PCR-based genotyping strategy on genomic DNA isolated from tail biopsies was used to follow the transmission of the mutant allele. Mice were backcrossed for a total of four generations to >90% C57BL/6 background for this study.
Generation of Mdm2P2/P2 knock-in allele. (A) Schematic representation of the targeting vector carrying mutations of the p53 response elements in the P2 promoter of the Mdm2 gene. Filled black boxes represent numbered exons, while the red ovals depict the mutations. The targeting construct contained Hsv-Tk and loxP-flanked (filled triangles) PGK-neo cassettes for negative and positive selection, respectively. Also shown are the 5′ and 3′ external probes used to genotype the ES cell lines. The side panel shows Southern blot analysis of the SpeI-digested genomic DNA from ES cell clones. (B) Partial P2 promoter sequence from the wild-type and mutant Mdm2 alleles. Asterisks denote the mutated nucleotides. (C) ChIP assays with p53 antibody to examine p53 binding at the Mdm2, p21, and Puma promoters 4 h after 6 Gy IR. n = 3, ±SEM. Acetylcholine receptor (AchR) promoter was used as negative control for the assay.
Mdm2P2/P2 mice are born in a normal Mendelian ratio
We intercrossed heterozygous Mdm2+/P2 mice to generate Mdm2P2/P2 homozygous mice. Surprisingly, Mdm2P2/P2 mice were born at an appropriate Mendelian ratio with no phenotypic aberrations (Supplemental Fig. 1). We sequenced the genomic DNA from an Mdm2P2/P2 homozygous mouse and confirmed the mutations in the germline (Fig. 1B). We next generated Mdm2P2/− mice with further reduced Mdm2 levels. Again, both Mdm2P2/P2 and Mdm2P2/− mice survived to adulthood lacking any distinctive phenotype. These results demonstrate that reduction in the P2-Mdm2 (Mdm2 expressed from the P2 promoter) level does not lead to lethal activation of p53. Thus, the autoregulatory loop is dispensable for normal development.
Mutations abrogate p53 binding at the P2-Mdm2 promoter
We next tested the specificity of P2-Mdm2 promoter mutations by performing in vivo chromatin immunoprecipitation (ChIP) assays. We isolated spleens from Mdm2P2/P2 and Mdm2+/+ mice after irradiation. Unirradiated Mdm2+/+ and Mdm2P2/P2 mouse spleens were obtained as controls at the same time. In vivo ChIP with a p53 antibody followed by real-time PCR analyses confirmed abrogation of p53 binding at the mutant Mdm2 promoter (Fig. 1C), while p53 still bound to the promoters of canonical targets p21 and Puma. Notably, p53 binding to the p21 promoter was significantly enhanced in irradiated Mdm2P2/P2 mouse spleens compared with Mdm2+/+ spleens for unknown reasons.
Degradation profile of p53 after ionizing radiation (IR) is not altered in Mdm2P2/P2 mice
According to current dogma, degradation of the accumulated p53 after DNA damage is attributed to its ability to transactivate Mdm2, which encodes the major E3 ubiquitin ligase for p53 (Wu et al. 1993; Barak et al. 1994; Zauberman et al. 1995; Haupt et al. 1997; Honda et al. 1997; Kubbutat et al. 1997). Therefore, we first investigated the role of Mdm2 in p53 degradation after IR in feedback-deficient Mdm2P2/P2 mice. We isolated spleens from irradiated Mdm2+/+ and Mdm2P2/P2 mice at different time points. Immunoblotting of protein lysate revealed the anticipated post-IR induction of p53 in both genotypes (Fig. 2A). A slight increase in p53 induction at the 4-h post-IR time point in Mdm2P2/P2 spleens in comparison with Mdm2+/+ spleens was noticeable (Fig. 2A). Nonetheless, the pattern of p53 degradation in Mdm2P2/P2 spleens remained similar to Mdm2+/+ spleens. In both genotypes, p53 was stabilized 2 h after IR, peaked by 4 h, and subsequently returned to baseline levels by 8 h. As expected, Mdm2 induction was visible only in Mdm2+/+ spleens in response to IR. To evaluate the role of the p53–Mdm2 feedback loop in p53 stability in other tissues, we further examined post-IR induction and degradation profiles of p53 in the skin, thymus, lungs, and kidneys of Mdm2+/+ and Mdm2P2/P2 mice (Fig. 2B). Again, a similar pattern of p53 induction and degradation was observed in the tissues of mice from either genotype (Fig. 2B). Of note, a slight delay in p53 decay was observed only in the Mdm2P2/P2 mouse skin. As expected, no induction of p53 was observed in the liver.
The Mdm2 generated from the P2 promoter is not involved in DNA damage-induced p53 degradation. (A) Western blot analysis for post-IR p53 levels in 6 Gy irradiated Mdm2+/+ and Mdm2P2/P2 mouse spleens at different time points. (−) Negative controls for p53/Mdm2 expression; (+) positive controls for p53/Mdm2 expression; (arrowhead) Mdm2 band; (*) nonspecific band. The bottom panels show dynamics of Mdm2, p21, and S18-p53 induction. (B) Western blot analysis for p53 levels in 6 Gy irradiated Mdm2+/+ and Mdm2P2/P2 mouse skin, lung, thymus, kidney, and liver tissues. (C) Western blot analysis for p53 kinetics in Mdm2P2/− and Mdm2P2/P2 mouse spleens. (D) Time course for p53 induction and degradation in MEF cells exposed to 50 J/m2 UV. Vinculin was used as loading control in these experiments. Blots are representative of three independent biological replicates. Numbers at the bottom denote p53 fold induction normalized to vinculin controls and relative to untreated Mdm2+/+ controls.
We also examined the impact of further reduced Mdm2 levels on p53 levels by analyzing p53 degradation in spleens from irradiated Mdm2P2/− mice. p53 induction was comparatively higher in Mdm2P2/− mouse spleens after IR due to minimal Mdm2 expression from a single allele containing only the P1 promoter (Fig. 2C). Nonetheless, the degradation profiles of p53 in Mdm2P2/− spleens were similar to that of Mdm2P2/P2 spleens (Fig. 2C). A dramatic decrease of p53 levels was observed 6 h after IR in both genotypes. However, p53 levels were not quite back to baseline at the 8-h time point in Mdm2P2/− spleens. Of note, while loss of Mdm2 stabilizes p53 in vivo (Francoz et al. 2006; Ringshausen et al. 2006; Xiong et al. 2006; Terzian et al. 2008), we did not observe any overt p53 stabilization in Mdm2P2/P2 or Mdm2P2/− tissues in the absence of DNA damage (Fig. 2), indicating that basal levels of Mdm2 expressed from the P1 promoter are sufficient to maintain normal p53 levels. The viability of these mice further supports this conclusion.
p53 degradation profile after ultraviolet radiation (UV) damage is not altered in Mdm2P2/P2 mice
We also tested whether another type of DNA damage could induce the feedback loop and alter p53 degradation. We used UV, which creates pyrimidine dimers to activate p53 (Saucedo et al. 1998). Mouse embryonic fibroblasts (MEFs) from Mdm2+/+ and Mdm2P2/P2 mice were exposed to 50 J/m2 UV and harvested at different time points. Protein lysates were analyzed by immunoblotting (Fig. 2D). Notably, p53 was stabilized in both sets after UV exposure, while an enhanced Mdm2 induction was restricted to Mdm2+/+ lysates. Moreover, the overall pattern of p53 induction and down-regulation remained similar in both MEF cell lines (Fig. 2D). These data emphasize the importance of basal Mdm2 levels in regulating p53 levels in response to DNA damage.
Tissues from Mdm2P2/P2 mice show higher levels of p53 activity after DNA damage
In vivo, loss of Mdm2 alone results in spontaneous p53 activation (Jones et al. 1995; Montes de Oca Luna et al. 1995; Mendrysa et al. 2003; Francoz et al. 2006; Ringshausen et al. 2006; Xiong et al. 2006). Therefore, we next examined whether p53 activity was altered in Mdm2P2/P2 mice. We isolated total RNA from unirradiated and irradiated Mdm2+/+ and Mdm2P2/P2 thymi and performed RT-qPCR analyses for p53 targets. A significant increase (P < 0.01) in p21 and Puma transcript levels was observed in Mdm2P2/P2 mice compared with Mdm2+/+ mice (Fig. 3A). Analogous analyses of p53 targets in RNA from spleens also showed modest increase in transcript levels of p53 target genes CyclinG1, Noxa, Puma, p21, and Bax in Mdm2P2/P2 mice (Fig. 3B). Notably, p21 levels were significantly higher in Mdm2P2/P2 mice, in agreement with the ChIP data (Fig. 1C) and Western blot analysis (Fig. 2A). No differences in basal transcript levels of these p53 target genes were observed between unirradiated Mdm2+/+ and Mdm2P2/P2 tissues (Fig. 3A,B). Additionally, no increase in Mdm2 transcripts was observed in either tissue from irradiated Mdm2P2/P2 mice, in conformity with the ChIP data (Fig. 3A,B). These results corroborate that the autoregulatory loop is engaged primarily under stress conditions (Mendrysa and Perry 2000) and also imply that in the absence of exogenous stress stimuli, basal levels of Mdm2 from the P1 promoter are sufficient for regulating p53 activity.
Enhanced p53 activity after DNA damage in specific tissues. (A) RT-qPCR analysis for p53 targets in thymi from Mdm2+/+ and Mdm2P2/P2 mice (n = 3, ±SEM). (B) RT-qPCR analysis of p53 transactivation function in spleens from Mdm2+/+ and Mdm2P2/P2 mice (n = 6, ±SEM) 4 h after 6 Gy IR. (C) Quantification of flow cytometry data for apoptosis in thymocytes from Mdm2+/+ and Mdm2P2/P2 mice (n = 5–6, ±SEM) 4 h after 6 Gy IR. (D) Representative forward/side scatter profiles of irradiated and unirradiated MEFs from Mdm2+/+ and Mdm2P2/P2 mice. (E) Quantification of cells in S phase from D. (F) RT-qPCR for p53 target gene p21 in MEFs. Three separate MEF lines per genotype were used for analyses. In all RT-qPCR experiments, mRNA expression was normalized to Rplp0 levels, and wild-type −IR was set to 1. (G) Cell growth analyses of Mdm2+/+ and Mdm2P2/P2 MEFs. P-values were calculated by unpaired Student's t-test, and a value <0.05 was considered significant. (*) P < 0.05; (**) P < 0.01; (***) P < 0.001.
Enhanced p53 functions in Mdm2P2/P2 mice
p53 functions to maintain genomic integrity by inducing apoptosis, cell cycle arrest, or senescence in damaged cells (Vousden and Lu 2002). To examine the acute impact of DNA damage on p53 function, we performed Annexin-V FITC flow cytometry analyses on thymocytes isolated from irradiated Mdm2+/+ and Mdm2P2/P2 mice. An increase in apoptotic response was observed in Mdm2P2/P2 thymocytes, although this was not statistically significant (P = 0.15) (Fig. 3C; Supplemental Fig. 2). To further examine the effect on cell cycle, we next used MEFs that preferentially undergo p53-dependent cell cycle arrest after DNA damage. We irradiated Mdm2+/+ and Mdm2P2/P2 MEFs and, after confirming p53 up-regulation (Supplemental Figure 3), analyzed them by flow cytometry. We observed a statistically significant decrease (P < 0.001) in the S-phase population of Mdm2P2/P2 MEFs compared with Mdm2+/+ in response to IR (Fig. 3D,E). Irradiated Mdm2P2/P2 MEFs also had significantly higher (P < 0.05) p21 mRNA levels as compared with Mdm2+/+ MEFs (Fig. 3F). Furthermore, as cell culture is itself a stressed system, we analyzed MEF growth using 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) assay (Fig. 3G). Again, in confirmation with high p53 activity, the cell growth rate of Mdm2P2/P2 MEFs was lower than Mdm2+/+ MEFs. These data suggest that a defective autoregulatory loop augments p53-dependent activities after stress.
Mdm2P2/P2 mice are extremely radiosensitive
Changes in Mdm2 levels impact radiation response and survival in mice (Mendrysa et al. 2003, 2006; Ringshausen et al. 2006; Terzian et al. 2007). To specifically investigate the importance of the p53–Mdm2 feedback loop long term, we irradiated mice with a sublethal dose of 6 Gy IR. While 100% of irradiated Mdm2+/+ mice survived 50 d (the duration of the experiment) with no significant pathology, 80% of irradiated Mdm2P2/P2 mutant mice died within 25 d (Fig. 4A). This is similar to the post-IR lethality of Mdm2+/− and Mdm2 hypomorphic mice (Mendrysa et al. 2003; Terzian et al. 2007). However, in contrast to these published studies, Mdm2P2/P2 mice have normal levels of basal Mdm2 from P1 promoter and only lack expression of p53-induced P2-Mdm2.
Absence of feedback loop results in extreme radiosensitivity in Mdm2P2/P2 mice. (A) Kaplan-Meier survival curves for 6 Gy irradiated Mdm2+/+ and Mdm2P2/P2 mice. (B) Kaplan-Meier survival curves for 6 Gy irradiated Mdm2P2/P2, Mdm2P2/−, Mdm2P2/P2Mdm4Δ2/+, Mdm2+/−, and Mdm4Δ2/+ mice. (C) Hematoxylin and eosin-stained BM sections of 6 Gy irradiated Mdm2+/+ and Mdm2P2/P2 mice at different time points (10× magnification). (D) Quantification of surviving CLPs/CMPs and HSCs (LSK) from 6 Gy irradiated and unirradiated Mdm2+/+ and Mdm2P2/P2 mouse BM 8 h after IR. n = 3, ±SEM; P-value was calculated by unpaired Student's t-test, (***) P < 0.001. (E) Peripheral blood leukocyte marker analysis of lethally irradiated recipient wild-type CD45.1 mice after competitive BM transplantation with 1:1 mix of unirradiated wild-type (CD45.1) and Mdm2P2/P2 cells (CD45.2) (left graph) or 4:1 mix of irradiated Mdm2P2/P2 (CD45.2) and unirradiated wild-type (CD45.1) BM cells (right graph) 16 wk after transplantation. n = 5, ±SEM. (F) Peripheral blood leukocyte marker analysis of lethally irradiated recipient Mdm2P2/P2 mice 16 wk after BM transplantation with wild-type CD45.1 cells. n = 4, ±SEM.
We next assessed the impact of gene dosage of Mdm2 and its homolog, Mdm4, in determining radiation sensitivity of Mdm2P2/P2 mice. We compared the survival of Mdm2P2/P2, Mdm2P2/−, and Mdm2P2/P2Mdm4Δ2/+ mice with Mdm2+/− and Mdm4Δ2/+ mice after 6 Gy IR (Fig. 4B). As anticipated, the reduction in Mdm2 and Mdm4 levels further enhanced the radiosensitivity of Mdm2P2/P2 mice. We also explored potential gender differences in determining radiosensitivity in Mdm2P2/P2 mice. However, no such difference was evident (Supplemental Fig. 4). These data suggest that the feedback loop is important for Mdm2-mediated inhibition of p53 activity and survival after DNA damage.
Irradiated Mdm2P2/P2 mice die due to p53-dependent bone marrow (BM) ablation
To identify the particular tissue type severely affected by radiation in the absence of the feedback loop and instigating the demise of irradiated Mdm2P2/P2 mice, we next performed a comparative histopathological examination of all major radiosensitive tissues from Mdm2P2/P2 and Mdm2+/+ mice. We observed mild to moderate hypoplasia of the spleen and thymus with atrophy of both the splenic white pulp and the thymic cortex (Supplemental Fig. 5). Areas of mild atrophy were also observed in the mucosa of the stomach, duodenum, and small/large intestine. A noted decline in cellularity was observed in BM sections from both genotypes, culminating in profound aplasia in Mdm2P2/P2 mice by day 12, while Mdm2+/+ mice recovered (Fig. 4C). Peripheral blood cell counts at this time point revealed severe thrombocytopenia and neutropenia as well as moderate anemia in Mdm2P2/P2 mice (Supplemental Fig. 6A). Additionally, BM differential analysis showed significant depletion of less differentiated elements of the white cell lineage in Mdm2P2/P2 mice (Supplemental Fig. 6B). Immunohistochemical analyses of BM sections with phospho-histone H3 and Ki-67 antibodies also revealed the absence of mitosis/proliferation in Mdm2P2/P2 compared with Mdm2+/+ irradiated mice (Supplemental Fig. 7A,B).
Overall, these results suggested that the death of Mdm2P2/P2 mice was likely a consequence of BM failure, a tissue overtly sensitive to radiation-induced p53 activity (Komarova et al. 2004; Ringshausen et al. 2006; Terzian et al. 2007; Wang et al. 2011). To further confirm that the BM aplasia in Mdm2P2/P2 mice was indeed a p53-dependent phenotype, we generated Mdm2P2/P2p53+/− and Mdm2P2/P2p53−/− mice and exposed them to 6 Gy IR. Both genotypes completely averted BM failure and survived the radiation dosage (Supplemental Fig. 8A,B).
Next, to characterize the particular cell types sensitized by loss of the p53–Mdm2 feedback loop in irradiated Mdm2P2/P2 mice, we performed flow cytometry analyses of BM constituent cells from Mdm2P2/P2 and Mdm2+/+ mice. Flow cytometry analysis revealed similar cell numbers in unirradiated mice of both genotypes but an acute and statistically significant (P < 0.001) depletion of common lymphoid progenitor cells (CLPs)/common myeloid progenitor cells (CMP) (Lin−veSca-1−ve/lowc-Kit+ve) and HSCs (Lin−veSca-1+vec-Kit+ve) in irradiated Mdm2P2/P2 mice (Fig. 4D).
p53 is important for maintaining HSC quiescence (Liu et al. 2009). To further confirm and characterize the role of the p53–Mdm2 feedback loop in HSC function, we examined the long-term engraftment potential of Mdm2P2/P2 HSCs by competitive and noncompetitive BM transplantation. We mixed BM cells from unirradiated Mdm2P2/P2 (expressing the CD45.2 leukocyte marker) and wild-type (expressing the CD45.1 leukocyte marker) mice in a 1:1 ratio and transplanted them into lethally irradiated wild-type recipient mice. Peripheral blood analysis after 16 wk confirmed normal engraftment and contribution to hematopoietic lineage by the Mdm2P2/P2 HSCs (Fig. 4E). However, in a similar transplantation experiment in parallel, a mix of irradiated Mdm2P2/P2 BM cells with unirradiated wild-type cells in a 4:1 ratio failed to engraft and contribute to hematopoiesis. This suggested that Mdm2P2/P2 HSCs are functionally normal in the absence of DNA damage. Furthermore, we also rescued the lethality of irradiated Mdm2P2/P2 mice by transplanting BM cells from unirradiated wild-type mice (Fig. 4F). Additionally, lethally irradiated wild-type mice transplanted with Mdm2P2/P2 BM cells in a noncompetitive assay acquired radiosensitivity (Supplemental Fig. 9). These results confirmed that the p53-dependent post-IR sensitivity in Mdm2P2/P2 mice was associated with stem cells per se and not the niche. Thus, the p53–Mdm2 feedback loop is critical for attenuating p53 response in the HSCs after genotoxic insults.
p53-mediated apoptosis causes HSC depletion in Mdm2P2/P2 mice
The p53 damage response predominantly initiates either cell cycle arrest or apoptosis. To identify the preferential p53 downstream pathway involved in the post-IR attrition of BM cells in feedback loop-deficient Mdm2P2/P2 mice, we first examined expression of two p53 transcriptional targets: p21, which encodes a cell cycle inhibitor, and Puma, a proapoptotic gene in BM cells of Mdm2P2/P2 mice. These genes have been previously implicated in radiosensitivity and HSC regulation (Cheng et al. 2000; van Os et al. 2007; Abbas et al. 2010; Shao et al. 2010; Yu et al. 2010). While the baseline levels were similar between Mdm2+/+ and Mdm2P2/P2 mice, we observed a significant and prolonged induction of these genes in irradiated Mdm2P2/P2 BM cells compared with irradiated Mdm2+/+ BM cells (Fig. 5A). As expected, Mdm2 mRNA levels were only induced in irradiated Mdm2+/+ BM cells (Supplemental Fig. 10).
Absence of feedback loop promotes p53-dependent apoptosis in irradiated Mdm2P2/P2 mice. (A) Relative p21 and Puma mRNA induction in Mdm2+/+ and Mdm2P2/P2 BM cells at different time points normalized to Rplp0 mRNA levels with −IR sample set to 1. n = 6, ±SEM; P-value was calculated by Student's t-test. (B) Hematoxylin and eosin-stained BM sections of Mdm2P2/P2, Mdm2P2/P2p21−/−, and Mdm2P2/P2Puma−/− mice 12 d after 6 Gy IR (10× magnification). (C) Kaplan-Meier survival curve for 6 Gy irradiated Mdm2P2/P2p21−/− and Mdm2P2/P2Puma−/− mice.
Next, we crossed Mdm2P2/P2 mice to p53515C/515C mice and generated Mdm2P2/P2p53515C/515C mice. Previously, our laboratory had shown that p53515C/515C mice express a mutant form of p53 (p53R172P) that activates only cell cycle arrest (partially) but not apoptotic programs (Liu et al. 2004). Interestingly, Mdm2P2/P2p53515C/515C mice survived exposure to 6 Gy IR (see Supplemental Fig. 8B). This suggested a predominance of the p53 apoptotic pathway in radiosensitization of Mdm2P2/P2 BM cells.
Next, to directly evaluate the role of apoptosis in post-IR BM attrition, we performed Annexin-V FITC flow cytometry analysis on BM cells from irradiated Mdm2+/+ and Mdm2P2/P2 mice (Supplemental Fig. 11). We could not detect a significant difference in apoptosis between the two genotypes using various doses of IR analyzed at different time points. It is possible that the rate of apoptosis essentially remains the same between the two genotypes but is prolonged in the case of Mdm2P2/P2 mice, causing increased loss of BM cellularity. A sustained induction of Puma (Fig. 5A) also supports this notion.
Finally, to segregate the in vivo role of apoptosis and cell cycle arrest pathways in radiation-induced BM aplasia, we crossed Mdm2P2/P2 mice with p21-null or Puma-null mice. Notably, Puma deficiency but not lack of p21 completely rescued the radiosensitivity (Fig. 5B,C). Furthermore, Puma heterozygosity also rescued the phenotype (Fig. 5C), thereby confirming that in the absence of the feedback loop, p53-mediated apoptosis is the principal pathway involved in HSC depletion.
Mdm2P2/P2 mice have a normal life span
Increase in p53 activity is also linked with increased genomic aberrations, stem cell depletion, and aging phenotypes (Tyner et al. 2002; Liu et al. 2010). To rule out the possibility that reduced levels of stress-induced Mdm2 in the feedback-defective Mdm2P2/P2 mouse could modulate stem cell function throughout life and impact survival, we monitored a cohort of Mdm2P2/P2, Mdm2P2/−, and Mdm2+/+ mice long term. No difference in survival was evident between the genotypes (Fig. 6A). Furthermore, the Mdm2P2/P2 and Mdm2P2/− mice reproduced and aged normally under standard nonstress laboratory conditions. This further confirms that the p53–Mdm2 feedback loop is dispensable for development and that its mere absence is not detrimental for normal functions. Moreover, basal Mdm2 levels from a single promoter (P1) are sufficient to regulate p53 and sustain life.
Lack of p53–Mdm2 feedback loop does not affect life span. (A) Kaplan-Meier survival curve for Mdm2+/+, Mdm2P2/P2, and Mdm2P2/− mice. (B) Kaplan-Meier survival curve for Mdm2+/+ and Mdm2P2/P2 mice after 3 Gy IR. Curves are censored at 800 d.
Finally, we examined whether exposure to a minor genotoxic insult capable of inducing p53 and initiating the feedback loop could alter the survival profile of these mice (Christophorou et al. 2006; Post et al. 2010; Gannon et al. 2012). To that end, we exposed Mdm2P2/P2 and Mdm2+/+ mice to 3 Gy IR and monitored them for survival (Fig. 6B). No difference in survival was evident in either group. In addition, the post-IR survival of Mdm2P2/P2 and Mdm2+/+ mice (Fig. 6B) was quite similar to the unirradiated mouse cohort (Fig. 6A). This suggests that a minor transient increase in p53 activity even in the absence of the p53–Mdm2 feedback loop is well tolerated and does not alter long-term stem cell functionality and overall survival.
Discussion
In the present study, using a mouse model with mutations at the p53-binding site in the Mdm2 P2 promoter, we provide the first in vivo characterization of the p53–Mdm2 feedback loop. In contrast to the prevailing paradigm, our results show that the p53–Mdm2 feedback loop is not essential for development, homeostasis, and longevity. Thus, constitutive Mdm2 levels expressed from the P1 promoter are sufficient for maintaining normal p53 protein levels. A second surprise is that after DNA damage, even in the absence of induced Mdm2 expression, the p53 degradation profile does not change. p53 levels appear at ∼2 h after radiation, peak at 4 h, and are barely detectable by 8 h in the radiosensitive spleen in normal mice and mice that lack the feedback loop. A similar pattern of p53 degradation is also evidenced in other tissues, such as the skin, thymus, lungs, and kidneys of these mice.
Still, p53 degradation is impeded in a Mdm2 mutant/null background (Ringshausen et al. 2006; Itahana et al. 2007). This suggests that additional factors may be involved in signaling p53 degradation. One possibility is that Mdm2 monoubiquitinates p53 and primes it for subsequent polyubiquitination and degradation in collaboration with other proteins (Grossman et al. 2003; Li et al. 2003). It is also possible that some as yet unknown E3 ligase is involved in p53 degradation after DNA damage in a p53-dependent fashion. Feedback-deficient Mdm2P2/P2 mice provide an excellent system to test these hypotheses in the future.
Regardless, the activity of p53 is compromised in Mdm2P2/P2 mice after DNA damage. While the absence of stress-induced Mdm2 does not significantly alter p53 protein levels, it does result in a modest increase in p53 activity that is well tolerated by most Mdm2P2/P2 mouse tissues. However, this increase in p53 activity turns catastrophic for the integrity of the hematopoietic system. Eighty percent of the Mdm2P2/P2 mice die due to hematopoietic failure, while all wild-type mice survive. The post-IR death of Mdm2P2/P2 mice could be simply averted by BM transplantation with wild-type donor cells. The hematopoietic system is in fact the most radiosensitive tissue and functions as a readout for small increases in p53 activity (Komarova et al. 2000; Liu et al. 2007; Terzian et al. 2007; Abbas et al. 2010; Wang et al. 2011). The importance of Mdm2 in inhibition of IR-induced p53 activity has been observed in heterozygous or hypomorphic Mdm2 mice (Mendrysa et al. 2003; Terzian et al. 2007). However, these mice have low levels of basal Mdm2 and correspondingly higher basal p53 activity, which is easily enhanced to lethal limits by IR. In contrast, Mdm2P2/P2 mice have normal levels of Mdm2 from the P1 promoter and normal p53 activity. Thus, the specific role of p53-induced Mdm2 could be clearly evaluated in these mice.
In the absence of the p53–Mdm2 feedback loop and in response to DNA damage, a modest increase in p53 activity promotes apoptosis of HSCs/CLPs/CMPs and impairs the normal dynamics of progenitor cell proliferation. Crosses with p53-null mice rescue the phenotype, suggesting p53 dependence. More importantly, crosses with Puma-null mice (an apoptotic gene that is a target of p53) but not p21-null mice (cell cycle arrest/senescent target of p53) completely rescue the stem cell phenotype. Thus, it is the p53-mediated apoptosis pathway that causes the demise of the animals after DNA damage. To our knowledge, this is the first study to simultaneously evaluate the role of p53-dependent apoptotic and cell cycle arrest pathways in determining radiation sensitivity in a mouse model.
Previously, mathematical modeling predicted that the autoregulatory loop controls the frequency and amplitude of p53 DNA damage response (Lahav et al. 2004). Our data suggest that it also regulates the duration of the p53 response in vivo. In particular, this is emphasized in the hematopoietic compartment after DNA damage, wherein the basal Mdm2 levels are likely insufficient. It will be interesting to test whether other tissue stem cells are also similarly sensitized in the Mdm2P2/P2 mice.
We also show that absence of the feedback loop does not promote aging or impede stem cell function under normal or low-dose IR (3 Gy) conditions. This suggests that it is dispensable for normal homeostasis and protection against minor stress conditions. Moreover, Mdm2 levels expressed from the single P1 promoter are sufficient in sustaining normal life. Overall, these results challenge our conventional view of the p53–Mdm2 feedback loop and question the accepted role of Mdm2 as the sole E3 ubiquitin ligase for stress-induced p53. Furthermore, these results imply that while basal levels of Mdm2 (from the P1 promoter) are required for p53 degradation, the primary function of increased Mdm2 levels (from the P2 promoter) pertains to regulating p53 activity.
These results have clinical relevance, as activation of wild-type p53 is currently being evaluated for therapeutic purposes in the clinic. Our results here suggest that pharmacological inhibition of the p53–Mdm2 interaction in combination with DNA damage can be explored as a potential therapeutic strategy for p53 activation in the hematopoietic system. In particular, such strategies can be used for sensitizing stem cells in hematological malignancies. Indeed, similar approaches are currently being tested in the clinic for leukemia treatment (Cheok et al. 2011). These strategies provide a relatively safer alternative in lieu of the high-dose radio/chemotherapy regimens and associated side effects.
Materials and methodsTargeting construct and generation of mice
The Xho1–Xho1 DNA fragment (1 kb) from intron 2 of mouse Mdm2 covering the P2 promoter and the p53 response elements was cloned into pBluescript vector. Site-directed mutagenesis was carried out using QuickChange kit (Stratagene). An EcoRV restriction enzyme site was created at the second p53 response element to enable us to distinguish wild-type and mutant alleles for genotyping. Mutated DNA was sequenced and cloned as part of the 5-kb 5′ homologous arm into pLG1 targeting vector backbone. A 1.2-kb homologous fragment was added as the 3′ arm. Lox-pGKNeo-Lox and Hsv-Tk1 cassettes were included for positive and negative selection, respectively (Fig. 1A). The targeting construct was sequenced completely and electroporated into TC1 mouse ES cells. G418-resistant clones were analyzed for correct homologous recombination by Southern blotting using 5′ and 3′ external probes (Fig. 1A). Two independent correctly targeted clones were expanded and injected into C57BL/6 blastocysts to generate Mdm2P2/+ chimeras. Male chimeras were backcrossed to C57BL/6 mice to secure germline transmission of the mutant allele.
Mouse breeding, maintenance, and genotyping
p53-null and CD45.1 mice were purchased from Jackson Laboratories. p21-null mice were a gift from Dr. T. Jacks (Massachusetts Institute of Technology), and Puma-null mice were from Dr. G. Zambetti (St. Jude Children's Research Hospital). Mdm2+/− and Mdm4Δ2/+ mice have been described previously (Montes de Oca Luna et al. 1995; Xiong et al. 2006). Mice were maintained in >90% C57BL/6 background. All mouse studies were conducted in compliance with Institutional Animal Care and Use Committee protocols. Genotyping was carried out either as described earlier (Post et al. 2010) or by PCR amplification over the p53 response elements with primer sets Mdm2-F (5′-GGTCCAGGAGGTGACAGGT-3′) and Mdm2-R (5′-ACGTCTTTCGGCAATAGCTC-3′) followed by EcoRV digestion and resolution on agarose gel.
Protein analysis
Protein lysates were prepared by lysing tissues or MEFs in NP-40 buffer. Protein estimation was carried out with BCA (Protein Assay kit, Pierce). Fifty micrograms of lysate was resolved on 8% SDS-PAGE and immunoblotted with antibodies against Mdm2 (1:500; 2A10, Calbiochem), Mdm4 (1:500; MX82, Sigma), p53 (1:1000; CM5, Vector Laboratories), S18-p53 (1:1000; 9284, Cell Signaling), Vinculin (1:1000; V9131, Sigma), and p21 (1:1000; 556431, BD Pharmingen). Western blots were repeated at least three times with biological replicates. p53 expression was quantitated using ImageJ software (National Institutes of Health [NIH]).
ChIP assay
Spleens harvested from irradiated and nonirradiated mice were washed with PBS and frozen-pulverized under liquid nitrogen. Chromatin was fixed with formaldehyde, and ChIP assay was carried out as described earlier (Jackson and Pereira-Smith 2006).
Quantitative RT–PCR
RNA isolation and quantitative RT–PCR were carried out as previously described in Pant et al. (2011).
MEF preparation and cell culture
MEFs prepared from 13.5 d post-coitum (dpc) embryos were maintained in Dulbecco's modified Eagle's medium (DMEM) (Invitrogen) supplemented with 10% FBS and penicillin (100 IU/mL)/streptomycin (100 μg/mL). Early passage MEFs (P2–P3) were used for analysis.
IR and UV studies
Mice were irradiated at 6 Gy in a cesium-137 irradiator and killed at different time points. Tissues were harvested and lysed in NP-40 buffer for protein or TRIzol for RNA analyses. MEFs cultured in a 100-mm tissue culture dish were irradiated at 10 Gy. For UV studies, MEFs were exposed to 50 J/m2 UV in a Stratalinker (Stratagene) without the medium. Fresh medium was added, and cells were incubated at 37°C before harvesting at different time points for experimental analyses.
BM harvesting and flow cytometry
BM cells were harvested from femurs and tibias of Mdm2P2/P2 and Mdm2+/+ mice in PBS. Cell suspension was passed through an 18.5-gauge needle and finally filtered through a 40-μm filter. All cellular suspensions were kept on ice until further processing. For flow cytometry, murine HSCs and CLPs/CMPs were identified using (eBioscience) fluorochrome-conjugated antibodies against ckit (PE-cyanine7) and Sca-1 (allophycocyanin) and for lineage depletion, we used CD4 (phycoerythrin [PE]), CD8a (PE), B220 (PE), Ter119 (PE), Mac-1 (PE), and Gr-1 (PE) in HBSS+. Cells were stained for 20 min in the dark on ice and then washed with HBSS+. We analyzed 1.25 million cells on BD LSRII System at the University of Texas M.D. Anderson Cancer Center Flow Cytometry Core Facility. Dead cells were excluded with DAPI staining (eBioscience).
BM transplantation
Donor BM cells were isolated from wild-type (CD45.1) or Mdm2P2/P2 (CD45.2) mice. For a noncompetitive assay, 2.5 × 105 cells from the donor mouse were tail vein-injected into lethally irradiated recipient mice. For a competitive transplantation assay, either 2.5 × 105 cells each from unirradiated wild-type (CD45.1) and unirradiated Mdm2P2/P2 (CD45.2) mice were mixed or 2.5 × 105 cells from unirradiated wild-type mouse (CD45.1) were mixed with 1 × 106 cells from irradiated Mdm2P2/P2 (CD45.2) mice and tail vein-injected into lethally irradiated recipient wild-type (CD45.1) mice. Eight weeks and 16 wk after transplants, retro-orbitally drawn peripheral blood was stained with leukocyte marker antibodies and analyzed by flow cytometry for donor reconstitution.
Apoptosis and cell cycle analyses
Thymocytes derived from thymi and BM cells harvested from femurs and tibias of irradiated/unirradiated mice were analyzed for apoptosis by Annexin-V FITC flow cytometry (Apolert Annexin-V FITC apoptosis kit, Clontech) per the manufacturer's instructions. For cell cycle, 10 Gy irradiated MEFs were incubated for 23 h at 37°C and processed as per the manufacturer's instructions (Click-iT EdU Alexa Fluor 488 flow cytometry assay kit, Invitrogen) for analysis on BD FACS Calibur System.
Histopathology and immunohistochemistry
Tissues harvested from Mdm2P2/P2 and Mdm2+/+ mice were fixed in 10% buffered formalin saline and paraffin-embedded. Five-micrometer sections were stained by hematoxylin and eosin and examined by light microscopy. Selected unstained sections were immunohistochemically analyzed with respective antibodies.
Statistical analysis
P-value was calculated by Student's t-test using Graphpad software, and values <0.05 were considered significant. (*) P < 0.05, (**) P < 0.01, and (***) P < 0.001.
Acknowledgments
Mice were made by the GEM Facility at M.D. Anderson Cancer Center. Studies were supported by Cancer Center support grant CA016672 and NIH grant CA47296 to G.L. V.P. was supported in part by a NIH training grant in Molecular Genetics of Cancer (CA009299) and the Brown Foundation. V.P. designed and performed most of the experiments and wrote the manuscript. S.X. performed thymocyte culture and apoptosis assays. J.G.J. performed ChIP assays. S.M.P. provided the vector for targeting and helped in BM isolation. H.A.A. performed LSK flow cytometry analysis. A.Q.C. performed BM differential counts and microscopy. A.N.H. is the pathologist associated with the study. V.P. and G.L conceived the project. G.L. guided the overall research work.
Supplemental material is available for this article.
Article published online ahead of print. Article and publication date are online at http://www.genesdev.org/cgi/doi/10.1101/gad.227249.113.
ReferencesAbbasHA, MaccioDR, CoskunS, JacksonJG, HazenAL, SillsTM, YouMJ, HirschiKK, LozanoG2010Mdm2 is required for survival of hematopoietic stem cells/progenitors via dampening of ROS-induced p53 activity.
7: 606–61721040902BarakY, JuvenT, HaffnerR, OrenM1993mdm2 expression is induced by wild type p53 activity.
12: 461–4688440237BarakY, GottliebE, Juven-GershonT, OrenM1994Regulation of mdm2 expression by p53: Alternative promoters produce transcripts with nonidentical translation potential.
8: 1739–17497958853ChengT, RodriguesN, ShenH, YangY, DombkowskiD, SykesM, ScaddenDT2000Hematopoietic stem cell quiescence maintained by p21cip1/waf1.
287: 1804–180810710306CheokCF, VermaCS, BaselgaJ, LaneDP2011Translating p53 into the clinic.
8: 25–3720975744ChristophorouMA, RingshausenI, FinchAJ, SwigartLB, EvanGI2006The pathological response to DNA damage does not contribute to p53-mediated tumour suppression.
443: 214–21716957739FrancozS, FromentP, BogaertsS, De ClercqS, MaetensM, DoumontG, BellefroidE, MarineJC2006Mdm4 and Mdm2 cooperate to inhibit p53 activity in proliferating and quiescent cells in vivo.
103: 3232–323716492744GannonHS, WodaBA, JonesSN2012ATM phosphorylation of Mdm2 Ser394 regulates the amplitude and duration of the DNA damage response in mice.
21: 668–67922624716GrossmanSR, DeatoME, BrignoneC, ChanHM, KungAL, TagamiH, NakataniY, LivingstonDM2003Polyubiquitination of p53 by a ubiquitin ligase activity of p300.
300: 342–34412690203HauptY, MayaR, KazazA, OrenM1997Mdm2 promotes the rapid degradation of p53.
387: 296–2999153395HondaR, TanakaH, YasudaH1997Oncoprotein MDM2 is a ubiquitin ligase E3 for tumor suppressor p53.
420: 25–279450543ItahanaK, MaoH, JinA, ItahanaY, CleggHV, LindstromMS, BhatKP, GodfreyVL, EvanGI, ZhangY2007Targeted inactivation of Mdm2 RING finger E3 ubiquitin ligase activity in the mouse reveals mechanistic insights into p53 regulation.
12: 355–36617936560JacksonJG, Pereira-SmithOM2006Primary and compensatory roles for RB family members at cell cycle gene promoters that are deacetylated and downregulated in doxorubicin-induced senescence of breast cancer cells.
26: 2501–251016537896JonesSN, RoeAE, DonehowerLA, BradleyA1995Rescue of embryonic lethality in Mdm2-deficient mice by absence of p53.
378: 206–2087477327KomarovaEA, ChristovK, FaermanAI, GudkovAV2000Different impact of p53 and p21 on the radiation response of mouse tissues.
19: 3791–379810949934KomarovaEA, KondratovRV, WangK, ChristovK, GolovkinaTV, GoldblumJR, GudkovAV2004Dual effect of p53 on radiation sensitivity in vivo: p53 promotes hematopoietic injury, but protects from gastro-intestinal syndrome in mice.
23: 3265–327115064735KubbutatMH, JonesSN, VousdenKH1997Regulation of p53 stability by Mdm2.
387: 299–3039153396LahavG, RosenfeldN, SigalA, Geva-ZatorskyN, LevineAJ, ElowitzMB, AlonU2004Dynamics of the p53–Mdm2 feedback loop in individual cells.
36: 147–15014730303LaneDP1992Cancer. p53, guardian of the genome.
358: 15–161614522LewandoskiM, WassarmanKM, MartinGR1997Zp3-cre, a transgenic mouse line for the activation or inactivation of loxP-flanked target genes specifically in the female germ line.
7: 148–1519016703LiM, BrooksCL, Wu-BaerF, ChenD, BaerR, GuW2003Mono- versus polyubiquitination: Differential control of p53 fate by Mdm2.
302: 1972–197514671306LiuG, ParantJM, LangG, ChauP, Chavez-ReyesA, El-NaggarAK, MultaniA, ChangS, LozanoG2004Chromosome stability, in the absence of apoptosis, is critical for suppression of tumorigenesis in Trp53 mutant mice.
36: 63–6814702042LiuG, TerzianT, XiongS, Van PeltCS, AudiffredA, BoxNF, LozanoG2007The p53–Mdm2 network in progenitor cell expansion during mouse postnatal development.
213: 360–36817893884LiuY, ElfSE, MiyataY, SashidaG, HuangG, Di GiandomenicoS, LeeJM, DeblasioA, MenendezS, AntipinJ, 2009p53 regulates hematopoietic stem cell quiescence.
4: 37–4819128791LiuD, OuL, ClemensonGDJr, ChaoC, LutskeME, ZambettiGP, GageFH, XuY2010Puma is required for p53-induced depletion of adult stem cells.
12: 993–99820818388MarineJC, FrancozS, MaetensM, WahlG, ToledoF, LozanoG2006Keeping p53 in check: Essential and synergistic functions of Mdm2 and Mdm4.
13: 927–93416543935MendrysaSM, PerryME2000The p53 tumor suppressor protein does not regulate expression of its own inhibitor, MDM2, except under conditions of stress.
20: 2023–203010688649MendrysaSM, McElweeMK, MichalowskiJ, O'LearyKA, YoungKM, PerryME2003mdm2 Is critical for inhibition of p53 during lymphopoiesis and the response to ionizing irradiation.
23: 462–47212509446MendrysaSM, O'LearyKA, McElweeMK, MichalowskiJ, EisenmanRN, PowellDA, PerryME2006Tumor suppression and normal aging in mice with constitutively high p53 activity.
20: 16–2116391230MomandJ, ZambettiGP, OlsonDC, GeorgeD, LevineAJ1992The mdm-2 oncogene product forms a complex with the p53 protein and inhibits p53-mediated transactivation.
69: 1237–12451535557Montes de Oca LunaR, WagnerDS, LozanoG1995Rescue of early embryonic lethality in mdm2-deficient mice by deletion of p53.
378: 203–2067477326OlinerJD, PietenpolJA, ThiagalingamS, GyurisJ, KinzlerKW, VogelsteinB1993Oncoprotein MDM2 conceals the activation domain of tumour suppressor p53.
362: 857–8608479525PantV, XiongS, IwakumaT, Quintas-CardamaA, LozanoG2011Heterodimerization of Mdm2 and Mdm4 is critical for regulating p53 activity during embryogenesis but dispensable for p53 and Mdm2 stability.
108: 11995–1200021730132PostSM, Quintas-CardamaA, PantV, IwakumaT, HamirA, JacksonJG, MaccioDR, BondGL, JohnsonDG, LevineAJ, 2010A high-frequency regulatory polymorphism in the p53 pathway accelerates tumor development.
18: 220–23020832750RingshausenI, O'SheaCC, FinchAJ, SwigartLB, EvanGI2006Mdm2 is critically and continuously required to suppress lethal p53 activity in vivo.
10: 501–51417157790SaucedoLJ, CarstensBP, SeaveySE, AlbeeLD2nd, PerryME1998Regulation of transcriptional activation of mdm2 gene by p53 in response to UV radiation.
9: 119–1309486848ShaoL, SunY, ZhangZ, FengW, GaoY, CaiZ, WangZZ, LookAT, WuWS2010Deletion of proapoptotic Puma selectively protects hematopoietic stem and progenitor cells against high-dose radiation.
115: 4707–471420360471TerzianT, WangY, Van PeltCS, BoxNF, TravisEL, LozanoG2007Haploinsufficiency of Mdm2 and Mdm4 in tumorigenesis and development.
27: 5479–548517526734TerzianT, SuhYA, IwakumaT, PostSM, NeumannM, LangGA, Van PeltCS, LozanoG2008The inherent instability of mutant p53 is alleviated by Mdm2 or p16INK4a loss.
22: 1337–134418483220TynerSD, VenkatachalamS, ChoiJ, JonesS, GhebraniousN, IgelmannH, LuX, SoronG, CooperB, BraytonC, 2002p53 mutant mice that display early ageing-associated phenotypes.
415: 45–5311780111van OsR, KammingaLM, AusemaA, BystrykhLV, DraijerDP, van PeltK, DontjeB, de HaanG2007A Limited role for p21Cip1/Waf1 in maintaining normal hematopoietic stem cell functioning.
25: 836–84317170062VousdenKH, LuX2002Live or let die: The cell's response to p53.
2: 594–60412154352WangYV, LeblancM, FoxN, MaoJH, TinkumKL, KrummelK, EngleD, Piwnica-WormsD, Piwnica-WormsH, BalmainA, 2011Fine-tuning p53 activity through C-terminal modification significantly contributes to HSC homeostasis and mouse radiosensitivity.
25: 1426–143821724834WuX, BayleJH, OlsonD, LevineAJ1993The p53–mdm-2 autoregulatory feedback loop.
7: 1126–11328319905XiongS, Van PeltCS, Elizondo-FraireAC, LiuG, LozanoG2006Synergistic roles of Mdm2 and Mdm4 for p53 inhibition in central nervous system development.
103: 3226–323116492743YuH, ShenH, YuanYXuFengR, HuX, GarrisonSP, ZhangL, YuJ, ZambettiGP, ChengT2010Deletion of Puma protects hematopoietic stem cells and confers long-term survival in response to high-dose γ-irradiation.
115: 3472–348020177048ZaubermanA, FlusbergD, HauptY, BarakY, OrenM1995A functional p53-responsive intronic promoter is contained within the human mdm2 gene.
23: 2584–25927651818oai:pubmedcentral.nih.gov:37782412014-03-01genesdevpmc-openGenes DevGenes DevGADGenes & Development0890-93691549-5477Cold Spring Harbor Laboratory PressPMC3778241PMC377824137782412401350124013501871166010.1101/gad.224386.113Research PaperThe C terminus of p53 regulates gene expression by multiple mechanisms in a target- and tissue-specific manner in vivoHamard et al.The C terminus regulates p53 activity in vivoHamardPierre-Jacques1BartheleryNicolas124HogstadBrandon124MungamuriSathish Kumar1TonnessenCrystal A.12CarvajalLuis A.12SenturkEmir12GillespieVirginia3AaronsonStuart A.12MeradMiriam12ManfrediJames J.125Department of Oncological Sciences,The Graduate School of Biomedical Sciences,Center for Comparative Medicine and Surgery, Mount Sinai School of Medicine, New York, New York 10029, USA
The C terminus of the tumor suppressor p53 is subjected to multiple post-translational modifications, suggesting that differing sets of modifications determine distinct cellular outcomes. Hamard et al. address this question by generating a Trp53 mutant mouse that constitutively expresses truncated p53. Intriguingly, the C terminus acts via three distinct mechanisms to control p53-dependent gene expression depending on the tissue. This study reconciles contradictory reports and delineates how regulation of target gene selectivity by p53 leads to alternate cellular outcomes.
The p53 tumor suppressor is a transcription factor that mediates varied cellular responses. The C terminus of p53 is subjected to multiple and diverse post-translational modifications. An attractive hypothesis is that differing sets of combinatorial modifications therein determine distinct cellular outcomes. To address this in vivo, a Trp53ΔCTD/ΔCTD mouse was generated in which the endogenous p53 is targeted and replaced with a truncated mutant lacking the C-terminal 24 amino acids. These Trp53ΔCTD/ΔCTD mice die within 2 wk post-partum with hematopoietic failure and impaired cerebellar development. Intriguingly, the C terminus acts via three distinct mechanisms to control p53-dependent gene expression depending on the tissue. First, in the bone marrow and thymus, the C terminus dampens p53 activity. Increased senescence in the Trp53ΔCTD/ΔCTD bone marrow is accompanied by up-regulation of Cdkn1 (p21). In the thymus, the C-terminal domain negatively regulates p53-dependent gene expression by inhibiting promoter occupancy. Here, the hyperactive p53ΔCTD induces apoptosis via enhanced expression of the proapoptotic Bbc3 (Puma) and Pmaip1 (Noxa). In the liver, a second mechanism prevails, since p53ΔCTD has wild-type DNA binding but impaired gene expression. Thus, the C terminus of p53 is needed in liver cells at a step subsequent to DNA binding. Finally, in the spleen, the C terminus controls p53 protein levels, with the overexpressed p53ΔCTD showing hyperactivity for gene expression. Thus, the C terminus of p53 regulates gene expression via multiple mechanisms depending on the tissue and target, and this leads to specific phenotypic effects in vivo.
The p53 tumor suppressor is a transcription factor that mediates a variety of cellular responses. The best characterized of these include cell cycle arrest, which, if sustained, leads to senescence and apoptosis (Vousden and Prives 2009). The p53 protein is kept inactive through interactions with several negative regulators, most notably Mdm2 (Luo et al. 2004; Kruse and Gu 2009; Vousden and Prives 2009; Manfredi 2010). It has been proposed that post-translational modifications of p53 are critical to disrupt its binding to Mdm2 and facilitate its interaction with cofactors, resulting in a p53 protein that is robust in its ability to transcriptionally regulate target gene expression (Kruse and Gu 2009; Manfredi 2010). The C terminus of p53 in particular is subjected to multiple and diverse post-translational modifications (Kruse and Gu 2009; Carvajal and Manfredi 2013) quite similar to that seen with histone tails (Kouzarides 2007). An attractive hypothesis is that differing sets of combinatorial modifications therein determine distinct cellular outcomes (Carvajal and Manfredi 2013).
The C terminus was originally characterized as playing a key role in sequence-specific DNA binding by p53 (Hupp et al. 1992). Its excess of basic residues contributes to its ability to bind to DNA, albeit nonspecifically. It had been proposed that this region of p53 interferes with the ability of the core domain to bind response elements with sequence specificity and therefore has a negative regulatory function (Anderson et al. 1997; Ahn and Prives 2001). This notion was discredited by studies suggesting that this may be an artifact of the particular in vitro DNA-binding assays being used (Espinosa and Emerson 2001), although this remains controversial (Luo et al. 2004; Kruse and Gu 2009; Vousden and Prives 2009; Carvajal and Manfredi 2013). Experiments in cell culture were more in line with a positive role for the C terminus (Chen et al. 1996; Kruse and Gu 2009; Hamard et al. 2012; Carvajal and Manfredi 2013). These supported a role in tracking of p53 on genomic DNA (McKinney et al. 2004; Kruse and Gu 2009; Carvajal and Manfredi 2013) as well as serving as recruitment sites for transcriptional cofactors (Barlev et al. 2001; Mujtaba et al. 2004). The challenge in interpreting such experiments is that they rely on ectopic expression of an exogenously introduced p53, often overexpressed at a nonphysiologically relevant level.
To gain more definitive insight into the role of this region of p53, a Trp53ΔCTD/ΔCTD mouse was generated that replaces the endogenous p53 with a truncated form of p53 that lacks the C-terminal 24 amino acids. In spite of presenting a wild-type phenotype immediately after birth, Trp53ΔCTD/ΔCTD mice quickly degenerate and die within 2 wk post-partum. Two main phenotypes are observed: a hematopoietic failure and an abnormal cerebellar development. The C terminus of p53 is shown to regulate gene expression at several levels: p53 accumulation, genomic site occupancy, or transcriptional activation. The precise mechanism depends on the tissue and target, thereby conferring specific phenotypic effects in vivo.
ResultsGeneration of the Trp53ΔCTD/ΔCTD mouse
To generate a Trp53ΔCTD/ΔCTD mouse that lacks the C-terminal 24 amino acids of p53, a knock-in strategy was used (Fig. 1A). To enable future studies with targeted expression of the deletion mutant, a Trp53NEO/NEO mouse was first generated by means of a targeting vector harboring two exon 11s separated by a NEO selection cassette (see the Supplemental Material for details on the construction of the targeting vector). The resulting targeted Trp53NEO allele contains a wild-type exon 11 in frame with the rest of the upstream gene and enables the expression of a full-length (FL), wild-type p53 protein along with the shorter mouse-specific alternative spliced form, p53AS (Bienz et al. 1984). Two loxP recombination sites within intron 10 and downstream from the NEO cassette, respectively, enable the excision of the first wild-type exon 11 along with the NEO cassette after recombination with the CRE recombinase. The resulting Trp53ΔCTD allele contains the mutated exon 11 in frame with the rest of the gene and enables the expression of a truncated p53 protein (p53ΔCTD) along with the shorter mouse-specific alternative spliced form (p53AS) as well. Two FRT recombination sites located within intron 10 and in the 3′ untranslated region (UTR) have been introduced for future studies using murine embryonic fibroblasts (MEFs) derived from Trp53NEO/NEO or Trp53ΔCTD/ΔCTD animals.
Generation of mice expressing a p53 protein devoid of its C-terminal domain. (A) Schematic of the targeting construct harboring two exon 11s separated by a NEO selection cassette. The resulting targeted Trp53NEO allele contains a wild-type exon 11 in frame with the rest of the gene and enables for the expression of a FL, wild-type p53 protein along with the shorter mouse-specific alternative spliced form (p53AS). After CRE recombination, the first wild-type exon 11 and the NEO cassette are excised, resulting in a Trp53ΔCTD allele that contains the second exon 11, which includes a STOP mutation at position 367. The Trp53ΔCTD allele encodes for a truncated p53 protein (p53ΔCTD) that lacks the last 24 amino acids as well as the p53AS isoform. The genotyping strategies are detailed in the Supplemental Material and Supplemental Figure 1. (Yellow triangles) loxP sites; (green triangles) FRT sites; (H) HindIII; (X) XhoI; (black bar) Southern blot probe; (a & b, c, and d) primers used for genotyping; (TK) thymidine kinase. (B) Immunoblot shows p53 protein expression in MEFs derived from Trp53+/+ and Trp53NEO/NEO animals before or after CRE recombination. The cells were infected with an adenovirus expressing the CRE recombinase at MOI = 200 to induce the expression of the truncated form. Twenty-four hours later, cells were treated with DOX at the final concentration of 0.2 μg/mL to reach appreciable levels of p53, and 24 h after treatment, cells were harvested, and immunoblotting analysis was conducted. β-Actin was used as a loading control. p53FL and p53ΔCTD are almost indistinguishable in size, and long running times and large gels were necessary to observe the difference in migration. (C) Immunoprecipitation of p53 in MEFs derived from Trp53+/+ and Trp53ΔCTD/ΔCTD animals with the p53CTD-specific monoclonal antibody PAb421. The cells were treated for 6 h with the proteasome inhibitor MG132 at a final concentration of 40 μM to reach appreciable levels of p53. The PAb421 antibody fails at precipitating the truncated p53 protein from Trp53ΔCTD/ΔCTD cells. A 5% input was immunoblotted for p53, and β-actin was used as a loading control.
Trp53NEO/+ heterozygous mice were viable and appeared phenotypically normal. They were intercrossed to generate Trp53NEO/NEO mice, which were also viable and indistinguishable from Trp53+/+ animals, and were born at the expected Mendelian ratio (Supplemental Fig. 1A). To ascertain that the CRE-driven recombination was functional, MEFs were derived from embryonic day 14.5 (E14.5) embryos and treated with increasing doses of a CRE-expressing adenovirus (Ad-CRE) (Supplemental Fig. 1B). Using genomic DNA from MEFs of +/+, NEO/+, and NEO/NEO genotypes, recombination was observed at a multiplicity of infection (MOI) as low as 100, and the nonrecombined DNA was undetectable in Trp53NEO/NEO MEFs at MOI = 500. To confirm that the recombination occurring at the genomic DNA level was giving rise to the shorter p53ΔCTD protein, MEFs were treated with Ad-CRE for 24 h and then treated with the DNA-damaging agent doxorubicin (DOX) for another 24 h at the indicated dose (Fig. 1B). Long running times were necessary to observe the difference in size between wild-type and p53ΔCTD proteins. Trp53NEO/NEO mice were crossed with Protamine-CRE (PrmCre) mice to generate Trp53NEO/+ CRE-expressing males. These males were intercrossed with wild-type C57BL/6J females to generate Trp53ΔCTD/+ heterozygous mice, which were born at the expected Mendelian ratio (data not shown), with no phenotypic difference from wild-type mice. Ultimately, these heterozygous mice were bred to obtain Trp53ΔCTD/ΔCTD homozygous animals. To confirm that these animals were expressing the truncated form of p53, MEFs derived from E14.5 Trp53+/+ and Trp53ΔCTD/ΔCTD embryos were treated with the proteasome inhibitor MG132 to attain discernable amounts of p53 protein and analyzed by immunoblotting. Similar to what was observed in NEO/NEO MEFs, a shorter form of p53 could be seen in ΔCTD/ΔCTD MEFs (Fig. 1C). The PAb421 monoclonal antibody has an epitope that is located within the C terminus of p53. An immunoprecipitation using this antibody confirmed that the protein expressed in the ΔCTD/ΔCTD cells lacked the PAb421 epitope (Fig. 1C).
Deletion of the C-terminal 24 amino acids from p53 leads to postnatal developmental defects and death within 2 wk post-partum
Mice homozygous for deletion of the p53 C terminus displayed a striking phenotype after birth. Although the mutant pups did not show any significant phenotypic difference from their +/+ and heterozygous littermates at day 1 post-partum (P1) (Fig. 2A), these animals were markedly reduced in size and weight by P10 (Fig. 2A,B) and died within 2 wk post-partum (Fig. 2C). They exhibited several developmental defects, including kinked tails, abnormal tail tip, and digit pigmentation, and significant ataxia suggestive of underlying neurological defects. Their organs were isolated and weighed at P1 and P10. The tibia, thymus, and spleen were dramatically reduced in size, while the liver and kidney were unaffected, and the heart was enlarged (Fig. 2D; Supplemental Fig. 2A,D). The observed cardiomegaly is reminiscent of that seen in Trp537KR/7KR mutants after irradiation (Wang et al. 2011) and is likely a compensatory consequence of the severe anemia these mutant mice present at P10. Mutant tibias were reduced in size and weight, and the red color associated with the bone marrow was not observed (Fig. 2D). Indeed, when assessed by complete blood count, Trp53ΔCTD/ΔCTD animals showed significant anemia, neutropenia, and thrombocytopenia (Fig. 2E). None of these defects were observed in the Trp53ΔCTD/+ heterozygous animals.
Deletion of the C-terminal 24 amino acids from p53 leads to postnatal developmental defects and death within 2 wk post-partum. (A) Pictures of whole animals at P1 show no difference between littermates of different genotypes (Trp53+/+, Trp53ΔCTD/+, and Trp53ΔCTD/ΔCTD). At P10, Trp53ΔCTD/ΔCTD animals are smaller than their wild-type counterparts and exhibit several developmental defects, including kinked tails, abnormal tail tip, and digit pigmentation. Bar, 1 cm. (B) Weight curve shows marked growth retardation for Trp53ΔCTD/ΔCTD animals but no significant difference between Trp53+/+ and Trp53ΔCTD/+ mice. (C) Survival curves spanning 30 d post-partum illustrate 100% death within 2 wk for Trp53ΔCTD/ΔCTD animals, whereas all other genotypes survive. (D) Pictures of P1 and P10 whole organs demonstrate no significant difference between all genotypes at P1 but marked anemia and reduction in size for most Trp53ΔCTD/ΔCTD organs at P10. Weight graphs of P10 whole organs show reduced weight for Trp53ΔCTD/ΔCTD spleen, thymus, and tibias but not the liver and no significant difference between Trp53+/+ and Trp53ΔCTD/+ organs. Bar, 0.4 cm. (E) Complete blood counts reveal a marked reduction of white and red blood cells, platelets, hematocrit percentage, and hemoglobin concentration in Trp53ΔCTD/ΔCTD animals at P10.
The terminal 24 amino acids of p53 do not include the tetramerization domain (amino acids 307–355). It is therefore likely that wild-type p53 and p53ΔCTD are capable of co-oligomerization. In order to discern whether the FL p53 protein was playing a dominant-negative role in Trp53ΔCTD/+ mice, these animals were crossed with Trp53−/− mice. The Trp53ΔCTD/− offspring were phenotypically indistinguishable from Trp53ΔCTD/+, Trp53+/−, and Trp53−/− animals up to 10 mo of age (data not shown). In contrast to Trp53−/− animals, which succumb to spontaneous lymphomas or sarcomas 6 mo after birth, Trp53ΔCTD/+ and Trp53ΔCTD/− mice were all tumor-free up to the age of 6 mo (data not shown). Thus, the truncated mutant p53 retains tumor suppressor activity. The rescue of Trp53ΔCTD/ΔCTD mice by deletion of one of the mutant alleles (Fig. 2C) also supports the idea that the level and/or activity of the p53 protein devoid of its C terminus (p53ΔCTD) are central to the observed phenotype.
Deletion of the C-terminal 24 amino acids from p53 induces hematopoietic failure post-partum
To investigate this further, histopathology studies using paraffin sections of P10 pups were conducted. No significant differences were observed in kidney, bladder, skin, or lungs at P10 (Supplemental Fig. 2C; data not shown). The hematopoietic compartment, however, was strongly affected by the deletion of the C terminus. Although no significant differences between genotypes were observed immediately before birth (E18.5) in bone marrow (Fig. 3A) and liver (Fig. 3B), Trp53ΔCTD/ΔCTD bone marrow cellularity started to decline as early as P1 and was virtually aplastic at P10 or P12 compared with its wild-type and heterozygous counterparts. No difference was noted between the bone marrow of Trp53+/+ and Trp53ΔCTD/+ animals at any age.
Deletion of the C-terminal 24 amino acids from p53 induces hematopoietic failure post-partum. (A) Bone sections stained with haematoxylin and eosin (H&E) from Trp53+/+, Trp53ΔCTD/+, and Trp53ΔCTD/ΔCTD animals of the indicated ages. No difference is observed at E18.5 among all phenotypes, but a progressive hematopoietic failure is seen as soon as P1 only in Trp53ΔCTD/ΔCTD animals. (B) Liver sections stained with H&E show no difference between E18.5 Trp53+/+, Trp53ΔCTD/+, and Trp53ΔCTD/ΔCTD animals and an impaired EMH (indicated by black arrowheads) starting as soon as P1 only in Trp53ΔCTD/ΔCTD animals. Centrilobular degeneration (yellow arrowheads) is observed over the regions surrounding the central vein (black asterisks). Bar, 200 μm.
Mice normally experience extramedullary hematopoiesis (EMH) in the liver at birth (Wolber et al. 2002), and it ceases around P9–P13. At E18.5, livers from all three genotypes exhibited abundant EMH, characterized by aggregates of mature and immature hematopoietic cells from all three lineages (erythrocytic, granulocytic, and megakaryocytic) (Fig. 3B). At P1, abundant EMH was still apparent in the Trp53+/+ mouse, with decreased amounts in the Trp53ΔCTD/ΔCTD mouse (Fig. 3B). At P10 and P12, scattered EMH was still present in the Trp53+/+ and Trp53ΔCTD/+ mice, but EMH was virtually absent in the Trp53ΔCTD/ΔCTD mice. There was also centrilobular vacuolar degeneration in the Trp53ΔCTD/ΔCTD mice at P10 and P12 that was not apparent in the Trp53+/+ and Trp53ΔCTD/+ mice (Fig. 3B). Centrilobular areas are the last to receive oxygenated blood in the liver and are more susceptible to hypoxic injury. The Trp53ΔCTD/ΔCTD mice are significantly anemic, and their attendant hypoxia might contribute to the observed centrilobular vacuolar degeneration.
The remaining hematopoietic organs were also affected by the deletion of the C terminus of p53. The spleens of Trp53+/+ and Trp53ΔCTD/+ mice were identical in size and cellularity (Figs. 2D, 9C, below) and exhibited abundant EMH in the red pulp and cellular white pulp. The spleens of the Trp53ΔCTD/ΔCTD mice, in contrast, were dramatically reduced in size and cellularity (Figs. 2D, 9C, below) and exhibited sparse EMH and less cellular white pulp (Fig. 9C, below). Likewise, although no differences were noted between the Trp53+/+ and Trp53ΔCTD/+ mice, the thymus was markedly less cellular in the Trp53ΔCTD/ΔCTD mice, especially prominent in the cortex (Fig. 6C, below). Gut-associated lymphoid tissue (GALT) and lymph nodes were not apparent in most of the Trp53ΔCTD/ΔCTD mice, and when present, these tissues were much less cellular than their Trp53+/+ and Trp53ΔCTD/+ counterparts (Supplemental Fig. 2C). Taken together, these data argue that the deletion of the C terminus of p53 has profound consequences for both medullary hematopoiesis and EMH.
Hematopoietic failure can have multiple causes but is often associated with hematopoietic stem cell (HSC) defects. To determine whether the C terminus of p53 played a role in HSC functions, the relative abundance of lineage-negative Lin− Sca1+ cKit+ (LSK) cells (in prenatal livers and P10 whole bone marrow) of Trp53+/+, Trp53ΔCTD/+, Trp53ΔCTD/ΔCTD, and Trp53−/− mice was measured by flow cytometry. Although no significant difference was observed in the relative frequency of LSK cells between the Trp53+/+, Trp53ΔCTD/+, and Trp53ΔCTD/ΔCTD mice in the fetal liver at E14.5 (Fig. 4A; Supplemental Fig. 3A), a statistically significant reduction in the absolute number of fetal liver LSK cells was noted in the Trp53ΔCTD/ΔCTD mice. In accordance with previous studies, the relative numbers of LSK cells were higher in Trp53−/− mice. By comparison, in the whole bone marrow at P10, both the relative frequency and absolute numbers of LSK cells were significantly decreased in the Trp53ΔCTD/+ and Trp53ΔCTD/ΔCTD mice in a dose-dependent manner (Fig. 4A; Supplemental Fig. 3B). In fact, the LSK population in the bone marrow of the Trp53ΔCTD/ΔCTD mice was nearly completely depleted. (Fig. 4A; Supplemental Fig. 3B). To compare the differentiation capacity of stem cells from the wild-type and mutant fetal liver and P10 bone marrow, a colony-forming unit assay was conducted. Interestingly, in accordance with the measurable presence of LSK cells by flow cytometry in the livers of the Trp53ΔCTD/ΔCTD mice, the liver cells cultured from the Trp53+/+, Trp53ΔCTD/+, and Trp53ΔCTD/ΔCTD mice were capable of forming colonies (Fig. 4B), implying that the differentiation capacity of the Trp53ΔCTD/ΔCTD liver LSK cells is intact in vitro. Conversely, no colonies were observed using whole bone marrow cells from Trp53ΔCTD/ΔCTD mice (Fig. 4B; Supplemental Fig. 3C), in accordance with the observed ablation of LSK cells in Trp53ΔCTD/ΔCTD mouse bone marrow measured by flow cytometry. HSCs are highly sensitive to reactive oxygen species (ROS), which limit and impair their functions (He et al. 2009). It has also been shown that ROS can induce the p53 pathway. Nevertheless, cultivating the fetal liver cells in 5% oxygen did not rescue the phenotype observed at 20% oxygen (data not shown).
HSC homeostasis is perturbed in Trp53ΔCTD/ΔCTD mice. (A) Percentages of LSK cells from E14.5 fetal liver and P10 whole bone marrow show decreased numbers of HSCs in the bone marrow. Bars indicate SEM. (*) P < 0.05; (**) P < 0.01; (***) P < 0.001 (one-way ANOVA); (N.S.) nonsignificant. (B) Number of colonies formed by E14.5 fetal liver and P10 whole bone marrow cells from Trp53+/+, Trp53ΔCTD/+, Trp53ΔCTD/ΔCTD, and Trp53−/− animals cultivated on methylcellulose-based medium. Colonies originating from three types of progenitors were counted: erythroid progenitors (burst-forming unit-erythroid [BFU-E]), granulocyte and macrophage progenitors (colony-forming unit-granulocyte and macrophage [CFU-GM]), and multipotential granulocyte, erythroid, macrophage, and megakaryocyte progenitors (CFU-granulocyte, erythroid, macrophage, and megakaryocyte [CFU-GEMM]). (C) The percentage survival of lethally irradiated mice of mixed genetic background (BL6/129Sv) transplanted with E14.5 fetal liver cells from matching BL6/129Sv Trp53+/+, Trp53ΔCTD/+, and Trp53ΔCTD/ΔCTD embryos shows the decreased ability of ΔCTD/ΔCTD cells to recolonize the recipient bone marrow. Animals receiving no transplant were used as negative control. (D) Relative expression of two transcription factors essential for hematopoiesis (Gfi1 and Gfi1b) in P10 livers, bone marrow, spleens, and thymi from Trp53+/+, Trp53ΔCTD/+, Trp53ΔCTD/ΔCTD, and Trp53−/− animals by qRT–PCR reveals impaired activity for the p53ΔCTD mutant. Expression is normalized to Gapdh. In all panels, n = 5; bars indicate SEM; (*) P < 0.05; (**) P < 0.01 (Student's t-test).
To address whether the wild-type or mutant HSC repopulating capabilities were different, fetal liver cells from donor mice with varying genotypes (+/+, ΔCTD/+, or ΔCTD/ΔCTD) were transplanted into lethally irradiated recipient mice of matching mixed genetic background (BL6/129Sv). Less than 70% of the recipients who received either ΔCTD/ΔCTD cells survived, (Fig. 4C), demonstrating that the HSCs from the Trp53ΔCTD/ΔCTD fetal livers are capable of reconstituting the bone marrow of the irradiated recipient, although not as consistently as their wild-type counterparts. This capacity to reconstitute the wild-type recipient mice is in accordance with the in vitro colony-forming capacity of the Trp53ΔCTD/ΔCTD fetal liver cells (Fig. 4B). The reduced ability to reconstitute all irradiated recipients may indicate a cell-intrinsic defect in the Trp53ΔCTD/ΔCTD HSCs or simply reflect the reduced absolute LSK cell numbers among mutant fetal liver cells. These data collectively demonstrate that the deletion of the C terminus of p53 disturbs critical HSC functions, including differentiation and repopulating capacities.
Recent studies have shown that p53 plays a pivotal role in HSC self-renewal (Milyavsky et al. 2010) and quiescence (Liu et al. 2009), in part by activating a subset of specific and relevant target genes. Notably, p53 transcriptionally activates expression of the transcription factor Gfi1, which has been shown with its homolog, Gfi1b, to be one of the key regulators of hematopoiesis (Cellot and Sauvageau 2005; Duan and Horwitz 2005; Liu et al. 2009; van der Meer et al. 2010). Gfi1 and Gfi1b are both expressed in a tissue-specific manner. Although both are expressed in the bone marrow, Gfi1 is expressed exclusively in the thymus, while Gfi1b is expressed in the spleen (Tong et al. 1998). The expression levels of these genes in hematopoietic tissues across different genotypes were measured by quantitative RT–PCR (qRT–PCR) (Fig. 4D). Interestingly, basal levels of Gfi1 were controlled by p53 only in the thymus (Fig. 4D) and to a lesser extent in the spleen, even though the overall expression in the latter was significantly diminished compared with the thymus, as expected. In the bone marrow, on the other hand, Gfi1 basal levels were identical between Trp53+/+, Trp53ΔCTD/+, and Trp53−/− genotypes but were significantly reduced in the Trp53ΔCTD/ΔCTD animals. The same pattern was observed for Gfi1b, with p53 controlling basal levels of Gfi1b in the spleen but not in the bone marrow (Fig. 4D). Thus, in the bone marrow, the dominant source of hematopoiesis in newborn and adult mice, the deletion of the C terminus of p53 resulted in reduced expression of Gfi1 and Gfi1b, two of the most important regulators of hematopoietic homeostasis. These low levels of Gfi1 and Gfi1b might indicate that the pool of cells in which these two genes are normally expressed is absent in the mutant bone marrow. Furthermore, flow cytometric analysis of CD3+, CD19+, and CD11b+ populations revealed an aberrant cellular subset composition in the mutant bone marrow, with an accumulation of CD3-positive cells and no change in CD19+ or CD11b+ cells' relative frequency among all genotypes (Supplemental Fig. 7). It is also possible that abnormal expression of p53 target genes within different hematopoietic lineages may drive shifts in cell populations within organs, which may account for differences in whole-tissue gene expression readouts. This could explain, at least in part, the phenotypic difference observed between Trp53−/− and Trp53ΔCTD/ΔCTD mice and the hematopoietic failure observed in the latter after birth.
Deletion of the C-terminal 24 amino acids from p53 induces senescence in bone marrow cells
To gain further insight into the mechanism of hematopoietic failure in Trp53ΔCTD/ΔCTD mice, p53-dependent senescence and apoptosis were assessed in whole bone marrow cells isolated from P10 animals. First, whole bone marrow cells from P10 animals were stained for senescence-associated β-galactosidase (SA-β-Gal) activity (Fig. 5A,B). The number of SA-β-Gal-positive cells was greatly increased in Trp53ΔCTD/ΔCTD animals, while little was observed between the other genotypes. mRNA levels of several bona fide senescence markers were measured by qRT–PCR. Cdkn1a (p21) and Cdkn2b (p15INK4b), two potent cell cycle inhibitors that have been shown to be overexpressed in senescent cells (Collado and Serrano 2006), were up-regulated in the mutant bone marrow cells (Fig. 5C). Expression of Cdkn1a (p21) is clearly p53-dependent, as its level is markedly reduced in p53-null tissue. The increase in its expression is thus likely due to hyperactivity of p53ΔCTD. This was not the case for several other p53 target genes, including Mdm2, Bbc3 (Puma), Pmaip1 (Noxa), or Tigar (Supplemental Fig. 4). Expression of Cdkn2b (p15INK4b) is also similar between wild-type and null animals. Thus, in the case of Cdkn2b (p15INK4b), the increased levels that are observed (Fig. 5C) are likely an indicator of the enhanced senescence. Two other transcripts originating from the Cdkn2a locus (p19ARF and p16INK4a) did not show any statistically significant difference between Trp53+/+, Trp53ΔCTD/+, Trp53ΔCTD/ΔCTD, and Trp53−/− bone marrow. In addition, bone marrow cells were stained for activated caspase 3, an apoptotic marker (Fig. 5D). MEFs derived from Trp53+/+, Trp53ΔCTD/+, Trp53ΔCTD/ΔCTD, and Trp53−/− E14.5 embryos and treated with the DNA-damaging agent DOX for 24 h were used as positive and negative controls (Fig. 5D, inserts). Although the p53ΔCTD mutant protein retained its ability to induce apoptosis after DNA damage in these cells, no apoptosis was observed in bone marrow cells from all of the genotypes tested (Fig. 5D). These data suggest that although p53ΔCTD retains its ability to induce both senescence and apoptosis, only senescence is involved in the observed bone marrow phenotype.
Deletion of the C-terminal 24 amino acids from p53 induces senescence in bone marrow cells. (A) Quantitative assay of SA-β-Gal activity in P10 bone marrow cells from Trp53+/+, Trp53ΔCTD/+, Trp53ΔCTD/ΔCTD, and Trp53−/− animals reveals a marked increase in senescent cells in the mutant bone marrow. n = 5; bars indicate SEM; (***) P < 0.001 (Student's t-test). (B) Representative pictures of A. (C) Relative expression of several senescence markers in P10 whole bone marrow from Trp53+/+, Trp53ΔCTD/+, Trp53ΔCTD/ΔCTD, and Trp53−/− animals by qRT–PCR. The p53 direct target gene Cdkn1a (p21) is a bona fide senescence marker and is up-regulated in ΔCTD/ΔCTD bone marrow. Cdkn2b but not the two transcripts encoded by Cdkn2a is significantly up-regulated in the ΔCTD/ΔCTD bone marrow. Expression is normalized to Gapdh. In all panels, n = 5; bars indicate SEM; (*) P < 0.05; (**) P < 0.01 (Student's t-test); (N.S.) nonsignificant. (D) Whole bone marrow cells from P10 animals of the genotypes Trp53+/+, Trp53ΔCTD/+, Trp53ΔCTD/ΔCTD, and Trp53−/− stained with an activated caspase 3 antibody, a bona fide marker for apoptotic cells, did not show enhanced apoptosis. Insets represent MEFs derived from E14.5 Trp53+/+, Trp53ΔCTD/ΔCTD, and Trp53−/− embryos and treated with the DNA-damaging agent DOX at 0.2 μg/mL. Twenty-four hours after treatment, cells were fixed and stained with an anti-activated caspase 3. Unlike p53−/− cells, p53ΔCTD/ΔCTD MEFs are capable of undergoing apoptosis after DNA damage to the same extent as p53+/+ or p53ΔCTD/+ MEFs.
In the thymus, the C-terminal domain negatively regulates p53-dependent gene expression by inhibiting DNA binding
The thymi in the Trp53ΔCTD/ΔCTD mice are reduced in size (Fig. 2D) and show severe morphological differences at the microscopic level (Fig. 6C). To examine p53-dependent effects in this tissue, the gene expression of a set of well-characterized p53 targets was compared between wild-type and p53-null animals (Fig. 6A). All targets examined showed clear p53 dependence in their gene expression: Cdkn1a (p21), Mdm2, Bbc3 (Puma), Pmaip1 (Noxa), and Tigar (Fig. 6A). Of these, the wild-type and the p53ΔCTD mice showed little difference in expression of either Cdkn1a (p21), Mdm2, or Tigar. In contrast, both Bbc3 (Puma) and Pmaip1 (Noxa) showed significantly increased expression in the presence of the p53ΔCTD. The effects on p21 and Puma protein expression were confirmed by immunoblotting (Fig. 6B). These differences in gene expression were not due to altered p53 levels, determined by either immunoblotting of tissue extracts (Fig. 6B) or immunohistochemistry (IHC) (Fig. 6C). To gain mechanistic insight into the difference in gene expression, chromatin immunoprecipitation (ChIP) analysis was performed on selected genomic sites. Occupancy of p53 on all sites examined was only seen in the wild-type but not the null tissues (Fig. 6D). Although wild-type and p53ΔCTD were found equivalently associated with the two response elements in the Cdkn1a (p21) promoter, the p53ΔCTD showed markedly increased occupancy on the elements in the Bbc3 (Puma) and Pmaip1 (Noxa) genes (Fig. 6D). Thus, the enhanced Bbc3 (Puma) and Pmaip1 (Noxa) expression seen with p53ΔCTD is accompanied by increased occupancy of the relevant genomic sites. This suggests that in the thymus, the C terminus of p53 negatively regulates DNA binding at least on a subset of target genes.
The C-terminal domain is a negative regulator of p53 activity in the thymus. (A) Relative expression of five p53 direct target genes in P10 thymi from Trp53+/+, Trp53ΔCTD/+, Trp53ΔCTD/ΔCTD, and Trp53−/− animals by qRT–PCR demonstrates tissue-specific up-regulation of the two proapoptotic targets Bbc3 (Puma) and Pmaip1 (Noxa). Expression is normalized to Gapdh. In all panels, n = 5; bars indicate SEM; (*) P < 0.05 (Student's t-test). (B, top) Immunoblot shows comparable p53 expression between the wild-type and mutant thymi at P10. p21 expression is not changed, whereas Puma protein levels are increased in the mutant thymus. β-Actin was used as the loading control. (Bottom) Quantitation of p53 levels (relative intensity of p53 bands vs. actin bands), n = 5; bars indicate SEM; (P) Student's t-test; (N.S.) nonsignificant. (C) P10 Trp53ΔCTD/ΔCTD thymi display reduced size and cellularity but no difference in basal p53 levels, as assessed by H&E staining and IHC, respectively. Bar, 200 μm. (D) ChIP on P10 thymus protein extracts followed by qPCR demonstrates enhanced binding of p53ΔCTD on its response elements within Bbc3 (Puma) and Pmaip1 (Noxa) genes.
Hyperactive p53 lacking the C-terminal domain induces apoptosis but not senescence in thymocytes
To determine whether this enhanced expression of proapoptotic targets has a corresponding phenotypic outcome, thymocytes were stained for either SA-β-Gal activity (as a marker for senescence) or activated (cleaved) caspase 3 (as an indicator of apoptosis) (Fig. 7). Although little SA-β-Gal activity could be detected (Fig. 7B), there is clear evidence of activated caspase 3 in the thymi from ΔCTD/ΔCTD mice but not those harvested from either the wild-type or heterozygous animals (Fig. 7A). Taken together, this indicated that in the thymus, the C terminus of p53 prevents DNA binding to Bbc3 (Puma) and Pmaip1 (Noxa) but not Cdkn1a (p21). This in turn leads to differential gene expression that is consistent with an enhanced apoptosis in these tissues. It is likely, then, that this increased cell death is responsible for the smaller thymus size in the homozygous mutant mice.
Hyperactive p53 lacking the C-terminal domain induces apoptosis but not senescence in thymocytes. (A) IHC on thymus sections from P10 Trp53+/+, Trp53ΔCTD/+, and Trp53ΔCTD/ΔCTD animals reveals a significant increase of activated caspase 3 staining in the mutant thymus. Black arrowheads show activated caspase 3-positive cells. (B) SA-β-Gal activity in P10 thymocytes isolated from Trp53+/+, Trp53ΔCTD/+, and Trp53ΔCTD/ΔCTD animals reveals no difference in the number of senescent cells between wild-type, heterozygous, or homozygous mutant thymi.
In the liver, the C-terminal domain of p53 is required for target gene expression at a step subsequent to DNA binding
Livers from the Trp53ΔCTD/ΔCTD mice are unaffected in size (Fig. 2D) but show altered colorization, most likely due to altered blood cell production (Fig. 2E). To examine p53-dependent effects in this tissue, p53 target gene expression was again compared between wild-type and p53-null animals (Fig. 8A). Cdkn1a (p21), Mdm2, Pmaip1 (Noxa), and Tigar showed clear p53 dependence in their gene expression, while Bbc3 (Puma) did not (Fig. 8A). As in the thymus, comparison of wild-type and p53ΔCTD mice showed little difference in expression of Cdkn1a (p21). In contrast, Mdm2, Pmaip1 (Noxa), and Tigar all showed significantly reduced expression in the presence of the p53ΔCTD. The effects on p21 protein expression were confirmed by immunoblotting (Fig. 8B). These differences in gene expression were not due to altered p53 levels, determined by either immunoblotting of tissue extracts (Fig. 8B) or IHC (Fig. 8C). ChIP analysis showed that occupancy of p53 on all sites examined was only seen in the wild-type but not the null tissues (Fig. 8D). There were no significant differences in promoter occupancy by wild type or p53ΔCTD for either the Cdkn1a (p21), Mdm2, or Pmaip1 (Noxa) genes (Fig. 8D). Intriguingly, the p53ΔCTD showed markedly reduced occupancy on the Tigar gene (Fig. 8D). Thus, the reduced gene expression seen with p53ΔCTD is not reflected in changes in occupancy of the relevant genomic sites. This argues that in the liver, the C terminus of p53 is needed for a subsequent step in transcriptional regulation beyond DNA binding, perhaps at the level of coactivator recruitment.
In the liver, p53 lacking the C-terminal domain is expressed at wild-type levels but is defective in gene expression. (A) Relative expression of five p53 direct target genes in P10 livers from Trp53+/+, Trp53ΔCTD/+, Trp53ΔCTD/ΔCTD, and Trp53−/− animals by qRT–PCR demonstrates tissue-specific down-regulation of all targets except p21. Expression is normalized to Gapdh. In all panels, n = 5; bars indicate SEM; (*) P < 0.05; (**) P < 0.01; (***) P < 0.001; (N.S) nonsignificant (Student's t-test). (B, top) Immunoblot shows comparable p53 expression between wild-type and mutant livers at P10. β-Actin was used as the loading control. (Bottom) Quantitation of p53 levels (relative intensity of p53 bands vs. actin bands); n = 5; bars indicate SEM; (P) Student's t-test; (N.S.) nonsignificant. (C) P10 Trp53+/+ and Trp53ΔCTD/ΔCTD livers display comparable size and cellularity and no difference in basal p53 levels, as assessed by H&E staining and IHC, respectively. Bar, 200 μm. (D) ChIP on P10 liver protein extracts followed by qPCR. In all panels, n = 4; bars indicate SEM; (*) P < 0.05; (**) P < 0.01; (N.S.) nonsignificant (Student's t-test).
In the spleen, p53 lacking the C-terminal domain is hyperactive due to its overexpression
Spleens from the Trp53ΔCTD/ΔCTD mice are reduced in size (Fig. 2D) and show morphological differences at the microscopic level (Fig. 9C). In this tissue, p53-dependent expression of Cdkn1a (p21), Pmaip1 (Noxa), and Tigar is seen, but not for Mdm2 or Bbc3 (Puma) (Fig. 9A). Mice expressing p53ΔCTD showed enhanced expression of Cdkn1a (p21) and Pmaip1 (Noxa) but reduced mRNA levels of Tigar (Fig. 9A). The effects on p21 protein expression were confirmed by immunoblotting (Fig. 9B). In contrast to the thymus and liver, p53ΔCTD levels were substantially higher, as seen by immunoblotting of tissue extracts (Fig. 9B) or IHC (Fig. 9C). There was no detectable evidence of senescence or apoptosis using appropriate assays (Fig. 9D). Thus, the increased protein levels of p53ΔCTD are accompanied by enhanced gene expression on some (Cdkn1a and Pmaip1) but not other (Tigar) genes. The molecular basis for this striking difference remains to be determined. p53ΔCTD is similarly overexpressed in fetal livers (Supplemental Fig. 5A,B). As in the spleen, this is accompanied by enhanced gene expression of Cdkn1a (p21) and Pmaip1 (Noxa) and reduced levels of Tigar mRNA (Supplemental Fig. 5C). In contrast to the thymus and adult liver, in the spleen and fetal liver, the C terminus is needed to maintain appropriate p53 protein levels.
In the spleen, p53 lacking the C-terminal domain is overexpressed and hyperactive in a target gene-selective manner. (A) Relative expression of five p53 direct target genes in P10 spleens from Trp53+/+, Trp53ΔCTD/+, Trp53ΔCTD/ΔCTD, and Trp53−/− animals by qRT–PCR demonstrates target-specific regulation. Expression is normalized to Gapdh. In all panels, n = 5; bars indicate SEM; (*) P < 0.05; (***) P < 0.001; (N.S.) nonsignificant (Student's t-test). (B, top) Immunoblot shows increased p53 expression in the mutant spleen compared with wild type at P10. β-Actin was used as the loading control. (Bottom) Quantitation of p53 levels (relative intensity of p53 bands vs. actin bands); n = 5; bars indicate SEM; (***) P < 0.001 (Student's t-test). (C) Histology of the spleen of P10 animals shows no difference between Trp53+/+ and Trp53ΔCTD/+ organs but reduced size and cellularity in Trp53ΔCTD/ΔCTD animals. IHC on the same tissues reveals a significant and dose-dependent increase of p53 protein expression. Bar, 200 μm. (D, top) SA-β-Gal activity in P10 splenocytes isolated from Trp53+/+, Trp53ΔCTD/+, and Trp53ΔCTD/ΔCTD animals reveals no difference in senescent cell numbers between wild-type, heterozygous, or mutant thymi. (Bottom) IHC on spleen sections from P10 Trp53+/+, Trp53ΔCTD/+, and Trp53ΔCTD/ΔCTD animals reveals no significant increase of activated caspase 3 staining in the mutant spleen.
The p53 C-terminal domain is required for proper cerebellum development in vivo
Aside from hematopoietic failure, another major phenotype observed in Trp53ΔCTD/ΔCTD mice was pronounced ataxia, which suggested the presence of neurological defects. Thus, the brains of Trp53+/+ and Trp53ΔCTD/ΔCTD animals were compared. Overall, the brains of homozygous mutant mice were smaller than those of their wild-type or heterozygous counterparts (Supplemental Fig. 6A; data not shown). The cerebellar vermis in the homozygous mutant brain was virtually nonexistent, revealing more of the colliculi (Supplemental Fig. 6A). In addition, the cerebellar folia appeared shallower with a perturbed foliation pattern, while no significant differences were noted in the cerebrum or brainstem (Supplemental Fig. 6B). Previous studies have shown that p53 levels are critical in the development of the external granular layer (EGL) and internal granular layer (IGL) of the cerebellum (Malek et al. 2011). Accordingly, there was an apparent decrease in the thickness and cellularity of both the EGL and IGL of the Trp53ΔCTD/ΔCTD cerebellum compared with the Trp53+/+ and Trp53ΔCTD/+ counterparts (Supplemental Fig. 6B,C). Of note, the Purkinje cells of the homozygous mutant cerebellum failed to form a uniform monolayer, as observed in the wild-type or heterozygous counterparts (Supplemental Fig. 6C). Interestingly, when assessed by immunoblot, p53 protein levels were increased in the cortex of both Trp53ΔCTD/+ and Trp53ΔCTD/ΔCTD animals compared with Trp53+/+, whereas no difference was observed in the cerebellum of these three genotypes (Supplemental Fig. 6D). This finding argues that the C-terminal domain regulates p53 functions in the cerebellum without interfering with its level of expression. These data collectively demonstrate the importance of C-terminal regulation of p53 in cerebellar development. Further studies will be needed to determine the precise mechanism by which p53 intervenes in this process.
Discussion
The p53 tumor suppressor is a critical mediator of the cellular response to stress and is mutated in >50% of human tumors (Brosh and Rotter 2009). Trp53−/− mice were originally described as developmentally normal but rapidly succumb to spontaneous cancers (mainly lymphomas and sarcomas) within 6–9 mo after birth (Donehower et al. 1992; Jacks et al. 1994). p53 has also been implicated in stem cell maintenance and homeostasis (Zheng et al. 2008; Cicalese et al. 2009; Liu et al. 2009; Spike and Wahl 2011; Bonizzi et al. 2012). These p53 functions may be of particular relevance to tissues that are known to contain a significant pool of adult stem cells (Uccelli et al. 2008; Doulatov et al. 2012;). As p53 has also been shown to regulate reprogramming of somatic cells into induced pluripotent stem cells (iPSCs), these findings argue in support of the idea that one of the roles of p53 is to tightly regulate homeostatic adult tissues. Similarly, p53 has been implicated in brain development (Liu et al. 2007; Terzian et al. 2007; Malek et al. 2011; Mendrysa et al. 2011). Although originally described as developmentally normal, Trp53−/− mice were later shown to have development defects, including impaired maternal reproduction (Hu et al. 2007), aberrant mesenchymal differentiation programs (Molchadsky et al. 2008), and exencephaly (Armstrong et al. 1995; Sah et al. 1995). The molecular basis for these developmental functions of p53 clearly is an important area for further study.
In the present study, it was shown that the deletion of the C-terminal domain of the p53 protein is sufficient to provoke striking postnatal developmental phenotypes in the absence of induced DNA damage, including hematopoietic failure and brain defects. Previous reports of C-terminal Trp53K6R and Trp537KR mice in which six or seven lysines have been mutated to arginines failed to show any significant phenotype compared with wild type (Feng et al. 2005; Krummel et al. 2005). Recently, however, a re-evaluation of the Trp537KR mouse revealed that the C-terminal lysines may be important for HSC survival, although this is only revealed after DNA damage-dependent activation (Wang et al. 2011). While the present study was under consideration, Simeonova et al. (2013) reported that mice expressing a truncated p53 lacking the C-terminal 31 amino acids presented a severe aplastic anemia, leading to death within weeks of birth, similar to what was observed with the p53ΔCTD-expressing mice reported here. These findings and the current data support the hypothesis that p53 C-terminal post-translational modifications are likely to be indispensable for proper p53 function and strongly argue for a central role of p53 in the maintenance and homeostasis of the adult hematopoietic compartment through its basic C-terminal domain. This central and novel role for the C terminus is supported by the observation that the hematopoietic phenotype is triggered by the mere deletion of the C terminus of p53 in the absence of genotoxic stress. Moreover, the data presented here shed light on the mechanisms leading to the hematopoietic failure in Trp53ΔCTD/ΔCTD animals.
According to these data, p53 activity is critical in regulating hematopoietic homeostasis in vivo, in agreement with several previous studies (Liu et al. 2007, 2009; Abbas et al. 2010; Wang et al. 2011; Ceccaldi et al. 2012). Both fetal and postnatal hematopoietic tissues are affected by the C-terminal deletion, although to differing extents. Although the absolute number of fetal liver HSCs is reduced in the mutant animals, these cells are still capable of properly differentiating and, to a certain extent, reconstituting the bone marrow of lethally irradiated recipients. Flow cytometry analysis of gross CD3+, CD19+, and CD11b+ populations in P10 organs revealed aberrant cellular subset composition in postnatal hematopoietic tissues, with an overall increase in CD3+ T-cell populations and decrease in CD19+ B-cell populations of mutant p53 mice (Supplemental Fig. 7). Consistent with their observed colony-forming capacity (Fig. 4B), p53 mutant HSCs were capable of differentiating toward major lymphoid and myeloid blood lineages, yet subsequent lineage development and/or homeostasis is somehow disrupted. Hematopoiesis fails by P10 in these mutant p53 mice, leading to death within a time frame coincident with the transition from liver to bone marrow-predominated hematopoiesis. HSCs migrate from the fetal liver to the bone marrow niche, which provides the appropriate cellular and molecular microenvironment for their self-renewal and differentiation (Suda et al. 2005). While the bone marrow of the Trp53ΔCTD/ΔCTD mice is virtually devoid of HSCs, fetal liver cells from Trp53ΔCTD/ΔCTD mice are capable of engrafting and rescuing lethally irradiated recipient mice, although not as efficiently as wild-type cells. These data raise the intriguing possibility that p53 might also play important roles in such non-cell-autonomous functions as formation or maintenance of the bone marrow stem cell niche. The migration of HSCs to the bone marrow as well as the maintenance of the bone marrow stem cell niche are critically dependent on the chemokine CXCL12 (Suda et al. 2005; Ding and Morrison 2013; Greenbaum et al. 2013). As p53 activation has been demonstrated to attenuate cancer cell migration through repression of CXCL12 (Moskovits et al. 2006), it is tempting to speculate that p53ΔCTD hyperactivity within the bone marrow niche could cause CXCL12 repression, thereby ablating the CXCL12 chemokine gradient and undermining the bone marrow stem cell niche in the Trp53ΔCTD/ΔCTD mice. Further studies will be needed to determine the extent of p53 contribution in these processes.
Mdm2 negatively regulates p53 protein levels through ubiquitination and subsequent proteasomal degradation. Cell-based studies initially demonstrated that the p53 C-terminal lysines are required for this process (Kubbutat et al. 1997, 1998). Studies in vivo later contradicted these results, however, by showing that mutation of these C-terminal lysines that had been shown to be targeted by ubiquitination had no effect on p53 stability (Feng et al. 2005; Krummel et al. 2005). This could reflect species differences between humans and mice as well. Here it is shown that p53ΔCTD protein levels are differentially regulated in a tissue-specific manner. In the thymus, cerebellum, and adult liver, the wild-type and truncated p53 proteins are expressed at comparable levels. In contrast, in the spleen, brain cortex, and fetal liver, the p53ΔCTD protein is found at significantly higher levels. This is reminiscent of previous studies showing that in the setting of a hypomorphic Mdm2 (Malek et al. 2011) or haploinsufficiency for Mdm2 and Mdm4 (Terzian et al. 2007), p53 shows tissue-specific hyperactivity in the hematopoietic compartment and the cerebellum. These findings and the current data support a model in which p53 activity is tightly regulated in specific organs through both its C-terminal domain and its negative regulators, Mdm2 and Mdm4. This is corroborated by one study in which the deletion of the C-terminal domain of p53 decreased the Mdm2–p53 interaction (Poyurovsky et al. 2010). Further investigations will clearly be needed to clarify the relationship between p53 and its negative regulators in Trp53ΔCTD/ΔCTD animals.
Abnormal levels of several relevant markers (including the cell cycle-dependent kinase inhibitors Cdkn2b/p15 and Cdkn1a/p21, the latter being a direct p53 target gene) are correlated with high levels of senescence in the bone marrow of mutant animals. Previous studies have shown that HSC quiescence is maintained by Cdkn1a (p21) (Cheng et al. 2000). These results are substantiated by other studies that have proposed that p53 counteracts stem cell reprogramming by activating a senescence-promoting stress response (Hong et al. 2009; Utikal et al. 2009). Indeed, a similar phenotype was found in mice that express an endogenous mutant p53 (172P) in the absence of Mdm2 (Liu et al. 2007). This particular mutant p53 protein lacks the ability to mediate an apoptotic response but can still up-regulate p21 and cause cell cycle arrest (Ryan and Vousden 1998). This is consistent with the hematopoietic phenotype being mediated through hyperactivity of the p53-dependent cycle arrest pathway via p21. Interestingly, Cdkn2b (p15) has been shown to be repressed by Gfi1, and its knockout in mice leads to monocytosis and predisposition to myeloid leukemia (Basu et al. 2009; Bies et al. 2010). In the Trp53ΔCTD/ΔCTD mice, down-regulation of Gfi1 in bone marrow cells is correlated with higher expression of Cdkn2b (p15) and senescence. It is tempting to speculate that the senescent cells observed in the mutant bone marrow might correspond to HSCs. In fact, the percentage of senescent cells in the Trp53ΔCTD/ΔCTD bone marrow roughly equals the commonly admitted percentage of stem cells among whole bone marrow cells in mice (∼0.05%) (Fig. 5A). Careful analysis of the bone marrow cell composition will be needed to substantiate this hypothesis.
In addition to controlling levels of expression, the C-terminal domain of p53 has been extensively studied in transcriptional regulation (Kruse and Gu 2009). It has alternatively been shown to exert either a negative or positive effect depending on the particular study and conditions (Kruse and Gu 2009; Carvajal and Manfredi 2013). Its highly basic amino acid content is consistent with an interaction with DNA. Indeed, studies have suggested that it can negatively regulate by interfering with sequence-specific DNA binding by the core domain (Hupp et al. 1992; Anderson et al. 1997; Ahn and Prives 2001). Alternatively, it is required to serve as a means for p53 to track along DNA (McKinney et al. 2004). Post-translational modifications within this domain have also been implicated in the binding of specific transcriptional cofactors (Barlev et al. 2001; Mujtaba et al. 2004). Studies with the Trp53ΔCTD/ΔCTD serve to reconcile these apparently contradictory findings in the literature in that the C terminus appears to mediate each of these effects but in a tissue-specific and target gene-specific manner.
In the thymus, wild-type and truncated p53 are not expressed at significantly different levels. Here, the expression of Bbc3 (Puma) and Pmaip1 (Noxa) is enhanced in the homozygous mutant mice (Fig. 6A). This correlates with an increased occupancy on the corresponding genomic sites by ChIP (Fig. 6D). Such a finding is consistent with the idea that in this tissue, the C terminus somehow interferes with sequence-specific DNA binding and thereby attenuates gene expression. This is selective for Bbc3 (Puma) and Pmaip1 (Noxa), as this effect is not seen with Cdkn1a (p21) (Fig. 6). In contrast, in the liver, where, again, wild-type and mutant p53 are not expressed at appreciablely different levels, the mRNA levels for Tigar are significantly decreased (Fig. 8A), and this is associated with less occupancy on the endogenous Tigar gene by the mutant p53 as compared with the FL wild-type protein (Fig. 8D). This latter finding supports a mechanism in which the C terminus is required for sequence-specific DNA binding and subsequent gene activation. Finally, in the liver, Mdm2 shows reduced gene expression in mutant tissues (Fig. 8A), although gene occupancy is comparable between wild-type and truncated p53 proteins (Fig. 8D). This suggests that the differential regulation is at a step subsequent to DNA binding, most likely because of impaired cofactor recruitment. How does one reconcile such tissue- and target-specific effects? It is possible that existing epigenetic landscapes are established in a p53-independent but tissue-specific manner. This combined with the efficiency of post-translational modifications of the C terminus being different between tissues may thus lead to distinct requirements for target gene expression.
It is intriguing that the overt phenotypic effects of the deletion of the C terminus were restricted primarily to the hematopoietic compartment in homozygous mutant mice. (Figs. 2–4). Since many p53 activities are mediated through the modulation of the expression of target genes, an attractive hypothesis is that such tissue-specific phenotypes rely on cell type-specific target genes. The expression pattern of Gfi1 and Gfi1b, two essential regulators of hematopoiesis (Hock et al. 2004; van der Meer et al. 2010), is highly restricted to hematopoietic organs (Tong et al. 1998). Both genes are down-regulated in Trp53ΔCTD/ΔCTD hematopoietic tissues (Fig. 4D). One of these regulators (Gfi1) was previously shown to be a direct p53 target gene involved in the maintenance of HSC quiescence in a p53-dependent manner (Liu et al. 2009). More recently, Gfi1 was shown to directly interact with p53 and impair its proapoptotic functions in thymocytes, suggesting a possible negative feedback loop (Khandanpour et al. 2013). This would support a model in which Gfi1 interacts with the C terminus of p53, thereby impairing p53 functions. Further characterization of the p53/Gfi1b relationship would be informative. In fact, a rapid survey of the promoter and the first intron of this gene revealed several potential bona fide p53-responsive elements in these regions (data not shown).
Recently, small peptides derived from the p53 C terminus have been proposed as anti-tumorigenic therapeutic agents (Snyder et al. 2004). The data here raise concerns about the use of therapeutic approaches based on small molecules that disrupt the Mdm2–p53 interaction (Vassilev 2007). Increases in p53 protein activity may have unintended consequences on the normal hematopoietic tissues of patients.
In summary, the C-terminal domain of p53 was shown to play key roles in the postnatal homeostasis of the hematopoietic compartment and development of the brain in vivo. Further studies are required to establish the precise molecular pathways at play. Nevertheless, the data indicate that HSCs are highly sensitive to p53 activity and that deletion of the C terminus of p53 can affect key functions and ultimately induce abnormal senescence in mutant bone marrow. Furthermore, the exquisite tissue and target specificity that is observed in these mutant p53 mice have important implications for the role of therapies leading to p53 activation in treatment of tumors of differing origins.
Materials and methodsGeneration of the Trp53NEO/NEO and the Trp53ΔCTD/ΔCTD mice
The Trp53NEO/NEO mouse was generated by means of a targeting vector harboring two exon 11s separated by a NEO selection cassette (see Fig. 1A; Supplemental Material). Trp53NEO/NEO mice were crossed with PrmCre mice (a gift from Dr. Philippe Soriano) (O'Gorman et al. 1997) to generate Trp53NEO/+ CRE-expressing males. These males were intercrossed with wild-type C57BL/6J females to generate Trp53ΔCTD/+ heterozygous mice. Ultimately, these heterozygous mice were bred to obtain Trp53ΔCTD/ΔCTD homozygous animals.
Histopathology and IHC
Mice were placed in 10% formalin for 48 h and then decalcified before sectioning into ∼3- to 4-mm coronal sections. Sections were processed and embedded en bloc and stained with haematoxylin and eosin (H&E). One femur from each genotype (Trp53+/+, Trp53ΔCTD/+, and Trp53ΔCTD/ΔCTD) was used to harvest bone marrow for bone marrow brush smears and stained with Diffquik for cytologic evaluation. IHC was performed using the anti-p53 CM5 antibody (Leica Microsystems).
qRT–PCR and immunoblotting
Whole organs were harvested at the indicated times and processed as follows. For RNA preparation, a small sample was excised and submerged in RNAlater RNA stabilization reagent (Qiagen) for later use. The tissue was disrupted and homogenized (PowerGen 125 homogenizer, Fisher Scientific). Total RNA and cDNA were prepared as previously described (Hamard et al. 2012). For protein preparation, a small sample was excised and homogenized into a lysis buffer composed of 50 mM HEPES (pH 7.5), 1% Triton X-100, 150 mM NaCl, 1 mM MgCl2, 1 mM phenylmethylsulfonyl fluoride (PMSF), 5 mg/mL leupeptin, and 50 mg/mL aprotinin. One-hundred micrograms of total protein was resolved by SDS-PAGE. Immunoblot analysis was conducted using the following antibodies: anti-p53 (CM5, Leica Microsystems) and anti-β-actin (Sigma).
Flow cytometry and HSC and fetal liver cell isolation
For determining the percentage of LSK cells, total bone marrow cells were preincubated with 5% rat serum, followed by incubation with c-Kit-FITC, Sca-1-PE, and a biotin-conjugated cocktail containing anti-CD3, anti-CD11b, anti-CD45R/B220, anti-Ly6G, and anti-TER119 (all from BD Biosciences), followed by APC-conjugated streptavidin labeling (eBiosciences). At the end of incubation, the stained cells were treated with DAPI to exclude dead cells and were acquired on either a LSR-II or Fortessa (BD Biosciences). For fetal liver cell staining, pregnant mice were sacrificed at E13.5, fetal liver cells were isolated, and single-cell suspensions were prepared by passing through a cell strainer (70 μM). For fetal liver LSK analysis, anti-CD11b was excluded from the multilineage cocktail. All of the data were analyzed with FlowJo 9.5 software (Tree Star).
Colony-forming unit assay
About 2 × 105 fetal liver cells from E14.5 fetuses of different genotypes were grown on methylcellulose-based medium (MethoCult M3434, Stem Cell Technologies) and maintained in minihumidity chambers in 20% or 5% oxygen tissue culture incubators. After 2 wk, colonies were classified and enumerated based on their morphology. Three types of colonies were counted: colony-forming unit-granulocyte and macrophage (CFU-GM), burst-forming unit-erythroid (BFU-E), and CFU-granulocyte, erythroid, macrophage, and megakaryocyte (CFU-GEMM).
SA-β-Gal assay
Bone marrow cells from P10 whole bone marrow were collected and cytospun on slides. Cells were stained with Senescence-Associated β-Galactosidase Staining kit (Cell Signaling) according to the manufacturer's protocol. Using an inverted microscope and a camera, the number of blue cells were counted and averaged, and a representative picture was taken.
Activated caspase 3 assay
Bone marrow cells from P10 whole bone marrow were collected and cytospun on slides. MEFs derived from E14.5 Trp53+/+, Trp53ΔCTD/+, Trp53ΔCTD/ΔCTD, or Trp53−/− embryos were grown on coverslips and treated with DOX (Sigma) for 24 h. Both cell types were stained with an anti-cleaved caspase 3 antibody (Cell Signaling, no. 9661S), and immunofluorescence analysis was conducted as previously described (Hamard et al. 2012).
Transplantation experiments
C57/B6 mice were irradiated with a split dose of 6 Gy, 4 h apart. Three hours after the second dose, 0.5 × 106 fetal liver cells from E14.5 Trp53+/+, Trp53ΔCTD/+, Trp53ΔCTD/ΔCTD, or C57/B6 embryos were injected retro-orbitally into ketamine-sedated mice. Mice were monitored for survival on a daily basis. For irradiation control, four mice were irradiated without transplantation for each experiment.
Statistics
Data are represented as means and SEM. Unless otherwise indicated, all experiments were performed in triplicate. A two-tailed student's t-test was used for comparison between two groups. P < 0.05 was considered significant. One-way ANOVA was used for comparison between more than two groups. P < 0.05 was considered significant.
ChIP assay on tissues
Briefly, the organs were dissected from P10 animals, and a single-cell suspension was prepared by crushing them through a 70-μm nylon cell strainer (BD Falcon, catalog no. 352350). Cells were cross-linked in 10 mL of 1% formaldehyde (EM Science, no. FX0415-5) for 10 min at room temperature. The cross-linking reaction was stopped by adding 0.5 mL of 2.5 M glycine to a final concentration of 125 mM for 5 min at room temperature. Cells were spun at 2000 rpm for 5 min, washed once in 1× PBS, and spun again. The remainder of the protocol was previously described (Carvajal et al. 2012) and adapted from Espinosa et al. (2003) using Protein A Dynabeads (Invitrogen) for the pull-down.
Acknowledgments
We thank L. Resnick-Silverman for her help in many aspects of this study. The generation of the embryonic stem cell clones and mouse model was performed in the Transgenic Shared Cored Facility at Mount Sinai. In that regard, K. Kelley is thanked for his expertise in generating the mouse model and his expert advice regarding the maintenance and manipulation of the animals. This work would not have been possible without the enthusiastic advice of P. Soriano and G. Lozano. This work was supported by P01 CA080058 from the National Cancer Institute to J.J.M. and S.A.A. L.A.C. and E.S. were supported by T32 CA078207, also from the National Cancer Institute.
Supplemental material is available for this article.
Article is online at http://www.genesdev.org/cgi/doi/10.1101/gad.224386.113.
Heterochromatin spreading leads to gene silencing, and boundary elements constrain such spreading. IRC inverted repeats are required for boundary function at centromeric heterochromatin in fission yeast. Jia and colleagues now identify BET family homolog Bdf2 as required for heterochromatin boundary function at IRCs. Bdf2 interacts with boundary protein Epe1, recognizes acetylated histone H4 tails, and antagonizes Sir2-mediated deacetylation of histone H4K16. This study illustrates a mechanism for establishing chromosome boundaries through recruitment of a factor that protects euchromatic histone modifications.
Heterochromatin spreading leads to the silencing of genes within its path, and boundary elements have evolved to constrain such spreading. In fission yeast, heterochromatin at centromeres I and III is flanked by inverted repeats termed IRCs, which are required for proper boundary functions. However, the mechanisms by which IRCs prevent heterochromatin spreading are unknown. Here, we identified Bdf2, which is homologous to the mammalian bromodomain and extraterminal (BET) family double bromodomain proteins involved in diverse types of cancers, as a factor required for proper boundary function at IRCs. Bdf2 is enriched at IRCs through its interaction with the boundary protein Epe1. The bromodomains of Bdf2 recognize acetylated histone H4 tails and antagonize Sir2-mediated deacetylation of histone H4K16. Furthermore, abolishing H4K16 acetylation (H4K16ac) with an H4K16R mutation promotes heterochromatin spreading, and mimicking H4K16ac by an H4K16Q mutation blocks heterochromatin spreading at IRCs. Our results thus illustrate a mechanism of establishing chromosome boundaries at specific sites through the recruitment of a factor that protects euchromatic histone modifications. They also reveal a previously unappreciated function of H4K16ac in cooperation with H3K9 methylation to regulate heterochromatin spreading.
BETH4K16acetylationboundaryheterochromatin
In eukaryotes, genomic DNA is folded with histone and nonhistone proteins in the form of chromatin. Chromatin regulates diverse DNA-based processes such as gene transcription, recombination, and DNA damage repair (Bannister and Kouzarides 2011; Zentner and Henikoff 2013). Based on the level of compaction, chromatin is categorized as euchromatin or heterochromatin. Euchromatin is gene-rich and usually associated with active transcription, whereas heterochromatin is gene-poor and highly compacted and limits the access of transcription and recombination machinery (Grewal and Jia 2007; Beisel and Paro 2011). Heterochromatin has the intrinsic ability to spread in a sequence-independent manner, inactivating genes over long distances (Talbert and Henikoff 2006; Grewal and Jia 2007; Moazed 2011). Thus, to maintain stable gene expression patterns, it is essential to protect euchromatin from encroachment by heterochromatin (Gaszner and Felsenfeld 2006; Valenzuela and Kamakaka 2006).
Heterochromatin spreading depends on the self-propagation of heterochromatin-associated histone modifications and chromatin proteins (Rusche et al. 2003; Talbert and Henikoff 2006; Moazed 2011). Histones within heterochromatin are methylated on H3 Lys9 (H3K9me), which serves as a signal for the recruitment of heterochromatin protein 1 (HP1) family proteins (Rea et al. 2000; Bannister et al. 2001; Nakayama et al. 2001). HP1 recruits the H3K9 methyltransferase SUV39 (Schotta et al. 2002; Stewart et al. 2005), leading to methylation of adjacent nucleosomes. This positive feedback loop induces heterochromatin spreading outwards from its nucleation sites (Talbert and Henikoff 2006). Histones within heterochromatin regions are also hypoacetylated. In budding yeast, which lacks H3K9me and HP1 proteins, the Sir2/Sir3/Sir4 complex mediates heterochromatin spreading in a similar fashion, where Sir2/Sir4-mediated deacetylation of histone H4K16 and recruitment of Sir3 form a similar self-propagating mechanism for heterochromatin spreading (Kurdistani and Grunstein 2003; Rusche et al. 2003; Moazed 2011). Given that H4K16 acetylation (H4K16ac) status directly affects the compaction levels of nucleosome arrays in vitro (Shogren-Knaak et al. 2006), modulation of H4K16ac levels could potentially be a universal mechanism to regulate heterochromatin assembly and spreading. However, functional homologs of Sir3 and Sir4 are absent from organisms using H3K9me/HP1 systems to assemble heterochromatin, and the role of H4K16ac in regulating heterochromatin spreading outside of budding yeast is unknown.
Spreading of heterochromatin is limited by the availability of silencing proteins and competition between positive and negative regulatory components. As a result, the heterochromatin–euchromatin border is often not precisely defined but instead exhibits an extended transition zone (Kimura and Horikoshi 2004). In certain cases, heterochromatin is confined by specific DNA sequences termed boundary elements, which are characterized by sharp transitions in histone modification profiles (Gaszner and Felsenfeld 2006; Valenzuela and Kamakaka 2006). When the functions of such boundaries are disrupted, heterochromatin spreads and silences genes located outside of boundaries. Thus, boundary elements are essential for maintaining stable gene expression patterns. The mechanisms that specify heterochromatin boundaries are diverse, but most function by either recruiting histone-modifying enzymes to directly antagonize heterochromatin-associated histone modification cycles or tethering chromatin to the nuclear envelope to physically separate different chromatin domains (Gaszner and Felsenfeld 2006; Valenzuela and Kamakaka 2006).
Heterochromatin assembly has been extensively studied in fission yeast. Constitutive heterochromatin is mainly present at the pericentric, subtelomeric, and silent mating type regions of the fission yeast genome. All of these regions contain similar DNA elements that serve as nucleation centers of heterochromatin assembly through the RNAi pathway (Moazed 2009; Lejeune and Allshire 2011; Castel and Martienssen 2013). Additional DNA-binding proteins such as Atf1/Pcr1 and Taz1 promote heterochromatin nucleation within the silent mating type region and telomeres, respectively (Jia et al. 2004; Kanoh et al. 2005). These nucleation sites recruit the histone methyltransferase Clr4 complex (CLRC), leading to H3K9me and the recruitment of the HP1 homolog Swi6 (Nakayama et al. 2001; Hong et al. 2005; Horn et al. 2005; Jia et al. 2005). Once initiated, heterochromatin spreads from these nucleation centers into neighboring regions (Hall et al. 2002; Kanoh et al. 2005). The silent mating type and pericentric regions are surrounded by special DNA sequences that correspond to sharp transitions of histone modification profiles (Noma et al. 2001; Cam et al. 2005). At the mating type region, heterochromatin is flanked by two inverted repeats that recruit TFIIIC (Noma et al. 2006), the machinery required for RNA polymerase III (Pol III)-mediated transcription (Paule and White 2000). TFIIIC tethers its binding sites to the nuclear periphery, a process that may contribute to formation of heterochromatin boundaries (Noma et al. 2006). Some pericentric heterochromatin borders correlate with the presence of tRNA gene clusters (Cam et al. 2005), which are critical for boundary function (Scott et al. 2006). Given that Pol III transcribes tRNA genes, it is possible that these tRNA genes mediate boundary formation through a TFIIIC-based mechanism. There are also inverted repeats flanking centromeres I and III termed IRCs, which are also required for boundary activity (Noma et al. 2006), but the mechanism by which IRCs define heterochromatin boundaries is unknown.
Epe1 is a JmjC domain-containing protein required for boundary function (Ayoub et al. 2003). It is recruited to all heterochromatin domains in a Swi6-dependent manner (Zofall and Grewal 2006; Trewick et al. 2007; Sadaie et al. 2008) but is concentrated at borders of heterochromatin, such as IRCs, due to the degradation of Epe1 in the middle of heterochromatin by the proteasome, mediated by the Cul4–Ddb1 E3 ubiquitin ligase complex (Braun et al. 2011). Although JmjC domain proteins are generally involved in histone demethylation, Epe1 does not possess such activity (Tsukada et al. 2006), and the mechanism by which Epe1 functions in blocking heterochromatin spreading is still unknown.
By systematically screening a fission yeast deletion library for mutations that result in heterochromatin boundary defects, we identified a novel protein, Bdf2, required for heterochromatin boundary function at IRCs. Bdf2 interacts with Epe1 and is recruited to IRCs in an Epe1-dependent manner. Through its double bromodomains, Bdf2 preferentially interacts with acetylated histone H4 tails and protects them from Sir2-mediated deacetylation of H4K16. Our results demonstrate that protection of a euchromatic histone modification by a chromatin modification binding protein at a specific DNA element is sufficient to establish a heterochromatin boundary. We also uncovered a previously unappreciated role of H4K16ac in regulating heterochromatin spreading in a system where heterochromatic silencing is dominated by H3K9me and HP1.
ResultsIRC1 is a heterochromatin boundary
The right border of pericentric heterochromatin on chromosome I corresponds to the presence of an IRC element (IRC1R). There are neither any tRNA genes nor any detectable TFIIIC enrichment (Noma et al. 2006), making it an ideal location to study IRC function. We inserted a ura4+ reporter gene to the right of IRC1R (IRC1R∷ura4+) (Fig. 1A). Because heterochromatin spreading depends on the dosage of heterochromatin proteins, overexpression of Swi6 has been used to improve heterochromatin spreading at the silent mating type region and centromeres (Noma et al. 2001, 2006). Therefore, we inserted an additional copy of swi6+ at the ade6 locus to enhance heterochromatin spreading. In wild-type cells, IRC1R∷ura4+ was fully expressed, and cells could not grow on counterselective medium containing 5-fluoroorotic acid (FOA), which is toxic to cells expressing Ura4 (Fig. 1B). However, when IRC1R was deleted, the spreading of heterochromatin led to silencing of this reporter, as evidenced by increased growth on FOA medium (Fig. 1B). This silencing was eliminated in clr4Δ cells (Fig. 1B), suggesting that Clr4-mediated H3K9me is critical for heterochromatin spreading.
Bdf2 is required for boundary function. (A,C,G) Schematic diagrams of the IRC1R∷ura4+, L5-IRC1R-ura4+, and IRC3L∷ura4+ reporters. (B,D,H) Tenfold serial dilution analyses of the indicated yeast strains were grown on the indicated medium to measure the spreading of heterochromatin into the ura4+ reporter. (E,I) ChIP analysis of H3K9me2 and Swi6 levels at the ura4+ reporters, normalized to act1. The data are averages of three experiments, and error bars represent standard deviation. (F) qRT–PCR analysis of emc5 transcript levels, normalized to act1. Data presented are averages of three experiments, and error bars represent standard deviation.
Because the effect of IRC1RΔ on heterochromatin spreading might be a result of moving the ura4+ reporter closer to heterochromatin, we also tested whether IRC1R is sufficient to block heterochromatin spreading. Ectopic insertion of a part of a pericentric repeat (L5) induces silencing of an adjacent ura4+ reporter gene (Partridge et al. 2002; Sadaie et al. 2008). Interestingly, when we inserted IRC1R between L5 and ura4+ (Fig. 1C), heterochromatin spreading was blocked, as indicated by the failure of cells to grow on FOA medium (Fig. 1D). Thus, IRC1R is both necessary and sufficient for boundary function.
Bdf2 is required for proper boundary function
To identify factors required for boundary function, we screened a fission yeast haploid deletion library containing ∼3500 individual deletions of nonessential genes (Kim et al. 2010) for mutants that showed heterochromatin spreading (Supplemental Fig. S1). Our screen identified epe1Δ, which is known to be required for boundary function (Ayoub et al. 2003; Zofall and Grewal 2006; Trewick et al. 2007), as well as an uncharacterized null mutant of SPAC631.02, which has been named bdf2+ (Garabedian et al. 2012) and nrc1+ in PomBase. The Bdf2 protein contains two bromodomains and is homologous to the mammalian bromodomain and extraterminal (BET) family of double bromodomain proteins (Supplemental Fig. S2; Belkina and Denis 2012; Prinjha et al. 2012), which were recently identified as therapeutic targets for a number of cancers (Dawson et al. 2011; Zuber et al. 2011).
We confirmed that bdf2Δ resulted in heterochromatin spreading into IRC1R∷ura4+, as indicated by growth on FOA plates (Fig. 1B). Furthermore, heterochromatin hallmarks such as H3K9 dimethylation (H3K9me2) and Swi6 spread to the reporter gene, as indicated by chromatin immunoprecipitation (ChIP) analyses (Fig. 1E). Histone 3 levels at the reporter were unaffected by bdf2Δ (Supplemental Fig. S3A), indicating that the change in H3K9me2 levels was specific for this modification. Moreover, the expression level of emc5, which is the gene immediately adjacent to IRC1R, was reduced in bdf2Δ cells (Fig. 1F), suggesting that heterochromatin spreading regulates the expression of neighboring genes.
Similarly, bdf2Δ cells showed heterochromatin spreading into ura4+ inserted outside of the centromere III heterochromatin boundary (IRC3L∷ura4+) (Fig. 1G–I) and the silent mating type region heterochromatin boundary (IRR∷ura4+) (Supplemental Fig. S4A,B; Singh and Klar 2008). Given that IRs and IRCs do not share sequence homology, our results indicate that Bdf2 is required for proper boundary functions at different locations, independent of the presence of IRCs.
Bdf2 is enriched at heterochromatin boundaries
To examine the cellular functions of Bdf2, we generated a C-terminally Flag-tagged version of Bdf2 expressed from its endogenous chromosomal location. Bdf2-Flag was functional, as no defects in boundary function were observed (Supplemental Fig. S5A). We performed ChIP–chip analyses of Bdf2-Flag and H3K9me2 using an Agilent whole-genome microarray with additional customized probes that provide high coverage of centromeres. Bdf2 was enriched at all IRC elements, which correspond to sharp decreases in H3K9me2 levels (Fig. 2A,E). The localization of Bdf2 to IRCs was further confirmed by ChIP and quantitative real-time PCR (qPCR)-based analyses (Fig. 2B,F). Bdf2 also localized at the right border of centromere II heterochromatin, which correlates with a sharp drop of H3K9me2 levels but does not contain any IRC element (Fig. 2C,D). Interestingly, Bdf2 was not enriched at the left side of the centromere II heterochromatin boundary (Fig. 2C). We also found that Bdf2 localized at the mating type region boundary IR, although at lower levels compared with centromere boundaries (Supplemental Fig. S4C). Moreover, we found that Bdf2 also localized selectively at the promoter regions of a small group of genes (Supplemental Fig. S6A,B), indicating a possible role of Bdf2 in transcriptional regulation in addition to its role in boundary function.
Bdf2 is enriched at IRCs. (A,C,E) ChIP–chip analysis of Bdf2-Flag and H3K9me2 levels across centromeres I, II, and III. The data are averages of two microarrays. Black arrows in the diagrams indicate the positions of IRCs. Bars indicate tRNA genes. Note that centromere II does not contain IRCs. (B,D,F) ChIP-qPCR analysis of Bdf2-Flag levels at boundary regions, normalized to act1. The positions of amplified regions used for qPCR quantification are indicated. The data are averages of three experiments, and error bars represent standard deviation.
Bdf2 interacts with Epe1
To investigate further the function of Bdf2, we performed affinity purification of Bdf2-Flag and identified interacting proteins by mass spectrometry (Supplemental Fig. S7A; Supplemental Table S1). Significantly, among the proteins identified that specifically associated with Bdf2 but were absent in control purifications was the boundary protein Epe1 (Fig. 3A), consistent with Bdf2's boundary function. Bdf2 was also associated with a number of TAFs (TBP-associated factors) (Supplemental Fig. S7B), which might explain the promoter localization of Bdf2 in euchromatic regions. ChIP analyses of Taf10-Flag indicated that TAFs are not enriched at heterochromatin boundaries (Supplemental Fig. S7C). Conversely, affinity purification of Epe1-Flag also identified Bdf2 (Fig. 3A; Supplemental Table S1). To confirm the interaction between Bdf2 and Epe1, we performed coimmunoprecipitation analysis with extracts from cells expressing both Bdf2-myc and Epe1-Flag from their endogenous chromosomal loci. When Epe1-Flag was immunoprecipitated with anti-Flag-agarose resin, Bdf2 specifically copurified as well, demonstrating that Bdf2 interacts with Epe1 in vivo (Fig. 3B). Moreover, the interaction persisted when the cell lysates were treated with ethidium bromide (Fig. 3B), suggesting that the interaction between Epe1 and Bdf2 is direct rather than DNA-mediated.
Epe1 recruits Bdf2 to IRCs. (A) Multidimensional protein identification technology (MudPIT) mass spectrometry analyses of purified Bdf2-Flag and Epe1-Flag complexes. The number of Bdf2 and Epe1 peptides identified and the percentage of each protein that these peptides cover are indicated. (B,G) Cell extracts from the indicated strains were incubated with Flag-agarose beads to isolate Epe1-Flag. Bound proteins were resolved by SDS-PAGE, and Western blot analyses were performed with Myc and Flag antibodies. (C,I) Serial dilution analyses were performed to measure heterochromatin spreading outside of IRC1R∷ura4+. (D,H) ChIP analysis of Bdf2-Flag and Epe1-Flag levels at IRCs, normalized to act1. The data are averages of three experiments, and error bars represent standard deviation. (E) Diagram of Bdf2 constructs used in yeast two-hybrid analysis. (F) Yeast two-hybrid analysis of Bdf2 with Epe1. Bdf2 was fused with the Gal4 DNA-binding domain, and Epe1 was fused with an activation domain. Interaction between Bdf2 and Epe1 resulted in the activation of a HIS3 reporter gene, allowing cells to grow in the absence of histidine.
Epistasis analysis showed that a bdf2Δ epe1Δ double mutant resulted in heterochromatin spreading similar to that of epe1Δ, suggesting that Bdf2 and Epe1 function in the same pathway (Fig. 3C). We also consistently observed that epe1Δ resulted in a slightly greater increase in heterochromatin spreading than bdf2Δ, suggesting additional roles for Epe1 in regulating boundary functions (Fig. 3C).
To examine the relationship between Bdf2 and Epe1, we performed ChIP analysis of Bdf2-Flag in an epe1Δ strain. Interestingly, Bdf2 localization at IRC1 and IRC3 was lost or significantly reduced in epe1Δ cells (Fig. 3D). Bdf2 protein levels and Bdf2 localization to the promoter of oca8 were not affected by epe1Δ (Supplemental Figs. S6C, S8), indicating that Epe1 only regulates the localization of Bdf2 at IRCs. Epe1 is recruited to heterochromatin by Swi6 (Zofall and Grewal 2006; Trewick et al. 2007; Sadaie et al. 2008; Braun et al. 2011). As expected, Bdf2 was also delocalized from IRC1 and IRC3 in swi6Δ cells (Fig. 3D). On the other hand, the localization of Epe1 to IRCs was not affected by bdf2Δ (Fig. 3D). These results suggest that Epe1 functions upstream of Bdf2 for localization to IRCs.
We further demonstrated that Bdf2 and Epe1 interacted in a yeast two-hybrid assay (Fig. 3E,F), suggesting that the interaction between these two proteins might be direct. To determine the region of Bdf2 that mediates interaction with Epe1, we generated Bdf2 fragments that lack the N-terminal region (ΔN), the C-terminal region (ΔC), or both (BD) (Fig. 3E). We found that Bdf2 lacking the C-terminal region failed to interact with Epe1 in yeast two-hybrid assays (Fig. 3F). Moreover, coimmunoprecipitation showed that Bdf2-ΔC-myc interaction with Epe1-Flag was strongly reduced (Fig. 3G), suggesting that the C-terminal region of Bdf2 is required for Bdf2–Epe1 interaction in vivo. As expected, Bdf2-ΔC localization to IRC1 was reduced (Fig. 3H), concomitant with heterochromatin spreading into IRC1R∷ura4+ (Fig. 3I).
The bromodomains of Bdf2 are required for boundary function
Bromodomains are well known for their ability to bind acetylated lysines (Mujtaba et al. 2007). Bdf2 contains two bromodomains (Fig. 4A), but their binding specificities are unknown. Binding assays of recombinant GST-Bdf2-BD (amino acids 229–497, encompassing both bromodomains) with an array of histone tail peptides containing different combinations of post-translational modifications showed that Bdf2 preferentially binds to multiply acetylated histone H4 tail peptides (Supplemental Fig. S9). Such results are consistent with those of a systematic study of BET family bromodomains, which indicates that they usually have broad specificity and preferentially bind to multiply acetylated histone tails (Filippakopoulos et al. 2012). To confirm this finding, we generated recombinant His-Bdf2-BD and examined its binding to a tetra-acetylated histone H4 tail peptide (K5, K8, K12, and K16) and a diacetylated histone H3 peptide (K9 and K14). Bdf2-BD bound strongly to the acetylated H4 peptide but not to its unacetylated form (Fig. 4B). Moreover, mutations of two amino acids that are predicted to form acetyl-lysine-binding sites, one in each of the bromodomains (Y268A and Y430A, hereafter denoted as 2YA) (Supplemental Fig. S2; Mujtaba et al. 2007), abolished this binding, indicating that Bdf2 binds to acetylated histone tails through its bromodomains (Fig. 4B). Bdf2 had little affinity for the diacetylated histone H3 peptide or any of the singly acetylated H4 peptides (Fig. 4B; data not shown), so it seems to prefer multiply acetylated H4 tails.
The bromodomains of Bdf2 are required for boundary function. (A) Schematic diagram of Bdf2 protein. The positions of mutations are indicated. (B) Peptide pull-down assays of the indicated recombinant proteins with biotinylated H4 tail (2–24) unmodified and tetra-acetylated at the K5, K8, K12, and K16 (Ac-H4) peptides as well as H3 tail (1–21) unmodified and diacetylated at the K9 and K14 (Ac-H3) peptides. (C) Serial dilution analysis to measure heterochromatin spreading outside of IRC1R∷ura4+. (D,E) ChIP analysis of H3K9me2 and Swi6 levels at IRC1R∷ura4+ and Bdf2-Flag levels at IRC1R and IRC3L, normalized to act1. The data are averages of three experiments, and error bars represent standard deviation. (F) Western blot analysis of Bdf2-Flag levels. (G) Schematic diagram of the L5-gbs-ura4+ reporter. (H) Serial dilution analysis to measure heterochromatin spreading to ura4+. All strains used contain epe1Δ to rule out the possibility that Bdf2 recruits Epe1 to establish heterochromatin boundaries. (I) ChIP analysis of H3K9me2 and Swi6 levels at L5-gbs-ura4+ and Bdf2-GBD levels at 3xgbs, normalized to act1. The data are averages of three experiments, and error bars represent standard deviation. All strains used contain epe1Δ.
To examine further the role of bromodomains of Bdf2 in boundary function, we introduced the 2YA mutations at the endogenous bdf2+ chromosomal location. Interestingly, boundary function was compromised to a degree similar to that of bdf2Δ, as indicated by silencing of IRC1R∷ura4+ (Fig. 4C) and the spreading of H3K9me2 and Swi6 into the reporter gene (Fig. 4D). These results suggest that binding of acetylated histone H4 tails by Bdf2 is critical for boundary function. The mutations had no effect on Bdf2 protein levels (Fig. 4F), and Bdf2 still localized at IRCs, although at lower levels compared with that of wild type (Fig. 4E). This result suggests that binding of Bdf2 to acetylated histones also stabilizes Bdf2 at IRCs. Neither the Y268A nor the Y430A single mutant affected heterochromatin boundary function (Supplemental Fig. S5B), suggesting that the two bromodomains have redundant roles, at least in regulating boundary formation at IRCs.
Tethering Bdf2 to DNA is sufficient to establish a heterochromatin boundary
We further tested whether recruiting Bdf2 was sufficient to establish a heterochromatin boundary independent of Epe1. We inserted three copies of the Gal4-binding site (gbs) between L5 and ura4+ to create the L5-gbs-ura4+ reporter (Fig. 4G). In the absence of Epe1, Bdf2-GBD, but not Bdf2-2YA-GBD, was able to block the spreading of heterochromatin from L5 to ura4+, as indicated by silencing assays (Fig. 4H) and ChIP analyses of H3K9me2 and Swi6 (Fig. 4I). Bdf2-GBD and Bdf2-2YA-GBD were equally enriched at gbs (Fig. 4I), indicating that it is not the binding of Bdf2 per se but rather its ability to bind acetylated histone H4 tails that is essential for boundary activity.
The spreading of heterochromatin from nucleation sites depends on not only Clr4-mediated H3K9me and binding of Swi6, but also Sir2-mediated histone deacetylation (Hall et al. 2002; Shankaranarayana et al. 2003; Zhang et al. 2008; Buscaino et al. 2013;). However, sir2Δ has only minor effects on silencing of reporter genes inserted at pericentric heterochromatin (Shankaranarayana et al. 2003; Freeman-Cook et al. 2005; Alper et al. 2013; Buscaino et al. 2013), and it is not known whether Sir2 is required for heterochromatin spreading outside IRCs. We generated bdf2Δ sir2Δ and epe1Δ sir2Δ strains with the IRC1R∷ura4+ reporter. In both cases, heterochromatin failed to spread to the reporter gene, as indicated by silencing assays and ChIP analyses (Fig. 5A–D), suggesting that heterochromatin spreading outside of its boundaries depends on Sir2.
Bdf2 counteracts Sir2-mediated histone deacetylation. (A,C) Serial dilution analysis to measure heterochromatin spreading outside of IRC1R∷ura4+. (B,D,E) ChIP analysis of H3K9me2 and Swi6 levels at IRC1R∷ura4+ and H4K16ac levels at IRC1, normalized to act1. The data are averages of three experiments, and error bars represent standard deviation. (F, left) Coomassie-stained gel of recombinant Sir2. (Right) HDAC assays were performed with recombinant Sir2, a tetra-acetylated histone H4 peptide, and recombinant Bdf2. The production of nicotinamide was measured via the generation of nicotinic acid and free ammonia by nicotinamidase.
The role of H4K16ac in regulating heterochromatin spreading
The major substrates of Sir2 are H3K9ac and H4K16ac in vitro (Shankaranarayana et al. 2003; Alper et al. 2013). Since Clr4-mediated H3K9me is critical for heterochromatin assembly (Rea et al. 2000; Nakayama et al. 2001), it is conceivable that Sir2 functions to generate deacetylated H3K9 for subsequent action by Clr4. However, whether deacetylation of H4K16 is also involved in heterochromatin spreading is unknown. The fact that Bdf2 and Epe1 counteract Sir2 for boundary function and that Bdf2 binds to acetylated histone H4 instead of H3 tail suggests that H4K16ac also plays a role in regulating heterochromatin spreading. Indeed, ChIP analysis showed that H4K16ac levels at IRC1 were significantly reduced in epe1Δ and bdf2Δ cells (Fig. 5E). Moreover, H4K16ac levels were increased in sir2Δ cells but restored to near wild-type levels in epe1Δ sir2Δ and bdf2Δ sir2Δ cells (Fig. 5E). Furthermore, recombinant Sir2 showed robust activity in an in vitro histone deacetylase (HDAC) assay when a tetra-acetylated histone H4 peptide was used as substrate, and this activity was strongly reduced when recombinant Bdf2, but not Bdf2-2YA, was added to the reaction (Fig. 5F).
If the function of Bdf2 is to protect H4K16ac, we expect that H4K16ac levels will be higher at heterochromatin boundaries where Bdf2 is present. To test this idea, we performed ChIP–chip analyses of H4K16ac at heterochromatin regions. H4K16ac levels were low in the middle of heterochromatin, which is consistent with the fact that heterochromatin is generally devoid of histone acetylation. Interestingly, however, H4K16ac levels spiked at heterochromatin boundaries, corresponding to Bdf2 peaks (Fig. 6A–C).
H4K16ac modulates heterochromatin boundary function. (A–C) ChIP–chip analysis of H4K16ac levels across centromeres I, II, and III. The Bdf2-Flag profile is also shown for comparison. (D) Serial dilution analysis to measure heterochromatin spreading outside of IRC1R∷ura4+. All strains used are in an h4.1Δ/h4.3Δ background. (E,F) ChIP analysis of H3K9me2 and Swi6 levels at IRC1R∷ura4+ and Bdf2 levels at IRC1, normalized to act1. The data are averages of three experiments, and error bars represent standard deviation. All strains used are in an h4.1Δ/h4.3Δ background.
To test directly whether deacetylation of H4K16 is required for heterochromatin spreading in fission yeast, we generated an H4K16Q mutation, which mimics H4K16ac. Fission yeast contains three pairs of H3/H4 genes, and deleting two of the H3/H4 pairs (h4.1Δ/h4.3Δ) allows the production of histone H4 from the remaining single copy (Mellone et al. 2003). In this background, histone H4 levels were reduced, as indicated by Western blot analyses (Supplemental Fig. S10A). Heterochromatin still spread into IRC1R∷ura4+ in epe1Δ cells albeit at lower efficiency, indicating that histone dosage also contributes to heterochromatin spreading (Supplemental Fig. S10B). Interestingly, in bdf2-2YA H4K16Q cells, heterochromatin failed to spread into IRC1R∷ura4+, as indicated by serial dilution analyses and ChIP analyses of H3K9me2 and Swi6 (Fig. 6D,E), suggesting that deacetylation of H4K16 is indeed required for heterochromatin spreading. Moreover, a H4K16R mutation, which abolished H4K16ac, resulted in heterochromatin spreading into IRC1R∷ura4+ (Fig. 6D,E). Similar results were obtained with epe1Δ and bdf2Δ (Supplemental Fig. S11). The localization of Bdf2 to IRC1 was reduced in H4K16R cells but slightly increased in H4K16Q cells (Fig. 6F). In contrast, the localization of Bdf2-2YA to IRC1 was little affected by H4K16 mutations (Fig. 6F). These results demonstrate that Bdf2 is critical for regulating H4K16ac levels at boundaries and heterochromatin spreading.
Mst1 regulates H4K16ac levels at IRCs
Among the histone acetyltransferases (HATs) in fission yeast, Mst1 is a member of the highly conserved MYST family of HATs, and it forms a complex similar to the NuA4 complex of budding yeast (Gomez et al. 2005, 2008; Shevchenko et al. 2008), which acetylates H4K16 (Lee and Workman 2007). We found that a mst1-1 allele, which contains a temperature-sensitive L344S mutation (Gomez et al. 2008), resulted in heterochromatin spreading past IRC1R even at a permissive temperature, as indicated by the silencing of IRC1R∷ura4+ and the accumulation of H3K9me2 and Swi6 at the reporter (Fig. 7A,B). Moreover, in mst1-1 H4K16Q cells, heterochromatin spreading is abolished (Fig. 7A,B). Furthermore, mst1-1 resulted in a significant decrease of H4K16ac levels and reduced recruitment of Bdf2 at IRC1 (Fig. 7C,D). Thus, Mst1 directly regulates H4K16ac levels and the recruitment of Bdf2 at IRCs for proper boundary formation.
(A) Serial dilution analysis to measure heterochromatin spreading outside of IRC1R∷ura4+. All strains used are in an h4.1Δ/h4.3Δ background. (B–D) ChIP analysis of H3K9me2 and Swi6 levels at IRC1R∷ura4+ and H4K16ac and Bdf2 levels at IRC1, normalized to act1. The data are averages of three experiments, and error bars represent standard deviation. (E) A model by which H4K16ac and Bdf2 regulate heterochromatin spreading and boundary function. Heterochromatin spreading is mediated by cycles of Clr4-mediated H3K9me and the recruitment of Swi6. Sir2 mediates the deacetylation of H3K9, which is required for creating the substrate for Clr4, as well as the deacetylation of H4K16, which facilitates nucleosome compaction to bring the adjacent nucleosome closer to Clr4. Bdf2 is recruited to IRCs through Epe1, which is highly enriched at IRCs. Mst1 mediates H4K16ac at IRCs, which further stabilizes binding of Bdf2 to IRCs through its bromodomains. Bdf2 protects H4 tails from deacetylation by Sir2, preventing further heterochromatin spreading.
Discussion
Proper formation of boundaries between chromosomal domains is essential for maintaining stable gene expression patterns (Gaszner and Felsenfeld 2006; Valenzuela and Kamakaka 2006). Here we showed that a double bromodomain protein of the BET family, Bdf2, is both necessary and sufficient for the formation of boundaries that limit the spreading of heterochromatin. We further showed that the bromodomains of Bdf2, which preferentially bind to multiply acetylated histone H4 tails, are critical for Bdf2 boundary function. Moreover, we demonstrated that Bdf2 directly protects histones from Sir2-mediated deacetylation and that the deacetylation of H4K16 is necessary for H3K9me and Swi6-mediated heterochromatin spreading. Below, we discuss the implications of our findings for heterochromatin spreading and boundary functions.
The role of Bdf2 in boundary formation
The formation of heterochromatin has long been regarded as a paradigm for the assembly of self-propagating chromatin structures (Moazed 2011). The mechanism of heterochromatin spreading is complex and not well understood (Talbert and Henikoff 2006). A simplified view is that heterochromatin spreads by a positive feedback loop of histone modification and chromatin protein binding, the reiteration of which leads to the spreading of these modifications/proteins from nucleation sites (Talbert and Henikoff 2006; Grewal and Elgin 2007). There are two ways heterochromatin spreading is curbed. One is through competition between positive and negative forces, which results in the formation of extended transition zones that fluctuate when the heterochromatin–euchromatin balance changes (Kimura and Horikoshi 2004). These borders are termed “negotiable borders,” since they are not defined by DNA sequence but move depending on the dosage of opposing activities. For example, in the classical example of position effect variegation (PEV) in Drosophila, where the white gene is juxtaposed to pericentric heterochromatin due to chromosome rearrangement, it is stochastically silenced by heterochromatin in a portion of cells due to variegation in the distance of heterochromatin spreading, resulting in mottled eyes (Elgin and Reuter 2007). Most of the genes identified that affect PEV are involved in regulating chromatin function in general rather than boundary functions specifically (Elgin and Reuter 2007). The more reliable way to stop heterochromatin from spreading is by specific DNA elements that actively establish borders to stop the spreading of heterochromatin (Gaszner and Felsenfeld 2006; Valenzuela and Kamakaka 2006). These DNA elements recruit histone-modifying enzymes to directly modify histones to counteract heterochromatin spreading (Oki et al. 2004; West et al. 2004). Alternatively, they tether chromosomal regions to the nuclear periphery, resulting in physical separation of chromosomal domains (Ishii et al. 2002; Noma et al. 2006). Regardless of mechanism, the key to boundary formation is to break up the self-propagation of heterochromatic histone modifications. We showed here that Bdf2 disrupts this chain reaction by protecting modified histone tails from heterochromatic modifications through the recruitment of a histone tail-binding protein to the boundary region (Fig. 7E).
Bdf2 is recruited to IRCs through boundary protein Epe1 (Fig. 3D). This recruitment is dependent on the C-terminal region of Bdf2, which mediates its interaction with Epe1 (Fig. 3F–H). Loss of Epe1 resulted in the delocalization of Bdf2 from IRCs without affecting the localization of Bdf2 to gene promoters elsewhere in the genome (Fig. 3D; Supplemental Fig. S6C). However, Epe1 exhibits a broad localization pattern across entire heterochromatin domains, although Epe1 levels at IRC elements are higher compared with those in the middle of heterochromatin (Zofall and Grewal 2006; Braun et al. 2011). This suggests that high concentrations of Epe1 are required for Bdf2 localization to IRCs. Alternatively, other factors present at IRCs might enhance the interaction between Epe1 and Bdf2 to promote the localization of Bdf2 to IRCs. We favor the latter idea because even in ddb1Δ cells, which exhibit a significant increase of Epe1 at the body of heterochromatin, there is only marginal enrichment of Bdf2 at the heterochromatic repeats (Supplemental Fig. S12). It is possible that acetylated histone H4 stabilizes Bdf2 binding to chromatin, as indicated by lower levels of Bdf2 at IRCs when the bromodomains are compromised or when H4K16ac levels are reduced (Figs. 4E, 6F, 7D). The lack of Bdf2 enrichment at pericentric repeats in ddb1Δ cells is also consistent with this idea, since histones within heterochromatin regions are generally hypoacetylated (Kurdistani and Grunstein 2003; Grewal and Elgin 2007). The localization of Bdf2 at the right side of the centromere II heterochromatin boundary (Fig. 2B,E), which does not show elevated Epe1 levels (Zofall and Grewal 2006), is also consistent with the idea that additional mechanisms contribute to Bdf2 recruitment to heterochromatin boundaries. The high concentration of Epe1 at heterochromatin boundaries is achieved through Cul4–Ddb1-mediated degradation of Epe1 in the middle of heterochromatin (Braun et al. 2011) and thus should be independent of DNA sequences. However, heterochromatin borders at centromeres and the silencing mating type region are well defined, suggesting that boundary DNA elements have independent contributions to shape the localization of Epe1 and Bdf2.
Overexpression of Epe1 results in defects in pericentric heterochromatin assembly (Zofall and Grewal 2006; Trewick et al. 2007) even in bdf2Δ cells (Supplemental Fig. S13), suggesting that Epe1 negatively regulates heterochromatin assembly independently of its role in recruiting Bdf2. Such an activity may also contribute to the boundary function of Epe1. This idea is further supported by the fact that epe1Δ showed stronger spreading of heterochromatin than bdf2Δ. Although Epe1 has also been proposed to be a demethylase or a hydroxylase and mutations of putative active site residues affect Epe1 function (Trewick et al. 2005, 2007), the substrate of Epe1 has not been identified. Further characterization of the enzymatic activity of Epe1 is needed to reveal additional mechanisms by which Epe1 regulates heterochromatin spreading.
In budding yeast, the double bromodomain protein Bdf1 is also required to prevent Sir2-mediated heterochromatin spreading near telomeres through the binding of Bdf1 to acetylated histone tails (Ladurner et al. 2003). Unlike fission yeast Bdf2, budding yeast Bdf1 does not specifically localize to defined boundary elements but is recruited to the transition zone, in which a gradient of H4K16ac is formed by the actions of HAT SAS2 and HDAC Sir2 (Kimura et al. 2002; Suka et al. 2002; Ladurner et al. 2003). Bdf1 is also part of the Swr1 complex that is required for the deposition of histone variant H2A.Z (Krogan et al. 2003; Kobor et al. 2004; Mizuguchi et al. 2004), which is also involved in counteracting heterochromatin spreading (Meneghini et al. 2003). There is a Bdf1 homolog in fission yeast, which is part of the fission yeast Swr1 complex (Buchanan et al. 2009; Hou et al. 2010). However, fission yeast Bdf1 is not involved in boundary function at IRCs (Supplemental Fig. S14).
H4K16ac and heterochromatin spreading in fission yeast
The bromodomains of Bdf2 interact with tetra-acetylated but not singly acetylated histone H4 tail peptides in vitro. This is consistent with studies of human BET family proteins, which favor multiply acetylated substrates (Filippakopoulos et al. 2012). Mutational analysis of acetylated lysines on the histone H4 tail in budding yeast demonstrates that Lys5, Lys8, Lys12, and Lys16 are partially redundant and cumulatively regulate gene expression, heterochromatin assembly, maintenance of genome integrity, and cell cycle progression, although Lys16 acetylation has a more dominant role (Megee et al. 1990, 1995; Park and Szostak 1990; Durrin et al. 1991; Dion et al. 2005). The fact that fission yeast Sir2 preferentially deacetylates K16 among the H4 tail lysines (Shankaranarayana et al. 2003; Alper et al. 2013) and that Bdf2 counteracts Sir2 to regulate H4K16ac levels at IRCs suggests that acetylated H4K16 is a major target of Bdf2 in vivo. Indeed, the association of Bdf2 with IRCs and gene promoters is significantly reduced in H4K16R cells (Fig. 6F; Supplemental Fig. S6D).
H4K16ac directly regulates the compaction of nucleosomal arrays in vitro (Shogren-Knaak et al. 2006), and in budding yeast, H4K16ac is critical for regulating heterochromatin spreading in vivo (Kimura et al. 2002; Suka et al. 2002). In fission yeast, eliminating H4K16ac with a H4K16R mutation resulted in heterochromatin spreading outside of IRC1R, and mimicking hyperacetylation of H4K16 with a H4K16Q mutation effectively blocked heterochromatin spreading in bdf2Δ and epe1Δ cells. Such results suggest that deacetylation of H4K16 is an integral part of H3K9me and Swi6-mediated heterochromatin spreading. However, H4K16ac-mimicking mutations such as H4K16A and H4K16Q have little effect on silencing of reporter genes inserted inside pericentric repeats such as otr∷ade6+ (Supplemental Fig. S15; Mellone et al. 2003), indicating that additional heterochromatin assembly pathways at pericentric regions, such as RNAi, can overcome the requirement of Sir2 (Alper et al. 2013; Buscaino et al. 2013).
In both budding yeast and fission yeast, Sir2 is the conserved HDAC that regulates heterochromatin assembly (Kimura et al. 2002; Suka et al. 2002; Shankaranarayana et al. 2003; Freeman-Cook et al. 2005; Buscaino et al. 2013). In fission yeast, it was proposed that Sir2-mediated deacetylation of H3K9 is a prerequisite for H3K9me (Shankaranarayana et al. 2003). However, sir2Δ has only minor effects on H3K9me levels at heterochromatin nucleation centers such as pericentric repeats or the cenH sequence at the mating type region (Shankaranarayana et al. 2003; Freeman-Cook et al. 2005; Alper et al. 2013; Buscaino et al. 2013), suggesting that deacetylation of H3K9 in the absence of Sir2 can be accomplished by other HDACs. Our results that Sir2 regulates H4K16ac levels at heterochromatin boundaries and that H4K16 mutants modulate heterochromatin boundary function suggest that Sir2 also deacetylates H4K16 in regulating heterochromatin spreading. We hypothesize that the deacetylation of H4K16 increases chromatin compaction levels, thus bringing the adjacent nucleosomes closer to Clr4 for repeated cycles of H3K9me propagation. The fact that reducing histone dosage affects the ability of heterochromatin to spread (Supplemental Fig. S10) also supports such a hypothesis.
In budding yeast, Sas2 is the major H4K16 acetyltransferase that regulates silencing and boundary formation (Kimura et al. 2002; Suka et al. 2002; Shia et al. 2006). However, there is no Sas2 homolog in fission yeast. We discovered that the fission yeast MYST family acetyltransferase Mst1 is required for H4K16ac and the binding of Bdf2 at IRCs, and a mutation in Mst1 also resulted in defects in boundary formation at IRCs. Mst1 forms a complex with composition similar to that of the budding yeast NuA4, which acetylates H4K16 as well as other residues (Lee and Workman 2007; Gomez et al. 2008; Shevchenko et al. 2008). However, in budding yeast, mutations in Esa1, the catalytic subunit of NuA4, have no effect on telomeric heterochromatin assembly (Kimura et al. 2002; Suka et al. 2002). Thus, the fission yeast Mst1 complex has taken over the role of both Sas2 and NuA4 complexes in budding yeast, which might be attributed to the slight difference in the composition of Mst1 and NuA4 complexes (Shevchenko et al. 2008).
Overexpression of Swi6 enhances heterochromatin spreading
Multiple mechanisms cooperate to establish heterochromatin boundaries. In addition to Epe1, TFIIIC binding to IR elements contributes to boundary formation at the silent mating type region, and tRNA genes and transcription of IRC1 contribute to boundary activity at centromeres (Noma et al. 2006; Scott et al. 2006; Keller et al. 2013). As a result, heterochromatin spreading in epe1Δ cells is a stochastic event happening only in a small portion of cells. Although the reporter-based assays provide sensitive readouts of heterochromatin spreading in this subpopulation, it is difficult to perform biochemical assays with ensembles of cells, since the majority of cells do not show spreading. For example, our ChIP–chip analysis showed that in epe1Δ cells without Swi6 overexpression or reporter-based selection, H3K9me2 levels around IRC1 were similar to those of wild-type cells (data not shown), consistent with earlier analysis of H3K9me2 levels in epe1Δ cells at the mating type region and centromeres (Ayoub et al. 2003; Braun et al. 2011). The use of an additional copy of Swi6 provides a more profound effect on heterochromatin spreading at centromere boundaries, and similar approaches have been used to enhance heterochromatin spreading at the mating type region (Noma et al. 2001, 2006).
The functions of Bdf2 in other processes
In addition to its role in boundary function, Bdf2 also interacted with a number of TAFs, and our ChIP–chip analysis showed that Bdf2 localized at a small group of gene promoters (∼13% of all promoter probes in our microarray), suggesting that Bdf2 might be involved in transcriptional regulation of certain genes. IRCs were transcribed (Zofall and Grewal 2006; Keller et al. 2013), and IRC1 transcript levels were strongly reduced in bdf2Δ and epe1Δ cells (Supplemental Fig. S16), suggesting that Bdf2 regulates transcription of IRCs. However, targeting of Bdf2 to Gal4-binding sites is sufficient to block the spreading of heterochromatin at an ectopic site in a bromodomain-dependent manner, suggesting that the ability of Bdf2 to counteract heterochromatin spreading is through protection of H4K16ac by its bromodomains, although transcription of IRCs has an independent contribution to boundary function (Keller et al. 2013). It is also possible that Bdf2 regulates transcription of other factors involved in heterochromatin boundary functions, and it would be interesting to determine the role of Bdf2 in regulating gene expression at euchromatic regions in the future. Bdf2 has also been shown to regulate DNA damage response, which is attributed to its effect on regulating global H4 acetylation levels (Garabedian et al. 2012).
Implications of BET protein functions in other systems
The BET family of bromodomains in mammals includes BRD2, BRD3, BRD4, and BRDT. They associate with diverse protein complexes involved in chromatin modifications and play important roles in transcription regulation, cell cycle control, gene bookmarking, and viral genome transcription (Belkina and Denis 2012). Misregulation of BET proteins, especially Brd4, has been linked to a number of human cancers. For example, chromosome translocation between BRD4 and NUT (nuclear protein in testis) leads to a highly lethal form of carcinoma, and BRD4 is a therapeutic target in acute myeloid leukemia (French et al. 2003; Zuber et al. 2011). The identification of small molecules such as JQ1 and IBET that inhibit the binding of BET family bromodomains to chromatin (Filippakopoulos et al. 2010; Nicodeme et al. 2010) makes them the focus of drug-mediated epigenetic treatment of diverse types of cancers (Dawson et al. 2011; Delmore et al. 2011; Mertz et al. 2011; Zuber et al. 2011; Loven et al. 2013). Although the precise mechanism of these inhibitors is not well understood, the displacement of BET proteins from chromatin in part down-regulates oncogene c-Myc to promote differentiation (Dawson et al. 2011; Mertz et al. 2011; Zuber et al. 2011; Loven et al. 2013). However, mice treated with these drugs do not exhibit tissue-renewing defects associated with lower Myc activity, indicative of additional mechanisms (Filippakopoulos et al. 2010; Dawson et al. 2011; Delmore et al. 2011; Mertz et al. 2011; Zuber et al. 2011). The fact that BET domain proteins Bdf2 in fission yeast and Bdf1 in budding yeast regulate heterochromatin spreading suggests that BET inhibitors may affect the spreading of heterochromatin and reset the epigenetic landscape in cancer cells. In other multicellular organisms such as zebrafish, worms, and flies, BET proteins are essential for the determination of cell fate during development (Huang and Dawid 1990; Chang et al. 2007; Dibenedetto et al. 2008; Shibata et al. 2010). In particular, the Drosophila BET protein FSH (female sterile 1 homeotic) is classified as a Trithorax group gene that counteracts Polycomb group protein-mediated gene silencing (Chang et al. 2007; Kockmann et al. 2013). Thus, it seems that the function of BET proteins in antagonizing gene silencing is highly conserved across species.
Materials and methodsFission yeast strains and genetic analyses
The IRC1R∷ura4+ reporter was constructed by inserting ura4+ at the right side of IRC1R (chromosome I, 3790595). The query strain used for our screen contains a closely linked natMX6 cassette (chromosome I, 3804996). IRC1RΔ removed chromosome I: 3789581–3790711. The IRC3L∷ura4+ reporter was constructed by inserting ura4+ at the left side of IRC3L (chromosome III, 1067955). L5-IRC1R-ura4+ and L5-3gbs-ura4+ strains were constructed by inserting IRC1R or 3gbs sequences between L5 and ura4+ using L5-ura4+ (Sadaie et al. 2004) as a template. Bdf2-Flag, Bdf2-myc, and Epe1-Flag were constructed by a PCR-based module method. bdf2Δ, epe1Δ, and sir2Δ were derived from the Bioneer fission yeast deletion library, verified via PCR, and backcrossed. The bdf2-ΔC and bdf2-2YA strains were constructed by integrating a PCR fragment containing the mutations and a Flag-kanMX6 cassette into the endogenous bdf2+ locus. H4K16R and H4K16Q mutations were constructed and introduced into the h4.1Δ/h4.3Δ background as described previously (Mellone et al. 2003), with the exception of introducing a linked natMX6 cassette. Genetic crosses were used to construct all other strains. For serial dilution plating assays, 10-fold dilutions of a log-phase culture were plated on the indicated medium and grown for 3 d at 30°C. All strains used in heterochromatin spreading assays contain an extra copy of the swi6+ gene inserted at the ade6 locus and driven by the ade6 promoter to enhance heterochromatin spreading. A list of the yeast strains used is provided in Supplemental Table S3.
ChIP analysis
ChIP analyses were performed as described previously (Hou et al. 2010). Antibodies used were H3K9me2 (Abcam, ab1220), H4K16ac (Active Motif, 39167), H3 (Abcam, ab1791), H4 (Abcam, ab10158), and Flag (Sigma, A2220). Swi6 antibody was custom-made with full-length recombinant Swi6 protein and affinity-purified (Reddy et al. 2011).
ChIP–chip analysis was performed according to the Agilent Yeast ChIP-on-chip Analysis protocol. The microarray used was an Agilent Schizosaccharomyces pombe Whole-Genome ChIP-on-chip microarray (G4810A) with additional probes that encompass centromeres, which were originally absent from the array due to the repetitive nature of these DNA sequences. Blunt-end DNA was generated from immunoprecipitated chromatin fractions (ChIP) or whole-cell extract (WCE) with T4 DNA polymerase and then ligated to a linker. ChIP and WCE DNA were amplified from blunt-end DNA samples with primers annealing to the linker and were labeled by Cy5-dUTP or Cy3-dUTP, respectively, with random priming PCR (Invitrogen CGH kit). Three micrograms of Cy5-labeled ChIP DNA and the corresponding Cy3-labled WCE DNA were hybridized to the microarray. The slides were washed and processed in accordance with Agilent protocols and scanned with an Agilent scanner. Data were collected with the Agilent Feature Extraction program. The enrichment value for each probe was calculated by dividing normalized ChIP signal by WCE signal. The data presented are averages of two independent ChIP–chip experiments. ChIP–chip data have been submitted to the Gene Expression Omnibus under accession number GSE46430.
qPCR was performed with Maxima SYBR Green qPCR Master Mix (Thermo Scientific) in a StepOne Plus Real-Time PCR System (Applied Biosystems). DNA serial dilutions were used as templates to generate a standard curve of amplification for each pair of primers, and the relative concentrations of target sequence and a control act1 sequence in ChIP and WCE samples were calculated accordingly. The final enrichment was calculated as [(ChIP target)/(WCE target)]/[(ChIP act1)/(WCE act1)]. A list of the primers used is provided in Supplemental Table S4.
Protein purification, coimmunoprecipitation, and mass spectrometry analysis
Exponentially growing yeast cells were harvested, washed with 2× HC buffer (300 mM HEPES-KOH at pH 7.6, 2 mM EDTA, 100 mM KCl, 20% glycerol, 2 mM DTT, protease inhibitor cocktail [Roche]), and frozen in liquid nitrogen. Crude cell extracts were prepared by vigorously blending frozen yeast cells with dry ice using a household blender, followed by incubation with 30 mL of 1× HC buffer containing 250 mM KCl for 30 min. The lysate was cleared by centrifugation at 82,700g for 3 h. The supernatants were precleared with protein A agarose, then incubated with 200 μL of Flag-agarose overnight, and washed eight times with 1× HC containing 250 mM KCl. For mass spectrometry analysis, bound proteins were eluted with 200 μg/mL 3xFlag peptides followed by TCA precipitation. Multidimensional protein identification technology (MudPIT) mass spectrometry analysis was performed as described previously (Wang et al. 2009). For coimmunoprecipitation analysis, bound proteins were resolved by SDS-PAGE followed by Western blot analyses with Myc (Santa Cruz Biotechnology, sc-789) and Flag (Sigma, F7425) antibodies.
Peptide-binding assays
The bromodomains of Bdf2 (amino acids 229–497) were cloned into pGEX or pRSET bacterial expression vector. The 2YA mutant was generated by site-directed mutagenesis according to the manufacturer's protocol (Agilent). Recombinant GST and His-tagged Bdf2 proteins were purified with Glutathione Sepharose 4B (GE) and Talon Metal Affinity Resin (Clontech), respectively, according to manufacturer's protocols, followed by gel filtration with a Superdex 200 column. Binding of GST-Bdf2-BD to a Modified Histone Peptide Array (Active Motif, catalog no. 17-0756-01) was performed according to the manufacturer's protocol. The array contains 59 different post-translational modifications for histone acetylation, methylation, phosphorylation, and citrullination on the N-terminal tails of histones H2A, H2B, H3, and H4 (Supplemental Table S2). Each 19mer peptide may contain up to four modifications. Array Analyze software (Active Motif) was used to analyze spot intensity from the image. The results were graphed as a specificity factor, which is the ratio of the average intensity of all spots containing the mark divided by the average intensity of all spots not containing the mark. For peptide pull-down assays, 1 μg of recombinant proteins was incubated with 1 μg of biotinylated H3 or H4 histone peptides (Millipore, catalog nos. 12-403 for H3, 12-402 for Ac-H3, 12-372 for H4, and 12-379 for Ac-H4) in 1× HC buffer containing 200 mM KCl overnight at 4°C and washed extensively. Bound fractions were resolved by SDS-PAGE and visualized by silver staining.
Yeast two-hybrid assay
Full-length Bdf2, Bdf2-ΔN, Bdf2-ΔC, and Bdf2-BD were cloned into the XmaI/BamHI site of pGBT9 (Clontech) to generate fusion with the GAL4 DNA-binding domain. Epe1 was cloned into the PstI/BglII site of pGAD424 (Clontech) to generate fusion with the GAL4 activation domain. Both plasmids were transformed into the budding yeast strain pJ69-4A, and transformants were selected on medium lacking tryptophan and leucine to maintain both plasmids. The interaction of the two proteins was indicated by the activation of a HIS3 reporter, allowing growth on medium lacking histidine.
RNA extraction and RT–PCR
Total cellular RNA was isolated from log-phase cells using MasterPure yeast RNA purification kit (Epicentre) according to the manufacturer's protocol. Quantification with real-time RT–PCR was performed with Power SYBR Green RNA-to-CT one-step kit (Applied Biosystems). RNA serial dilutions were used as a template to generate a standard curve of amplification for each pair of primers, and the relative concentration of the target sequence was calculated accordingly. An act1 fragment served as a reference to normalize the concentration of samples. The concentration of each target gene in wild type was arbitrarily set to 1 and served as a reference for other samples.
Sir2 HDAC assay
Full-length Sir2 cDNA was cloned into the pRSET bacterial expression vector (Invitrogen). Recombinant His-tagged Sir2 protein was purified with Talon Metal Affinity Resin (Clontech) according to the manufacturer's protocol. Sir2 HDAC assay was performed using the SIRTainty Class III HDAC assay kit (Millipore) according to the manufacturer's protocol. Briefly, 5 μg of recombinant Bdf2 or Bdf2-2YA proteins was incubated with β-NAD, nicotinamidase, tetra-acetylated H4 peptides, and different amounts of recombinant Sir2 for 1 h at 37°C followed by further incubation with developer reagent for 1 h at room temperature. Fluorescence intensity was measured with a BioTek Synergy 4 Hybrid Microplate Reader with filter set to excitation at 420 nm and emission at 470 nm.
Acknowledgments
We thank Anudari Letian for technical assistance; Robin Allshire, Amar Klar, Jun-ichi Nakayama, Janet Partridge, and the National BioResource project in Japan for yeast strains; Michael Keogh for communicating unpublished results; and James Manley, Scott Kallgren, Allison Cohen, and Eric Wei for critical reading of the manuscript. This work is supported by National Institutes of Health grants R01-GM085145 to S.J., the National Center for Research Resources (P41-RR011823), and National Institute of General Medical Sciences (P41-GM103533). X.T. is a Fulbright Scholar.
Supplemental material is available for this article.
Article is online at http://www.genesdev.org/cgi/doi/10.1101/gad.221010.113.
This study investigates the role of the putative splicing regulator and transcriptional cofactor PRMT5 in mammalian development. Using mouse models, Bezzi et al. show that PRMT5 deletion affects splicing through its ability to reduce Sm protein methylation. The authors identify the Mdm4 pre-mRNA as a key PRMT5 target; Mdm4 pre-mRNA senses defects in the spliceosomal machinery and consequently activates the p53 response. This study uncovers a new role for PRMT5 in splicing and identifies Mdm4 pre-mRNA as a potential anti-cancer target.
The tight control of gene expression at the level of both transcription and post-transcriptional RNA processing is essential for mammalian development. We here investigate the role of protein arginine methyltransferase 5 (PRMT5), a putative splicing regulator and transcriptional cofactor, in mammalian development. We demonstrate that selective deletion of PRMT5 in neural stem/progenitor cells (NPCs) leads to postnatal death in mice. At the molecular level, the absence of PRMT5 results in reduced methylation of Sm proteins, aberrant constitutive splicing, and the alternative splicing of specific mRNAs with weak 5′ donor sites. Intriguingly, the products of these mRNAs are, among others, several proteins regulating cell cycle progression. We identify Mdm4 as one of these key mRNAs that senses the defects in the spliceosomal machinery and transduces the signal to activate the p53 response, providing a mechanistic explanation of the phenotype observed in vivo. Our data demonstrate that PRMT5 is a master regulator of splicing in mammals and uncover a new role for the Mdm4 pre-mRNA, which could be exploited for anti-cancer therapy.
Arginine methylation is a post-translational modification known to play a key role in both transcription and post-transcriptional RNA processing by mediating epigenetic control of chromatin and functionally regulating RNA-binding proteins and components of the splicing machinery (Cheng et al. 2007; Kouzarides 2007; Bedford and Clarke 2009; Migliori et al. 2010).
Initial attempts to identify arginine-methylated proteins have generated lists of putative protein arginine methyltransferase (PRMT) targets (Boisvert et al. 2003; Ong et al. 2004). These studies failed to identify residues methylated by specific PRMT family members and to distinguish between symmetric and asymmetric dimethylation. However, they did shed light on the relevance of arginine methylation in splicing regulation by identifying as targets key components of the constitutive splicing machinery (e.g., Sm proteins) as well as several additional regulators of alternative splicing (e.g., FUS/TLS, SF2, and members of the heterogeneous nuclear ribonucleoprotein [hnRNP] family).
The splicing complexity occurring in the mammalian brain is a remarkable product of evolution and distinctively distinguishes the human species from others (Barbosa-Morais et al. 2012; Dillman et al. 2013). Proper functioning of all splicing-associated proteins allows a significant increase in the complexity of the cell proteome. On the other hand, mutations in proteins involved in RNA processing have been causally linked to neurodegenerative disorders such as spinal muscular atrophy (SMA) and amyotrophic lateral sclerosis (ALS), to mention a few (Zhang et al. 2008; Vance et al. 2009; Da Cruz and Cleveland 2011). These disease-causing mutations emphasize the pivotal role of RNA processing in higher vertebrate brain development, urging researchers to investigate the role of novel splicing regulators in the CNS.
PRMT5 is a type II arginine methyltransferase able to symmetrically dimethylate several nuclear and cytoplasmic proteins (Bedford and Clarke 2009; Karkhanis et al. 2011). In the cytoplasm, PRMT5 acts together with pICln and WDR77 as part of the methylosome, which mainly methylates Sm proteins (SmB/B′, SmD1, and SmD3), increasing their affinity for the Tudor domain of spinal motor neuron 1 (SMN1) (Friesen et al. 2001a,b, 2002; Meister et al. 2001). SMN1 deficiency results in early lethality (embryonic day 3.5 [E3.5]) in the mouse embryo, similarly to Prmt5 constitutive deletion (Hsieh-Li et al. 2000; Tee et al. 2010), while lack of SMN1 in SMA mouse models leads to splicing defects in not only spinal motor neurons, but several organs (Zhang et al. 2008). Given the direct connection between PRMT5 and SMN1 and the relevance of arginine methylation in regulating splicing proteins, it is of extreme relevance to assess the role, if any, of PRMT5 in the CNS.
Here we demonstrate that selective deletion of PRMT5 in the CNS leads to the death of the animal 14 d after birth. We first show that genetic deletion of p53 in a Prmt5-null background partially rescues the developmental defects. Second, we show that the absence of PRMT5 in neural stem/progenitor cells (NPCs) leads to defects in the core splicing machinery. This results in reduced constitutive splicing and the alternative splicing of specific mRNAs, which have weak 5′ donor sites. We identify Mdm4 as one of these mRNAs that acts as a sensor of the splicing defects. Specifically, the Mdm4 alternative splicing event results in the generation of the unstable Mdm4s product, the reduction of the full-length protein, and the transduction of the p53 signaling cascade. We finally expand our findings to other cell types and tissues, demonstrating that Mdm4/Mdm4s alternative splicing senses the absence of PRMT5 also in mouse embryonic fibroblasts (MEFs) in several organs during embryo development and in human cancer cell lines.
We believe our data provide an underlying mechanism for many observations on PRMT5 biology (Jansson et al. 2008; Scoumanne et al. 2009) and, more in general, on perturbation of the splicing machinery (Allende-Vega et al. 2013) and their link to the p53 pathway that were previously ignored.
ResultsPRMT5 deficiency in the CNS results in early postnatal lethality
To address the effect of PRMT5 depletion in mammals, we made use of a conditional knockout mouse (White et al. 2013) harboring LoxP (F/F) sequences flanking exon 7 in the Prmt5 gene and studied the effect of its conditional deletion in the CNS. We used a Nestin-Cre (Nes) transgenic mouse strain that expresses Cre recombinase under a neural-specific enhancer of the Nestin promoter, leading to an efficient recombination event in precursors of neurons and glia starting at E10.5 (Graus-Porta et al. 2001). All of the Prmt5F/FNes mice were obtained from Prmt5F/F × Prmt5+/FNes crosses, and, as expected, the Prmt5+/FNes mice were viable and fertile, and we could not observe any evident defects. Single-site insertion was verified by Southern blotting, and CNS-specific deletion of PRMT5 was confirmed by genomic PCR and Western blotting (Supplemental Fig. S1).
Prmt5F/FNes transgenic mice were born at the expected Mendelian frequency but displayed balance disorders, tremors, and akinesis and all died within 14 d after birth. CNS development was impaired, as evident from differences in brain size and weight, which was detectable starting at E17.5 (Fig. 1A). At postnatal day 10 (P10), the external granular layer (EGL) of the cerebellum, an actively proliferating area at this age, was missing in mutant mice, as evident from both sagittal and coronal sections. The lateral ventricles were morphologically enlarged and disrupted, and the thickness of the cortex was reduced in size (Fig. 1B). We next focused on two earlier developmental stages: E15.5 and P0. The cortex of P0 Prmt5F/FNes brains had a lower cellularity count in both the cortical plate (CP) and the ventricular zone/subventricular zone (VZ/SVZ) (Fig. 1C) and a lower number of SOX2/Ki67-positive proliferating NPCs (Fig. 1D) as opposed to controls (PRMT5F/F). We could not observe any appreciable differences in E15.5 brains (Supplemental Fig. S1E).
PRMT5 deficiency in the CNS results in early postnatal lethality. Nestin-Cre-induced deletion of the PRMT5 gene in the CNS. (A) Weight in milligrams of wild-type (Prmt5F/F) and Prmt5-deleted (Prmt5F/FNes) brains at three different time points (E17.5, P0, and P10). Brain sizes of P10 Prmt5F/F and Prmt5F/FNes mice are shown as an example in the right panel. (B) Hematoxylin and eosin (H&E)-stained sagittal and coronal sections of P10 cerebrum (Cr) and the cerebellum (Cb) from Prmt5F/F (right) and Prmt5F/FNes (left). (C) Coronal sections of P0 brains. Cellularity of the cortical plate (CP) and VZ/SVZ are indicated in wild-type (wt; black) and mutant (red) brains. (MZ) Marginal zone; (SP) subplate; (IZ) intermediate zone. (D) SOX2 and Ki67 immunohistochemistry (IHC) staining in P0 brains. (E) Cleaved Caspase 3 (CC3) staining is shown in both the cortex and the gangliomic eminence.
The data suggested that the absence of PRMT5 resulted in either premature differentiation, cell death of NPCs, or a combination of the two. To test the former hypothesis, we extracted proteins from Prmt5F/F and Prmt5F/FNes P0 and P10 brains and tested the expression of NPC markers (SOX2) and intermediate progenitor markers (TBR2) as well as neuronal and glia markers (TBR1/TuJ and GFAP, respectively). We did observe a significant decrease of SOX2 and TBR2 levels upon PRMT5 deletion, while the levels of differentiated neurons and glia markers were similar in both control and mutant brains (Supplemental Fig. S1F). To test the occurrence of cell death, we stained brain sections for cleaved Caspase 3 (CC3) (Kuida et al. 1996). Despite the fact that changes in brain size are not evident at E15.5 (data not shown), we did detect apoptotic death, specifically in the VZ/SVZ and the ganglionic eminence, both areas containing proliferating NPCs, suggesting that this could be the cause for the reduced brain size of Prmt5F/FNes animals (Fig. 1E).
PRMT5 is required for NPC homeostasis
To further define whether PRMT5 is required for normal cell cycle regulation and survival and to protect cells from apoptosis, we derived NPCs from the dorsal telencephalon of E14.5 mice. NPCs can be efficiently grown in vitro as neurospheres. The number of primary neurospheres was significantly reduced in Prmt5F/FNes as opposed to controls. Furthermore, the number of cells in Prmt5F/FNes neurospheres was markedly reduced (Fig. 2A), and, importantly, their self-renewal potential was impaired, as highlighted by the fact that, following replating, virtually no secondary neurospheres could be derived (Fig. 2B). To test whether PRMT5 catalytic activity was necessary for the observed phenotype, we infected primary NPCs derived from Prmt5F/FNes mice with wild-type human PRMT5 (hPRMT5) or a catalytically inactive mutant (hPRMT5AAA). Only cells infected with hPRMT5 were able to grow and could be propagated into secondary neurospheres, and when expanded into tertiary neurospheres, cells expressing hPRMT5 grew as efficiently as NPCs derived from Prmt5F/F control litters (Fig. 2C). To confirm the results obtained in vivo, we first counted the percentage of pyknotic nuclei and then stained Prmt5F/FNes-derived NPCs for CC3 to verify that they were undergoing apoptosis, confirming the requirement for PRMT5 to suppress cell death (Fig. 2D).
PRMT5 is required for NPC homeostasis. (A) Number of primary neurospheres and total number of cells from cultures of E14.5 dorsal telencephalon NPCs derived from Prmt5F/F and Prmt5F/FNes embryos. Each bar represents an average of at least three experiments. (B) Number of secondary neurospheres, as in A. (C) Primary neurospheres (left panel) from Prmt5F/FNes mice infected with empty vector (EV), wild-type PRMT5 (hPRMT5), or a catalytically inactive PRMT5 mutant (hPRMT5AAA) and passaged to derive secondary and tertiary neurospheres (right panel). (D) Neurospheres derived from Prmt5F/F or Prmt5F/FNes NPCs were stained with DAPI and CC3, and the percentage of pyknotic nuclei was counted. (E) Protein levels upon treatment with OHT and subsequent PRMT5 depletion for 4 d. The antibodies used are indicated on the right of each panel. As a positive control, p53 and the DDR were induced by treating cells with 10 μM etoposide for 2 h.
To understand the molecular mechanism underpinning the observed apoptotic phenotype, we next performed a gene expression analysis of Prmt5F/FNes NPCs. Approximately 2500 genes were differentially expressed when compared with control, showing up-regulation of the p53 pathway and down-regulation of genes involved in cell cycle progression and replication (Supplemental Fig. S2).
We then generated a second conditional knockout strain by crossing the Prmt5F/F mice to the ROSA26:CreERt2 (ER) mice, which allowed the triggering of a recombination event in both live animals and, ex vivo, primary cells by using 4-hydroxytamoxifen (OHT). We switched to the Prmt5F/FER system for three main reasons: First, it allowed us to look at cell-autonomous defects. Second, we could derive a much larger number of cells amenable for further mechanistic studies. Third, it allowed us to focus on early time points after PRMT5 depletion in order to detect causal defects. In all experiments described hereafter in which Prmt5F/FER-derived cells, tissues, or embryos were analyzed, we always used the ER counterparts as negative controls, making sure that the addition of OHT or tamoxifen (TAM) was not toxic (data not shown).
p53 is a transcription factor that drives the expression of several downstream targets in response to a variety of stimuli, including activation of the DNA damage response (DDR) (Lane 1992). Much is known about the regulation of p53 by post-translational modifications, and many of them, including phosphorylation and acetylation, are known to regulate its protein stability, leading to transcriptional activation. We first checked whether, upon PRMT5 deletion, we could detect DDR activation and whether p53 would be stabilized. We did observe a modest p53 protein stabilization and p53 phosphorylation (P-p53) as well as basal levels of H2AX phosphorylation (γH2AX). As a positive control, we used a DNA-damaging agent (etoposide), which, as expected, greatly stabilized p53 and increased the levels of γH2AX. Notably, despite a minor activation of the DDR, the absence of PRMT5 caused an even greater induction of the p53 target gene p21 as compared with etoposide (Fig. 2E, cf. lanes 2 and 3).
The data indicated that PRMT5 deficiency triggered a p53 response and that the phenotypic outcome in NPCs led to cell death. To formally prove this conclusion, we crossed Prmt5F/FNes mice into a p53-null background. Prmt5F/FNes;p53−/− mice survived, on average, 1 wk longer than Prmt5F/FNes;p53wt, while mice heterozygous for p53 (Prmt5F/FNes;p53+/−) displayed an intermediate phenotype (Fig. 3A). When stained for activated Caspase 3, E15.5 Prmt5F/FNes;p53−/− embryos showed a complete rescue of the apoptotic response, with levels of staining similar to wild type (Fig. 3B). Importantly, the number of SOX2-positive cells in the VZ/SVZ of Prmt5F/FNes;p53−/− embryos was increased when compared with Prmt5F/FNes;p53wt brains (Fig. 3C). However, we did not observe a significant rescue of proliferating KI67-positive cells, suggesting a p53-independent impairment in cell cycle progression, which most likely accounts for the lethality of the animals 20–22 d after birth (Fig. 3C, right panel). Indeed, when we derived NPCs from mice with different p53 backgrounds and cultured them as neurospheres, p53 deficiency led to a significant but not complete rescue in the number of proliferating cells (Fig. 3D). The same was true in P10 brains, where defects in EGL morphogenesis in the cerebellum and, in general, in brain development were only partially rescued in the absence of p53 (Fig. 3E). We can conclude that p53 plays an important role in regulating the apoptotic response in Prmt5F/FNes cells. The fact that we still observed death of the animals, although significantly delayed, however, pointed at additional proliferative defects in targeted cells. To gain further insight, we first checked by RT-qPCR the level of transcriptional up-regulation of p53 targets in both Prmt5F/FNes and Prmt5F/FER in a p53−/− background. Activation of cell cycle inhibitor p21, proapoptotic Noxa, Puma (Akhtar et al. 2006), and all other target genes was completely muted in the absence of p53, excluding compensation by other transcription factors such as p53 family members p63 and p73 (Fig. 3F; Supplemental Fig. S3A). In the absence of p53, PRMT5 depletion led to the reduction of the number of BrdU-positive cells and their exit from the cell cycle, and, consistently, we observed a reduction in the levels of apoptotic cells (Fig. 3G; Supplemental Fig. S3B). These data confirm that, despite inactivation of the p53 response, a second checkpoint mechanism prevents these cells from proliferating.
p53 deletion partially rescues Prmt5F/FNes developmental defects. (A) Kaplan-Meier survival analysis of Prmt5F/FNes mice in a p53wt (n = 14), p53+/− (n = 24), or p53−/− (n = 14) background. (B) Nestin-Cre-induced deletion of the PRMT5 gene in the CNS of p53−/− embryos. Coronal sections of E15.5 brains stained for CC3 (B) and P0 brains stained for SOX2 and Ki67 (C) to identify stem cells and assess their proliferation status. The antibodies used are indicated for each panel. (D) Total number of NPC cells grown as primary neurospheres derived from Prmt5F/FNes;p53wt, Prmt5F/FNes;p53+/−, and Prmt5F/FNes;p53−/− as indicated. (E) H&E-stained coronal brain sections of PRMT5F/FNes mice with different p53 backgrounds. The cerebellum is shown at a higher magnification in the inset. (F) Expression of p53 up-regulated target genes in NPCs from different genotypes as indicated. The activation of the genes is expressed as the average fold change of three embryos/NPCs, normalized against Prmt5F/F;p53wt and HK. (G) NPCs treated with OHT to delete Prmt5 were stained with propidium iodide and subjected to FACS. Bars indicate the increase in sub-G1/apoptotic cell populations, normalized to EtOH-treated cells. P53 genotypes are indicated.
PRMT5 loss leads to malfunction of the constitutive splicing machinery and to alternative splicing events
We looked for defects that could mechanistically underpin both the activation of the p53 pathway and the additional proliferation defects. In Drosophila and HeLa cells, PRMT5 is known to symmetrically dimethylate Sm proteins (Gonsalvez et al. 2006, 2007; Deng et al. 2010; Sanchez et al. 2010). We first tested whether this was also relevant during mammalian development. We treated Prmt5F/FER NPCs with either ethanol (EtOH = wild-type PRMT5 levels) or OHT (OHT = PRMT5-depleted) and observed that despite constant levels of SmD1 and SmD3 proteins (Fig. 4A, top panels), there was a reduction in the levels of symmetric arginine dimethylation by day 4, as detectable by two independent antibodies (SYM10 and Y12) (Fig. 4A, bottom panels). We analyzed cells at this early time point for further experiments. Consistent with the fact that the SMN1 Tudor domain binds arginine-methylated SmB/B′, SmD1, and SmD3, we observed a reduced binding of SMN1 to SmD1 and SmD3 (Fig. 4B), suggesting that PRMT5-depleted NPCs would have suboptimal small nuclear ribonucleoprotein (snRNP) maturation. This is indeed the case, since we observed a clear reduction of assembled Sm proteins by 35S pulse-chase assay (Supplemental Fig. S4A). In order to mechanistically understand what could link the splicing defects to apoptosis, we then generated libraries for pair-end RNA sequencing (RNA-seq) from samples treated with EtOH or OHT in order to delete PRMT5. We identified 2416 genes being differentially expressed between the two conditions (Supplemental Table S1). Consistently, the functional annotation of the up-regulated and down-regulated genes looked similar to the one from Prmt5F/FNes cells and showed the activation of the p53 pathway as the top up-regulated category (Supplemental Fig. S2B).
PRMT5 loss leads to malfunction of the constitutive splicing machinery and to alternative splicing events. (A) PRMT5, SmD1, SmD3, and SMN1 levels were assessed in Prmt5F/FER NPC cells depleted of PRMT5 after 2, 3, and 4 d after OHT treatment. Levels of symmetric arginine dimethylation were assessed by staining SmB/B′, SmD1, and SmD3 with SYM10 and Y12 antibodies. (B) Coimmunoprecipitation between SMN, SmD3, and SmD1, as indicated, in the presence ([E] EtOH) or absence ([O] OHT) of PRMT5. (C) Total number of reads in introns (red) or genes (blue) expressed as fold change of the events in NPCs lacking PRMT5 over control (wild-type PRMT5). A smooth density estimate is drawn as calculated by a Gaussian kernel. (D) Number of genes affected by alternative splicing events in each NPC population (derived from independent embryos). (Right panel) (Snapshot; full figure is in Supplemental Fig. S4B.) Network representation of the differentially spliced genes upon Prmt5 deletion in NPCs. The gene ontology (GO) terms are represented as nodes based on their κ scores. The edges represent the relationships between the GO terms and the shared genes. (E) Shapiro (CV) score of 5′ donor sites of the RI events in NPCs identified by MATS. A smooth density estimate is drawn as calculated by a Gaussian kernel. The top panels depict the sequence logo of the 5′ donor of all RI events (left) and the 5′ donor of the RI events detected upon PRMT5 deletion (right, indicated by the red arrow). The CV score of the downstream donor site is displayed for direct comparison. (F) Same as in E. Shapiro (CV) score of 5′ donor sites of the SE events (in red). The CV scores of the exclusion site (left, indicated by the blue arrow) and the downstream donor site (right, indicated by the green arrow) are displayed for direct comparison.
In contrast to what was reported in plants and Drosophila, where PRMT5 regulates splice site selection without greatly affecting constitutive pre-mRNA splicing (Sanchez et al. 2010), we observed that the compiled number of reads in introns was elevated in the absence of PRMT5, with 1682 introns being significantly affected (Fig. 4C). We then proceeded to characterize the splicing defects in more detail using Multivariate Analysis of Transcript Splicing (MATS) (Fig. 4D–F; Supplemental Fig. S4B; Shen et al. 2012). Prmt5F/FER mice were not on a pure C57BL/6 genetic background; hence, we sequenced three independent NPC populations and first checked the variability in splicing among embryos. In the absence of PRMT5, we observed an overlap of 320 genes affected by alternative splicing in two out of three embryos. These alternatively spliced genes are not random but belonged to specific biological pathways. Importantly, network analysis revealed that these genes are involved in post-transcriptional RNA processing, membrane organization, and negative regulation of cell cycle processes (Fig. 4D; Supplemental Fig. S4C). The latter included transduction of the p53 signaling pathway, suggesting that early problems with the core splicing machinery can be sensed by key alternatively spliced mRNAs to instruct cell cycle arrest or apoptosis (Fig. 4D).
In the absence of PRMT5, we observed a majority of retained intron (RI) and skipped exon (SE) events in all three embryos (Supplemental Fig. S4B), and we could validate 18 of 20 SE events (Supplemental Fig. S4D) and 21 of 21 RI events (Supplemental Fig. S4E), confirming that despite the observed embryo-to-embryo variability, we identified a robust set of conserved alternatively spliced events. Both the RI (Fig. 4E) and SE (Fig. 4F) events detected in the absence of PRMT5 are characterized by a weak 5′ donor site, as quantified by their low CV score (Shapiro score) (Shapiro and Senapathy 1987), their low MaxEntScan (Yeo and Burge 2004) and H-Bond (Freund et al. 2003) scores (Supplemental Fig. S4F), and an overall randomization of the key bases at positions −1, −2, +4, and +5. What distinguishes SE from RI events is the length of the affected intron, which is significantly shorter in the case of RI events (Supplemental Fig. S4G). Hence, absence of PRMT5 leads to selective retention of introns and skipping of exons with weak donor sites.
The Mdm4 alternative splicing event is a sensor of PRMT5 depletion and defects in the constitutive splicing machinery
MDM4 (also known as MDMX) has been reported to be down-regulated upon direct depletion of spliceosome components (Allende-Vega et al. 2013), and perturbation of its levels stood out as potentially recapitulating the activation of the p53 response that we observed in vivo. Importantly, the phenotype of the Mdm4−/− conditional CNS deletion is remarkably similar to what was observed for Prmt5F/FNes, and the most up-regulated gene in the absence of PRMT5 is Ptprv (Supplemental Fig. S2), which was originally identified as deregulated in Mdm4−/− embryos (Doumont et al. 2005).
We thus decided to focus our attention on the alternative splicing of Mdm4 (Supplemental Fig. S5) for the rest of the study. MDM4 is a direct regulator of p53 activity; it binds to p53 and inhibits its function by blocking its transactivation capabilities (Francoz et al. 2006; Xiong et al. 2006). Mdm4 undergoes alternative splicing at exon 7 in Prmt5F/FER OHT-treated cells, resulting in the production of a shorter MDM4 isoform that has been previously described as MDM4S (Supplemental Fig. S5A; Rallapalli et al. 2003; Lenos and Jochemsen 2011). Mdm4 exon 7 is located within a 1-kb genomic region that is highly conserved in vertebrates (as assessed by PhyloP) (Supplemental Fig. S5B), suggesting a common mechanism to regulate the abundance of the differentially spliced isoform. To verify that the alternative splicing event of Mdm4/Mdm4s was not a consequence, rather than a cause, of p53 activation and apoptosis, we derived Prmt5F/FNes NPCs with different p53 backgrounds. Reassuringly, the degree of alternative splicing was even greater in p53−/− cells (Fig. 5A). This suggests that the cells in which Mdm4 alternative splicing takes place are rapidly eliminated due to p53 activation. Importantly, all alternative splicing events that we validated were not a consequence of p53 activation or apoptosis (Supplemental Fig. S4D,E).
Mdm4 alternative splicing event is a sensor of PRMT5 depletion and defects in the constitutive splicing machinery. (A) PCR validation and relative quantification of the alternative splicing event taking place on the Mdm4 mRNA upon PRMT5 deletion in different p53 genetic backgrounds. (B) Semiquantitative PCR of the indicated transcripts upon CHX (100 μg/mL) treatment to block NMD. Cells were pretreated for 3 h and then for the indicated time with 5 μg/mL Actinomycin D to block transcription. (C) MDM4 full-length protein levels are reduced upon PRMT5 deletion. (O) OHT. Tubulin was used as a loading control. (D) PCR detecting both Mdm4 and Mdm4s in wild-type (wt) and mutant NPCs upon inhibition of the core splicing machinery, 100 μM TG003, and 30 ng/mL Spliceostatin A (SSA), or p53 stabilization (Nutlin and 5 μM etoposide). (D) DMSO; (M) MetOH. (E) Full-length Mdm4, re-expressed in PRMT5-depleted NPCs, is able to partially rescue the activation of the p53 response. PCR quantification of p53 target genes upon PRMT5 deletion in cells re-expressing full-length Mdm4 (gray and blue bars) or negative control, empty vector plasmid (black and red bars). A representative experiment of three is shown as an example. (F) NPCs infected with a retroviral vector stably expressing MDM4 or empty vector (EV) control. Prmt5 was deleted (OHT), and cells were stained with propidium iodide and subjected to FACS. Bars indicate the increase in sub-G1/apoptotic cell populations, normalized to EtOH-treated cells.
The literature on the MDM4S protein isoform is quite controversial (Rallapalli et al. 2003; Lenos and Jochemsen 2011), with some reports suggesting its possible role as a potent p53 inhibitor, and others stating that the MDM4S product is unstable. Notably, all of the data are based on forced overexpression experiments and negative results (failure to detect the endogenous protein product). We thus decided to address this issue by looking at the Mdm4s mRNA stability. We performed polysome profiling of cells upon PRMT5 deletion and noted that the full-length Mdm4 product was present in the polysome fractions (F4–F5), while the Mdm4s mRNA was associated with fractions containing significantly fewer polysomes (F3–F4) (Supplemental Fig. S5C). This result suggested two possibilities: either a low level of translation of the Mdm4s RNA or the fact that this RNA would be targeted for nonsense-mediated decay (NMD) (Chiu et al. 2004). To test the latter possibility, we treated cells with cyclohexamide (CHX), an inhibitor of protein synthesis known to block NMD-mediated mRNA degradation, and later with Actinomycin D, which blocks RNA polymerase II (Pol II) transcription. The data in Figure 5B demonstrate that the Mdm4s isoform is less stable than the full-length product and is targeted for NMD.
Our data suggest that upon PRMT5 depletion, the Mdm4 mRNA undergoes alternative splicing, giving rise to the unstable Mdm4s product. Indeed, this leads to the reduction of the full-length MDM4 protein (Fig. 5C). To extend our findings beyond perturbation of the splicing machinery through PRMT5, we treated NPCs with well-characterized splicing inhibitors (TG003 and Spliceostatin A) (Muraki et al. 2004; Kaida et al. 2007) and consistently observed Mdm4/Mdm4s alternatively splicing. As controls, neither p53 activation by Nutlin nor the induction of DNA damage by etoposide generated similar results (Fig. 5D). These results are in contrast with previous reports (Allende-Vega et al. 2013) and provide a direct mechanistic link between perturbation of the splicing machinery and downstream activation of p53.
To confirm our hypothesis, we demonstrated that the p53 transcriptional response (Fig. 5E) and the induction of apoptosis (Fig. 5F) caused by PRMT5 deletion could be rescued by reintroducing full-length MDM4 into NPCs. Not surprisingly, the rescue was only partial due to other alternative splicing events induced by the absence of PRMT5 (Supplemental Fig. S4).
To further prove that loss of PRMT5 leads to reduced levels of spliceosomal snRNPs and that this leads to changes in splicing patterns, we decided to deplete snRNPs alternatively by directly knocking down SmB/B′. We obtained efficient knockdown in NPCs with four independent shRNA-expressing vectors (Supplemental Fig. S5D). Importantly, these phenocopy the loss of PRMT5 and induce reduction of full-length Mdm4 (Supplemental Fig. S5E), activation of p53 target genes (Supplemental Fig. S5F), and increased apoptosis (Supplemental Fig. S5G).
PRMT5 depletion triggers Mdm4 alternative splicing and p53 activation in multiple tissues
So far, we dissected the role of PRMT5 in the developing CNS and showed that it plays a key role in ensuring the proper splicing of Mdm4 in proliferating NPCs. We next asked what the effect would be of deleting PRMT5 in other cell types or in different organs in the mouse embryo. First, we could confirm in MEFs most of what was observed in NPCs (Supplemental Fig. S6). The most notable differences were that MEFs displayed less splicing defects when compared with NPCs and phenotypically did not undergo cell death but rather exited the cell cycle. Despite these differences, the overlap of genes with increased intronic reads was remarkable (57%), and the p53-independent exon skipping event on Mdm4 was fully conserved between the two cell types (Supplemental Fig. S6; Supplemental Table S1).
To assess the effect of PRMT5 depletion during organogenesis, Prmt5 was selectively deleted in Prmt5F/FER embryos from pregnant Prmt5F/F females crossed to Prmt5+/FER males following TAM injection. CRE-ER was activated efficiently in different organs (Fig. 6A), and mutant embryos were readily recognizable by their smaller size, pale color, and growth retardation (Fig. 6B). Upon Prmt5 deletion, we did observe a switch in the ratio of the full-length over the Mdm4s isoform in most samples (Fig. 6C, bottom panel). Importantly, the degree of Mdm4 alternative splicing upon PRMT5 deletion correlates with the up-regulation of p53 targets (Fig. 6C, top panel). The effect was more pronounced in actively proliferative organs such as the lung and liver. The latter, at this stage of development, is populated by hematopoietic progenitor cells, recognizable by their dark-purple color in the hematoxylin and eosin (H&E) staining (Fig. 6D), and the impairment of their homeostasis is consistent with the pale color of the embryos (Fig. 6B). Phenotypically, we observed both activation of the apoptotic response (CC3 staining) and exit from the cell cycle (reduced Ki67 staining) (Fig. 6D).
PRMT5 depletion triggers Mdm4 alternative splicing and p53 activation in multiple organs. (A) Experimental strategy used to delete PRMT5 at mid-gestation (E10.5). Embryos were analyzed at E15.5 and E17.5. Upon TAM injection, no pups were born alive. (Bottom panel) Efficiency of CRE recombination taking place in different organs. (B) Weight of PRMT5 wild-type (EtOH) or PRMT5-deleted (TAM) whole embryos at E15.5 and E17.5. (Right panel) Representative example of E15.5 embryos with wild-type (left) or deleted PRMT5 (right). (C) Quantitative PCR (qPCR) quantification of p53 targets in the indicated organs upon PRMT5 deletion. (Bottom panel) PCR validation and relative quantification of the alternative splicing event taking place on the Mdm4 mRNA upon PRMT5 deletion in the same organs. (D) H&E staining of wild-type and knockout E15.5 lung and liver sections. In the liver, light-purple-stained hepatocytes and dark-purple-stained hematopoietic precursor cells are easily detectable. Note the dramatic loss of the latter and the corresponding loss of ki67 staining. Below each H&E staining are the IHC stainings of lung and liver sections from a representative embryo. CC3 was used to detect apoptotic cells, and Ki67 was used to detect proliferating cells. Mdm4 pre-mRNA senses defects in the spliceosomal machinery in cancer lines. (E) PCR quantification of the alternative splicing event taking place on the Mdm4 mRNA upon PRMT5 knockdown (KD). Scramble shRNA was used as a control (Scr). Treatment with the splicing inhibitor TG003 (or with DMSO vehicle control) was used as an alternative way of perturbing the splicing machinery. The experiments were performed in the indicated human cancer cells (shown at the top). GAPDH was used as a loading control. (F) Quantification of Mdm4fl/Mdm4s splicing levels 4 d after infection and 2-d selection in 1 mg/mL puromycin upon knockdown with three different shRNA lentiviral constructs (Sh1–Sh3) in U2OS cells. (Scr) Scrambled control shRNA.
Finally, we cloned the region surrounding mouse exon 7 into a minigene construct (Supplemental Fig. S7). Exon 7 is skipped upon PRMT5 depletion (OHT) in MEFs, recapitulating what was observed at the endogenous level.
Mdm4 pre-mRNA senses defects in the spliceosomal machinery in cancer lines
Activation of the p53 pathway is important in cell homeostasis as well as in development but is certainly best known for its aberrant deregulation in human cancer. PRMT5 has been described as a potential oncogene in human malignancies (Karkhanis et al. 2011). Given the high degree of conservation of the region around the alternative splicing exon 7 on mouse Mdm4 (Supplemental Fig. S5B), we tested whether the orthologous human exon 6 conserved a similar sensing mechanism. Upon Prmt5 knockdown and treatment with the splicing inhibitor TG003, we observed a similar alternative splicing event occurring on the human Mdm4 transcript in cancer cell lines derived from different tissues (osteosarcoma, gastric, breast, and glioma) (Fig. 6E). Notably, in accordance with what was observed in NPCs (Supplemental Fig. S5D–G), knockdown of SmB/B′ in U2OS cells leads to Mdm4 alternative splicing (Fig. 6F).
Overall, our data uncover a key mechanism of PRMT5 function through the regulation of Mdm4 splicing that is conserved in mammalian cells during development and relevant to human cancer (Fig. 7).
Mdm4 pre-mRNA senses defects in the spliceosomal machinery in cancer lines. Schematic model of the data presented in the study: Upon PRMT5 deletion (or reduction), we observed a loss of symmetric arginine dimethylation at key components of the splicing machinery (SmB/B′, SmD1, SmD3, and possibly others). This leads to aberrant snRNP maturation. The consequence is the activation of a sensing mechanism, which is linked to alternative splicing of key mRNAs (mainly RIs and SEs). As an example, we show Mdm4, which induces a potent p53 transcriptional activation. (Bottom) Other alternative splicing events might be equally important and will ultimately result in a p53-independent cell cycle arrest.
DiscussionRole of PRMT5 in splicing
The initial aim of our project was to dissect the role of PRMT5 in mouse CNS development. We showed by using a conditional knockout mouse model and a combination of in vitro and in vivo approaches that PRMT5 is an essential regulator of splicing in mammals. The first key finding is that these defects in splicing lead to activation of the p53 pathway in the CNS and, in general, in mammalian cells, as opposed to what was observed in other organisms. The role of PRMT5 in vivo has indeed been studied in Planaria, Drosophila, and plants. However, the lack of PRMT5 and, subsequently, symmetrically dimethylated Sm proteins results in distinct phenotypes in different systems. In Planaria, PRMT5 depletion results in defects in homeostasis, growth, and regeneration (Rouhana et al. 2012). In flies, it leads to problems in germ cell specification and circadian rhythm (Gonsalvez et al. 2006; Sanchez et al. 2010), while in plants, it leads to defects in flowering and circadian rhythm (Deng et al. 2010; Sanchez et al. 2010).
PRMT5 is not the only type II symmetric arginine methyltransferase in higher eukaryotes. PRMT7 is an essential gene in Drosophila (unlike PRMT5) (Gonsalvez et al. 2008), while in HeLa cells, it has been shown to methylate Sm proteins and have nonredundant functions in cytoplasmic snRNP biogenesis (Gonsalvez et al. 2007). Our data suggest that, at least in development, PRMT7 is not compensating for the absence of PRMT5, but to what extent, if any, PRMT7 plays a role in mammalian splicing in vivo remains to be addressed.
Splicing disorders are the cause of several neurodegenerative diseases (Dredge et al. 2001; Tollervey et al. 2011), such as SMA, in which the expression of the SMN1 gene is perturbed. Interestingly, also in the case of SMA, there is apoptosis of motor neurons. Although SMA is often referred to as a motor neuron disease, recent evidence suggests that the splicing defects are present in multiple organs (Zhang et al. 2008), and this can lead to disease-relevant phenotypes in cells other than motor neurons (Hayhurst et al. 2012). Surprisingly, neurospheres derived from the Smn−/−;SMN2 mice, which represent a severe model of human SMA, did not differ in number when compared with wild type (Shafey et al. 2008). Our data place PRMT5 upstream of SMN in the maturation cycle of Sm proteins in vivo. This discrepancy can be reconciled by two considerations. First, we are looking at a full deletion of PRMT5, as opposed to a hypomorphic SMA model. Indeed, the full SMN knockout has an early embryonic phenotype similar to the PRMT5 full knockout (Hsieh-Li et al. 2000; Tee et al. 2010). Second, PRMT5 might control splicing more broadly than simply through regulating the maturation of snRNPs. In this respect, the splicing regulator CA150 (Cheng et al. 2007) has been identified as a PRMT5 target, and this could be the case for other splicing proteins (Boisvert et al. 2003; Ong et al. 2004).
Role of PRMT5 in stem cell biology
A previous study has reported the constitutive knockout of PRMT5 using a gene trap model (Tee et al. 2010). Phenotypically, lack of PRMT5 leads to derepression of differentiation genes in mouse embryonic stem cells (mESCs) due, at least in part, to the lack of symmetric Arg3 methylation on histone H2A (H2AR3me2s). However, because of the early embryonic lethality, the investigators were not able to perform large-scale molecular experiments, and thus whether PRMT5 plays any role in controlling splicing in mESCs remains to be explored. It is of note that PRMT5 has been used to improve induced pluripotent stem cell (iPSC) derivation in combination with klf4 and Oct3/4 (Nagamatsu et al. 2011). Reducing p53 activity is known to be very important to enhance iPSC derivation (Hong et al. 2009; Kawamura et al. 2009). We thus believe that our data, which link PRMT5 methyltransferase activity to the regulation of MDM4 abundance by controlling its alternative splicing, will provide new insights into this expanding field of research.
Role of PRMT5 in regulating cell cycle progression and cell death
We described here how cells can sense general defects in the core splicing machinery (such as the one caused by PRMT5 depletion) by regulating the alternative splicing of a key p53 activator such as Mdm4. This alternative splicing event reduces the full-length MDM4 protein and gives rise to the unstable MDM4S product (Lenos and Jochemsen 2011), thus activating the p53 transcriptional program. Our findings describe for the first time the link between the methylosome (Friesen et al. 2001b), the core splicing machinery, alternative splicing, and activation of a p53 response in mammalian development.
What we uncovered here is indeed a much broader picture of how cells can activate the alternative splicing of sensor mRNAs (e.g., Mdm4) at key exons, characterized by weak 5′ donor sites. This occurs upon perturbation of the general splicing machinery—whether because of PRMT5 deletion or chemical inhibition (Fig. 5D)—to arrest growth and/or induce apoptosis. Besides Mdm4, there are other mRNAs that can potentially play a similar role. To mention a few, the SE event observed in the mRNA of 5′ cap-binding protein eIF4E, which is a rate-limiting component in the translation process, could affect genes involved in apoptosis and cell cycle arrest (Supplemental Fig. S4D; Mamane et al. 2004), while the RI events observed in Dvl1 might lead to the inactivation of the Wnt/Dvl1/β-catenin signaling pathway, which is known to support NPCs' growth and self-renewal potential (Supplemental Fig. S4E; Faigle and Song 2013).
We described here how the complete depletion of PRMT5 in mouse cells leads to a minor induction of γH2AX but a strong activation of p53 target genes such as p21 even when compared with etoposide. This is in contrast to what has been observed previously using siRNA/shRNA strategies to reduce the levels of PRMT5 in human cancer lines (Jansson et al. 2008; Scoumanne et al. 2009). This discrepancy could be due to differences between mouse and human cells and between primary cells and cancer cells and the fact that PRMT5 levels have to fall below a certain threshold in order to exhibit an activation of the p53 pathway. The latter concept has already been observed for other splicing regulators such as SMN (Zhang et al. 2008).
Both the concept that perturbation of the core splicing machinery can lead to regulation of alternative splicing (Saltzman et al. 2011) and the concept that apoptosis is regulated by alternative splicing (Schwerk and Schulze-Osthoff 2005; Moore et al. 2010) have been described previously. The fundamental advance that we describe here is that the two pathways are directly linked because Mdm4, the target of alternative splicing, unlike Bcl2-like factors, caspases, death receptors, and proapoptotic ligands, is as a direct upstream regulator of p53.
Splicing disorders have been estimated to occur in 50% of tumors (Ritchie et al. 2008; David and Manley 2010; Ward and Cooper 2010). These can contribute to tumor progression by giving rise to alternative isoforms of oncogenes or tumor suppressors. At the same time, while the perturbation of the splicing machinery was known to activate the p53 pathway, the underlying mechanisms by which this occurred were unknown (Allende-Vega et al. 2013). Our data uncover the mechanism of p53 activation by identifying Mdm4 as a key sensor mRNA. We believe these data to be extremely relevant for the entire p53 field. Mdm4 is indeed up-regulated in several p53wt tumors (Gembarska et al. 2012), and our findings provide new therapeutic avenues to alter its protein levels by affecting its splicing pattern.
In conclusion, this study expands our understanding of the complex network regulating correct splicing and cell fate decisions in mammalian development as well as in human cancer lines, providing new possibilities to target the arginine methyltransferase family to treat neurodegenerative diseases (Dredge et al. 2001; Tollervey et al. 2011) and cancer (Ritchie et al. 2008; David and Manley 2010; Ward and Cooper 2010).
Materials and methodsMouse strains and genotyping
The PRMT5 knockout first mice were obtained from the European Conditional Mouse Mutagenesis Program (EUCOMM; http://www.knockoutmouse.org). To generate the PRMT5 FLOX allele, the bgal-neomycin cassette was removed by crossing PRMT5 knockout first mice with β-actin-Flpe transgenic mice (Rodriguez et al. 2000) [strain name: B6.Cg-Tg(ACTFLPe) 9205Dym/J; stock no. 005703, The Jackson Laboratory]. These were then crossed to Nestin-CRE [B6.Cg-Tg(Nes-cre)1Kln/J; JAX Laboratory]. 4-OHT-inducible conditional knockouts were created by crossing PRMT5F/F mice with ER transgenic mice (Hameyer et al. 2007) in a mixed C57BL/6 × 129S1/SvlmJ background. Southern blotting was carried out according to a standard protocol (Southern 2006).
Immunohistochemistry (IHC) staining
Automated IHC staining and counterstaining was performed on the Leica Bond-Max autostainer. Detailed protocols are available in the Supplemental Material.
Cell cultureNPCs
Neurosphere cultures were established as previously described (Lim and Kaldis 2012). Briefly, E14.5 embryos were harvested, and cortices were carefully dissected in ice-cold PBS and incubated in trypsin (Invitrogen, 25300120) for 10 min at 37°C. The tissue was then mechanically dissociated, and single cells were cultured into complete NSC medium (DMEM, Life Technologies, 11965118 + 2% B-27, Life Technologies, 17504-044), 1% penicillin/streptomycin (Life Technologies, 15140122), 20 ng/mL recombinant human epidermal growth factor (EGF) (Peprotech, 100-15), and 20 ng/mL recombinant human fibroblast growth factor-basic (FGF-2) (Peprotech, 100-18B). PRMT5F/F ER day 4 neurospheres were treated with either 50 nM 4-OHT (Sigma, H7904) or the equivalent volume of EtOH for 24 h before splitting to induce PRMT5 knockout.
MEFs
Primary MEFs were prepared from E14.5 embryos as previously described (Xu 2005) and maintained in a humidified 5% CO2 atmosphere at 37°C in DMEM (Life Technologies, 11965118) supplemented with 10% fetal bovine serum (FBS) (Hyclone, SH30070.03) and 1% penicillin/streptomycin (Life Technologies, 15140122). To induce PRMT5 knockout, MEFs (passage 1) were grown to confluence in 15-cm dishes, and either 50 nM 4-OHT (H7904; Sigma) or the equivalent volume of EtOH was added 24 h before splitting.
Human cell lines
HEK293T, Phoenix-Eco, A549, U87, U2OS, and HCT116 were obtained from American Type Culture Collection (ATCC) and propagated according to ATCC data sheets.
Microarray analysis
The expression data from quadruplicate Illumina MouseRef-8 V2 microarrays were quantile-normalized, and only probes with an absolute fold change >1.5-fold and a Q-value <0.01 were labeled as significantly differentially expressed.
RNA-seq library preparation and splicing analysis
For RNA-seq library preparation, we followed the Illumina TruSeq RNA Sample Preparation kit version 2 manual. At least 70 million, 51-base-pair (bp)-long paired-end reads were mapped to the NCBI37/mm9 version of the mouse genome per replicate. Genes were labeled as significantly differentially expressed if the P-value as called by cuffdiff (Trapnell et al. 2012) was <0.05.
To determine differential splicing events, MATS 3.0.6 beta (Shen et al. 2012) was used for counting junction reads and reads falling into the tested region within ENSEMBL version 65 gene definitions. Matching embryos were analyzed individually, and only significant events occurring in at least two replicates were considered. Splicing events were labeled significant if the sum of the reads supporting a specific event exceeded 10 reads, the P-value was <0.05, and the minimum inclusion level difference as determined by MATS was >0.2. All sequencing and microarray data have been submitted to the Gene Expression Omnibus repository and are available under accession number GSE45285.
Functional annotation and network analysis
The functional annotation of the significant microarray and RNA-seq genes was performed with the Database for Annotation, Visualization, and Integrated Discovery (DAVID) (Huang et al. 2009) using Kyoto Encyclopedia of Genes and Genomics (KEGG) pathway (Kanehisa et al. 2012) representations. Network representations were generated using Cytoscape 3.0.1 (Shannon et al. 2003) with the plug-in ClueGO (Bindea et al. 2009). Only gene ontology terms with at least five members and a κ score >0.3 were used.
Polysome purification
Polysomes were isolated and separated as previously described (Zhang et al. 2012).
TAM injections
A single pulse of 2 mg of TAM (Sigma, T5648) plus 1 mg of progesterone (Sigma, P8783) in mineral oil (Sigma, M5904) was given to pregnant females at E10.5. Embryos were harvested and analyzed at E15.5 and E17.5.
Competing interest statement
The research of Z.Q.B. was funded by Lilly Research Laboratories, and he is an employee of Eli Lilly and Company.
Acknowledgments
We thank P.R. Kaldis, S.H. Lim, K. Diril, and U. Surana for sharing protocols and for helpful discussions. We thank L. Tora and P.R. Kaldis for critically reading the manuscript; K. Rogers, S. Rogers, and E.W. Sim for help with histopathology work; and M. Al-Haddawi for help with the pathology description of the PRMT5F/FNes brains. We thank P. Kaldis for sharing the Nestin-CRE and ER transgenic animals, V. Tergaonkar for sharing the p53-null mice, D. Lane and C.F. Cheok for the kind gift of Nutlin, Y. Minoru for the kind gift of Spliceostatin A, E. Makeyev for sharing the RFP-minigene backbone construct, and the BRC Shared facilities for technical support. We are grateful to X. Ruan and the GIS Genome Sequencing Team for help with the Solexa high-throughput sequencing, and the entire E.G. laboratory for critical discussion. This work was supported by an AGA-SINGA (Singapore Graduate Award) fellowship to M.B. and by IMCB, A-STAR. E.G. acknowledges support from JCO-ASTAR grants number 1134c001 and number 11/03/FG/07/04.
Supplemental material is available for this article.
Article is online at http://www.genesdev.org/cgi/doi/10.1101/gad.219899.113.
A binary switch between telomerase-extendible and telomerase-nonextendible states determines telomere length homeostasis. Here, Qiao and coworkers address how shelterin complex component Tpz1 regulates telomere length in fission yeast. Separation-of-function mutant analyses indicate that Tpz1-mediated linkage within the shelterin complex defines the telomerase-nonextendible state. Interestingly, the authors show that Tpz1 also plays a role in the activation of telomeres to the extendible state via its interaction with shelterin component Ccq1. Thus, this study suggests that Tpz1 coordinates both positive and negative regulators of telomere length homeostasis.
Telomeres are nucleoprotein complexes comprising telomeric DNA repeats bound by the multiprotein shelterin complex. A dynamic binary switch between telomerase-extendible and telomerase-nonextendible telomeric states determines telomere length homeostasis. However, the molecular nature of the nonextendible state is largely unknown. Here, we show that, in fission yeast, Tpz1 (the ortholog of human TPP1)-mediated complete linkage within the shelterin complex, bridging telomeric dsDNA to ssDNA, controls the telomerase-nonextendible state. Disruption of this linkage leads to unregulated telomere elongation while still retaining the shelterin components on telomeres. Therefore, the linkage within the shelterin components, rather than the individual shelterin components per se, defines the telomerase-nonextendible state. Furthermore, epistasis analyses reveal that Tpz1 also participates in the activation of telomeres to the extendible state via its interaction with Ccq1. Our results suggest critical regulatory roles of Tpz1 in the telomere binary switch.
Telomeres are DNA–protein complexes that protect the ends of eukaryotic chromosome ends from degradation and prevent their recognition as DNA damage sites (Palm and de Lange 2008; de Lange 2009). Telomere integrity is essential for cell survival and proliferation (Artandi and Cooper 2009; Jain and Cooper 2010), and, accordingly, dysfunctional telomeres can initiate genomic instability, cellular senescence, and organismal aging. The telomeric DNA consists of short tandem DNA repeats, which are G-rich in one strand (called the G strand) and C-rich in the complementary stand (called the C strand). The G strand extends beyond the C strand and forms a single-stranded G overhang. The 3′ end of the G overhang acts as the substrate for telomerase—a reverse transcriptase (Lingner et al. 1997; Nakamura et al. 1997) with its intrinsic RNA containing the template—to extend the telomeric DNA (Autexier and Lue 2006; Collins 2006). The basic structure and function of telomeres are conserved among eukaryotes; this conserved telomere protein complex that interacts with specific telomere sequences and caps chromosome ends is called shelterin (de Lange 2005). Identification of fission yeast (Schizosaccharomyces pombe) Pot1, the telomere ssDNA-binding protein, immediately allowed the discovery of human POT1 (Baumann and Cech 2001). A protein interaction partner of human POT1, TPP1, was subsequently identified (Houghtaling et al. 2004; Liu et al. 2004a; Ye et al. 2004). The POT1/TPP1 complex binds single-stranded telomere DNA with higher affinity than POT1 itself, forming an ortholog of the ciliate TEBP-α/TEBP-β complex—the archetypal telomere ssDNA overhang-binding module (Horvath et al. 1998; Wang et al. 2007). Thanks to the recent development of the powerful proteomic techniques, a nearly complete suite of telomere-localized proteins has been identified in human and fission yeast (Liu et al. 2004b; de Lange 2005; Miyoshi et al. 2008). Similar to the human telomeres, the fission yeast telomere dsDNA repeats are covered directly by sequence-specific dsDNA-binding protein Taz1 (TRF1 and TRF2 in humans) (Cooper et al. 1997). A “protein bridge,” consisting of Rap1, Poz1, and Tpz1 (human TPP1 ortholog), connects the dsDNA and ssDNA regions of the telomere through their direct protein–protein interactions with Taz1 and Pot1, respectively (Miyoshi et al. 2008).
In telomerase-positive cells, such as human embryonic stem cells (Gunes and Rudolph 2013), adult germline cells (Tan et al. 2012), most cancer cells (Shay and Wright 1996, 2010), and single-celled eukaryotes (ciliated protozoa and yeasts) (Greider and Blackburn 1985; Cohn and Blackburn 1995), telomeres are not maintained at a defined length but instead within a rather species-specific range; this telomere length homeostasis is proposed to be mediated through dynamic switching of the chromosome ends between two states: telomerase-extendible and telomerase-nonextendible states (Teixeira et al. 2004). The accessibility of the telomere substrate to telomerase is the key difference distinguishing the two states. The shelterin complex is believed to regulate telomere accessibility and thus control the telomere length homeostasis (Smogorzewska and de Lange 2004). The shelterin complex is a group of interlinked telomere proteins; deleting any member of this group alters telomere length homeostasis. Most if not all telomere proteins are multifunctional; their many functions enable telomere regulation and telomere protection. In the mammalian systems, shelterin components are recruited to telomeres through the dsDNA binders TRF1/TRF2, and interactions between shelterin components are essential for their localizations to the telomeres (Loayza and de Lange 2003; Liu et al. 2004a; Ye et al. 2004; Takai et al. 2011; Sfeir and de Lange 2012). This interdependent telomere association behavior of the mammalian shelterin components hampers a precise understanding of how each individual component and the linkage between them separately contribute to telomerase regulation. S. pombe has a similar shelterin complex (Miyoshi et al. 2008); however, the recruitments of its shelterin components to telomeres are not solely dependent on dsDNA-binding protein Taz1. The Pot1 complex is found to associate with telomeres in a Taz1- or Rap1-independent manner (Miyoshi et al. 2008). This nifty feature of S. pombe shelterin makes it an ideal system to investigate how telomere length homeostasis is achieved through interactions among members of the telomere protein complex without the complications of their dissociations from the telomere due to the loss of interactions.
Among S. pombe telomere proteins, Tpz1 physically lies at the interface of telomeric dsDNA- and ssDNA-binding proteins and is functionally positioned between the positive and negative regulators of telomere elongation (Fig. 1A). In addition to its interactions with Poz1 and Pot1, Tpz1 is also associated with Ccq1—a telomerase recruiter and checkpoint response inhibitor (Flory et al. 2004; Miyoshi et al. 2008; Tomita and Cooper 2008; Jain et al. 2010; Moser et al. 2011; Webb and Zakian 2012; Yamazaki et al. 2012; Nandakumar and Cech 2013). The unique position of Tpz1 in the shelterin complex signifies its architectural role in shelterin complex assembly and implies its coordination roles in communicating the telomeric dsDNA length and structural information to the 3′ end of the G overhang—the ultimate destination of telomerase. Thus, studying Tpz1 could help us understand the molecular mechanism by which the shelterin complex regulates telomeres in the nonextendible state and how it is switched to the telomerase-extendible state—two central questions that await answers. However, previous studies showed that most tpz1Δ cells are inviable. Surviving haploid cells completely lose the telomeric signal and form self-circularized chromosomes (Miyoshi et al. 2008), indicating the critical role of Tpz1 in chromosome end protection. The dominant telomere deprotection consequence of tpz1+ deletion completely masks other important functions of Tpz1 in telomere length homeostasis, making it impossible to study Tpz1's role in telomerase regulation using tpz1Δ cells. Clearly, separation-of-function mutants of Tpz1 are necessary to comprehensively understand the multifunctionality of Tpz1 in telomerase regulation and telomere protection. Recent studies using the separation-of-function mutations on the TEL patch of human TTP1 successfully uncovered its roles in recruiting telomerase to telomeres and promoting telomerase processivity (Nandakumar et al. 2012; Sexton et al. 2012; Zhong et al. 2012). However, the important function of Tpz1 in regulating different telomeric states via its interactions with other shelterin components is still not clear.
Tpz1 interacts with Poz1 and Ccq1 simultaneously using two different patches. (A) Fission yeast shelterin complex. It is composed of telomeric sequence-specific dsDNA- and ssDNA-binding proteins Taz1 and Pot1, respectively, accompanied by their protein interaction partners, Rap1, Poz1, and Tpz1, forming a bridge between Taz1 and Pot1. Ccq1 was recently found to recruit telomerase to telomeres through Est1. If the deletion of a telomere protein causes telomere elongation, this protein is regarded as a negative regulator of telomere length and is therefore labeled “−”; otherwise, it is labeled “+”. For clarity, the stoichiometry of each individual component is not indicated in the figure; only one copy of each component is shown. (B) In vitro GST pull-down assays examining Tpz1–Poz1 and Tpz1–Ccq1 binary and Tpz1–Poz1–Ccq1 ternary interactions. Tpz1-CTD binds to Ccq1-NTD and Poz1 individually (lanes 3,5, respectively) and also at the same time (lanes 6–8). Competition experiments shown in lanes 7 and 8 demonstrate that even when the concentration of Ccq1-NTD is 100 times higher than that of Poz1, Tpz-CTD and Poz1 interaction remains stable. GST-Tpz1-CTD input (shown in lane 1) is one-tenth of the total protein used in the binding assays. Inputs shown in lanes 12–14 are one-fourth of the total Poz1, Ccq1-NTD, and Ccq1-CTD used in the binding assays. (C) Sequence alignment of Tpz1 from three different fission yeasts showing two conserved patches. Highly conserved residues are highlighted in black, and similar residues are highlighted in gray. (D) In vitro GST pull-down assays testing the binding of Tpz1 mutants to Poz1 and Ccq1-NTD. Two groups of mutants were identified that are defective in binding to Ccq1 (colored in blue) and Poz1 (colored in green), respectively. (E) Tpz1-L449A and Tpz1-I501R disrupt Tpz1–Ccq1 and Tpz1–Poz1 interactions, respectively, as evaluated by coimmunoprecipitation assays. Cdc2 is shown as the loading control. (Input) One-thirtieth of input WCE (whole-cell extract). (F) Schematic representation of Tpz1's interactions with Ccq1, Poz1, and Pot1. Tpz1 interacts simultaneously with Ccq1 and Poz1 via its CTD and with Pot1 via its NTD.
Here, we biochemically identified Tpz1 mutants that can individually but specifically disrupt its interactions with Poz1, Ccq1, or Pot1. Using these separation-of-function mutants of Tpz1, we found that the complete linkage between telomere dsDNA- and ssDNA-binding proteins within the shelterin complex is required for defining the telomerase-nonextendible state of telomeres. Disruption of the linkage on either the dsDNA binder or the ssDNA binder side of Tpz1 causes unregulated elongation of telomeres without delocalizing shelterin components from telomeres. Moreover, epistasis analyses of functional roles of Tpz1-centered interactions indicate that Ccq1 may activate the telomerase-nonextendible state of telomeres through its interaction with Tpz1, acting upstream of telomerase recruitment. Our results suggest that Tpz1 acts as the coordinator of positive and negative regulators of telomere length homeostasis and is also a critical regulatory target for controlling the competence of telomeres for elongation.
ResultsTpz1 interacts with Poz1 and Ccq1 simultaneously using two different but adjacent patches
We set out to mechanistically dissect the shelterin complex in S. pombe by focusing on Tpz1 due to its physical connections to multiple other shelterin components. Previous studies using yeast two-hybrid assays demonstrated that both Poz1 and Ccq1 bind to the Tpz1 C-terminal domain (Tpz1-CTD), a domain of ∼100 residues (Miyoshi et al. 2008). However, the investigation was unable to determine whether the two binding events happen simultaneously on Tpz1-CTD due to the limitations of the yeast two-hybrid assay. Using Escherichia coli-expressed, recombinant Tpz1-CTD (residues 406–508), Poz1, the Ccq1 N-terminal domain (Ccq1-NTD; residues 2–439), and Ccq1-CTD (residues 544–716), we performed GST pull-down assays to distinguish whether Poz1 and Ccq1 bind to Tpz1 simultaneously (as drawn in Fig. 1A) or their interactions with Tpz1 are mutually exclusive. As shown in Figure 1B, we found that GST-Tpz1-CTD interacted with Poz1 and Ccq1-NTD but not with Ccq1-CTD. Moreover, increasing the concentration of Ccq1-NTD to 100 times higher than that of Poz1 does not disassociate Poz1 from Tpz1-CTD. This experiment demonstrates that both Poz1 and Ccq1 can bind to Tpz1-CTD at the same time.
Simultaneous binding of Ccq1-NTD and Poz1 to Tpz1-CTD implies that there are two different surfaces on the Tpz1-CTD that mediate its interactions with Ccq1-NTD and Poz1, respectively. We then aimed to determine whether Tpz1–Poz1 interaction and Tpz1–Ccq1 interaction could be biochemically separated by distinct Tpz1 mutations. We hypothesized that functionally important residues should be evolutionarily conserved and thus represent leading candidates for residues to be targeted by mutagenesis. Sequence alignment of Tpz1 from S. pombe and two other fission yeast relatives, Schizosaccharomyces japonicus and Schizosaccharomyces octosporus (Rhind et al. 2011), reveals two clusters of relatively conserved patches (labeled with green and blue bars above the sequences in Fig. 1C). We hypothesized that these two patches correspond to the binding sites for Ccq1 and Poz1, respectively. Therefore, we made GST-Tpz1-CTD mutants and tested their binding ability to Poz1 and Ccq1-NTD. As predicted, mutations on one conserved patch (labeled in blue in Fig. 1C) disrupted Tpz1–Ccq1 binding while retaining Tpz1–Poz1 interaction, with the converse result for mutations on the other patch (labeled in green in Fig. 1C). As controls, we introduced mutations in the nonconserved residues between the two patches. These mutations disrupted neither Tpz1–Ccq1 nor Tpz1–Poz1 interaction. Given that the Tpz1 mutants on each conserved patch retain their binding to at least one protein, it is unlikely that the mutations cause global unfolding of Tpz1-CTD. Furthermore, using coimmunoprecipitation, we confirmed that mutations that abrogate Tpz1-CTD and Poz1 or Tpz1-CTD and Ccq1 interaction in vitro also disrupted the interactions between the full-length Tpz1 and Poz1 or between the full-length Tpz1 and Ccq1, respectively (Fig. 1E). Therefore, we identified two adjacent patches merely ∼50 residues away from each other on Tpz1-CTD that mediate its binding to Poz1 and Ccq1, respectively (as shown in Fig. 1F). Having obtained and validated Tpz1 separation-of-function mutations, we next used these mutants for functional dissection of the roles of Tpz1 in telomere maintenance in vivo.
Disruption of Tpz1–Poz1 interaction results in dramatically elongated telomeres
To elucidate the functional roles of Tpz1–Poz1 interaction in telomere maintenance, we made two S. pombe strains with tpz1 mutants encoding proteins defective in Poz1 binding: tpz1-I501R and tpz1-I501A/R505E. Both mutants had dramatically elongated telomeres, although not to the extent of telomeres in the poz1Δ cells (Fig. 2A; Miyoshi et al. 2008). This experiment indicates that the direct interaction of Poz1 with the telomeric ssDNA-binding complex Tpz1–Pot1 is required for the negative regulatory function of Poz1. To further investigate whether the negative regulatory role of Tpz1–Poz1 interaction in telomere length is imposed on telomerase, we deleted trt1+ in the tpz1-I501R mutant to make tpz1-I501R/trt1Δ and test whether telomere elongation resulting from loss of Tpz1–Poz1 interaction is telomerase-dependent. The classic “ever-shorter telomere” phenotype appears in the tpz1-I501R/trt1Δ double-mutant cells, indicating that telomerase is downstream from the defective Tpz1 function in tpz1-I501R (Fig. 2B). Consistent with the above, the telomere lengths of the tpz1-I501R/rad51Δ or tpz1-I501R/rad55Δ double mutant were not shortened upon deleting rad51+ or rad55+, which are required for homologous recombination (HR)-dependent telomere elongation (Fig. 2C). In addition, chromatin immunoprecipitation (ChIP) assays demonstrated that in both tpz1-I501R and tpz1-I501A/R505E mutant cells, Poz1 remains associated with telomeres (Fig. 2D), possibly via its interaction with Rap1. This result further indicates that it is the interaction between Tpz1 and Poz1, not Poz1 per se, that leads to the negative regulation of telomerase. Taken together, we conclude that the interaction between Tpz1 and Poz1 negatively regulates telomerase because disruption of this interaction results in telomerase-dependent, dramatically elongated telomeres. It is worth noting that the ∼1.5-kb more telomere elongation in poz1Δ cells compared with tpz1-I501R or tpz1-I501A/R505E cells suggests that Poz1 may have some additional roles in negatively regulating telomerase via a pathway that is separable from Poz1–Tpz1 interaction.
Tpz1–Poz1 interaction negatively regulates telomerase. (A, top) Diagram of the telomere region showing relative positions of restriction enzyme sites and the telomere DNA probe for Southern blots. (Bottom) Southern blot analysis of EcoRI-digested genomic DNA using the telomere DNA probe for the indicated tpz1-I501R and tpz1-I501A/R505E mutant strains from successive restreaks on agar plates. Disruption of Tpz1–Poz1 interaction causes elongated telomeres that are ∼1.5 kb shorter than poz1Δ. In the telomere length analysis Southern blots presented in this study, either the 2-log DNA ladder from New England Biolabs (for A and Fig. 3D) or the 1-kb plus marker from Invitrogen (the rest of the telomere blots) was used and is denoted as “M.” Wild-type cells are denoted as “wt” in the blot. (B) Telomere maintenance in tpz1-I501R cells is telomerase-dependent. (C) Telomere maintenance in tpz1-I501R cells is not HR-dependent. (D) Recruitment of Poz1 to telomeres in tpz1-wt, tpz1-I501R, and tpz1-I501A/R505E cells was monitored by ChIP assays. Telomere association of Poz1 in each genetic background was monitored by slot blot. After hybridization with the telomere probe, the same membrane was stripped and then hybridized with the rDNA probe. Telomeric enrichment of Poz1 was expressed as immunoprecipitate (IP)/whole-cell extract (WCE) from the telomere DNA probe. Error bars in the quantitation of the slot blot analysis represent standard deviations of two individual repeats.
Disruption of Tpz1–Pot1 interaction also elongates telomeres
In fission yeast, the telomeric dsDNA is linked to a single-stranded G overhang through a protein bridge composed of Taz1, Rap1, Poz1, Tpz1, and Pot1, as shown in Figure 1A. Deletion of Taz1, Rap1, or Poz1 causes telomere elongation (Cooper et al. 1997; Miyoshi et al. 2008), while deletion of either Tpz1 or Pot1 leads to telomere deprotection and shortening and the subsequent formation of circularized chromosomes (Baumann and Cech 2001; Miyoshi et al. 2008; Pitt and Cooper 2010). This strong phenotype masks the functional roles of Tpz1 and Pot1 in telomerase regulation. Our finding shows that just like deletion of taz1+, rap1+, or poz1+, disruption of Tpz1–Poz1 interaction also disrupts negative regulation of telomerase and leads to elongated telomeres. What these alterations share in common is that they all break the complete linkage between the telomeric dsDNA and ssDNA mediated by the shelterin components. It is reasonable to hypothesize that if any linkage point between the telomeric dsDNA- and ssDNA-binding proteins is removed or broken, negative regulation of telomerase will be abrogated, thereby keeping telomeres in a telomerase-extendible state constitutively. The full connection between telomeric dsDNA-binding protein and ssDNA-binding protein may define the telomerase-nonextendible state of the telomere. This model predicts that disruption of Tpz1–Pot1 interaction will also lead to elongated telomeres in a telomerase-dependent manner, just as disruption of Tpz1–Poz1 interaction does. To test this, we set out to identify a specific mutant of Tpz1 that disrupts its interaction with Pot1, as we did for Poz1 and Ccq1 above.
To this end, we resorted to the crystal structure of TEBP-α/β, the Oxytricha nova homolog of the S. pombe Pot1/Tpz1 complex (Horvath et al. 1998). Through a secondary structure prediction of Tpz1 and its homology with the TEBP-α/β structure (as shown in Supplemental Fig. S1), we identified a loop region in the Tpz1-NTD (residues 1–234) (Nandakumar and Cech 2012) as a strong candidate to mediate Tpz1–Pot1 interaction (colored green in Fig. 3A). Within that loop of Tpz1, we introduced point mutations individually to residues that are conserved among fission yeasts. The purified recombinant GST-Tpz1-NTD mutants, produced in E. coli, were then subjected to GST pull-down assays to evaluate their binding ability to Pot1. As shown in Figure 3B, among nine Tpz1-NTD mutants, Tpz1-I200R evidently abolished the in vitro association between Tpz1 and Pot1.
Tpz1–Pot1 interaction negatively regulates telomerase. (A) Structural representation of the O. nova TEBP-α/TEBP-β complex, an ortholog of the Pot1/Tpz1 complex. The loop region of TEBP-β that mediates protein–protein interaction between TEBP-α and TEBP-β is indicated by an arrow and colored green. (B) In vitro GST pull-down assays examining the binding of Tpz1-NTD mutants to Pot1. (C) Tpz1-I200R disrupts Tpz1–Pot1 interaction, as evaluated by coimmunoprecipitation assays. Cdc2 was shown as the loading control. (Input) One-thirtieth of input WCE (whole-cell extract). (D) Telomeres elongate in the tpz1-I200R mutant, as shown by Southern blot analysis. (E) Telomere maintenance in the tpz1-I200R mutant is telomerase-dependent but not HR-dependent.
Next, we confirmed that Tpz1-I200R mutation, which abrogates Tpz1-NTD and Pot1 interaction in vitro, also disrupts full-length Tpz1–Pot1 interaction in coimmunoprecipitation assays (Fig. 3C). Consistent with our model, telomeres in tpz1-I200R mutant cells are also elongated (Fig. 3D), reminiscent of the tpz-I501R mutant in which Tpz1–Poz1 interaction is disrupted. In addition, just like we observed in the tpz-I501R mutant, deletion of telomerase resulted in an “ever-shorter telomere” phenotype in the tpz1-I200R/trt1Δ double mutant, whereas both the tpz1-I200R/rad51Δ and tpz1-I200R/rad55Δ double mutants maintain elongated telomeres (Fig. 3E), indicating that telomere elongation in the tpz1-I200R mutant is telomerase-dependent but not HR-dependent.
Both Tpz1–Poz1 and Tpz1–Pot1 interactions function in the same pathway as Rap1 and Poz1 in negatively regulating telomerase
We further asked whether Tpz1–Pot1 interaction functions in the same pathway as Tpz1–Poz1 in telomerase regulation. To address this, we constructed a tpz1-I200R/I501R double-mutant strain and found that its telomeres were also elongated (Fig. 4A), similar to the single-mutant tpz1-I200R or tpz1-I501R but not additionally, indicating epistasis between the mutants. Hence, Tpz1–Pot1 interaction negatively regulates telomerase-mediated telomere elongation in the same pathway as Tpz1–Poz1 interaction. In the S. pombe shelterin complex, Tpz1, Poz1, and Rap1 connect the telomeric dsDNA-binding protein Taz1 to the G-overhang-binding protein Pot1. Deletion of rap1+ or poz1+causes telomerase-dependent telomere elongation, much like the phenotypes observed for Tpz1–Poz1 or Tpz1–Pot1 interaction-defective mutants. To determine whether the “connector” proteins Rap1 and Poz1 also act in the same pathway as Tpz1–Poz1 and Tpz1–Pot1 interactions to negatively regulate telomerase, we employed epistasis analysis and generated a set of double mutants in which rap1+ or poz1+ was individually deleted in tpz1-I200R and tpz1-I501R mutant strains. As shown in Figure 4, B and C, deletion of either rap1+ or poz1+ in both Tpz1–Pot1 interaction-defective (tpz1-I200R) and Tpz1–Poz1 interaction-defective (tpz1-I501R) mutant cells still produced elongated telomeres in the double-mutant cells; moreover, none of the double mutants had telomeres longer than those of the single mutants, suggesting that Rap1 and Poz1 act through Tpz1 and then Pot1 to prevent telomerase from elongating telomeres. Thus, we conclude that breaking any linkage within the shelterin complex connecting telomeric dsDNA and ssDNA leads to loss of negative regulation and therefore to elongated telomeres. Recent work showed that telomerase recruitment can be mediated by the interaction between telomerase subunit Est1 and Ccq1 when Thr93 of Ccq1 is phosphorylated by Tel1ATM and/or Rad3ATR. A nonphosphorylable mutant of Ccq1, Ccq1-T93A, cannot bind to Est1 and therefore fails to promote telomerase recruitment (Moser et al. 2011; Webb and Zakian 2012; Yamazaki et al. 2012). Taking advantage of this discovery, we constructed tpz1-I200R/ccq1-T93A and tpz1-I501R/ccq1-T93A double mutants and found that both double-mutant cells showed progressive telomere loss (Fig. 4D), suggesting that Tpz1-mediated negative telomerase regulation functions upstream of telomerase recruitment.
Tpz1–Poz1 and Tpz1–Pot1 interactions act in the same pathway as Rap1 and Poz1 to negatively regulate telomerase. (A) tpz1-I200R/I501R cells have elongated telomeres, similar to those of tpz1-I501R cells. (B) Double-mutant strains tpz1-I200R/poz1Δ and tpz1-I501R/poz1Δ both have telomere length similar to the rap1Δ single-mutant strain. (C) Double-mutant strains tpz1-I200R/rap1Δ and tpz1-I501R/rap1Δ both have telomere length similar to the poz1Δ single-mutant strain. (D) Double-mutant strains tpz1-I200R/ccq1-T93A, tpz1-I501R/ccq1-T93A, and poz1Δ/ccq1-T93A show progressive telomere shortening.
Taken together, our finding implies that the complete linkage between the double-stranded and single-stranded telomeric DNA may define the telomerase-nonextendible telomeric state in which the telomere is a “dead” substrate for telomerase to elongate and has to be activated before telomerase can extend it. This mechanism provides a key molecular element about how the negative regulatory information is delivered from the telomere dsDNA side to 3′ of the G overhang—where telomerase works.
Loss of Ccq1–Tpz1 interaction causes telomere shortening and telomere maintenance via HR
Ccq1, a more recently identified factor in fission yeast telomere maintenance, has been shown to be required for telomerase recruitment and inhibition of DNA damage-induced checkpoint activation at telomeres (Flory et al. 2004; Miyoshi et al. 2008; Tomita and Cooper 2008). Using our biochemically identified Ccq1–Tpz1 interaction-defective mutant (tpz1-L449A) in hand (Fig. 1D), we explored the functional significance of this interaction in telomere maintenance. We found that S. pombe cells containing the tpz1-L449A mutation elongated progressively with successive generations (Fig. 5A), a phenotype characteristic of DNA damage checkpoint activation. When tpz1-L449A cells reach 50 generations, ∼50% of the cells are twice as long as wild-type cells, reminiscent of ccq1Δ cells. This observation suggested that disruption of Ccq1–Tpz1 interaction triggers a checkpoint pathway, similar to the deletion of ccq1+. In addition, telomere length in tpz1-L449A cells, as shown in Figure 5B and Supplemental Figure S2A, is stable but ∼150 base pairs (bp) shorter than that in wild-type cells directly following sporulation from tpz1-L449A heterozygous diploid cells (tpz1-L449A/+). Similar to ccq1Δ cells (Tomita and Cooper 2008), tpz1-L449A cells appear to stably maintain short telomeres many more generations than the trt1Δ cells, in which telomeres shorten progressively. Furthermore, the telomere maintenance in tpz1-L449A cells is not dependent on telomerase (Fig. 5C); instead, it is achieved via a HR mechanism because deletion of either rad51+ or rad55+ immediately abrogates the stably maintained short telomeres in tpz1-L449A cells (Fig. 5D) and eventually generates survivals with circular chromosomes (Fig. 5E) that have lost ∼10-kb-long subtelomeric regions (Supplemental Fig. S2B). Since Rad55 is not required for telomere maintenance in ccq1Δ cells (Tomita and Cooper 2008) but is required for cells with defective Tpz1–Ccq1 interaction, Ccq1 might have additional roles in regulating HR-based telomere maintenance. In conclusion, shortened telomeres and loss of telomerase-mediated telomere elongation in Tpz1–Ccq1 interaction-defective mutant tpz1-L449A cells indicates that Ccq1 requires interaction with Tpz1 to carry out its functions as a positive regulator of telomerase.
Loss of Ccq1–Tpz1 interaction causes telomere shortening and telomere maintenance via HR. (A) tpz1-L449A cells show elongated cell shape. (B) Telomeres in tpz1-L449A cells are ∼150 bp shorter than those of the wild-type cells, similar to those in ccq1Δ cells. (C) Telomere maintenance in tpz1-L449A cells is not dependent on telomerase. (D) Telomere maintenance in tpz1-L449A mutants is dependent on HR. (E) Deletion of either rad51+ or rad55+ in tpz1-L449A cells leads to chromosome circularization.
Loss of Ccq1–Tpz1 interaction does not affect the association of telomerase with telomeres or cause telomere deprotection
As mentioned above, recent studies demonstrated that Ccq1 is critical for telomerase recruitment via its interaction with Est1; this interaction is regulated by the phosphorylation of Ccq1-Thr 93 by Tel1ATM and/or Rad3ATR (Moser et al. 2011; Yamazaki et al. 2012). One possible outcome of disrupting Tpz1–Ccq1 interaction is that Ccq1 might not be able to localize to telomeres and thus fails to recruit Est1 and Trt1 to telomeres. To test this possibility, we performed ChIP assays to examine whether the associations of Ccq1 and/or Trt1 with telomeres were abolished in tpz1-L449A cells. To our surprise, we found that both Ccq1 and Trt1 bound to telomeres at levels similar to those in the wild-type strain, just as in the other two Tpz1 mutant strains (tpz1-I200R and tpz1-I501R) bearing elongated telomeres (Fig. 6A,B). In contrast, we determined from the same ChIP assay that Trt1 fails to localize to telomeres in ccq1Δ and ccq1-T93A cells (both are known to affect telomerase recruitment) (Fig. 6B). In addition, we found that, in all three Tpz1 interaction mutant strains, Ccq1 specifically bound to the telomere region or the first ∼300 bp of the subtelomeric region close to the telomere but not to the rest of the subtelomeric regions covering ∼20 kb from the chromosome end (Fig. 6C). This result indicated that telomere shortening in tpz1-L449A cells was not due to the failure to recruit telomerase to telomeres. This observation implies that Ccq1 and therefore telomerase may also be recruited to telomeres through another functional patch on Tpz1 or by other shelterin components.
Loss of Ccq1–Tpz1 interaction does not affect the association of telomerase with telomeres or cause telomere deprotection. (A) Telomere association of Ccq1 in the indicated tpz1 mutants was assayed by ChIP assay. Immunoprecipitated DNA was applied to a slot blot for hybridization. After hybridization with the telomere probe, the same membrane was stripped and then hybridized with the rDNA probe. Error bars in the quantitation of the slot blot analysis represent standard deviations of two individual repeats. (B) ChIP analysis of Trt1. Telomere association of Trt1 in each background was monitored by quantitative real-time PCR with a primer pair against subtelomeric region I. The same PCR reaction using a primer pair against an fbp1 gene fragment was carried out as background control. Plots show mean values ± SD for two independent experiments. (C) Binding of Ccq1 to the telomeric and subtelomeric regions was evaluated by quantitative real-time PCR-ChIP analysis. (D) Telomere association of Pot1 in the indicated tpz1 mutant cells was monitored by a dot blot ChIP assay. Error bars in the quantitation of the dot blot analysis represent standard deviations of two individual repeats.
We next addressed whether disruption of Ccq1–Tpz1 interaction caused the dissociation of single-stranded telomere overhang-binding protein Pot1 from the telomeres, which would result in telomere deprotection and subsequent telomere shortening. As shown in Figure 6D, ChIP assays suggested that Pot1 interacted with telomeres of tpz1-L449A cells at a level similar to those in wild-type cells or Tpz1 mutants tpz1-I200R and tpz1-I501R. Furthermore, ChIP assays examining the telomeric association of Poz1 (Supplemental Fig. S3A) or Tpz1 (Supplemental Fig. S3B) showed that their presence at telomeres was not affected by tpz1-L449A mutation or the other two Tpz1 mutations (tpz1-I200R and tpz1-I501R) (Supplemental Fig. S3B,C). Altogether, we conclude that telomere integrity is not compromised in Ccq1–Tpz1 interaction-defective cells.
Ccq1 activates the telomerase-nonextendible state of telomeres via its interaction with Tpz1
Since Ccq1–Tpz1 interaction does not affect telomerase recruitment to telomeres, we hypothesized that this interaction may stimulate telomerase action through the substrate side; namely, Ccq1 may interact with Tpz1 to induce a breakdown of Tpz1-mediated complete linkage between telomeric dsDNA and ssDNA, switching telomeres from a telomerase-nonextendible to a telomerase-extendible state. To investigate this possibility, we took advantage of a Tpz1 mutant strain characterized earlier in this study, tpz1-I501R, in which telomeres stay constitutively in the telomerase-extendible state due to the loss of complete linkage between the telomeric ssDNA and dsDNA by shelterin. If our hypothesis were correct, the tpz1-I501R mutant would bypass the requirement for the telomere activation step mediated by Ccq1–Tpz1 interaction. Indeed, tpz1-L449A/I501R double-mutant cells had telomeres elongated similarly to the tpz1-I501R single mutant (Fig. 7A). Moreover, we carried out the same experiment in the poz1Δ mutant background, which is similarly defective in negative regulation of telomere extension. poz1Δ/tpz1-L449A double-mutant cells showed telomere length identical to poz1Δ cells (Fig. 7B), further confirming bypass of Ccq1–Tpz1 interaction for telomere extension when telomeres are constitutively extendible. Not surprisingly, both the tpz1-I200R/L449A double-mutant cells and the tpz1-L449A/I501R/ poz1Δ triple-mutant cells also bypassed the necessity of Ccq1–Tpz1 interaction for telomerase-mediated telomere elongation (Supplemental Fig. S4A,B). However, tpz1-I501R cannot suppress ccq1Δ due to the lack of telomerase recruitment through Ccq1; instead, tpz1-I501R/ccq1Δ double-mutant cells appear to have the telomere deprotection phenotype (Supplemental Fig. S4C), similar to poz1Δ/ccq1Δ double-mutant cells (Miyoshi et al. 2008).
Ccq1–Tpz1 interaction activates the telomerase-nonextendible state of telomeres. (A) tpz1-L449A/I501R cells have elongated telomeres, which are the same length as tpz1-I501R cells. (B) tpz1-L449A/poz1Δ cells have elongated telomeres, which are the same length as poz1Δ cells. (C,D) Telomere maintenance in the tpz1-L449A/I501R mutant (C) and tpz1-L449A/poz1Δ mutants (D) is telomerase-dependent. (E) A model for controlling telomerase-nonextendible telomeric state by the shelterin linkage and its switching to the extendible state by Ccq1–Tpz1 interaction.
Furthermore, we confirmed that in both tpz1-L449A/I501R and tpz1-L449A/poz1Δ cells, telomere elongation was mediated by telomerase because deletion of trt1+ in both strains led to telomere shortening (Fig. 7C,D). Thus, we conclude that Ccq1–Tpz1 interaction acts upstream of Tpz1–Poz1 interaction, Tpz1–Pot1 interaction, or likely all other negative regulators (such as Poz1) that retain telomeres in the nonextendible state. Therefore, Ccq1–Tpz1 interaction is necessary for the activation of telomeres to the extendible state for telomerase to elongate. Thus, as shown in Figure 7E, we propose that in addition to its telomerase recruitment role, Ccq1 is also a negative regulator of the shelterin-composed dsDNA–ssDNA negative regulatory bridge for telomerase-mediated telomere elongation; its interaction with Tpz1 is required to antagonize the negative force transmitted through Taz1–Rap1–Poz1–Tpz1–Pot1 (red curved line in Fig. 7E) on telomere extension. In other words, Ccq1 interacts with Tpz1 to induce the activation of the shelterin-controlled nonextendible state of telomeres, switching it to the extendible state.
Discussion
The equilibrium between telomerase-extendible and telomerase-nonextendible states is regulated by telomere length; short telomeres are elongated by telomerase more frequently than long telomeres (Teixeira et al. 2004). This equilibrium contributes to telomere length homeostasis, which in turn defines the telomere length in a species-specific range. Understanding the molecular nature of these two telomeric states depends on elucidating the functional roles of each individual molecular interaction among the telomere proteins. As telomere proteins are all interconnected with more than one interacting partner, separation-of-function mutants of them, mostly identified genetically (such as cdc13-1, cdc13-2, etc.), have been instrumental in revealing their multifunctionality in telomere maintenance (Evans and Lundblad 1999; Qi and Zakian 2000; Chandra et al. 2001; Pennock et al. 2001). However, in this “genetics first” approach, the biochemical properties of some of the mutants are difficult to clarify. In this study, we focused on fission yeast Tpz1, which physically lies in the interface of telomeric ssDNA- and dsDNA-binding proteins and is functionally positioned between the positive and negative regulators of the telomere elongation. It is almost impossible to study Tpz1's role in telomerase regulation because tpz1+ deletion immediately leads to circular chromosomes, presumably due to the loss of the telomere protection function of Tpz1. To overcome this complication, we biochemically identified Tpz1 mutants, which can individually but specifically disrupt its interaction with Pot1, Poz1, or Ccq1 while maintaining critical telomere protection functions. Coupled with epistasis analyses, these Tpz1 separation-of-function mutants allowed us to further dissect the multifaceted roles of Tpz1 in regulating telomerase-extendible and telomerase-nonextendible states.
A model for the nonextendible telomeric state
Shelterin is believed to have an inhibitory effect on telomerase (Bianchi and Shore 2008). Earlier studies, primarily using genetic deletions in the yeast systems and RNAi knockdown in the human system, have revealed that most of the shelterin components act as negative regulators of telomerase, which includes TRF1 and TRF2 (homologs of Taz1), RAP1, and TIN2 (homolog of Poz1). Telomere length phenotypes of POT1/TPP1 knockdown in human cells are controversial (Colgin et al. 2003; Veldman et al. 2004), and deletion of pot1+ or tpz1+ in fission yeast leads to telomere deprotection. These observations support the proposal that inhibition of telomerase through integration of telomere length information is transduced from the dsDNA-binding TRF1 complex, including TRF2, TIN2, and TPP1, to the telomere terminus through recruiting POT1 to the very end, thereby controlling telomere accessibility to telomerase (Marcand et al. 1997; Loayza and de Lange 2003; Barrientos et al. 2008; Kendellen et al. 2009). Alternatively, one or more shelterin components could also directly act on telomerase to enforce their negative roles. Deletion or knockdown of whole components cannot distinguish between these two possibilities. Here, by examining the telomere length of our Tpz1 separation-of-function mutants, in which only one residue of a protein is altered, likely preserving all the other functions, we can unambiguously evaluate the contributions of specific protein–protein interactions to telomerase regulation and avoid the complications from simultaneously losing other functional interactions mediated by the same protein. Our data together with previous work (Cooper et al. 1997; Miyoshi et al. 2008; Chen et al. 2011) demonstrate that the complete linkage between the double-stranded telomeric DNA and single-stranded G overhang through shelterin, but not individual shelterin components per se, controls telomeres in the telomerase-nonextendible state (Fig. 7E). The importance of the linkage in controlling telomeres in the nonextendible state is evident in both tpz1-I200R and tpz1-I501R mutants, in which shelterin components are disconnected due to the loss of Tpz1–Pot1 and Tpz1–Poz1 interactions, respectively, but are still associated with telomeres (shown by the ChIP assays in Figs. 2D, 6D; Supplemental Fig. S3). In particular, we show through the tpz1-I200R mutant that fission yeast Pot1 itself cannot inhibit telomerase even if it is associated with telomeres. The complete linkage from Pot1 to Taz1 is an essential element to prevent telomerase from elongating telomeres. Structurally, either the T-loop or G-overhang fold-back model can render telomerase-nonextendible telomeres because in both structures the very 3′ end of the G overhang is sequestered from being accessible to the telomerase. Importantly, the linkage between the dsDNA and ssDNA telomere binders is a shared key requirement for both structures. The longer the telomere, the more telomere dsDNA-binding proteins will be on the telomere and therefore more likely to form the linkage with the ssDNA-binding protein via the bridging proteins. Thus, our findings help explain why long telomeres tend to fall in the telomerase-nonextendible state.
The physical association of telomerase with telomeres is necessary but not sufficient to elongate telomeres, as we observed in the Tpz1–Ccq1 interaction-defective tpz1-L449A cells, in which Trt1 still binds to the telomere (Fig. 6B), but new telomere addition by telomerase does not occur (Fig. 5B) unless the negative regulation of telomeres resulting from the shelterin linkage is removed. Telomeres in the nonextendible state are basically “dead” substrates for telomerase and need to be activated. It is not hard to envision that the model that we propose for fission yeasts is applicable to the human system as well, given the conservation of the functional counterparts and pairwise interactions in both systems. In human cancer cells, telomerase is suggested to be prepositioned on telomeres. Under telomere steady-state maintenance conditions with normal telomere length, most telomeres are elongated only one round by telomerase every cell cycle; however, after telomeres are artificially shortened, the extendible telomeric state (also called open conformation) lasts longer, and therefore more telomerase molecules were observed to carry out additional rounds of telomere extensions in order to rapidly elongate telomeres to be above the critical length to avoid the activation of DNA damage signaling (Zhao et al. 2009, 2011).
Activation of the telomere substrate acts upstream of the telomerase recruitment
To be elongated by telomerase, the very 3′ end of the telomere has to become accessible. How does the telomere switch from the telomerase-nonextendible state to the telomerase-extendible state? Our in-depth epistasis analyses of Tpz1 mutants provide a clue. Without Ccq1–Tpz1 interaction (in the tpz1-L449A background), telomerase can still localize to telomeres but cannot elongate them; however, when poz1+ is deleted or Tpz1–Poz1 interaction is disrupted in this background, telomeres can be elongated again (Fig. 7A,B). This result indicates that Ccq1–Tpz1 interaction acts upstream of the shelterin linkage, the negative force keeping telomeres in the telomerase-nonextendible state. In other words, Ccq1 interacts with Tpz1 to activate the telomerase-nonextendible state. Ccq1 is therefore an inhibitor of nonextendible telomeres (net activation of extension) along with its role as a telomerase recruiter. In addition to the telomerase activation step (Taggart et al. 2002), our observation suggests that telomeres also need to be activated to make the elongation happen. This finding opens a brand new route to explore the biochemical mechanisms whereby the nonextendible telomeres are activated and become optimal substrates for telomerase. One possibility is that Ccq1–Tpz1 interaction may induce post-translational modifications of Tpz1 (such as phosphorylation by a kinase). The modified Tpz1 then loses its interaction with either Poz1 or Pot1, switching the telomere to the telomerase-extendible state.
In summary, our study reveals a key mechanistic aspect of the telomerase-nonextendible telomeric state and provides the first genetic evidence of how it is activated to the extendible state. Telomerase enzyme inhibitors have been developed as promising anti-cancer drugs (Harley 2008). Our mechanistic understanding of the nature of the telomerase-nonextendible state makes the telomere, the substrate, also “druggable.” Locking telomeres in the nonextendible state represents a possible new therapeutic approach in addition to inhibiting the enzyme.
Materials and methodsYeast strains, gene tagging, and mutagenesis
Fission yeast strains used in this study are listed in Supplemental Table S1. Single-mutant strains were constructed by one-step gene replacement of the entire ORF with the selectable marker. Double-mutant and triple-mutant strains were produced by mating, sporulation, dissection, and selection followed by PCR verification of genotype. Genes were fused to specific epitope tags at the C terminus by HR; the pFA6a plasmid modules were used as a template for the PCR reaction (Bahler et al. 1998; Sato et al. 2005). Point mutations were made by mutagenesis PCR using the high-fidelity polymerase Pfu. All mutations were confirmed by DNA sequencing (Eton).
Protein expression and purification
The constructed plasmids were transformed into Rosetta-BL21(DE3) cells; protein expressions were induced by adding IPTG to a final concentration of 0.3–0.4 mM for 4–5 h at 30°C or 0.1–0.2 mM IPTG overnight at 16°C. Cells were harvested by centrifugation at 5000 rpm, and pellets were resuspended in lysis buffer (25 mM Tris-HCl at pH 8.0, 350 mM NaCl, 5 mM β-mercaptoethanol, 2 mM PMSF). Cells were disrupted by sonication, and the supernatant was incubated with equilibrated Ni-NTA (Qiagen) resin for 1 h. After centrifugation at 2000 rpm for 2 min, the resin was washed twice with B350 wash buffer (25 mM Tris-HCl at pH 8.0, 350 mm NaCl, 15 mM imidazole, 2 mM β-mercaptoethanol), and the protein was stepwise-eluted with elution buffer containing up to 300 mM imidazole.
GST pull-down assay
Fifteen microliters of 1 mg/mL GST fusion protein was incubated with 20 μL of glutathione sepharose beads for 1 h at 4°C. After incubation, the beads were washed twice with 800 μL of GST pull-down buffer (50 mM Tris-HCl at pH 8.0, 200 mM NaCl, 10 mM β-ME, 0.05% Tween-20). The bound proteins were then incubated with 20 μL of target protein (1 mg/mL) for 1 h at 4°C with gentle rocking. After washing three times with 800 μL of GST pull-down buffer, the supernatants were removed by centrifugation at 3,000 rpm for 30 sec and then boiled for 5 min in 15 μL of 2× SDS loading buffer. Eluted proteins were resolved by 10% SDS-PAGE and then visualized by Coomassie blue staining.
Coimmunoprecipitation
Frozen S. pombe cells cells were cryogenically disrupted using CryoMill (Retsch) and then resuspended in ice-cold lysis buffer (50 mM Tris-HCl at pH 7.5, 200 mM NaCl, 2 mM EDTA, 0.1% Triton X-100, Complete proteinase inhibitor [Roche], 1 mM DTT, 2 mM PMSF, 2 mM benzamidine, 1 mM Na3VO4, 1 mM NaF). Extracts were clarified, and a final concentration of extracts was adjusted to 10 mg/mL. Anti-Flag M2 affinity gel (from Sigma) was equilibrated and washed twice with the same lysis buffer. Immunoprecipitations were performed for 4 h at 4°C and washed; proteins were eluted from the beads by incubating for 10 min at room temperature with 30 μL of 0.1 M glycine solution (pH 2.0) followed by the addition of 2 μL of 1 M Tris-HCl (pH 8.0). Eluted proteins were resolved by 10% SDS-PAGE and then subjected to Western blotting. Western blot analysis was performed using monoclonal anti-Flag (M2-F1804, from Sigma), monoclonal anti-Myc (9E10, from Covance), anti-Cdc2 (y100.4, from Abcam), or monoclonal anti-HA (3F10, from Roche). Whole-cell extracts were prepared using either trichloracetic acid (TCA) or urea lysis buffer with protease and phosphatase inhibitors.
Pulsed-field gel electrophoresis and Southern blotting for telomere length analysis
S. pombe cells grown in 2 mL of YEAU medium were used to extract chromosomal DNA, which was then digested by NotI. The digested DNA in plugs was subjected to pulsed-field gel electrophoresis as described (Moser et al. 2011). For telomere length analysis by Southern blotting, EcoRI-digested genomic DNA from 2 mL of YEAU S. pombe culture was separated on 1% agarose gel and probed with a telomeric DNA probe as previously described (Moser et al. 2011).
ChIP
S. pombe cells were grown at 32°C in YEAU to OD600 0.5–0.6, shifted for 1 h to 20°C prior to 20-min fixation with an 11% formaldehyde solution (11% formaldehyde, 100 mM NaCl, 1 mM EDTA at pH 8.0, 0.5 mM EGTA, 50 mM Tris-HCl at pH 8.0). After addition of 125 mM glycine and incubation for 5 min at 20°C, cells were chilled on ice, washed with ice-cold 1× TBS, and resuspended in 400 μL of lysis buffer (50 mM Hepes at pH 7.5, 140 mM NaCl, 1 mM EDTA, 1% Trition X-100, 0.1% sodium deoxycholate, Complete proteinase inhibitor [Roche], 1 mM PMSF, 1 mM benzamidine, 1 mM Na3VO4, 1 mM NaF). Crude extracts were prepared by four pulses (60 sec) of bead-beating in FastPrep MP with Cryo-adaptor until 90% of cells were broken. Extracts were sonicated three times for 30 sec in 18 cycles using a Bioruptor until chromatin was sheared to an average size of less than ∼300 bp and subsequently cleared of insoluble cell debris by centrifugation at 15,000 rpm for 10 min. Ten microliters of the whole-cell extract was saved as an input control. Immunoprecipitation was performed for 2 h with antibody-conjugated beads (anti-Flag M2 affinity gel from Sigma; agarose-conjugated HA-probe F-7 or c-Myc 9E10 from Santa Cruz Biotechnology was used according to the tag on the protein). Precipitates were washed twice with 800 μL of lysis buffer, 800 μL of lysis buffer plus salt (lysis buffer with 500 mM NaCl), 800 μL of wash buffer, and 800 μL of 1× TE buffer, respectively. After the addition of 100 μL of 10% Chelex100 resin into the input controls and precipitates, respectively, those samples were boiled for 15 min at 100°C and then cooled at room temperature. Each sample was incubated with 2 μL of proteinase K (10 mg/mL) for 30 min at 55°C with gentle shaking. After denaturing with 0.4 M NaOH, ChIP and input samples were then transferred to a Hybond-XL membrane by using a slot or dot blot module. The membrane was hybridized with a probe specific for the telomeric sequence and then reprobed with rDNA after stripping. The hybridization signals were quantified using ImageQuant software. In addition, quantitative real-time PCR was used to analyze the same ChIP and input samples. Fold enrichment values were calculated based on ΔCt between ChIP and input samples after performing independent duplicate SYBR Green-based real-time PCR (Bio-Rad) using primer pairs of subtelomere and an fbp1 (fructose-1,6-bisphosphatase) gene fragment (as the background control); the values were expressed as immunoprecipitate/whole-cell extract (subtelomere) divided by immunoprecipitate/whole-cell extract (fbp1+) (Dehe et al. 2012).
Amplification of subtelomeric regions
Genomic DNAs were analyzed using primers selected to amplify specific subtelomeric regions by PCR (Moser et al. 2011). PCR products were loaded onto 2% agarose gels. Agarose gels stained with ethidium bromide were visualized using a Bio-Rad imaging system.
Acknowledgments
We thank Toru Nakamura, Fuyuki Ishikawa, Julie Cooper, Virginia Zakian, and Takashi Toda for providing plasmids and strains; Toru Nakamura for sharing detailed protocols of ChIP assay and pulsed-field gel electrophoresis; Wei Yao and Songtao Jia for sharing yeast protocols; and Craig Kaplan, Peter Kaiser, Songtao Jia, Jayakrishnan Nandakumar, Suzanne Sandmeyer, and Dorothy Shippen for comments on the manuscript and helpful discussions. The work in the laboratory of F.Q. is supported by a Basil O'Connor Starter Scholar Research Award from March of Dimes, a Beginning Grant-in-Aid from American Heart Association, and National Institutes of Health grant R01 GM098943.
Supplemental material is available for this article.
Article is online at http://www.genesdev.org/cgi/doi/10.1101/gad.219485.113.
A key question in stem cell biology is how distinct cell types arise from common multipotent progenitor cells. It is unknown how liver and pancreas cells diverge from a common endoderm progenitor population and adopt specific fates. Using RNA-seq, Spagnoli and colleagues define the gene expression programs of liver and pancreas progenitors and identify the noncanonical Wnt pathway as a potential developmental regulator of this fate decision. Furthermore, this study provides a framework for lineage-reprogramming strategies to convert adult hepatic cells into pancreatic cells.
Understanding how distinct cell types arise from multipotent progenitor cells is a major quest in stem cell biology. The liver and pancreas share many aspects of their early development and possibly originate from a common progenitor. However, how liver and pancreas cells diverge from a common endoderm progenitor population and adopt specific fates remains elusive. Using RNA sequencing (RNA-seq), we defined the molecular identity of liver and pancreas progenitors that were isolated from the mouse embryo at two time points, spanning the period when the lineage decision is made. The integration of temporal and spatial gene expression profiles unveiled mutually exclusive signaling signatures in hepatic and pancreatic progenitors. Importantly, we identified the noncanonical Wnt pathway as a potential developmental regulator of this fate decision and capable of inducing the pancreas program in endoderm and liver cells. Our study offers an unprecedented view of gene expression programs in liver and pancreas progenitors and forms the basis for formulating lineage-reprogramming strategies to convert adult hepatic cells into pancreatic cells.
liverpancreasmouseRNA-seqWnt signaling
The liver and pancreas share many common aspects of their early embryonic development as well as some adult functional properties and both are endowed with essential metabolic functions (Slack 2007; Zaret 2008). This close relationship makes the liver a very attractive tissue source for generating new pancreatic β cells by lineage reprogramming, which might be used in the context of cell-based therapy of diabetes. However, how liver and pancreas cells diverge from a common endoderm progenitor population and adopt specific fates remains elusive.
During embryonic development, the pancreas originates from distinct outgrowths of the dorsal and ventral regions of the foregut endoderm (Spagnoli 2007; Puri and Hebrok 2010). Subsequently, the two buds fuse to form a single organ containing both pancreatic exocrine and endocrine cells (Spagnoli 2007; Puri and Hebrok 2010). The liver originates solely from the ventral foregut endoderm, adjacent to where the ventral pancreas (vpa) emerges, and the two cell fates are specified concomitantly by mouse embryonic day 8.5 (E8.5) (Deutsch et al. 2001). Previous studies showed that the ventral foregut endoderm is multipotent for the hepatic and pancreatic programs (Deutsch et al. 2001; Tremblay and Zaret 2005; Miki et al. 2012). Hepatic and pancreatic endoderm express a common set of transcription factors, such as the FoxA and GATA factors, and are exposed to the same extrinsic signals: fibroblast growth factor (FGF) and BMP (Deutsch et al. 2001; Chung et al. 2008; Spagnoli and Brivanlou 2008; Zaret 2008). However, the developmental regulators of the fate choice between liver and pancreas are poorly understood. It is also unknown how the pancreatic program is stably repressed in nascent hepatic progenitors and how the hepatic program is repressed in pancreatic progenitors. This knowledge will aid in not only the programming of stem cells to pancreatic and hepatic lineages, but also the discovery of the mechanisms underlying cellular plasticity between liver and pancreas.
Studies conducted on various stem cell types or whole organisms (Tumbar et al. 2004; Wang et al. 2004; Dequéant et al. 2006; Mitiku and Baker 2007) have clearly demonstrated that significant insights into the regulation of the cell fate decision are obtained through global analysis of gene expression events. Thus, to search for developmental regulators that are involved in the control of the pancreas and liver cell fate decision, it is fundamental to have a global picture of the regulation of gene expression in the mammalian hepatic–pancreatic endoderm lineage. To this aim, we performed sequencing-based expression profiling (RNA sequencing [RNA-seq]) on hepatic and pancreatic progenitors in mice at distinct developmental stages. We devised a strategy to purify liver and pancreas progenitor populations directly from the endoderm of Tg(Prox1-EGFP [enhanced green fluorescent protein]) reporter mouse embryos before and after the onset of organogenesis. By integrating the temporal and spatial gene expression profiles, we found mutually exclusive signaling signatures in hepatic and pancreatic progenitors. Importantly, we identified the noncanonical Wnt pathway as a potential developmental regulator of the pancreas versus liver fate decision, since it is expressed in the foregut endoderm before the cell fate choice is made and then is maintained in pancreas progenitors but is absent in hepatic progenitors. Moreover, when assayed in Xenopus embryos and mammalian cells, activation of the noncanonical Wnt pathway is able to promote pancreatic fate, suggesting an ancient mechanism for controlling the pancreas versus liver fate choice. Overall, our study offers an unprecedented view of gene expression programs in liver and pancreas progenitors during the defined period of their lineage divergence and provides novel insights into key mechanisms that may underpin cellular plasticity and reprogramming between the two cell types.
ResultsRNA-seq on FACS-purified hepatic and pancreatic progenitors
To explore the transcription program associated with liver and pancreas progenitors in vivo at the time of their fate divergence, we performed RNA-seq on distinct mouse endoderm progenitor populations isolated at two developmental stages. For in vivo monitoring of hepatic and pancreatic progenitor cells, we used the transgenic mouse line Tg(Prox1-EGFP)Gsat/Mmcd that carries the reporter gene EGFP into the Prox1 homeobox gene locus (Gong et al. 2003). Prox1 is the earliest specific marker in common between hepatic and pancreatic endoderm from gastrulation onward (Burke and Oliver 2002; Wandzioch and Zaret 2009) and therefore is ideally suited for isolating both hepatic and pancreatic progenitors (Fig. 1). We showed that Prox1-EGFP transgenic mouse embryos reproduced the endogenous pattern of Prox1 expression in the endoderm from E7.5 onward (Fig. 1A,B; Burke and Oliver 2002; data not shown). In both ventral and dorsal foregut endoderm of Tg(Prox1-EGFP) embryos, the localization of EGFP perfectly mirrored endogenous Prox1 expression and overlapped with other endodermal genes, such as Foxa2, at E8.5 and with tissue-specific genes, such as Pdx1 in the pancreas or Liv2 in the liver, from E9.5 onward (Fig. 1A,B; Watanabe et al. 2002; Lee et al. 2005; Wandzioch and Zaret 2009; Puri and Hebrok 2010). Thus, the Tg(Prox1-EGFP) in vivo model enabled us to visualize hepatic and pancreatic progenitors under fluorescence microscopy before organogenesis had started. Distinct regions of the prospective hepatic and pancreatic endoderm were manually microdissected, and Prox1-EGFP+ cells were FACS-purified and subjected to RNA-seq analysis to define their transcriptomes (Fig. 1C–E). Previous fate mapping studies in mouse and chick embryos suggested differential locations of hepatic and pancreatic progenitors in the ventral foregut at the early somite stage (Tremblay and Zaret 2005; Miki et al. 2012). To determine whether regional identity within the foregut is associated with differential gene expression, we collected Prox1-EGFP+ cells of the whole ventral foregut (referred to as fg) and exclusively of the medial ventral foregut (referred to as mfg) from E8.5 embryos at the same somite stage (seven to nine somites) (Fig. 1C, top). At E10.5, budding sites of the liver and pancreas, including both the dorsal pancreas (dpa) and vpa, were readily distinguishable, and the three distinct progenitor populations were isolated and processed for the analysis (Fig. 1C, bottom).
In vivo isolation and RNA-seq profiling of endoderm progenitor cells from Tg(Prox1-EGFP) mouse embryos. (A) Representative maximum confocal Z-projections of immunofluorescence analysis in Tg(Prox1-EGFP) E8.5 mouse embryos shows EGFP reporter expression in both ventral (arrows) and dorsal (asterisk) foregut cells, mirroring endogenous Prox1 expression (see the inset for overlay) and overlapping with Foxa2 expression domains at the three-somite (3S) stage. Embryos are presented in ventral view. (B) In Tg(Prox1-EGFP) E9.5 mouse embryos, EGFP expression was detected in the hepatic and both dorsal and ventral pancreatic buds, mirroring the endogenous Prox1 (see the inset for overlay). EGFP colocalized with Pdx1 in both pancreatic buds and with Liv2 solely in the liver bud. Embryos are presented in lateral view. (C–E) Schematic representation of cell sampling and the RNA-seq procedure. (C) EGFP+ fg and mfg endoderm at the seven- to nine-somite (7–9S) stage/E8.5 and liver (lv), ventral (vpa), and dorsal (dpa) pancreas at E10.5 were microdissected from Tg(Prox1-EGFP) embryos. Cells were dissociated and subjected to FACS. (D) Representative diagram of the EGFP+ cell fraction isolated by FACS. The dashed box indicates EGFP+-gated cells, and cells negative for EGFP are in purple. Transcript expression was profiled by RNA-seq. (E) Example illustrating the RNA-seq read coverage profile of the pancreatic-specific gene Pdx1. The Y-axis indicates the number of read counts in each cell population (Y-axis scale is 0 1000 counts). FPKM (fragments per kilobase of exon per million fragments mapped) values for Pdx1 in each data set are included on the right. Exons are depicted as gray boxes at the bottom on the X-axis. As expected, a large number of reads was found in pancreatic progenitors. Bars: A, 100 μm; B, 50 μm.
Given the small number of progenitor cells, we applied a submicrogram RNA-seq method and used paired-end sequencing technology (Adamidi et al. 2011). A comparable number of high-quality raw reads was obtained from each sample and used to estimate the relative abundance of transcripts (Table 1; Trapnell et al. 2010). Principal component analysis (PCA) on the nonredundant gene expression data showed that individual samples perfectly clustered according to their embryonic stage and/or region of origin (Fig. 2A). Known hepatic and pancreatic tissue-specific transcripts were found in the E10.5 liver and pancreas RNA-seq profiles, respectively, reflecting the high reliability and accuracy of the data obtained (Fig. 1E; Supplemental Fig. 1).
Summary of sequencing data and annotation information
Temporal and spatial integration analysis of the RNA-seq-derived transcriptome profiles. (A) PCA shows that the RNA-seq-derived transcriptome profiles are characteristic of different progenitor cell types (for detailed description, see the Materials and Methods). (B) Venn diagrams showing the number of unique and common highly expressed transcripts between progenitor cells at different developmental stages. To further focus our analysis on subsets of genes with distinct expression patterns, we divided the working data set into two groups based on a cutoff for high expression (defined as FPKM = 10, at approximately the 50th percentile of RNA-seq expression for each sample). Of 14,053 genes, 8110 could be categorized in the defined Venn regions. For example, 5437 genes exhibited relative abundance values of >10 FPKM in all samples, while 517 genes were highly expressed in the foregut, vpa, and dpa but not in the liver (referred to as group FP). In contrast, 89 genes were highly expressed in the foregut and liver but not in the vpa and dpa. Three-hundred-sixty-three transcripts were present (>10 FPKM) only in the fg but not in the liver, dpa, and vpa (not shown in the diagram). As shown in the PCA plot in A, fg and mfg were highly similar; therefore, these samples were combined together as “foregut” to simplify the visualization. (C,D) Levels of Wnt signaling pathway gene expression across the fg, liver (lv), vpa, and dpa progenitors. FPKM values (Y-axis) were plotted against the different progenitors cell types (X-axis). (*) Wnt factors present in the 150 FP group transcripts that showed significant differential regulation between the pancreas and liver.
Integrating temporal and spatial transcription profiles of the progenitor populations
To identify developmental regulators of pancreatic and/or hepatic fate, we integrated the temporal and spatial analyses. Both hepatic and pancreatic progenitors arise from foregut endoderm cells (Spagnoli 2007; Zaret 2008). Therefore, the set of genes expressed in the foregut (fg or mfg) is expected to include regulators of both early pancreatic and hepatic fate specification as well as the choice between the two fates (liver and pancreas). We compared those genes with the genes expressed later in the E10.5 liver, vpa, and dpa and found a number of genes specific to either liver or pancreas that were already expressed at the earlier time point in the E8.5 foregut (89 and 517 genes, respectively) (Fig. 2B, left). Of interest, these numbers suggest a higher divergence between the foregut and liver when compared with the number of genes whose expression is instead maintained in the pancreas (vpa and dpa) (Fig. 2B, left; Supplemental Fig. 2). This is in line with the hypothesis of the pancreas being the default fate of the ventral foregut endoderm (Deutsch et al. 2001). In contrast, genes initially showing low expression in the foregut but high expression later in E10.5 hepatic or pancreatic progenitors represent cell type-specific markers of differentiation and possibly factors involved in late aspects of organogenesis (Fig. 2B, right; Supplemental Fig. 3). At this later time point, the number of genes that are unique to either the dpa or vpa (211 vs. 257) (Fig. 2B, right) indicates a large divergence between the two buds, reflecting the dual embryological origin of the pancreas.
Different Wnt signaling signatures correlate with pancreatic or hepatic progenitor states
To study the mechanisms underlying the pancreas versus liver fate decision, we focused on the transcripts whose expression was detected in the E8.5 foregut and maintained exclusively in the pancreatic rudiments and therefore possibly was endowed with a role in this cell fate restriction. We hereafter refer to this cluster as FP (foregut pancreas) (Fig. 2B, left). A gene ontology (GO) term analysis of our results revealed significant enrichment for Wnt-related “Molecular Function” categories in the group FP, such as Wnt protein-binding (GO:0017147; seven genes; P-value < 0.00001.9) and Wnt-activated receptor (GO: 0042813; five genes; P-value < 0.00035) activities. Among the Wnt-associated factors were genes such as Frizzled 2 (Fzd2), Fzd4, Ror2, Sfrp5, Sfrp2, and Wls (Supplemental Table 1). Also, Wnt-related “Biological Process” categories were primarily enriched in the group FP, such as establishment of epithelial cell polarity (GO: 0090162), Wnt receptor signaling pathway (GO: 0007223), and establishment of planar polarity (GO: 0001736) (Supplemental Table 2). Of interest, no other regions of the Venn diagram displayed such a significant enrichment for Wnt-related categories as the group FP, suggesting a unique signaling signature of the pancreas versus liver fate in the mammalian endoderm.
We decided to further characterize the expression dynamics of Wnt signal transducers. Interestingly, in our data set, we found that intracellular transducers of the canonical Wnt/β-catenin signaling (Grigoryan et al. 2008; Clevers and Nusse 2012)—such as β-catenin (Ctnnb1), APC, Axin, and TCFs—as well as the common canonical and noncanonical component Dishevelled 1 (Dvl1) (Angers and Moon 2009; Kikuchi et al. 2009) were expressed at stable levels in the different progenitor populations (Fig. 2C), whereas transcripts encoding ligands, receptors, and coreceptors exhibited lineage-specific induction or repression (Fig. 2D). In particular, only noncanonical Wnt ligands (Wnt5a, Wnt5b, and Wnt7b) (Kikuchi et al. 2009; Grumolato et al. 2010) were found in the foregut and pancreas and all of them were strongly down-regulated in liver progenitors (Supplemental Fig. 4). Additional determinants of noncanonical/planar cell polarity (PCP) Wnt specificity displayed similar expression profiles, including the receptors Fzd2 and Fzd7 and the PCP core membrane proteins Stbm/Vangl2, Celsr2, and Fat1 (Fig. 2D). Finally, the coreceptors Lrp5 and Ror2 showed divergent expressions—one strongly induced in the liver, and the other one induced exclusively in pancreas progenitors (Fig. 2D). These results suggest a cell type-dependent noncanonical Wnt activation in the foregut and pancreatic progenitors that is ensured by not only distinct classes of ligands, but also recruitment of unrelated coreceptors, as previously described in other contexts (Grumolato et al. 2010). For instance, the differential expression of Ror2 coreceptor might sustain noncanonical Wnt activation in foregut and pancreas progenitors but not in liver progenitors.
Noncanonical Wnt signaling signature underlies the pancreas versus liver fate decision
If multiple components of a pathway are expressed and the expression is developmentally regulated, it is very likely that the pathway is active. Therefore, we undertook several approaches to study the possible role of the Wnt noncanonical pathway in the process of pancreas versus liver cell fate specification. For further analysis, we prioritized FP group transcripts that showed significant differential regulation between the pancreatic and hepatic endoderm (150 transcripts out of 517; Cufflinks, P-value < 0.05) (Figs. 2D, 3A). We performed quantitative RT-qPCR and immunofluorescence staining to confirm that the transcripts identified as differentially expressed by RNA-seq indeed showed significant enrichment in the pancreatic cell lineage (Fig. 3B,C; Supplemental Fig. 5). In Drosophila as well as some mammalian tissues, the PCP proteins are initially enriched in the apical cell membrane prior to their asymmetric distribution at either the proximal or distal side of the cell (Vichas and Zallen 2011; Wallingford 2012). Their distribution within the plane of the foregut epithelium has not been previously reported. We found Fzd2 receptor, Ror2 coreceptor, and PCP proteins such as Celsr2 and Fat1 to be enriched at the cell surface of both foregut epithelial and pancreatic progenitor cells with no obvious asymmetric distribution but completely absent in hepatic progenitors (Fig. 3C; Supplemental Fig. 5A).
Analysis of candidate regulators of the pancreatic versus hepatic fate decision. (A) Heat map view of the FP group transcripts that were differentially expressed between any of two samples (150 transcripts out of 517; Cufflinks, P-value < 0.05). Colors represent high (red) or low (blue) expression values based on Z-score normalized FPKM values for each gene. White represents the average between red (high) and blue (low) expression values. Dashed boxes highlight gene sets validated by either RT-qPCR or immunofluorescence analyses. (B) RT-qPCR validation of a subset of differentially expressed genes of the FP group. Data were normalized to that of succinate dehydrogenase (SDHA) and are represented as fold change compared with the E8.5 foregut sample (set to 1 as calibrator). Error bars represent ±SEM. (C) Immunofluorescence analysis validated the exclusive localization of Celrs2, Claudin 4 (Cldn4), CK19, Fat1, Fzd2, and Ror2 in E8.5 foregut endoderm (see arrows in the insets) and pancreatic progenitors and their absence in the liver (see arrows). Micrographs show cross-sections of E8.5 and E10.5 mouse embryos. Bars, 50 μm. (duo) Duodenum; (hep) hepatic progenitors; (lv) liver; (nf) neural folds.
In several systems, there is direct evidence that noncanonical Wnt signaling regulates cell adhesion and cytoskeleton components (Karner et al. 2009). Accordingly, in our data set, several adhesion molecules—including Claudin 4, laminin, integrin a3 (Itga3), fibronectin leucine-rich transmembrane protein 2 (Flrt2), and Flrt3—as well as cytoskeleton components such as cytokeratins CK19 and CK7 and Shroom3 showed similar differential expression, and were down-regulated in hepatocyte progenitors after lineage divergence (Fig. 3B,C; Supplemental Fig. 5B). Taken together, these results indicate that active remodeling of the foregut epithelium accompanies cell fate determination between liver and pancreas, resulting in differences in the relative adhesiveness of the cells. Moreover, our findings suggest that the noncanonical Wnt signaling might influence this cell fate decision within the foregut endoderm.
One prediction from this hypothesis is that upon exposure to noncanonical Wnt, endodermal cells acquire a pancreatic-specific differentiation program and fail to induce hepatic genes expression. To test this prediction, we first used one of the most intensively studied noncanonical Wnts, Wnt5a, which we also found to be cell-autonomously produced by the E8.5 foregut endoderm (Supplemental Table 1; Supplemental Fig. 4A–C). Subsequently, at E10.5, Wnt5a expression was maintained at lower levels in pancreatic progenitor cells and was absent in hepatoblasts (Supplemental Fig. 4B,C). In addition to the endodermal expression, we found Wnt5a to be abundant in the surrounding mesenchyme at both E8.5 and E10.5, suggesting potential autocrine and/or paracrine ligand activities (Supplemental Fig. 4C). The other noncanonical Wnts, Wnt5b and Wnt7b, were found to be abundant in pancreatic progenitor cells and not in hepatoblasts starting from E10.5, while they were expressed at low levels in foregut cells (Supplemental Fig. 4A,B).
Perturbations of Wnt5a signaling in Xenopus, zebrafish, and mice strongly suggest that Wnt5a activates a conserved pathway that controls cell fate, movements, and polarity during development (Moon et al. 1993; Ho et al. 2012). Moreover, the expression pattern appears conserved across species because Wnt5a is expressed in endodermal as well as surrounding mesodermal cells in the Xenopus embryo at the time of foregut organogenesis (Supplemental Fig. 4D; Zhang et al. 2013). Therefore, we first used the Xenopus laevis embryo as a model system and exposed anterior endodermal cells, which are fated to become either liver or pancreas, to soluble Wnt5a protein. In addition to the expected morphogenetic defects, Wnt5a induced pancreatic progenitor gene expression, including Pdx1 and Ptf1a, and, to a lesser extent, pancreatic differentiation markers, such as Glucagon (Gcg) and Trypsin (Fig. 4A,B). Concomitantly, Wnt5a treatment repressed expression of the hepatic genes Hex, For1, and Albumin in the same endodermal cells without affecting other endodermal cell fates, as judged by unchanged Nkx6.2 (duodenum) and Sox2 (stomach) expression levels (Fig. 4B; Chalmers et al. 2000; Dichmann and Harland 2011). Similar effects were observed when Wnt5a mRNA was microinjected into either anterior or posterior endodermal cells of the Xenopus embryo, resulting in expanded Ptf1a expression beyond its normal boundaries (Fig. 4C). Furthermore, we found that activation of the noncanonical Wnt pathway using a different noncanonical ligand, the Wnt5b recombinant protein, also induced a strong expression of pancreatic progenitor transcription factors in anterior endodermal cells (Fig. 4D). Finally, we examined whether the exposure to the noncanonical Wnt pathway has an effect on cell proliferation in the anterior endoderm. Immunostaining of phospho-histone H3 (PHH3) to mark cells undergoing mitosis revealed no significant differences in proliferating cell number in the foregut between Wnt5a-injected and uninjected early tadpole stage embryos (Supplemental Fig. 4E).
Noncanonical Wnt5a activity promotes pancreatic versus hepatic fate in the anterior endoderm. (A) Xenopus embryos injected with Wnt5a mRNA showed a shortened and mildly bent body at the tailbud stage, as previously described (Kim et al. 2005). (B) Endodermal explants were cultured in the presence of Wnt5a recombinant protein from stage 10, collected at the tadpole stage, and assayed for expression of the indicated pancreatic, hepatic, and duodenum/stomach genes by RT-qPCR analysis. Untreated anterior endodermal explants were used as control. Data were normalized to that of ornithine decarboxylase (ODC) and are represented as fold changes compared with untreated endoderm sample (set to 1 as calibrator). Error bars represent ±SEM. (C) Whole-mount double in situ hybridization analysis of Hex (light blue) and Ptf1a (purple) in control and Wnt5a-injected Xenopus embryos at the tadpole stage. The arrow indicates Hex expression in the liver bud, and arrowheads indicate Ptf1a expression in the two pancreatic buds (dorsal and ventral buds). Dashed lines mark expanded Ptf1a expression in the injected embryos. Total number of injected embryos = 61; 41% showed visible expansion of Ptf1a. (AE) Anterior endoderm; (PE) posterior endoderm. (D) RT-qPCR analysis of endodermal explants treated with Wnt5b (200 ng/mL) recombinant protein. Data were normalized to that of ODC and are represented as fold changes compared with untreated endoderm sample (set to 1 as calibrator). (E) RT-qPCR analysis of endodermal explants treated with 500 ng/mL Wnt3a recombinant protein. Data were normalized to that of ODC and are represented as fold changes compared with untreated endoderm sample (set to 1 as calibrator). (F) RT-qPCR analysis of direct downstream target genes of the Wnt/β-catenin pathway in endodermal explants treated with 500 ng/mL Wnt5a or 500 ng/mL Wnt3a recombinant protein. Data were normalized to that of ODC and are represented as fold changes compared with untreated endoderm sample (set to 1 as calibrator). (G) TOPFLASH and ATF2-luc reporter assays in Xenopus embryos. Four-cell stage embryos were injected into the vegetal blastomeres with 50 pg of TOPFLASH or 100 pg of ATF2-luc plus 25 pg of Renilla luciferase reporter plasmids. Endodermal explants were dissected at stage 9 and either left untreated as control (CTRL) or exposed to 500 ng/mL Wnt5a or 500 ng/mL Wnt3a recombinant protein, as indicated. Luciferase reporter assays were carried out in explants lysed at gastrula and early tailbud stages. (H) Western blot analysis of dissected anterior endodermal explants either left untreated as control (CTRL) or exposed to Wnt5a or Wnt3a recombinant protein. The relative ABC/tubulin levels in the treated explants compared with the control, which was set to 1.0, are indicated. (β-cat) Total β-catenin; (tub) α-tubulin. (*) P < 0.05; (**) P < 0.01, as determined by the REST program statistical analysis (Pfaffl et al. 2002).
On the other hand, the establishment of proper levels of canonical Wnt/β-catenin signaling and its temporal sequential activation in the anterior endoderm (Fig. 4) are known to be essential for foregut identity and organ formation (Ober et al. 2006; Li et al. 2008; Angers and Moon 2009; Puri and Hebrok 2010; Zhang et al. 2013). After adding Wnt3a recombinant protein, a well-known canonical Wnt activator (Grigoryan et al. 2008; Clevers and Nusse 2012), to anterior endodermal cells, we observed that the levels of expression of hepatic and pancreatic genes were unchanged when compared with control endoderm, ruling out an apparent choice of cell type between liver or pancreas (Fig. 4E).
The noncanonical Wnt signaling is able to antagonize the canonical signaling in certain biological contexts (Grumolato et al. 2010; Ho et al. 2012). To distinguish between these two possible Wnt5a activities (e.g., noncanonical Wnt activity or antagonist effect on Wnt/β-catenin) in the context of liver and pancreas lineage divergence, we performed a series of quantitative assays and biochemical analysis in the Xenopus endoderm. First, we analyzed the transcriptional responses to both pathways in endoderm cells treated with Wnt5a by (1) examining direct downstream target genes of Wnt/β-catenin pathway and (2) using specific luciferase reporter assays based on TCF/LEF (TOPFLASH) and ATF2 response elements for canonical and noncanonical Wnt, respectively (Ohkawara and Niehrs 2011; Clevers and Nusse 2012). RT-qPCR analysis showed that Wnt3a treatment increased the expression levels of direct Wnt target genes such as Cyclin-D1 (Ccdn1), Lef-1, Myc, and Axin 2, whereas Wnt5a treatment does not affect their expression in anterior endodermal cells (Fig. 4F). This is in line with the RNA-seq profiles of mammalian pancreatic and hepatic progenitor cells displaying comparable expression levels of these downstream target genes (Fig. 2; Supplemental Table 2). Furthermore, we found that exposure to Wnt5a does not suppress endogenous canonical Wnt activity in the endoderm, as monitored by the Wnt/β-catenin TOPFLASH reporter assay, whereas it induces the noncanonical Wnt ATF2 reporter activity (Fig. 4G). Notably, endogenous canonical Wnt/β-catenin transcriptional activity accumulates in the Xenopus endoderm from gastrulation onward (Fig. 4G; Angers and Moon 2009; Ohkawara and Niehrs 2011). This results in elevated basal levels of TOPFLASH luciferase activity in the untreated endodermal cells (control [CTRL]) and is consistent with the mild induction of the TOPFLASH reporter observed upon exposure to Wnt3a-soluble protein (Fig. 4G).
Next, we conducted biochemical assays on endoderm explants treated with Wnt5a and Wnt3a to analyze the status of activation of β-catenin signaling. Western blot analysis showed that levels of total and active β-catenin (ABC) protein (dephosphorylated on Ser37 and Thr41) (van Noort et al. 2002) remained unchanged in Wnt5a-treated anterior endodermal cells when compared with control endoderm (Fig. 4H), whereas Wnt3a moderately increased dephosphorylated β-catenin levels. As expected, Wnt5a induced JNK phosphorylation, reflecting noncanonical Wnt activation (Fig. 4H). Taken together, these results rule out Wnt5a function as an antagonist of the canonical Wnt pathway and support a role of the noncanonical Wnt signaling in the pancreatic versus hepatic fate decision in a β-catenin-independent manner.
Last, we sought to expand the functional analysis on the noncanonical Wnt pathway directly to mammalian endodermal cells. First, we used the mouse embryonic stem cell (mESC) system to model ex vivo endoderm development (Rossant 2011). In particular, to induce pancreatic specification, mESCs were stimulated using a step-wise protocol in a monolayer culture adapted from previously published studies (D'Amour et al. 2006; Nostro et al. 2011; Wang et al. 2011; Chen et al. 2013). RT-qPCR analysis revealed first up-regulation of definitive endoderm transcription factors such as Sox17 and Foxa2 followed by sequential induction of pancreatic progenitor genes, including Pdx1, Sox9, Pax6, and Islet1 (Isl1) (Fig. 5A). Consistent with pancreatic differentiation, the levels of expression of Sox9, Pax6, and Isl1 increased with time and upon exposure to additional cytokines (e.g., noggin, retinoic acid, and cyclopamine), while no or minimal induction of the hepatic marker Albumin was detected (Fig. 5A). Most differentiation protocols use FGF10 to pattern the definitive endoderm-induced population toward the pancreatic endoderm fate within the foregut and the subsequent pancreatic progenitor fate (D'Amour et al. 2006; Nostro et al. 2011). To determine whether the noncanonical Wnt pathway would enhance pancreatic specification, we examined the consequences of treating definitive endoderm cultures generated from ESCs with Wnt5a in the presence or absence of FGF10. Interestingly, Wnt5a enhanced the levels of Pdx1 expression compared with the standard pancreatic endoderm differentiation conditions (Fig. 5B). In addition, the positive effect of Wnt5a was even more evident in definitive endoderm cells cultured in the absence of FGF10, as its addition restored the induction of Pdx1 expression to pancreatic endoderm standard levels (Fig. 5C). Importantly, the duodenal marker Cdx2 was not induced in pancreatic endoderm cultures differentiated in the presence of Wnt5a, and the slight induction of Albumin expression in pancreatic endoderm cultures was down-regulated by Wnt5a treatment, supporting a specific effect on pancreatic fate (Fig. 5B). In addition, we found that other pancreatic progenitor transcription factors, including Nkx6.1 (at the pancreatic endoderm stage), Sox9, and Pax6 (at the pancreatic progenitor stage), were strongly induced in both the absence and presence of Wnt5a, although a further enhancement of their transcript levels (as seen for Pdx1) was not detectable in these culture conditions (Fig. 5B). This finding suggests that patterning of anterior endoderm with noncanonical Wnt signaling might be important for optimal pancreatic specification in protocols for directed differentiation of ESCs.
Conserved Wnt5a activity in promoting pancreatic fate. (A) Directed differentiation of mESC monolayer cultures into pancreatic progenitors. RT-qPCR analysis evaluating definitive endoderm (DE), pancreatic endoderm (PE), and pancreatic progenitor (PP) gene expression at different stages of differentiation. Untreated mESCs were used as control (d0). Data were normalized to that of SDHA and are represented as fold changes compared with control (d0) mESCs (set to 1 as calibrator). (B) Day 5 and day 8 mESC cultures were analyzed by RT-qPCR for the expression of the indicated genes following either standard pancreatic endoderm and pancreatic progenitor culture conditions or in the presence of Wnt5a recombinant protein (PE + Wnt5a and PP + Wnt5a). RT-qPCR data were normalized to that of SDHA and are represented as fold changes compared with control (d0) mESCs (set to 1 as calibrator). Error bars represent ±SEM. (C) Day 5 mESC cultures were analyzed by RT-qPCR for the expression of the pancreatic gene Pdx1 following standard pancreatic endoderm culture conditions in the absence of FGF10 (PE − FGF10), the presence of Wnt5a (PE + Wnt5a), or the presence of Wnt5a but without FGF10 (PE − FGF10 + Wnt5a). RT-qPCR data were normalized to that of SDHA and are represented as fold changes compared with standard pancreatic endoderm condition (set to 1 as calibrator). Error bars represent ±SEM. (D) BAML liver cells cultured in the presence of 200 ng/mL Wnt5a for 2 wk were assayed for expression of the indicated pancreatic and hepatic genes by RT-qPCR analysis. Untreated BMAL liver cells were used as control. Data were normalized to that of SDHA and are represented as fold changes compared with untreated liver cells (set to 1 as calibrator). Error bars represent ±SEM. (*) P < 0.05; (**) P < 0.01, as determined by the REST program statistical analysis (Pfaffl et al. 2002). (Alb) Albumin.
Next, we asked whether modulation of the noncanonical Wnt pathway might promote the pancreatic program in differentiated liver cells. To this aim, we used a mammalian ex vivo model system, the nontransformed BAML hepatic cell line (Fougère-Deschatrette et al. 2006). The BAML cells were established from adult mouse livers and were shown previously to retain hepatic differentiation hallmarks and repopulate the liver in vivo (Fougère-Deschatrette et al. 2006). We cultured the liver cells in their standard hepatocyte culture medium in the presence or absence of Wnt5a. After prolonged exposure to Wnt5a, we observed significant induction of the pancreatic genes, including Pdx1, Pax6, and MafA, in liver cells, whereas the level of expression of the liver markers Albumin and a1-antitrypsin (A1AT) and the liver-specific transcription factor HNF4α1 was not affected (Fig. 5C). This suggests that Wnt5a might also facilitate fate conversion of liver cells into pancreatic fate but does not itself possess the ability to suppress hepatic identity.
All together, these results implicate a novel and conserved role for the noncanonical Wnt signaling pathway in promoting the pancreas versus liver fate decision in the endoderm. In particular, Wnt5a appears to exert specific regulation on Pdx1 gene expression, which is conserved in both the mouse and Xenopus endoderm (Figs. 4, 5). An entire set of noncanonical Wnt transducers was identified in our analysis (Figs. 2, 3); whether Wnt5a and Wnt5b activities in this cell lineage decision are due to signaling through components of the PCP, Ror2/JNK, or Calcium/NFAT pathways (Kim et al. 2005; Ober et al. 2006; Li et al. 2008; Puri and Hebrok 2010) deserves further investigation.
Spatially distinct expression patterns mark the progenitor populations
Our data provide an opportunity to systematically examine additional in vivo expression patterns within the hepatic and pancreatic endoderm. For example, we focused on identifying transcriptional differences between discrete domains of the foregut endoderm at E8.5 (fg vs. mfg), before organ rudiments were formed (Fig. 6). Only 21 transcripts displayed statistically significant differential expression between these two regions (Fig. 6A). The small number of differentially expressed transcripts might reflect the active movements and intermingling of cells across the foregut at this stage (Tremblay and Zaret 2005). Among them were only a few genes that had been previously reported in the endoderm, including Pyy, Otx2, and Sox2 (Hou et al. 2007), but the majority was not known to be expressed in this territory (Fig. 6A–D). The expression of some of these foregut genes was maintained in either liver progenitors (E2f2), ventral pancreatic progenitors (Celsr2 and Nr1h5), or both (Cdkn1c), representing interesting candidates for lineage tracing studies to establish the exact contribution of progenitor domains within the endoderm.
Identification of distinct spatial patterns within the mouse foregut endoderm. (A) Heat map view of the transcripts that showed significant differential expression between E8.5 fg and mfg (21 transcripts). Colors represent high (red), low (blue), or average (white) expression values based on Z-score-normalized FPKM values for each gene. (B) RT-qPCR validation of a subset of the foregut differentially expressed genes. The XLOC_019271 is not supported by any spliced ESTs or Genscan predictions. By exon junction analysis, we predicted a gene model and validated the expression of this novel transcript in the fg and dpa by RT-qPCR (see also Fig. 3B; Supplemental Fig. 8). Data were normalized to that of SDHA and are represented as fold change compared with the E8.5 fg sample (set to 1 as calibrator). Error bars represent ±SEM. (C–H) Immunofluorescence and in situ hybridization analyses validated the expression of the indicated genes in the E8.5 foregut endoderm (see arrowheads) and/or dorsal pancreatic rudiments (demarcated by dashed line). At E8.5, Otx2 expression was detected in the mfg (C), which coexpressed Sox17 and E-cadherin (Ecad), as shown by immunofluorescence staining on serial section (C′). FoxD3 expression was detected in the E8.5 fg endoderm (D; see arrowheads) and dpa cells at E10.5 (E). Arrows in E indicate FoxD3/Pdx1-double-positive cells. At E10.5, Hox gene expression was detected in the dorsal pancreatic rudiment (F–H), which coexpressed Pdx1 and E-cadherin (Ecad), as shown by immunofluorescence staining on serial section (F′). Bars, 50 μm. (I) RT-qPCR validation of the indicated Hox genes showed differential expression between the vpa and dpa. Data were normalized to that of SDHA and are shown as expression ratio (2-log values) of dpa sample versus vpa sample. Error bars represent ±SEM. (J) The heat map shows the expression of 100 mouse genes found to be expressed in both the vpa and dpa (>10 FPKM) and whose human orthologs are also expressed in human pancreatic “progenitor-like” cells (Micallef et al. 2012).
Notably, in the ventral foregut epithelium, we detected Hox genes encoded by the HoxD clusters (Hoxd3 and Hoxd4) as well as a spliced noncoding RNA that is transcribed antisense to the Hoxb5 and Hoxb6 genes (0610040B09Rik), which might be relevant for their regulation (Fig. 6A; Supplemental Fig. 6; Dinger et al. 2008). A clear spatial difference in the expression of Hox genes was evident also at later stages in the pairwise comparison between E10.5 vpa and dpa data sets (Supplemental Fig. 7). Hox genes such as Hoxa3, Hoxa4, Hoxa7, Hoxb3, Hoxb6, Hoxb7, and Hoxb8 were abundantly expressed in the dorsal pancreatic region but were absent or at undetectable levels in the vpa and liver (Fig. 6A; Supplemental Fig. 7). Using RT-qPCR and in situ hybridization, we confirmed the differential expression between the dpa and vpa of selected Hox genes, including Hoxa7, Hoxb6, and Hoxb8 (Fig. 6F–I). Notably, Hoxa7 and Hoxb6 seem to be expressed in a heterogenous manner in the dpa, possibly marking specific cell types (Fig. 6F,G). Taken together, our findings suggest that a specific Hox code exists in the developing foregut as well as within the pancreatic territory, and its biological significance needs to be assessed (Grapin-Botton and Melton 2000; Iimura et al. 2009).
Besides the Hox genes, we detected large differences in gene expression between the ventral and dorsal pancreatic buds at the onset of pancreas organogenesis (192 transcripts; absolute value of log2 fold change > 3, P-value < 0.05) (Supplemental Fig. 7). To determine whether there was any recognizable biological relevance to the expression patterns, we analyzed GO terms with respect to molecular functions and found significantly enriched annotations in each cluster. Strong association with extracellular binding activities (GO: 0005488; 62 genes; P-value < 0.000129) was found in the vpa cluster, whereas the dpa cluster showed strong enrichment for DNA-binding transcription factor function (GO: 0003700; 25 genes; P-value < 9.41×10−18).
Our comprehensive analysis further illustrates how different the dorsal and ventral pancreatic progenitors are from each other. Despite these differences at E10.5, both pancreatic rudiments give rise to endocrine and exocrine cells at later stages and eventually fuse to form the definitive pancreas (Puri and Hebrok 2010). Our data set will enable a clear understanding of how distinct transcriptional programs might lead to the same cell type, allowing, for instance, the identification of alternative modes that can be used to program stem cells toward pancreatic endocrine fates.
Concordant gene signatures between mouse and human pancreatic progenitors
Next, we sought to evaluate the evolutionary importance of our transcriptome analysis of pancreatic progenitor cells and its relevance to human pancreas development. Given the fact that genomic analysis on equivalent developmental stage human material is not available, we compared our murine RNA-seq data sets with available microarray data obtained from human ESC (hESC)-derived pancreatic-like cells (Micallef et al. 2012).
Human pancreatic progenitor state was defined by the top 400 up-regulated genes in hESC-derived pancreatic cells versus undifferentiated hESCs (Micallef et al. 2012). Importantly, we restricted our analysis only to those genes that were up-regulated in hESCs differentiated into a specific pancreatic differentiation stage that precedes the onset of overt β-cell differentiation, representing the closest gene signature of human pancreatic progenitors. This corresponds to the fraction of “FACS-sorted insulin-GFP-negative cells” obtained after application of published differentiation protocols to undifferentiated INSGFP/w hESCs (Micallef et al. 2012). Mouse genes from the RNA-seq experiments were matched to their corresponding human homologs using the NCBI database HomoloGene (http://www.ncbi.nlm.nih.gov/homologene; see also the Materials and Methods). Importantly, we found that that 25% of the 400 human homolog genes are also markers of pancreas progenitors in mice, displaying high gene expression in the E10.5 vpa and dpa data sets despite the methodological and temporal differences (Fig. 6J).
Discussion
In conclusion, we devised a strategy to purify liver and pancreas progenitors directly from the endoderm of Tg(Prox1-EGFP) reporter mouse embryos. Using RNA-seq, we profiled the purified progenitor cells and identified transcripts that are differentially expressed across the pancreatic and hepatic endoderm at distinct developmental stages. Importantly, we uncovered a unique noncanonical Wnt signaling signature in the emergence of pancreas versus liver from endoderm progenitors that appears conserved across phylogenetically distant species. Our results provide an invaluable resource for lineage tracing analysis to pinpoint the exact origin of the hepato-pancreatic lineage and isolation of transient progenitor cell populations. Moreover, our transcriptional data lay the foundation for further targeted functional studies of developmental regulators of the liver and pancreas fate decision that will be relevant to humans. Indeed, when we compared our RNA-seq results with those previously obtained using microarray in hESCs differentiated into pancreatic-like progenitors (Micallef et al. 2012), we identified concordant gene signatures (e.g., similar markers of pancreatic progenitors: Wls, FLRTs, and Meis). These results suggest direct implications of our findings in developing novel strategies to generate pancreas progenitors and β cells for clinical transplantation from cellular programming of stem cells or induced pluripotent stem cells (iPSCs) and from reprogramming of adult hepatic cells. Finally, our RNA-seq analysis might be used as an in vivo reference for direct comparison with diseased human tissues such as pancreas dysplasia or agenesis or congenital liver defects.
Materials and methodsMouse embryo dissection and FACS sorting
The transgenic mouse line Tg(Prox1-EGFP)Gsat/Mmcd was obtained from the Heintz Laboratory-Gensat Project and generated as previously described (Gong et al. 2003). E8.5 and E10.5 Prox1-EGFP-positive embryos were selected and dissected using an epifluorescence stereomicroscope (Discovery V12, Zeiss). Eyebrow knives or tweezers were used for cutting the GFP-positive embryonic regions. The dissected embryonic regions were dissociated by trypsin (trypsin/EDTA 0.25% solution) digestion into single-cell suspension. To stop the reaction, DMEM (Invitrogen) was added to the cell suspensions and centrifuged at 300g at 4°C, and pellets were suspended and PBS/DEPC-treated. Before sorting, propidium iodide was added to select only live cells, and the cell suspension was filtered through a BD Falcon tube with cell strainer cap (BD 352235). After dead cell exclusion (SSC-A/PI-A), GFP-expressing cells were sorted at 4°C using a FACS Aria III flow cytometer (BD Biosciences) using a GFP filter and by setting the gate on the GFP fluorescence intensity. Conditions of sorting were as follows: 70-μm nozzle and sheath pressure of 70 psi. Sorted cells were collected directly in TRIzol reagent (Invitrogen) for RNA extraction. Cells isolated by FACS were used for RNA isolation and RNA-seq and subsequently for RT-qPCR validation. Approximately, a total of 10,000 cells per sample were pooled to isolate 100 ng of high-quality total RNA. All animal experimentation was conducted in accordance with the local ethics committee for animal care.
RNA-seq
RNA-seq was performed using ∼70 ng of total RNA quantified by Agilent RNA 6000 Pico kit (Agilent Technologies). The quality of RNA samples prior to library preparation was determined using an Agilent Bioanalzyer, and only samples with high RIN (RNA integrity number) scores (>8.5) were further processed. Briefly, poly(A) RNA was isolated by two rounds of oligo(dT)25 Dynabeads (Invitrogen) purification. Purified poly(A) RNA was fragmented for 3.5 min at 94°C using 5× fragmentation buffer (200 mM Tris-acetate at pH 8.1, 500 mM KOAc, 150 mM MgOA) as described previously (Adamidi et al. 2011). The fragmented RNA was purified by Agencourt RNAClean XP SPRI beads (Agencourt) and converted to first strand cDNA using random hexamer primers (Invitrogen) and SuperScript II reverse transcriptase (Invitrogen) followed by second strand cDNA synthesis with Escherichia coli DNA polymerase I (Invitrogen) and RNase H (Invitrogen) according to the manufacturer's intstructions. The 76-nucleotide (nt) paired-end sequencing library was prepared using New England Biolabs Next DNA Library preparation kit following the Illumina mRNA-Seq library preparation protocol. The minimum necessary number of cycles of amplification was used to minimize amplification biases (15 cycles with no overamplification observed). The quality of cDNA libraries for sequencing was assessed using Agilent Technologies 2100 Bioanalyzer. The prepared sequencing library was subsequently sequenced on Illumina HiSeq for 2 × 100 cycles following the standard protocol (Adamidi et al. 2011).
Bioinformatics analysis
Mapped reads from TopHat (version 1.33) were used as input for Cufflinks (version 1.3.0) (Trapnell et al. 2010) for transcript assembly and differential expression using the University of California at Santa Cruz mm9 reference annotation. To obtain a working data set for the purposes of comparing transcriptomes between different samples, we filtered the Cufflinks gene set for (1) successful deconvolution of fragments per kilobase of exon per million mapped fragments (FPKM), (2) FPKM > 0 in at least two samples, and (3) lower confidence level of FPKM > 0 in at least two samples. In the case that multiple Cufflinks genes for one gene symbol existed, Cufflinks genes were further selected for greatest transcript length and highest variability in the FPKM across all samples. This resulted in a nonredundant gene set of 14,053 Cufflinks genes.
Unless otherwise noted, bioinformatic analysis was carried out using R statistical environment (version ≥2.15; http://www.r-project.org). For the PCA of the nonredundant working data set, a ceiling FPKM value was set as the 90th percentile of all expression values in the data set. The first two principal components exhibited 48% and 35% of the total variability for PC1 and PC2, respectively. The Venn diagram was generated using the R packages limma (Ritchie et al. 2007) and VennDiagram (Adamidi et al. 2011). Enriched GO categories were calculated using the Bioconductor package TopGO (Alexa et al. 2006). In each case, the enrichment of terms in the group FP was compared with all genes in the working data set with annotations in their respective ontologies.
To demonstrate relevance with human pancreatic development, we obtained normalized microarray data from a previously published study of hESC differentiation to pancreatic cells (Micallef et al. 2012). Differential gene expression (at a false discovery rate [FDR] adjusted P-value < 0.05) between the differentiated pancreas-like cells (“Nostro protocol”) and hESC lines was determined using the R statistical package limma (Smyth 2005). Mouse genes from the RNA-seq experiments were matched to their corresponding human homologs using the NCBI database HomoloGene (http://www.ncbi.nlm.nih.gov/homologene). The RNA-seq data sets have been deposited in NCBI's Gene Expression Omnibus (GEO) and are accessible through accession number GSE40823.
Immunohistochemistry and in situ hybridization
Mouse embryos were fixed in 4% paraformaldehyde from 2 h to overnight at 4°C. Subsequently, samples were equilibrated in 20% sucrose solution and embedded in OCT compound (Sakura). In situ hybridization on cryostat sections was done as in Schaeren-Wiemers and Gerfin-Moser (1993). Cryosections (10 μm) were incubated with TSA (Perkin Elmer) blocking buffer for 1 h at room temperature and afterward with primary antibodies at an appropriate dilution (Supplemental Table 4). All confocal images were acquired with an LSM 700 confocal laser-scanning microscope (Zeiss).
RT-qPCR
For RNA isolation, embryonic tissues were microdissected, FACS-sorted, and collected in TRIzol reagent (Invitrogen). Subsequently, RNA was extracted from a minimum of 3000 cells according to the manufacturer's instructions (Roche). Briefly, total RNA was resuspended in 10 μL of DEPC H2O and processed for reverse transcription using Transcriptor First Strand cDNA Synthesis kit (Roche). A mix of anchored oligo(dT)18 and random hexamer primers was used to generate the cDNA. Thermolabile nuclease from the Real-Time Ready Cell Lysis kit (Roche) was added to the reverse transcription reaction to degrade dsDNA. Real-time PCR reactions were carried out using the SYBR Green Master mix (Roche) on the ABI StepOne Plus system. Succinate dehydrogenase (SDHA) was used as reference gene. All of the values were normalized to the reference gene and calculated using the REST program (Pfaffl et al. 2002). Data were determined in triplets. All experiments were repeated at least three times unless otherwise stated. Primer sequences can be obtained on request.
Xenopus embryo experiments
Xenopus embryo manipulations and dissections were performed as described (Spagnoli and Brivanlou 2008). The SP64T-xWnt5a plasmid used for microinjection was previously described (Moon et al. 1993). For whole endodermal explant experiments, endodermal cells were microdissected at stage 9 and cultured in 0.1× MMR in the presence of the indicated recombinant proteins until the desired stage. For anterior endodermal explants, dorsal vegetal cells were microdissected at stage 10 and cultured as above. Ornithine decarboxylase (ODC) was used as a reference gene in Xenopus RT-qPCR analysis. All of the values were normalized to the reference gene and calculated using the REST program (Pfaffl et al. 2002). For luciferase reporter assays in Xenopus, embryos were injected with ATF2-luc (Ohkawara and Niehrs 2011) or TOPFLASH and Renilla-TK plasmid DNA (Clevers and Nusse 2012). Three pools of 15 explants each were lysed with passive lysis buffer and assayed for luciferase activity using the Dual Luciferase system (Promega). Recombinant mouse Wnt5a, Wnt5b, and Wnt3a (R&D System) proteins were used at the indicated concentrations in both Xenopus and cell culture experiments. All experiments were repeated at least three times unless otherwise stated.
Cell culture experiments
mESCs (R1 mESC line) were maintained on gelatin-coated plates with mouse embryonic fibroblasts (MEFs) in standard mESC medium: DMEM (Invitrogen), 2 mM glutamax (Invitrogen), 1 mM sodium pyruvate (Invitrogen), 0.1 mM nonessential amino acids (Invitrogen), 15% fetal bovine serum (FBS) (PAN Biotech), 0.1 mM β-mercaptoethanol (Sigma), and 1000 U/mL leukemia inhibitory factor (ESGRO). For differentiation, cultures were MEF-depleted and seeded in mESC medium at high confluency on gelatin-coated dishes. Monolayer differentiation was carried out as described previously (D'Amour et al. 2006; Nostro et al. 2011). Briefly, definitive endoderm medium to day 2 consisted of RPMI medium (Invitrogen) and 0.2% FBS supplemented with 50 ng/mL Activin A and 25 ng/mL Wnt3a at day 1 and Activin A at day 2. Pancreatic endoderm medium to day 5 consisted of RPMI medium and 2% FBS supplemented with 3 ng/mL Wnt3a and 50 ng/mL FGF10. Pancreatic progenitor medium to day 8 consisted of DMEM with 1% (v/v) B27 supplement (Invitrogen), 50 μg/mL ascorbic acid (Sigma), 0.25 μM KAAD-cyclopamine (Toronto Research Chemicals), 2 μM all-trans retinoic acid (Sigma), and 50 ng/mL noggin. All recombinant proteins were purchased from R&D System unless otherwise stated. The BAML cells were cultured as previously described (Fougère-Deschatrette et al. 2006).
Western blot analysis
For Western blot analysis, endodermal cells were lysed as described (Spagnoli and Brivanlou 2008). Immunoblots were incubated with anti-ABC monoclonal antibody, which recognizes the dephosphorylated β-catenin on Ser37 and Thr41 (Millipore); anti-β-catenin polyclonal antibody (Santa Cruz Biotechnology); anti-pJNK monoclonal antibody (Santa Cruz Biotechnology); and anti-α-tubulin monoclonal antibody (Sigma) and analyzed using the LI-COR Odyssey system.
Statistical tests
All results are expressed as mean ± standard errors. The significance of differences between groups was evaluated with Student's t-test. P < 0.05 was considered statistically significant. Statistical tests relevant to the RNA-seq data set analysis are described in the “Bioinformatic Analysis” section.
Acknowledgments
We thank W. Birchmeier for discussion and critical reading of the manuscript, and all members of the Spagnoli, Andrade, and Chen laboratories for helpful discussions. We thank the Max Delbrück Center FACS Core Facility and Dr. Hans-Peter Rahn for support and technical assistance. We are grateful to T. Uemura for the Celsr2 antibody, Y. Minami for the Ror2 antibody, and C. Niehrs for the ATF2-luc plasmid. This research was supported by institutional funds from the Helmholtz Association and the Bundesministerium für Bildung und Forschung (BMBF). F.M.S. is supported by the European Research Council (ERC)-Starting Hepatopancreatic Grant 243045.
Supplemental material is available for this article.
Article is online at http://www.genesdev.org/cgi/doi/10.1101/gad.220244.113.
Spermatogonial stem cells (SSCs) have the potential to acquire pluripotency under specific culture conditions. Takashima et al. report that global DNA hypomethylation triggered by Dnmt1 depletion induces pluripotent cell derivation. Dnmt1 depletion down-regulates Dmrt1, a gene involved in sexual differentiation. Dmrt1 depletion up-regulates Sox2, which in turn up-regulates Oct4 and produces pluripotent cells. These results suggest that the Dmrt1–Sox2 axis plays a crucial role in repression of SSC pluripotency.
Spermatogonial stem cells (SSCs) present the potential to acquire pluripotency under specific culture conditions. However, the frequency of pluripotent cell derivation is low, and the mechanism of SSC reprogramming remains unknown. In this study, we report that induction of global DNA hypomethylation in germline stem (GS) cells (cultured SSCs) induces pluripotent cell derivation. When DNA demethylation was triggered by Dnmt1 depletion, GS cells underwent apoptosis. However, GS cells were converted into embryonic stem (ES)-like cells by double knockdown of Dnmt1 and p53. This treatment down-regulated Dmrt1, a gene involved in sexual differentiation, meiosis, and pluripotency. Dmrt1 depletion caused apoptosis of GS cells, but a combination of Dmrt1 and p53 depletion also induced pluripotency. Functional screening of putative Dmrt1 target genes revealed that Dmrt1 depletion up-regulates Sox2. Sox2 transfection up-regulated Oct4 and produced pluripotent cells. This conversion was enhanced by Oct1 depletion, suggesting that the balance of Oct proteins maintains SSC identity. These results suggest that spontaneous SSC reprogramming is caused by unstable DNA methylation and that a Dmrt1–Sox2 cascade is critical for regulating pluripotency in SSCs.
DNA methylationpluripotencyreprogrammingspermatogoniateratoma
Germ cells are thought to have pluripotency potential because they form teratomas. Teratomas are spontaneously formed in mice on a 129 background, and this strain was used to derive embryonic stem (ES) cells (Evans and Kaufman 1981; Stevens 1984). Although the frequency of spontaneous teratoma development was found to be limited to ∼2%, studies have shown that primordial germ cells (PGCs) in the fetus can be induced to form teratomas by transplanting the genital ridges of mid-gestational embryos into ectopic locations in other animals (Stevens 1984). In the most successful cases, teratomas were found in ∼80% of the PGC transplants. PGCs have pluripotency potential up until 12.5 d post-coitum (dpc). Although these classic experiments provided the foundation for teratoma studies, the mechanism of teratoma development has remained unknown. However, in 1992, two groups (Matsui et al. 1992; Resnick et al. 1992) demonstrated that PGCs can become ES-like pluripotent embryonic germ (EG) cells in vitro when they are cultured with Kit ligand (KitL), fibroblast growth factor2 (FGF2), and leukemia inhibitory factor (LIF). EG cells are similar to ES cells, except in their DNA methylation patterns, and formed chimeras when the cells were transferred into blastocysts (Matsui et al. 1992; Labosky et al. 1994). The frequency of EG cell formation decreases gradually during development, and such potential pluripotency is no longer found in the germline later than 12.5 dpc (Labosky et al. 1994).
In 2003, a long-term culture system for spermatogonial stem cells (SSCs), in which SSCs proliferated in vitro in the presence of FGF2 and glial cell line-derived neurotrophic factor (GDNF), was developed (Kanatsu-Shinohara et al. 2003). Cultured SSCs, designated as germline stem (GS) cells, form grape-like clusters of spermatogonia in vitro but reinitiate spermatogenesis when transplanted into seminiferous tubules of infertile testes. Although GS cells are unipotent and produce sperm, they were subsequently found to transform into ES-like cells (Kanatsu-Shinohara et al. 2004). These ES-like cells, called multipotent GS (mGS) cells, often appear as sheets of epiblast-like cells, which transformed into ES-like compact colonies upon passaging. Despite their spermatogonial origin, they proliferate without GDNF and produce teratomas in seminiferous tubules but are able to contribute to the blastocyst, thereby producing germline chimeras. Additional studies revealed that GS cells directly transform into mGS cells (Kanatsu-Shinohara et al. 2008), and similar cells were also derived from other species, including humans, although some of these results are being questioned (Ko et al. 2010b; Tapia et al. 2011).
While these results showed the pluripotency potential of SSCs, several critical issues remain to be answered. One is their low derivation frequency: mGS cells develop in only one out of every ∼30 testes during GS cell derivation (Kanatsu-Shinohara et al. 2004). Although several groups also reported the derivation of pluripotent cells from postnatal male germ cells, discrepancies exist among these studies in terms of the nature and efficiency of reprogramming (Geijsen and Hochedlinger 2009). For example, one study showed the usefulness of LIF and obtained four lines from 21 mice (Guan et al. 2006). However, multipotent adult GS (maGS) cells produced in this study were unique in that they contribute to not only spermatogenesis but also embryogenesis, but the result of this study is being questioned (Geijsen and Hochedlinger 2009). Although beneficial effects of coculturing on testicular CD34+ stromal cells in pluripotency induction were reported in another study (Seandel et al. 2007), even better results were obtained by low-cell-density culture on mouse embryonic fibroblasts (MEFs), and six ES-like germline-derived pluripotent stem (gPS) cells were derived from eight testes within 4 wk in the most successful case (Ko et al. 2012). However, the quality of MEFs is still a critical factor, and an additional 3–4 wk are required when ES-like cells do not appear by low-cell-density cultures (Ko et al. 2010a). In addition to the variation in induction procedure, the difference in DNA methylation patterns in the differentially methylated region (DMR) of H19 is also pointed out. While gPS cells show androgenetic DNA methylation patterns, maGS cells exhibit somatic cell DNA methylation patterns (Guan et al. 2006; Ko et al. 2009). It also should be noted here that many of these studies claimed derivation of ES-like cells from 129 or C57BL/6 (B6) mice, whose SSCs never proliferate without augmenting GDNF signal by GFRα1 supplementation (Kubota and Brinster 2008).
Such low or inconsistent derivation efficiency has made it difficult to study the molecular mechanism underlying pluripotency induction. We initially noticed that mGS cells often develop during initiation of GS cell cultures and that p53 deficiency improves their derivation (Kanatsu-Shinohara et al. 2004). We also found that mGS cells occasionally appear after freezing–thawing or electroporation (Kanatsu-Shinohara et al. 2005, 2008). Unexpectedly, GS cells were resistant to transfection of Yamanaka factors and did not become pluripotent (Morimoto et al. 2012). However, the mechanism of pluripotency regulation in SSCs has remained unknown. Thus, there is clearly a need to develop a fast and efficient system to induce SSC reprogramming, which will enable us to dissect the molecular mechanism involved in this process.
Here, we report a critical role of Dmrt1 (a gene involved in sex determination) (Raymond et al. 2000) in GS cell reprogramming. We found previously that mGS cells often exhibit abnormal DNA methylation in DMRs of imprinted genes (Kanatsu-Shinohara et al. 2004). Because Dnmt1 is responsible for maintaining genomic methylation, we depleted Dnmt1 and found that Dnmt1 knockdown induces H19 demethylation and mGS cell formation. Furthermore, Dnmt1 knockdown in GS cells was accompanied by the down-regulation of Dmrt1, a gene involved in sex differentiation (Raymond et al. 2000). Although this gene is thought to be responsible for meiotic induction in spermatogonia (Matson et al. 2010), Dmrt1 knockdown in GS cells up-regulates Sox2 and efficiently induces mGS cells, suggesting that Dmrt1 plays a crucial role in repression of pluripotency in SSCs. We also propose a model in which spermatogonial identity is regulated by the balance of Oct proteins.
ResultsReprogramming of GS cells by induction of DNA demethylation
Global methylation of genomic DNA in GS cells is significantly higher than those in mGS and ES cells (Fig. 1A). Because DNA demethylation is often found in DMRs of H19-imprinted genes after reprogramming into mGS cells, we directly examined the role of Dnmt1, which maintains DNA methylation. To induce DNA demethylation, we used a Dnmt1 knockdown vector. When we induced demethylation by Dnmt1 knockdown, the majority of cells underwent p53-dependent apoptosis (Fig. 1B), but no mGS cells were found. However, because Dnmt1 is a maintenance methylase and passive demethylation due to Dnmt1 depletion requires multiple cell divisions (Jackson-Grusby et al. 2001), poor proliferation of GS cells in these cultures probably prevented efficient global DNA demethylation at 7 d. Therefore, GS cells from p53 knockout mice were transduced with a Dnmt1 knockdown vector and kept for ∼1 mo. Although extensive apoptosis occurred after knockdown, some cells divided slowly and formed GS cell colonies. Using this protocol, mGS cells were obtained within 4 wk after Dnmt1 knockdown (Fig. 1C; Table 1; Supplemental Table S1). The morphology and the proliferation pattern of the Dnmt1 knockdown-induced mGS (Dnmt1-mGS) cells were indistinguishable from those of cells that developed spontaneously. Real-time PCR confirmed down-regulation of Dnmt1 expression after knockdown (Supplemental Fig. S1A). Examination of global DNA methylation showed 3.7% ± 0.6% reduction in total methylcytosine levels by Dnmt1 knockdown 2 wk after transfection (n = 3; P < 0.05 by t-test). Combined bisulfite restriction analysis (COBRA) showed progressive demethylation of H19 (Fig. 1D).
Development of mGS cells after Dnmt1 knockdown (KD). (A) Global DNA methylation levels in GS, mGS, and ES cells (n = 3). (B) Suppression of Dnmt1 knockdown-induced apoptosis by p53 knockdown. For each cell type, at least 99 cells were counted 7 d after transfection (n = 5). pSicoR was used as a control. (C) Development of mGS cells from p53 knockout (KO) GS cells after Dnmt1 knockdown. (D) COBRA of cultured cells. Open arrowheads indicate the sizes of the unmethylated DNA fragments, and closed arrowheads denote the sizes of the methylated DNA fragments. PCR products were digested with the indicated enzymes. Percent methylation is shown below the gel. pSicoR was used as a control. (E) Development of mGS cells from wild-type (WT) ROSA26 GS cells after double knockdown of Dnmt1 and p53. (F) RT–PCR analysis. Bars: C,E, 100 μm.
Summary of mGS induction efficiency
To examine whether the wild-type GS cells also transform into mGS cells, we established GS cells from transgenic pups carrying an enhanced green fluorescent protein (EGFP) gene (green mouse). Logarithmically growing GS cells were transduced with Dnmt1 and p53 knockdown vectors. Real-time PCR confirmed down-regulation of both p53 and Dnmt1 (Supplemental Fig. S1B,C). Wild-type GS cells produced mGS cells within 4 wk (Fig. 1E; Supplemental Table S1). RT–PCR confirmed Nanog expression in Dnmt1-mGS cells, which was accompanied by loss of Nanos3 expression, suggesting that GS cells lost their spermatogonial identity and became ES-like cells (Fig. 1F). We did not find mGS cells using 5-azacytidine treatment using both wild-type and p53 knockout GS cells.
Dmrt1 knockdown induces mGS cells
Because Dnmt1 knockdown causes tumors in somatic cells without p53 (Gaudet et al. 2003), we hypothesized that DNA demethylation might have changed the expression of genes responsible for germ cell tumor (GCT) development. We therefore examined the impact of 14 GCT candidate genes by deregulating their expression in a p53 knockout GS cell line. Dnmt1 knockdown down-regulated the expression of several genes, including Dnd1 and Dmrt1, both of which are implicated in the formation of teratomas from PGCs (Fig. 2A; Supplemental Fig. S2A,B; Gilbert et al. 2011). When we carried out knockdown experiments, knockdown of Dnd1 or Dmrt1 yielded mGS cell colonies within 4 wk (Supplemental Table S1). However, none of the other genes showed evidence of conversion.
Development of mGS cells after Dmrt1 knockdown (KD). (A) Real-time PCR analysis of GCT candidate gene expression in p53 knockout (KO) GS cells after Dnmt1 knockdown (n = 3). pSicoR was used as a control. (B) Suppression of mGS cell development by Dmrt1 overexpression (OE) 28 d after Dnmt1 knockdown (n = 9). CSII-EF1α-IRES2-Venus was used as a control. (C) Suppression of Dmrt1-induced apoptosis by p53 or Bax knockdown. For each cell type, at least 68 cells were counted 7 d after transfection (n = 5).
To confirm the effects of Dnd1 and Dmrt1 knockdown on wild-type GS cells, we cotransfected each knockdown vector with a p53 knockdown vector. Although we obtained no mGS cells with Dnd1 knockdown, Dmrt1 knockdown successfully yielded mGS cells (Table 1). Compared with Dnmt1 knockdown, Dmrt1 knockdown induced mGS cells at a higher frequency with a shorter incubation time, and colony development was observed as early as 12 d. Concurrent overexpression of Dmrt1 cDNA with a Dnmt1 knockdown vector significantly reduced mGS cell development (Fig. 2B), which suggests that Dmrt1 down-regulation is responsible for Dnmt1 knockdown-induced pluripotency. As in Dnmt1-mGS cells, simultaneous p53 knockdown was necessary because cells underwent apoptosis due to Dmrt1 knockdown (Fig. 2C). Although Bax knockdown was able to rescue cells from Dmrt1 knockdown-induced apoptosis, we were unable to observe mGS cells (Fig. 2C), suggesting that p53 knockdown has an additional function to promote pluripotency.
Because of the relatively high efficiency of mGS cell formation, we used this system to characterize the reprogramming process. As previously noted for spontaneously developed mGS cells, Dmrt1-mGS cell development is often accompanied by sudden generation of an epiblast-like sheet, which has Nanog expression in the center (Supplemental Fig. S3A). All Dmrt1-mGS cell colonies expressed Nanog on the day when colonies were discovered. This protocol was useful in deriving mGS cells from all seven tested wild-type GS cell lines, and mGS cells were obtained from not only pups but also adults (Supplemental Fig. S3B). These results suggest that Dmrt1 regulates pluripotency in GS cells.
Characterization of Dnmt1- and Dmrt1-mGS cells
RT–PCR and flow cytometry indicated that both Dnmt1- and Dmrt1-mGS cells have typical features of ES cells (Supplemental Fig. S4A,B). They expressed high levels of Oct4 and Sox2 and were positive for Nanog in immunohistochemical analyses (Supplemental Fig. S4C). However, several lines of Dnmt1- or Dmrt1-mGS cells expressed Neurog3 or Sohlh1, which suggested that the cells retain a spermatogonial phenotype (Supplemental Fig. S4A). We carried out bisulfite sequencing analysis to check the DNA methylation patterns in imprinted genes. Consistent with previous studies (Kanatsu-Shinohara et al. 2004), GS cells showed typical androgenetic DNA methylation patterns: hypermethylation in H19 and hypomethylation in Igf2r (Supplemental Fig. S4D). In contrast, Dnmt1- and Dmrt1-mGS cells showed demethylation in H19. Although we occasionally observed Igf2r methylation in mGS cells, no apparent difference in Igf2r was found, regardless of the induction method.
Functional characterization of Dnmt1- and Dmrt1-mGS cells
To determine whether Dnmt1- and Dmrt1-mGS cells are functionally similar to ES cells, we examined their differentiation potential. In these experiments, Dnmt1-mGS cells were produced by shRNA transfection, and the shRNA was removed by cre treatment (Supplemental Fig. S4E,F). We first induced neuroectodermal differentiation in adherent monocultures (Ying and Smith 2003). The cultured cells successfully differentiated into cells expressing βIII-tubulin, a neuronal marker. We then generated embryoid bodies (EBs) in vitro and examined whether the cells could differentiate into other lineages (Fig. 3A). After 5 d of culture on gelatin-coated dishes, the EBs were stained with several antibodies. Some cells stained positively for α-smooth muscle actin, a mesodermal marker, while others expressed cells with α-fetoprotein expression, suggesting that they belong to the endodermal lineage. We found no obvious difference between Dnmt1- and Dmrt1-mGS cells in terms of differentiation patterns and efficiencies. We also transplanted both cell types under the subcutaneous tissues of nude mice to examine their teratoma-forming potential. All clones produced tumors within 4 wk after transplantation. Histological analyses showed that the transplanted cells produced teratomas with the ectoderm (neural tube), mesoderm (chondrocytes), and endoderm (gut epithelium) (Supplemental Fig. S4G).
Differentiation potential of Dnmt1- and Dmrt1-mGS cells. (A) Neuronal or EB-mediated differentiation of mGS cells. (B,C) A live chimera formed by microinjection of Dnmt1-mGS cells (B) or Dmrt1-mGS cells (C) into ICR or B6 blastocyst, respectively. The cinnamon coat color represents the contribution of the Dnmt1- or Dmrt1-mGS cells on a DBA/2 background. (D) Contribution of Dmrt1-mGS cells to various organs, as indicated by EGFP fluorescence. Bar: A, 100 μm.
To test whether these mGS cells contribute to embryonic development, we microinjected them into blastocysts (Supplemental Table S2). In this experiment, GS cells from ROSA26 mice were used to derive Dnmt1-mGS cells, which were treated with cre to remove shRNA for Dnmt1 and p53 before the chimera experiment. On the other hand, we used green GS cells and transfected siRNAs for Dmrt1 and p53 to avoid vector integration for Dmrt1-mGS cell production. Both Dnmt1- and Dmrt1-mGS cells could produce chimeric offspring, albeit at low efficiency (Fig. 3B–D; Supplemental Table S2). Low efficiency of chimera production could be due to residual expression of Neurog3 in these cells (Supplemental Fig. S4H). Nevertheless, both offspring were apparently normal, and Dmrt1-mGS-derived F1 offspring were born by natural mating. No abnormalities in H19 were found by COBRA in these offspring (Supplemental Fig. S4I).
Identification of Dmrt1 target genes
In the next set of experiments, we sought to identify target genes regulated by Dmrt1. A previous study showed that Dmrt1 regulates >1400 genes in spermatogonia (Murphy et al. 2010). Among the diverse types of genes, we focused on those associated with cell cycle regulation (p18 and p19) and pluripotency (Nr5a1, Utf1, Sox2, and Zic3) because we reasoned that deregulation of these genes can potentially stimulate GCT formation. We analyzed the expression levels of these candidate genes in p53 knockout GS cells by real-time PCR after Dmrt1 knockdown (Fig. 4A). The analyses revealed that while Sox2 and Utf1 were up-regulated by Dmrt1 knockdown, Nr5a1 and Zic3 were down-regulated, which suggests that Nr5a1 and Zic3 are dispensable for reprogramming. Both p18 and p19 cyclin-dependent kinase inhibitors (CDKIs) were down-regulated, which may facilitate teratoma formation. Although similar results were obtained using wild-type GS cells with p53 knockdown, we failed to find statistically significant down-regulation of p19 (Supplemental Fig. S5).
Development of mGS cells after Sox2 or Oct4 overexpression (OE). (A) Real-time PCR analysis of Dmrt1 target genes in p53 knockout (KO) GS cells after Dmrt1 knockdown (KD) (n = 3). pLKO.1 EGFP was used as a control. (B) Suppression of mGS cell development by Sox2 knockdown 28 d after Dmrt1 knockdown (n = 13). pLKO.1 EGFP was used as a control. (C) Real-time PCR analysis of Dnmt1, Dmrt1, Sox2, and Oct1/4 expression in p53 knockout GS cells 14 d after Dmrt1 knockdown (n = 4). pSicoR was used as a control. For concurrent Dmrt1 overexpression, CSII-EF1α-IRES2-Venus was a control. (D) Real-time PCR analysis of Oct4 expression in p53 knockout GS cells 14 d after Sox2 overexpression (n = 3). CSII-EF1α-IRES2-hKO1 was used as a control. (E) Reduced Sox2-mGS cell development 28 d after Oct4 knockdown (n = 9).
Based on this observation, we screened for genes that trigger pluripotency in GS cells. We overexpressed pluripotency-related genes or depleted the expression of CDKIs in wild-type GS cells. All transfection experiments were carried out by cotransfection with a p53 knockdown vector. Although Utf1 transfection did not result in mGS cell development, transfection of Sox2 yielded mGS cells using three different wild-type GS cell lines (Table 1; Supplemental Table S1; Supplemental Fig. S2C). Many cells underwent p53-dependent apoptosis after Sox2 overexpression (Supplemental Fig. S6A), but the remaining cells transformed into mGS cells, a pattern similar to that for Dmrt1 knockdown. Sox2 knockdown inhibited Dmrt1-mGS cell development (Fig. 4B; Supplemental Fig. S2A). Neither p18 nor p19 knockdown resulted in mGS cell production. Because concurrent Dmrt1 overexpression suppressed Sox2 expression induced by Dnmt1 depletion, the Dmrt1–Sox2 axis also appears to operate in Dnmt1-mGS cell development (Fig. 4C; Supplemental Fig. S2C).
Sox2 is necessary to regulate transcription factors that affect Oct4 expression in ES cells (Masui et al. 2007). Although GS cells express Oct4, its level is significantly lower than that found in ES cells (Imamura et al. 2006). However, we noted that Oct4 mRNA and protein expression were up-regulated in Sox2 transfected GS cells (Fig. 4D; Supplemental Fig. S7). Moreover, Oct4 knockdown prevented mGS cell development after Sox2 overexpression (Fig. 4E). Although apparent changes in Oct4 expression were not observed after Dnmt1 knockdown (Fig. 4C), we reasoned that this could be due to the relatively variable timing and small number of pluripotent colonies by Dnmt1 knockdown. Because Sox2 maintains the requisite level of Oct4 in ES cells and Oct4 overexpression can sustain Sox2-null ES cells (Masui et al. 2007), we tested the possibility that Oct4 overexpression may abolish the need for Sox2 transfection. Transfection of Oct4 induced mGS cell formation using three different wild-type GS cell lines (Table 1; Supplemental Table S1). Unlike Sox2 overexpression, Oct4 overexpression did not induce GS cell apoptosis (Supplemental Fig. S6B) but did require p53 knockdown for mGS cell formation. These results suggest that Sox2 is responsible for Dmrt1-mGS cell generation and that increased Sox2 expression up-regulates Oct4, thereby inducing the formation of pluripotent cells.
Increased mGS cell development following Oct1 knockdown
Although these results show the importance of Oct4 regulation, the role of this protein in SSCs has been controversial. One study reported that Oct4 down-regulation induces GS cell apoptosis (Dann et al. 2008), while another study showed that Oct6, but not Oct4, is indispensable for SSCs (Wu et al. 2010). However, we noted that GS cells express Oct1 (Fig. 5A). Oct1 mRNA was down-regulated upon FGF2 or GDNF treatment, and Western blot analysis also confirmed this result (Fig. 5B,C). Although we did not observe apparent changes in Oct4 and Oct6 expression at mRNA levels, Western blotting showed that both Oct4 and Oct6 protein levels were slightly up-regulated by cytokine treatment. These results suggested that regulation of Oct1 and Oct4/6 expression is different. Because Oct1 often has the same targets as Oct4 (Kang et al. 2009), we reasoned that the relative balance of Oct proteins is important for regulating pluripotency in SSCs.
Development of mGS cells after Oct1 knockdown (KD). (A) RT–PCR analysis of Oct gene expression in GS cells during logarithmic growth phase. (B) Real-time PCR analysis of Oct expression after cytokine stimulation. GS cells were starved for 3 d and restimulated with the indicated cytokines for 24 h before sample collection (n = 6). (C) Western blot analysis of Oct expression after cytokine stimulation. GS cells were starved for 3 d and restimulated with the indicated cytokines for 24 h before sample collection. (D–G) Real-time PCR analysis of Oct1/4 and Sox2 expression 14 d after Oct4 overexpression (OE) (D), Oct1 knockdown (E), Oct4 knockdown (F), or Oct1 overexpression (G) (n = 3). CSII-EF1α-IRES2-Venus was used as a control for Oct1 and Oct4 overexpression. pLKO.1 was a control for Oct1 and Oct4 knockdown. (H) Real-time PCR analysis of Oct1 and Sox2 expression 14 d after Sox2 overexpression (n = 3). CSII-EF1α-IRES2-hKO1 was used as a control.
To test this hypothesis, we examined the impact of Oct knockdown in GS cells. Knockdown of Oct1 or Oct4 induced GS cell apoptosis in a p53-dependent manner (Supplemental Fig. S6C). Oct6 knockdown also caused apoptosis, which was not rescued by p53 deficiency. We transfected wild-type GS cells with Oct1 or Oct6 knockdown vectors together with a p53 knockdown vector. Knockdown of Oct1, but not Oct6, yielded mGS cells using two different wild-type GS cell lines (Table 1). The efficiencies of Oct4 overexpression and Oct1 knockdown were comparable, but combining Oct4 overexpression and Oct1 knockdown synergistically improved mGS cell derivation efficiency (P < 0.05 by ANOVA), and mGS cells were obtained (Table 1). While Oct4 overexpression increased Sox2 expression (Fig. 5D), Oct1 knockdown increased not only Sox2 expression but also Oct4 expression (Fig. 5E). On the other hand, Oct4 knockdown decreased Sox2 expression, suggesting that Oct4 is positively regulating Sox2 (Fig. 5F). Although Oct1 overexpression decreased Oct4 and Sox2 expression (Fig. 5G), it was not possible to suppress Oct4-mGS cell development, suggesting that Oct1 is necessary but not sufficient for suppressing pluripotency. Because neither Oct4 nor Sox2 overexpression influenced Oct1 expression (Fig. 5D,H), it is likely that Oct1 is regulated independently of Oct4 and Sox2. When we examined whether changes in Oct1/4 expression in GS cells can influence genes that are regulated by Oct4 in ES cells (van den Berg et al. 2010), several genes changed expression by Oct1 overexpression or knockdown (Supplemental Fig. S8). In particular, Utf1, one of the Dmrt1 target genes, was down-regulated by Oct1 overexpression, while it was up-regulated by Oct1 knockdown. There results suggest that Oct1 suppresses pluripotency in GS cells by competing with Oct4 for several pluripotency-related target genes.
Characterization of Sox2- and Oct4-mGS cells
Both Sox2- and Oct4-mGS cells were phenotypically similar to Dnmt1- and Dmrt1-mGS cells and were able to differentiate into cells of the three germ layers in vitro (Supplemental Fig. S9A–E). However, their global gene expression profiles were slightly different from those of ES cells (Fig. 6A; Supplemental Fig. S10). To confirm their pluripotency, we introduced Sox2 or Oct4 cDNA into a floxed vector that was transfected with a p53 knockdown vector. After transformation into mGS cells, the transgenes were removed by cre-mediated deletion (Supplemental Fig. S9F,G). The ability of the resulting cells to form chimeras by blastocyst injection was then tested (Supplemental Table S2). Chimeric offspring were obtained from both Sox2- and Oct4-mGS cells (Fig. 6B,C). Both types of mGS cells showed germline transmission by natural mating. These results suggest that Sox2- and Oct-4 mGS cells are pluripotent.
Development of normal offspring from Sox2- or Oct4-mGS cells. (A) Hierarchical clustering of gene expression by microarray. (B,C) A live chimera formed by microinjection of Sox2-mGS cells (B) and Oct4-mGS cells (C) into B6 and ICR blastocysts, respectively. (Inset) Offspring produced after germline transmission. While Sox2-mGS cell origin was identified by the EGFP expression (B), Oct4-mGS cell origin was confirmed by the cinnamon coat color (C). (D) A model to explain the mechanism of pluripotency regulation in SSCs.
Discussion
The molecular mechanism of culture-induced pluripotency is not well understood. While DNA methylation patterns in ES cells often change within several months, GS cells maintain a stable karyotype and DNA methylation patterns in imprinted genes for >2 yr (Kanatsu-Shinohara et al. 2005). Because demethylation of H19 DMRs is almost always found in spontaneously developed mGS cells, we reasoned that DNA demethylation might have contributed to the reprogramming of GS cells. To test our hypothesis directly, we used Dnmt1 knockdown to induce hypomethylation in GS cells. Unlike ES cells, which can proliferate without Dnmt genes (Tsumura et al. 2006), GS cell survival depends critically on Dnmt1 expression (Takashima et al. 2009). Although Dnmt1 knockdown resulted in apoptosis of GS cells, simultaneous p53 knockdown successfully induced mGS cell formation, suggesting that DNA demethylation and suppression of p53 function are necessary for dedifferentiation of GS cells into pluripotent cells.
In our previous study, we did not observe mGS cells by Dnmt1 knockdown, probably because of the higher virus titer and shorter incubation time (Takashima et al. 2009), and H19 DMR demethylation was not confirmed. The higher virus titer induced extensive apoptosis of GS cells, and the shorter incubation time (∼3 wk) was probably insufficient to induce passive demethylation by cell replication because the current study shows that most of the mGS cell colonies developed after 3 wk. Because deficiency of Dnmt1 causes somatic cell tumors (Gaudet et al. 2003), acquisition of pluripotency in GS cells is suggested to share similarities with tumorigenesis. Consistent with this notion, H19 is often demethylated in human GCTs (Looijenga et al. 1998). In particular, seminomas and nonseminomas generally show biallelic expression of the H19 gene. Moreover, we also recently showed H19 demethylation in ES-like colonies produced by in vitro transformation of mouse SSCs with activated Ras, c-myc, and p53 dominant-negative constructs (Morimoto et al. 2012). Although the roles of imprinted genes in GCTs are unknown, similarities between GCT formation and GS cell reprogramming led us to examine whether GCT candidate genes are deregulated as a result of global DNA hypomethylation.
Dmrt1 was one of the several candidate genes that were influenced by Dnmt1 knockdown, and Dmrt1 knockdown successfully induced mGS cell formation. Dmrt1 is one of a group of conserved transcriptional regulators of sexual differentiation that share a Doublesex/Mab-3 (DM) domain DNA-binding motif and is required for testicular development in vertebrates (Raymond et al. 2000). In mice, this gene is expressed only in the gonad and is essential for differentiation of germ cells and Sertoli cells. Strikingly, testes without Dmrt1 show ovarian differentiation even at the adult stage (Matson et al. 2011). Humans lacking one copy of Dmrt1 exhibit testicular dysgenesis and in some cases are feminized (Krentz et al. 2009). In germ cells, this gene is responsible for the formation of teratomas from PGCs, but it is limited to the 129 background (Krentz et al. 2009). In the postnatal testis, Dmrt1 has been considered as a transcriptional gatekeeper that controls mitosis versus meiosis in germ cells (Matson et al. 2010). Undifferentiated spermatogonia without Dmrt1 showed precocious entry into meiosis and reached meiotic prophase by skipping amplifying divisions of the differentiating spermatogonia population, but no tumor formation was reported in postnatal animals without Dmrt1. In contrast, Dmrt1 overexpression is thought to cause spermatocytic seminomas in human adults by increasing Ret expression (Krentz et al. 2009).
Although regulation of pluripotency by Dmrt1 in spermatogonia has not been reported, we speculate that the discrepancy between our results and previous studies is due to two factors. One is the suppression of apoptosis by p53 knockdown. Dmrt1 is involved in germ cell survival, and while Dmrt1 deficiency in PGCs caused teratomas on a 129 background, germ cell loss was observed in non-129 strains (Krentz et al. 2009). Apoptosis of germ cells is probably caused by reduced GDNF signaling because analyses of mutant testes showed that both GFRα1 and Ret are down-regulated, which suggests that Dmrt1 influences the responsiveness to GDNF. In the present study, we also observed that simple Dmrt1 knockdown induced apoptosis of GS cells, but simultaneous p53 knockdown rescued them from apoptosis, resulting in mGS cell formation. The other factor is the differentiation status of the target cells. In the previous study, analysis of Dmrt1 function in spermatogonia was carried out using cre driven by a Neurog3 promoter (Matson et al. 2010). However, Neurog3 is thought to be expressed in a subpopulation of SSCs in a reversible manner (Yoshida 2010). Therefore, some SSCs may not have undergone Neurog3-cre-mediated Dmrt1 deletion, and the function of Dmrt1 in the remaining SSC population is not known. The differentiation level of the target cell population is important for germ cell transformation because cells enriched for SSCs showed a higher frequency of development into transformed cells than committed spermatogonia after transfection of oncogenes or Yamanaka factors (Morimoto et al. 2012).
Knockdown of two GCT candidate genes, Dnd1 and Dmrt1, induced pluripotency in p53 knockout GS cells. Dnd1 is an RNA-binding protein that regulates PGC viability and suppresses teratoma formation (Zhu et al. 2007). It is similar to Dmrt1 in that it only causes teratomas on a 129 background. However, the mechanism of pluripotency regulation seems to differ between these two genes. Although we were able to obtain mGS cells by Dnd1 or Dmr1 knockdown using p53 knockout GS cells, only Dmrt1 was useful for deriving mGS cells from wild-type cells. In fact, studies on PGCs also suggest that they have independent functions (Krentz et al. 2009). For example, Dnd1 mutants undergo a severe loss of germ cells before 11.5 dpc, whereas those in Dmrt1 mutants survive until birth. Although Pten deficiency induces teratomas in PGCs on non-129 backgrounds, Dmrt1 deficiency did not change levels of Pten or Akt phosphorylation (Kimura et al. 2003; Krentz et al. 2009). These results suggested that Dmrt1 acts either independently of Pten or downstream from the Pten pathway in PGCs. Although p53 deficiency also causes teratomas on non-129 backgrounds, it is distinct from Pten in that Pten-deficient gonocytes and spermatogonia do not show pluripotency (Goertz et al. 2011). Therefore, p53 and Dmrt1 seem to be guardians of pluripotency at both embryonic and postnatal stages.
Of the Dmrt1 target genes, our functional screening suggested that Sox2 acts downstream from Dmrt1 to induce reprogramming. Sox2 mRNA is expressed from the early stages of PGC development, but Sox2 protein decreases from 13.5 dpc to 17.5 dpc in fetal gonocytes and is absent in spermatogonia (Imamura et al. 2006; Campolo et al. 2013). Dmrt1 protein expression is absent by 15.5 dpc but is re-expressed in spermatogonia (Lei et al. 2007). This suggests that Dmrt1 protein expression is discrepant with Sox2 protein expression, which apparently involves additional molecules. On the other hand, although the Oct4 protein is expressed in GS cells, its level is only 10% of that found in ES cells; Oct4/Sox2 sites in GS cells are not occupied by Oct4 and Sox2 despite their hypomethylated status, possibly due to the smaller amount of Oct4 and the absence of Sox2 (Imamura et al. 2006). Nevertheless, our results showed that Sox2 and Oct4 expression levels are closely correlated in GS cells. Although the role of Oct4 in SSCs is still under debate (Wu et al. 2010), high expression of Sox2 appears to overcome Sox2 repression at the translational level and tip the balance toward pluripotency by up-regulating Oct4. Although changes in Oct4 levels were modest, this may have a large impact on pluripotency regulation, considering that the threshold for inducing differentiation in ES cells is set at 50% above or below the normal Oct4 expression (Niwa et al. 2000). Because Oct1 knockdown also induced up-regulation of Oct4, Sox2, and Utf1, relative reduction in Oct1 may perhaps induce pluripotency genes by increasing the access of Oct4 to target genes to regulate several pluripotency-related genes (Fig. 6C). However, given our production of mGS cells by Dnd1 knockdown in p53 knockout GS cells and the fact that Oct1 overexpression could not suppress Oct4-mGS cell development, suppression of pluripotency probably involves additional molecules. Identifying other genes that modulate the response of GS cells to Sox2 overexpression is warranted.
Because mGS cells can be derived without genetic manipulation, SSCs represent a unique resource for deriving pluripotent cells. However, conflicting reports on the induction method and nature of SSC-derived PS cells need to be reconciled. In particular, it still remains unknown why mGS cell development is often accompanied by epiblast-like colony formation, while similar colonies were not reported in other studies (Seandel et al. 2007; Geijsen and Hochedlinger 2009). Such discrepancies could be due to the differences in the cell of origin and/or mode of reprogramming. This may also explain why mGS cell formation occurred at relatively lower frequency compared with gPS cells (0.0047% vs. 0.01%), which gradually form ES-like colonies over several weeks. We also do not know how p53 is involved in pluripotency regulation. Because mGS cell development often occurred after freezing–thawing or electroporation, we think that such changes in culture conditions induce cellular stress, which may influence p53 levels. However, because Bax knockdown could not induce mGS cells, the role of p53 knockdown is not simply rescuing apoptosis but has an additional role in pluripotency regulation. Because we can reproducibly obtain mGS cells with our new protocol, analysis of reprogramming mechanisms is now possible, which will improve our understanding of how pluripotency is suppressed in germ cells despite their similarity to ES cells.
Materials and methodsCell culture
GS cells were derived from the transgenic mouse lines C57BL6/Tg14(act-EGFP-OsbY01) and B6-TgR(ROSA26)26Sor, as previously described (Kanatsu-Shinohara et al. 2003, 2004, 2011). GS cells were also established from the transgenic mouse line Tg(Nanog-GFP, Puro)1Yam (Okita et al. 2007). GS cells from p53 knockout mice were previously described (Kanatsu-Shinohara et al. 2004). Culture medium was based on StemPro-34 SFM (Invitrogen) as previously described (Kanatsu-Shinohara et al. 2003). We also used mGS cells on a DBA/2 background that spontaneously developed during green GS cell derivation (Kanatsu-Shinohara et al. 2004). Growth factors used were 10 ng/mL human FGF2 and 15 ng/mL recombinant rat GDNF (both from Peprotech). ES cells (R1) were a generous gift from Dr. M. Ikawa (Osaka University, Suita, Japan). ES cells (BRC1 and B6-6) were provided by RIKEN BRC.
All pluripotent cell lines were maintained on MEFs in Dulbecco's modified Eagle's medium (DMEM) supplemented with 15% fetal bovine serum (FBS), 1000 U/mL LIF (ESGRO; Merck Millipore), nonessential amino acid mixture (Invitrogen), and 0.1 mM 2-mercaptoethanol (2-ME). We also used 2 μM PD0325901 (Selleck Chemicals), 3 μM CHIR99021 (Biovision, Inc.), and 1000 U/mL LIF in N2B27 medium to maintain these cells for chimera production (Ying and Smith 2003).
For differentiation into neuronal lineages, cells were cultured on gelatin-coated plates for 8 d in N2B27 medium. Neural cell differentiation was induced by replating on a LAB-TEK chamber slide (Thermo Fisher Scientific) coated with 0.2 mg/mL fibronectin (Invitrogen) and maintenance in N2B27 medium supplemented with 20 ng/mL FGF2 for 10 d (Ying and Smith 2003). For EB formation, cells were suspended in DMEM supplemented with 20% FBS, 10 μM 2-ME, and nonessential amino acid mixture and plated on a low-cell-binding plate at a density of 7.5 × 104 cells per 9.6 cm2. Two days after culture, the serum concentration was reduced to 15% FBS, and the cells were cultured for 8 d. EBs were then transferred to a LAB-TEK chamber slide coated with 0.1% gelatin and cultured for 5 d.
Statistical analyses
The results are presented as the mean ± SEM. Significant differences between means for single comparisons were identified using Student's t-test. Multiple comparison analyses were performed using ANOVA followed by Tukey's HSD test.
Accession number
Raw data sets have been submitted to the Gene Expression Omnibus database (http://www.ncbi.nlm.nih.gov/geo) and are available under the accession number GSE43850.
Acknowledgments
We thank Ms. Y. Ogata for technical assistance. This research was supported by Japan Science and Technology Agency (CREST), and the Ministry of Education, Culture, Sports, Science, and Technology (MEXT), Japan.
Supplemental material is available for this article.
Article published online ahead of print. Article and publication date are online at http://www.genesdev.org/cgi/doi/10.1101/gad.220194.113.
Briso et al. find that c-fos expression in the mouse epidermis is sufficient to promote inflammation-mediated epidermal hyperplasia. c-Fos transcriptionally controls mmp10 and s100a7a15 expression in keratinocytes, promoting CD4 T-cell recruitment to the skin. Combining c-fos expression with the carcinogen DMBA leads to the development of highly invasive SCCs, which was prevented by the anti-inflammatory drug sulindac. Human SCCs display a correlation between c-FOS and MMP10 and S100A15 proteins as well as CD4 T-cell infiltration. This work reveals promising therapeutic strategies to treat SCCs.
Skin squamous cell carcinomas (SCCs) are the second most prevalent skin cancers. Chronic skin inflammation has been associated with the development of SCCs, but the contribution of skin inflammation to SCC development remains largely unknown. In this study, we demonstrate that inducible expression of c-fos in the epidermis of adult mice is sufficient to promote inflammation-mediated epidermal hyperplasia, leading to the development of preneoplastic lesions. Interestingly, c-Fos transcriptionally controls mmp10 and s100a7a15 expression in keratinocytes, subsequently leading to CD4 T-cell recruitment to the skin, thereby promoting epidermal hyperplasia that is likely induced by CD4 T-cell-derived IL-22. Combining inducible c-fos expression in the epidermis with a single dose of the carcinogen 7,12-dimethylbenz(a)anthracene (DMBA) leads to the development of highly invasive SCCs, which are prevented by using the anti-inflammatory drug sulindac. Moreover, human SCCs display a correlation between c-FOS expression and elevated levels of MMP10 and S100A15 proteins as well as CD4 T-cell infiltration. Our studies demonstrate a bidirectional cross-talk between premalignant keratinocytes and infiltrating CD4 T cells in SCC development. Therefore, targeting inflammation along with the newly identified targets, such as MMP10 and S100A15, represents promising therapeutic strategies to treat SCCs.
c-FosAP-1inflammationcancerskinCD4 T cell
Skin squamous cell carcinomas (SCCs) are the second most common type of human nonmelanoma skin cancers (NMSK), with an incidence of 16 out of 100,000 people in Europe (Brantsch et al. 2008). Unlike basal cell carcinomas (BCCs), SCCs are characterized by an increased risk of metastasis (Ratushny et al. 2012). SCCs arise from keratinocytes of the epidermis and oral mucosa and are most commonly found in sun-exposed areas. Besides ultraviolet (UV) light, other risk factors have been associated with skin SCCs, such as arsenic exposure, tobacco, and human papilloma virus infection. SCCs typically manifest as a spectrum of progressively advanced malignancies, ranging from a precursor lesion like actinic keratosis to SCCs to, eventually, invasive SCCs.
Inflammatory processes often facilitate cancer development by fostering infiltration of immune cells and promoting stromal remodeling (Hanahan and Coussens 2012). In humans, inflammatory skin diseases such as lupus vulgaris or chronic skin ulcers predispose patients to develop cutaneous SCCs (Baldursson et al. 1993; Motswaledi and Doman 2007). In addition, deregulated expression of oncogenes or tumor suppressors in cancer cells can promote immune cell recruitment to the tumor microenvironment by activating the expression of cytokines or metalloproteases that control proinflammatory pathways (Kerkela and Saarialho-Kere 2003; Zenz et al. 2005; Soucek et al. 2007). Interestingly, CD4 T-cell depletion delayed the appearance of tumors in a skin SCC-prone genetic mouse model (Daniel et al. 2003).
A number of signaling pathways have been described as important in the development of SCCs, such as RAS, p53, Notch, and AP-1 (Guinea-Viniegra et al. 2012; Ratushny et al. 2012). Expression of a dominant-negative AP-1 transgene in the skin protects mice from UV-induced, chemically induced, and papillomavirus-induced tumor formation, indicating that AP-1 activity is essential for tumor development (Cooper et al. 2003). In skin physiology, c-Fos is required for the development of RAS-induced malignant papilloma or squamous cell lesions in the background of an epidermal-specific or complete c-fos knockout mouse, while it is dispensable for mouse skin development and homeostasis (Saez et al. 1995; Guinea-Viniegra et al. 2012). Moreover, keratinocyte-specific c-Fos deficiency promotes skin tumor suppression in a p53/TACE-dependent cell-autonomous manner (Guinea-Viniegra et al. 2012). Importantly, increased levels of c-Fos have been found in human SCCs (Sachdev et al. 2008).
In inflammatory diseases such as arthritis, c-Fos was shown to promote disease development (Aikawa et al. 2008; Shiozawa and Tsumiyama 2009). In contrast, c-Fos can act as a negative regulator of proinflammatory responses in myeloid and lymphoid cell lineages (Ray et al. 2006). It was reported that the AP-1 transcription factor regulates a number of genes affecting the microenvironment in the epidermis, including matrix proteins and other secreted factors (Eferl and Wagner 2003; Wagner and Eferl 2005). Whether c-Fos plays a role in epithelial inflammation and in an epithelial/immune cross-talk remains unknown.
Using inducible and regulatable mouse models, we show that c-Fos expression in the epidermis is sufficient to promote inflammation-dependent development of preneoplastic lesions. This phenotype seems to be induced by recruitment of proinflammatory CD4 T cells to the skin through the induction of novel transcriptional c-Fos target genes mmp10 and s100a7a15 in epithelial cells. We further show that this cross-talk between c-Fos expression in the epidermis and CD4 T-cell recruitment may also be conserved in human SCCs.
ResultsInducible epidermal c-fos expression in adult mice leads to epidermal hyperplasia
Increased c-Fos levels have been observed in SCCs arising from different stratified squamous epithelia (Sachdev et al. 2008; Guinea-Viniegra et al. 2012). To investigate whether c-Fos expression is sufficient to promote SCC development in the skin, we generated a doxycycline (Dox)-inducible mouse model. To achieve inducible, keratinocyte-specific expression of c-fos in vivo, keratin 5 (K5)-rtTA mice were crossed to colTetO-Fos mice (Supplemental Fig. 1A). Double-transgenic mice are referred to as c-FosEp-tetON mice, and wild-type or single-transgenic littermates are referred to as controls. Following 4 wk of Dox treatment, adult c-FosEp-tetON mice showed reduced body size, hair loss, and limited life span compared with controls (Fig. 1A,B; Supplemental Fig. 1B). The mutant mice did not show enlarged spleen size indicative of a systemic inflammatory disease (Supplemental Fig. 1C). The affected skin was thickened, with cornified epidermis appearing as dry, scaly lesions (Fig. 1B). Expression of inducible c-fos was detected in the epidermis 2 wk after Dox induction by RT-qPCR and immunohistochemistry (IHC) (Fig. 1C,D). At 2 wk, c-Fos was mainly localized to the hair follicles, while, at 4 wk, it was clearly visible in the basal layer of the epidermis and outer root sheath of the hair follicles (Fig. 1D, arrows). Other K5-expressing epithelial tissues, such as esophagus, tongue, and forestomach, expressed inducible c-fos, showing a mild phenotype only in the forestomach (Supplemental Fig. 1D), while K5-negative tissues, such as spleen or liver, did not express exogenous c-fos (data not shown). Histological analyses of c-FosEp-tetON back skin revealed a complete disruption of skin architecture, leading to the development of preneoplastic lesions characterized by massive epidermal hyperplasia, hyperkeratosis, parakeratosis, and cell/nuclear atypia (Fig. 1E). Strikingly, most hair follicles were absent, and foci of central keratinization within concentric layers of abnormal squamous cells were often observed (Fig. 1E). Proliferation and survival pathways were increased, as assessed by Ki67 and p-STAT3 stainings, in the basal layer of the epidermis upon inducible c-fos expression as early as after 2 wk of Dox treatment (Fig. 1F,G). Interestingly, the presence of a pronounced leukocyte infiltrate, as compared with resident leukocyte infiltrates in wild-type skin, was observed at 4 wk upon c-fos expression, as assessed by CD45 immunostainings (Fig. 1H). Furthermore, epidermal differentiation marker analyses revealed a dramatic expansion of the different epidermal layers in mutant mice. Multiple cell layers were positive for K5 and K14 (basal compartment), and the K1 (spinous layer) and the loricrin (granular layer) were also expanded (Supplemental Fig. 1E). Furthermore, calcium-induced keratinocyte differentiation, as measured by impaired Keratin10, Notch1, and Hes1 expression, was slightly impaired upon inducible c-fos expression (Supplemental Fig. 1F–K). These results show that inducible expression of c-fos in the epidermis is sufficient to trigger skin hyperplasia.
Inducible expression of c-fos in the epidermis of adult mice promotes epidermal hyperplasia with increased proliferation. (A) Scheme of the experimental design: Six-week-old (6w) mice were fed with Dox (0.5 g/L) in the drinking water, and the analyses were performed after 2 and 4 wk (2w and 4w). (B) Picture of a representative control (n = 11) (left) and a c-FosEp-tetON (n = 17) (right) mouse is shown at 4 wk of inducible c-fos expression with Dox in the drinking water. (C) Total c-Fos and exogenous c-Fos-Flag mRNA expression analyses in the back skin of control and c-FosEp-tetON mice at 4 wk of inducible c-fos expression (n = 3). Mean ± SD. (D, left) c-Fos immunostainings of back skin of control (littermate) and c-FosEp-tetON mice at 2 and 4 wk (n = 3) of inducible c-fos expression. (Right panel) Epidermal c-Fos positive nuclei IHC quantification. (E, left) H&E staining of the back skin of c-FosEp-tetON and control mice at 2 and 4 wk of inducible c-fos expression. (Right panel) Epidermal thickness quantification. (F, left) Ki67 immunostainings of the back skin of control and c-FosEp-tetON mice at 2 wk (n = 3) and 4 wk (n = 3) of inducible c-fos expression. (Right panel) Epidermal Ki67-positive nucleus IHC quantification. Mean ± SD. (G, left) p-STAT3 immunostainings of the back skin of control and c-FosEp-tetON mice at 2 wk (n = 3) and 4 wk (n = 3) of inducible c-fos expression. (Right panel) Epidermal p-STAT3-positive nucleus IHC quantification. Mean ± SD. (H) CD45 immunostainings of the back skin of control and c-FosEp-tetON mice at 2 wk (n = 3) and 4 wk (n = 3) of inducible c-fos expression.
To analyze whether the observed skin phenotype is a direct consequence of c-Fos-induced keratinocyte proliferation and whether c-Fos can promote proliferation cell-autonomously, primary FostetON keratinocytes were employed. Inducible expression of c-Fos was detected in vitro upon Dox treatment (Supplemental Fig. 2A). However, no increased proliferation—as assessed by keratinocyte numbers (Supplemental Fig. 2B), EdU incorporation (Supplemental Fig. 2C), colony formation assays (Supplemental Fig. 2D), and enhanced CyclinD1/2 or CyclinE2 expression—was observed (Supplemental Fig. 2E–G). Furthermore, no changes in the Fos-dependent p53/TACE differentiation pathway were observed upon c-Fos expression (Supplemental Fig. 2I,J). These data demonstrate that c-Fos does not promote proliferation in a cell-autonomous manner.
In addition to the epidermal hyperplasia, histological analysis of c-FosEp-tetON mouse skin revealed an accumulation of inflammatory cells, as shown in Figure 1H. To confirm the infiltration of immune cells in the skin of c-FosEp-tetON mice and determine the nature of these cells, we performed flow cytometry (FACS) analyses in skin homogenates from mice 2 wk after inducible expression of c-fos, when there is no microscopic skin phenotype. Although there were no significant changes in the relative presence of total leukocyte infiltration, as determined by CD45 cells, there was a pronounced accumulation of CD4 T cells as well as Gr1+ granulocytes in the c-FosEp-tetON mouse skin (Fig. 2A). Four weeks after inducible c-fos expression, when the epidermis shows a pronounced hyperplasia, there were increased leukocyte infiltrates in the skin of c-FosEp-tetON mice (Fig. 2B). Moreover, there was a clear accumulation of CD4 T cells in c-FosEp-tetON mice (Fig. 2B), while the number of Gr1+ cells was restored to normal (Fig. 2B). No changes were observed in the infiltration of CD8 T cells or B cells at either 2 wk or 4 wk, but macrophages were significantly reduced at 4 wk (Supplemental Fig. 3A,B). Thus, inducible c-fos expression in epidermis caused selective recruitment of CD4 T cells to the skin. Most CD4 T cells present in the skin of c-FosEp-tetON mutant mice were CD44high (data not shown) according to the expected activated status of the cells. Analyses of the CD4 T cells in the skin draining lymph nodes also revealed increased CD4+CD44high cells in c-FosEp-tetON mice compared with control mice (Fig. 2C), suggesting that accumulation of CD4 T cells in the skin of mutant mice could potentially be the result of the recruitment of activated CD4 T cells from the draining lymph nodes.
c-Fos expression induces skin inflammation characterized by chronic CD4 T-cell recruitment. (A) FACS (flow cytometry) analyses of the skin immune cell infiltrate (CD45+, CD45+CD4+, and CD45+Gr1+ populations) of control (n = 6) and c-FosEp-tetON (n = 9) mice after 2 wk of inducible c-fos expression. (B) FACS analyses of the skin immune cell infiltrate of control (n = 6) and c-FosEp-tetON (n = 9) mice at 4 wk of inducible c-fos expression. (C) FACS analyses of the immune cell infiltrate in the skin draining lymph nodes of CD4+/CD44+ cells from control and c-FosEp-tetON (n = 3) mice at 2 wk of inducible c-fos expression. (D) FACS analyses of the immune cell infiltrate (CD45+CD4+ population) in the skin of control and c-FosEp-tetON mice at 4 wk of inducible c-fos expression followed by 4 wk and 16 wk without Dox treatment. (E) Chemotaxis assay. CD4 T lymphocytes isolated from murine wild-type lymph nodes were activated for 5 h with anti-CD28 and anti-CD3 antibodies. Activated CD4 T lymphocytes were subjected to chemotaxis in a transwell assay using either medium alone, SDF-1 (10 ng/mL), or conditioned medium (CM) from control or c-FostetON keratinocytes cultured ±Dox and ±the recombinant protein S100a7a15. (F, right) Immunohistochemical analyses depicting H&E stainings of the back skin of control (Col-Fos+/+; K5rtTA+/+; Rag−/−) and c-FosEp-tetON [Col-Fos (+/KI); K5rtTA (+/T); Rag+/−] and seven c-FosEp-tetON Rag−/− mice [Col-Fos (+/KI); K5rtTA (+/T); Rag−/−] mice on Dox for 4 wk (n = 7, 4, or 10 males). Right: Epidermal thickness quantification of control, RAG1+/−, and RAG−/− c-FosEp-tetON mice. (G, right) Immunohistochemical analyses depicting Ki67 stainings of the back skin of control, c-FosEp-tetON; Rag+/−, and c-FosEp-tetON Rag−/− mice on Dox for 4 wk (n = 7, 4, or 10 males). (Left) Epidermal Ki67-positive nucleus quantification of control, c-FosEp-tetON; Rag+/−, and c-FosEp-tetON Rag−/− mice. (H) RT-qPCR expression analyses of mRNA of sorted CD4 T cells from the back skin of control and c-FosEp-tetON mice after 4 wk of Dox treatment (n = 3).
To demonstrate that the presence of these cells in the skin was dependent on c-fos expression in keratinocytes, Dox was removed after 4 wk of induction, and the mutant mice were analyzed after an additional 4 or 16 wk, when c-fos expression was shut off. The accumulation of CD4 T cells in the skin was restored to almost normal levels compared with control mice (Fig. 2D), and c-Fos levels were normalized (data not shown). The selective presence of CD4 T cells in the skin is therefore dependent on continuous epidermal c-fos expression.
To assess whether specific factors secreted by c-Fos-expressing keratinocytes can promote migration of CD4 T cells, chemotaxis assays were performed in vitro using wild-type murine primary lymph node-derived CD4 T cells. Supernatants from Dox-treated c-FostetON primary keratinocytes induced a twofold chemotactic migration of CD4 T cells, similar to SDF-1α, which was used as a positive control (Fig. 2E). These results indicate that specific proteins secreted upon c-Fos expression into the supernatants of Dox-treated c-FostetON keratinocytes are able to act as chemotactic molecules for CD4 T cells.
To address whether the selective accumulation of CD4 T cells in the skin of c-FosEp-tetON mice plays a functional role in the development of skin hyperplasia, c-FosEp-tetON mice were crossed to RAG1−/−-deficient mice (Mombaerts et al. 1992). Unlike c-FosEp-tetON mice, double-transgenic c-FosEp-tetON RAG1−/− mice showed significantly smaller skin lesions (Fig. 2F). Moreover, keratinocyte hyperproliferation and activation of the STAT3 pathway detected in FosEp-tetON mice was restored to normal levels in FosEp-tetON RAG1−/− mice, as assessed by Ki67 and p-STAT3 staining (Fig. 2G; Supplemental Fig. 3D), whereas c-Fos levels remained increased (Supplemental Fig. 3E). Thus, CD4 T cells accumulate in the skin upon c-fos expression in keratinocytes and contribute to c-Fos-mediated development of epidermal hyperplasia.
To understand how CD4 T cells could contribute to the development of the skin phenotype, CD4 T cells were FACS-sorted from the back skin of control and c-FosEp-tetON mice after 4 wk of Dox treatment. Interestingly, mRNA expression analyses of T-helper cell subpopulations revealed significant enhanced expression of IL-22 in CD4 T cells infiltrating the skin of c-FosEp-tetON mice when compared with controls (Fig. 2H). IL-22 has been shown to be involved in barrier defense and can promote keratinocyte survival by activating the STAT3 pathway (Boniface et al. 2005; Zhang et al. 2012).
c-Fos is necessary and sufficient to up-regulate the expression of mmp10 and s100a7a15 in keratinocytes
To understand the underlying molecular mechanisms and identify novel transcriptional targets of c-Fos, including putative secreted factors, we next employed genome-wide expression analyses using primary keratinocytes isolated from the c-FostetON mice. No major changes in the expression of genes involved in proliferation or differentiation were observed between c-FostetON keratinocytes upon inducible expression of c-fos (Supplemental Table 1). However, among a number of c-Fos-induced genes, two consistently up-regulated genes encoding secreted factors with a role in triggering inflammatory responses were identified: mmp10 and s100a7a15 (Fig. 3A). MMP10 is an MMP that facilitates the recruitment of leukocytes from the bloodstream by degrading components of the extracellular matrix and modulating cytokine and chemokine activity (Elkington et al. 2005). In addition, S100a7a15 is involved in the recruitment of CD4 T cells and granulocytes (Wolf et al. 2010).
MMP10 and S100a7a15 are novel transcriptional targets of c-Fos. (A) Schematic representation of the two most up-regulated genes in a microarray expression analysis comparing in vitro cultured c-FostetON [coltetO-Fos (+/KI), Rosa-rtTA (+/KI)] keratinocytes treated ±Dox (1 mg/mL; n = 3) for 6, 12, 24, 48, and 72 h. (B) RT-qPCR expression analyses of c-fos, mmp10, and s100a7a15 mRNA in c-FostetON keratinocytes treated ±Dox (1 μg/mL; n = 3). (C) Immunoblot depicting S100a7a15 and MMP10 protein levels as well as vinculin and GAPDH as loading controls in c-FostetON keratinocytes treated ±Dox for 48 h. (D) RT-qPCR expression analyses of c-fos, mmp10, and s100a7a15 mRNA in control and c-FosEp-tetON back skin at 4 wk of inducible c-fos expression (n = 3). Mean ± SD. (E) MMP10 immunohistochemical analyses of the back skin of control and c-FosEp-tetON mice after 2 and 4 wk of inducible c-fos expression (n = 3). (F) S100a7a15 immunofluorescence analyses in the back skin of control and c-FosEp-tetON mice after 4 wk of inducible c-fos expression (n = 3). (G) ChIP of c-Fos at the mmp10 promoter. (Left, top) Scheme of the TRE element at the mmp10 promoter where c-Fos binds. (Right) Endpoint qPCR fragments are shown together with the representation of the percentage of binding of c-Fos to mmp10 promoter. (Left, bottom) Chromatin was immunoprecipitated using c-Fos antibody from c-FostetON keratinocytes treated ±Dox for 24 h. (H) ChIP of c-Fos at the s100a7a15 promoter. (Left, top) Scheme of the TRE element at the mmp10 promoter where c-Fos binds. (Right) Endpoint qPCR-fragments are shown together with the representation of the percentage of binding of c-Fos to s100a7a15 promoters. (Left, bottom) Chromatin was immunoprecipitated using c-Fos antibody from c-FostetON keratinocytes treated ±Dox for 24 h.
The expression of these novel target genes was validated at the RNA and protein levels in vitro and in vivo. A significant increase in mRNA (Fig. 3B) and protein (Fig. 3C) expression levels of mmp10 and of s100a7a15 was observed upon c-Fos induction in cultured keratinocytes. Moreover, a significant increase of MMP10 and S100a7a15 was detected by ELISA in the serum of c-FosEp-tetON mice (Supplemental Fig. 4B; data not shown) and in vitro in the cell supernatants of c-FostetON keratinocytes (Supplemental Fig. 4C; data not shown). Importantly, no other MMPs or TACE were differentially expressed upon c-fos expression, as assessed in the microarray analyses or by RT-qPCR (Supplemental Figs. 4A, 2I). Both mmp10 and s100a7a15 mRNA levels were increased in the skin of c-FosEp-tetON mice compared with controls, as assessed by RT-qPCR (Fig. 3D), IHC, and immunofluorescence (Fig. 3E,F). Furthermore, the chemotactic effect observed using supernatants from c-FosEp-tetON keratinocytes was abrogated when conditioned medium was preincubated with an s100a7a15-blocking antibody (Supplemental Fig. 3C). In addition, CD4 T-cell chemotaxis was also observed upon stimulation with recombinant s100a7a15 protein (Supplemental Fig. 3C). These results suggest that S100a7a15 protein promotes CD4 T-cell recruitment in c-FosEp-tetON mice.
To further assess whether these genes are putative c-Fos target genes, c-fos-deficient keratinocytes treated with TPA were analyzed for the expression of mmp10 and s100a7a15. c-fos levels were gradually increased 2 h after TPA treatment (data not shown), and, after 6 h, 12 h, and 24 h, a gradual increase of mmp10 and s100a7a15 expression was observed. However, this induction was significantly abrogated in the absence of c-Fos (Supplemental Fig. 4D,E).
The promoter regions of mmp10 and s100a7a15 contain TPA response elements (TREs) to which AP-1 transcription factors can bind (Fig. 3G,H, top panel). To evaluate direct binding of c-Fos to both mmp10 and s100a7a15 promoters, chromatin immunoprecipitation (ChIP) assays were performed. Twenty-four hours after c-fos expression in primary c-FosEp-tetON keratinocytes, mmp10 and s100a7a15 showed binding of c-Fos to their promoters (Fig. 3G,H). Recruitment of c-Fos to mmp10 and s100a7a15 promoters was confirmed by ChIP analyses in c-fos-deficient keratinocytes stimulated with TPA for 3 h. Upon in vitro c-fos deletion, no binding of c-Fos to mmp10 and s100a7a15 promoters was detected (Supplemental Fig. 4F,G). These results show that mmp10 and s100a7a15 are direct transcriptional targets of c-Fos in keratinocytes and are likely responsible for the recruitment of CD4 T cells and consequent preneoplastic lesion development. In addition, increased MMP10 expression was also present in other K5-expressing tissues, such as the esophagus, forestomach, or tongue, and leukocyte recruitment was observed specifically in the forestomach (Supplemental Fig. 5A,B).
In addition, continuous TPA application in mice lacking c-fos in the epidermis showed that c-fos is necessary for MMP10 expression and, consequently, the recruitment of CD45+ and CD3+ inflammatory cells (Supplemental Fig. 6A–E). In addition, impaired expression of mmp10 and s100a7a15 was observed in keratinocytes lacking c-fos and stimulated for 4 h and 16 h with TPA (Supplemental Fig. 6F–H).
To address the question of whether MMP10 is an important contributor to the development of c-Fos-induced preneoplastic lesions, we blocked MMP10 activity by using a broad MMP inhibitor (TAPI-1) (Slack et al. 2001). The inhibitor was used in a preventive treatment and administered intraperitoneally (i.p.) three times per week to mice subjected to Dox treatment. As assessed by H&E, c-FosEp-tetON mice treated with the MMP inhibitor showed significant prevention of preneoplastic lesion development (Fig. 4A). However, even though TAPI-1-treated c-FosEp-tetON mice did not show a dramatic hyperplasia, the mice exhibited other features commonly developed by c-FosEp-tetON mice, such as hair follicle cysts (Fig. 4A). While c-FosEp-tetON mice still expressed MMP10 protein levels upon TAPI-1 treatment, the mutants showed no recruitment of T lymphocytes (Supplemental Fig. 7A,B). Keratinocyte proliferation and survival, as assessed by Ki67 and p-STAT3 immunostaining, were reduced compared with vehicle-treated mice (Fig. 4B; Supplemental Fig. 7C). Importantly, an MMP10 activity assay using skin protein lysates demonstrated that the drug reached the skin and inhibited MMP10 (Fig. 4C). mRNA levels of c-fos, s100a7a15, and mmp10 remained elevated upon TAPI treatment (Supplemental Fig. 7D). These results indicate that inhibition of broad MMP/ADAM signaling ameliorates the pathological features developed by the c-FosEp-tetON mice.
Blocking MMP10 signaling suppresses c-Fos-mediated epidermal hyperplasia. (A, left) Immunohistochemical analyses depicting H&E staining of c-FosEp-tetON mice [Col-Fos (+/KI); K5rtTA (+/T)] treated with vehicle or TAPI-1 (10 mg/kg, three times per week for 4 wk; n = 3 or 3). (Right panel) Epidermal thickness IHC quantification. Mean ± SD. (B, left) Immunohistochemical analyses depicting Ki67 stainings of the back skin of c-FosEp-tetON mice [Col-Fos (+/KI); K5rtTA (+/T)] treated with vehicle or TAPI-1 (10 mg/kg, three times per week for 4 wk), (n = 3 or 3). (Right panel) Ki67-positive nucleus IHC quantification. Mean ± SD. (C) MMP10 activity assay of the back skin of vehicle- and TAPI-treated control and c-FosEp-tetON mice.
7,12-dimethylbenz(a)anthracene (DMBA)-induced papilloma and SCC development is accelerated by c-Fos
Hras-activating mutations have been shown to render the epidermis susceptible to the development of skin cancers (Wong et al. 2013). To investigate whether c-Fos is sufficient to promote tumor development upon DMBA treatment, we applied an adapted version of the two-step skin carcinogenesis protocol (DMBA/TPA) (Bremner et al. 1994) to the c-FosEp-tetON mice. Hras mutations were induced by DMBA in the back skin of adult mice, and, instead of TPA, inducible c-fos expression was activated 1 wk later using a low dose of Dox (0.25 g/L) (Fig. 5A). Eight weeks after Dox treatment, c-FosEp-tetON mice developed papillomas, which became ulcerative at 11 wk (Fig. 5B–D). The tumor promoter TPA was included in a small set of animals as a positive control (Supplemental Fig. 8A–C). mRNA expression analyses of the whole tumor revealed increased levels of mmp10 as well as s100a7a15 compared with control skin (Fig. 5E). Histologically, tumors invaded the adjacent dermal and fat layers with features of SCCs, such as aberrant mitosis and atypia (Supplemental Fig. 8D; data not shown). IHC analyses revealed that expression of c-Fos was present throughout the tumor (Supplemental Fig. 8D). Furthermore, the tumors expressed high levels of MMP10 and Keratin 6, a keratin mainly expressed by proliferating keratinocytes (Supplemental Fig. 8D). Moreover, markers associated with cancer cell migration and invasiveness, such as podoplanin, a direct c-Fos target gene (Durchdewald et al. 2008), were up-regulated at the invading front of the tumor (Supplemental Fig. 8D).
c-Fos expression accelerates DMBA-induced papilloma and SCC development. (A) Scheme of the experimental design: Six-week-old (6w) mice were given a topical single dose of DMBA (0.5% [w/v] in acetone), and Dox was supplied 1 wk later (1w) in the water at a concentration of 0.25 g/L. (B) Representative picture of control (littermate) and c-FosEp-tetON (n = 10). (C) Quantification of the papilloma number after 8 wk of inducible c-fos expression. (D) Tumor size in c-FosEp-tetON female and male mice after DMBA and 8 and 11 wk of inducible c-fos expression. Mean ± SEM. (E) mRNA expression analyses of c-fos, mmp10, and s100a7a15 in control back skin and c-FosEp-tetON SCCs. (F) Scheme of the experimental design: Six-week-old mice were given a topical single dose of DMBA (0.5% [w/v]), and Dox was supplied 1 wk later (7w) in the food pellets. Sulindac was supplied in the drinking water at a concentration of 180 mg/L. (G, top panels) Representative pictures of DMBA-treated c-FosEp-tetON mice ± sulindac treatment after 7 wk of inducible c-fos expression. (Bottom)Immunohistochemical analyses depicting H&E and cleaved caspase 3 (cl.caspase 3) immunostaining of the back skin of DMBA-treated c-FosEp-tetON mice ± sulindac treatment after 7 wk of inducible c-fos expression. (H) Lesion number quantified when the mice were sacrificed, comparing controls and c-FosEp-tetON mice treated with DMBA ± sulindac. (I) Lesion size quantified when the mice were sacrificed, comparing controls and c-FosEp-tetON mice treated with DMBA ± sulindac. (J) Schematic representation of changes in the epidermis upon induction of c-Fos, leading to increased MMP10 and S100a7a15 and, subsequently, the recruitment of CD4 T cells. As a consequence, paracrine cytokine signaling (IL-22) promotes keratinocyte proliferation and survival, leading to epidermal hyperplasia and, when combined with DMBA, SCC development.
To test whether c-Fos promotes epidermal proliferation and, subsequently, SCC development by recruitment of inflammatory cells, specifically CD4 T cells, we aimed to inhibit inflammation by using the COX1/COX2 inhibitor sulindac. The adapted version of the two-step carcinogenesis protocol comparing sulindac-treated versus untreated mice was performed (Fig. 5F). After one dose of topical DMBA and 7 wk of inducible c-fos expression, mutant mice developed ∼100-mm2 tumors (Fig. 5G–I). Interestingly, sulindac-treated mice developed fewer tumors, and the ones developed were significantly smaller and did not show increased apoptotic rates (Fig. 5G). Therefore, we conclude that c-Fos is sufficient to promote SCC development and that pharmacological blockade of inflammatory responses, likely preventing CD4 T-cell recruitment, diminishes SCC incidence in c-FosEp-tetON mice.
Human SCCs express high c-FOS levels, correlating with high MMP10 and S100A15 expression levels and CD4 T-cell recruitment
To address whether the findings from the c-FosEp-tetON mouse model were applicable to human skin tumors, expression of c-FOS, MMP10, and the corresponding human orthologs S100A7 (psoriasin) and S100A15 (koebnerisin) was assessed. Sections from 85 BCC, 96 SCC, and 20 perilesional skin samples were analyzed by IHC (Fig. 6A,B). In BCCs, none of the tumors expressed c-FOS protein, which correlated with negative MMP10 staining (Fig. 6A). In contrast, 75% of the SCCs expressed high c-FOS protein levels, correlating with high MMP10 expression (Fig. 6A), and the remaining 25% percent of SCCs that expressed low or undetectable c-FOS levels had low or undetectable MMP10 levels (Fig. 6A). Therefore, a high correlation between c-Fos and MMP10 protein expression in human SCCs was observed (Fig. 6A). However, no correlation was observed between c-FOS and S100A7 protein levels in the same set of samples (data not shown), while 89% (41 positive for S100A15 out of 46 positive for c-FOS) of the S100A15 immunostainings correlated with c-FOS expression in human SCCs (Fig. 6B). Human SCC cell lines treated with a c-FOS/AP-1 inhibitor were used to confirm previous findings. mRNA expression analyses revealed decreased expression of MMP10 and S100A15 when SCC cell lines were cultured in the presence of the c-FOS/AP-1 inhibitor (Supplemental Fig. 9A,B). In addition, lentiviral infection of human epidermal keratinocytes (obtained from American Type Culture Collection) with a plasmid overexpressing c-FOS showed a good correlation between the levels of c-FOS and MMP10 but not with S100A15 (Supplemental Fig. 9C).
Correlating c-FOS levels and MMP10 and S100A15 in SCCs but not in BCCs or perilesional healthy skin. (A, top) c-FOS immunostainings on human “perilesional” skin (n = 20), BCCs (n = 85), and SCCs (n = 46). (Middle) MMP10 immunostainings on human “perilesional” skin (n = 20), BCCs (n = 85), and SCCs (n = 42). (Bottom) CD4 immunostainings on human “perilesional” skin (n = 20), BCCs (n = 85), and SCCs (n = 42). (Below) Correlation of c-FOS and MMP10 expression levels on human SCC immunostainings. (B) S100A15 immunostainings on human healthy skin (n = 20), BCCs (n = 85), and SCCs (n = 66).
Since c-FOS expression correlated with MMP10 and S100A15 in SCCs, the levels of CD4 T-cell infiltrates were assessed by immunostaining on human samples. Interestingly, 73% of the c-FOS-positive SCCs showed CD4 T-cell infiltrates, which was confirmed by CD3 immunohistochemical analyses (data not shown). In addition, a correlation between IL-22 expression and CD4 was observed in the tumor-infiltrating CD4 T lymphocytes (Supplemental Fig. 9D). Collectively, these data show a strong correlation between c-FOS expression in human SCCs, MMP10, S100A15, and CD4 T-cell infiltrates (Fig. 6A,B). These results suggest that the presence of CD4 T cells in human SCCs is likely the consequence of increased c-FOS, MMP10, and S100A15 protein levels.
Discussion
Here we show that inducible expression of c-fos in the basal layer of the epidermis leads to non-cell-autonomous keratinocyte proliferation and development of preneoplastic lesions caused by chronic recruitment of CD4 T cells (which secrete IL-22) through s100a7a15 and MMP10 signaling. Moreover, we show that upon a single dose of DMBA, c-Fos is sufficient to promote SCC development, which is ameliorated by anti-inflammatory treatment (Fig. 5J).
The expression of a number of proto-oncogenes in keratinocytes has been shown to favor the development of SCCs (Ratushny et al. 2012). However, in most cases, it is unclear whether this phenotype is solely due to an intrinsic effect on proliferation and/or survival of the targeted keratinocytes or an indirect effect on the microenvironment that feeds back to the keratinocytes. In this study, we show that c-Fos expression in adult epidermis is sufficient to cause skin hyperplasia. Nevertheless, it does not affect cell proliferation and/or survival of isolated keratinocytes, indicating that the effect of c-FOS on the development of skin hyperplasia and SCCs is non-cell-autonomous. Previous loss-of-function studies employed conditional targeted mice and keratinocytes lacking c-Fos to demonstrate a differentiation-induced inhibition of SCC development in a cell-autonomous manner (Guinea-Viniegra et al. 2012). Our study is the first one showing that the expression of c-Fos in keratinocyte requires additional signals from the microenvironment to cause the early signs of skin malignancy. In addition, we also show that paracrine external signals are provided by inflammatory cells that are recruited to the epidermis upon c-fos expression in keratinocytes.
The association of chronic inflammation and cancer initiation/progression has now been reported for a number of cancers (Colotta et al. 2009; DiDonato et al. 2012). In skin, certain inflammatory diseases such as lupus vulgaris (Motswaledi and Doman 2007) or chronic skin ulcers (Baldursson et al. 1993) predispose patients to NMSK (SCC). We show here that expression of c-Fos in keratinocytes is sufficient to trigger chronic accumulation of CD4 T cells in the epidermis with no effect on CD8 T cells and B cells. Since RAG deficiency prevents c-Fos-mediated skin hyperplasia, we propose that selective accumulation of CD4 T cells in the skin is directly promoting the development of the preneoplastic lesions in the epidermis. Mechanistically, increased intralesional/intratumoral IL-22-producing CD4 T cells could explain the epidermal hyperplasia or SCC development observed upon inducible c-Fos expression. The overall role of CD4 T cells in the incidence of preneoplastic lesions and SCCs in the skin has also been supported by other studies using the K14-HPV tumor-prone mice (Kemp et al. 1999; Daniel et al. 2003; de Visser et al. 2005). Unlike CD4 T cells, CD8 T cells seem to protect against skin carcinogenesis (Di Piazza et al. 2012), in agreement with the current view in cancer and inflammation assigning a protective effect to cytotoxic CD8 T cells and a tumor-promoting effect to helper CD4 T cells (Shiao et al. 2011; Brown et al. 2012). Nevertheless, there are some specific cases where CD4 T cells may have a tumor-protective role (Demehri et al. 2012). Although the presence of inflammatory cells in tumors has been previously reported in different types of cancer, it remains unknown when during development of the malignancy the inflammatory response occurs and what the extracellular signals that trigger this accumulation of inflammatory cells are. In this study, genome-wide expression analyses revealed two genes (s100a7a15 and mmp10) consistently up-regulated in c-Fos-expressing keratinocytes that could mediate the recruitment of CD4 T cells to the epidermis. We show here that mouse S100a7a15 is able to induce CD4 T-cell recruitment. Furthermore, S100a7a15, also named psoriasin in humans because of its involvement in psoriasis, is known to act as a chemoattractant molecule for granulocytes and CD4 T cells but not for CD8 T cells, B cells, or macrophages (Wolf et al. 2008, 2010). Altered expression of S100A7 and S100A15 has also been found during human epithelial tumorigenesis (Alowami et al. 2003; Kesting et al. 2009; Tiveron et al. 2012; Hattinger et al. 2013), although it is not known what is inducing their expression. Inducible expression of s100a7a15 in the mammary gland has also been shown to enhance tumor growth and metastasis (Nasser et al. 2012). MMP10, similar to other metalloproteases, can facilitate the recruitment of infiltrating cells by remodeling the extracellular matrix at the proximity of the c-Fos-expressing keratinocytes. Increased levels of MMP10 have been associated with cancer progression/initiation (Liu et al. 2012). MMP10 deficiency in a K-Ras-mediated lung cancer model leads to a decrease in tumor burden compared with controls (Regala et al. 2011). In addition, MMP10 has also been associated with inflammatory disease, such as arthritis (Barksby et al. 2006). Here we show that pharmacological blockers of MMPs in vivo delay the development of the preneoplastic lesions caused by c-fos expression, likely due to inhibition of MMP10, the only MMP/ADAM up-regulated upon c-Fos expression. Importantly, our studies also identify S100a7a15 and MMP10 genes as two novel direct targets of c-Fos in keratinocytes. Following DMBA-induced Hras mutations, inducible c-fos expression in the epidermis is sufficient to promote SCC development in mice with high levels of S100a7a15 as well as MMP10 proteins, further supporting the role of these proteins in mediating c-Fos functions in skin cancer.
Thus, our study in mouse models has uncovered a non-cell-autonomous function of c-Fos in the epidermis, linking proinflammatory responses to preneoplastic and neoplastic lesion development with a major contribution of CD4 T cells, likely due to the induction of newly uncovered target genes like s100a7a15 and MMP10. Importantly, our results also support that this novel pathway is relevant for human SCCs. In humans, there are two genes—S100A7 (psoriasin) and S100A15 (koebnerisin)—that correspond to the single s100a7a15 ortholog in mice. Psoriasin and koebnerisin have been previously shown to be up-regulated in human epithelial skin tumors (Moubayed et al. 2007; Hattinger et al. 2013). Here we show that S100A7 and S100A15 expression is increased in cutaneous SCCs compared with normal skin and epithelial BCCs. However, while c-FOS and S100A15 proteins are coexpressed in the SCCs, the c-FOS and S100A7 expression pattern are not overlapping. Thus, functionally, S100A15 in humans may resemble mouse s100a7a15 in SCCs. Similar to S100A15 and S100A17, increased MMP10 protein levels have also been reported in human epithelial tumors such as esophageal or oral SCCs (Stott-Miller et al. 2011; Liu et al. 2012). Indeed, tumoral and salivary MMP10 levels are being used as markers of oral SCCs (Stott-Miller et al. 2011). Our study shows a strong correlation between the expression of c-FOS and MMP10 in human SCCs. Interestingly, our results also revealed CD4 T-cell infiltrates in human SCCs, but not BCCs, and a positive correlation between high levels of c-FOS, MMP10, and CD4 T cells, and these cells secrete IL-22, which induces keratinocyte proliferation. Therefore, induction of MMP10 and S10015 expression by c-Fos in human SCCs could promote the selective accumulation of CD4 T cells and cancer progression. S100A15 and MMP10 may emerge as novel targets in human skin cancer progression as an alternative to intervene with CD4 T-cell responses in the skin.
Materials and methodsGeneration of c-FosEp-tetON mice
ColTetO-Fos mice were generated by knocking in a Flag-tagged c-Fos cDNA controlled by a tetracycline operator (tetO) downstream from the col1a1 gene (L Bakiri and EF Wagner, unpubl.). K5-rtTA mice were obtained from S. Gutkind. Control and mutants were identified by PCR genotyping on tail DNA. Mice were maintained in a C57BL/6x129sv background. For the carcinogenesis treatment, F1 FvB mice were used, since this background has been shown to accelerate tumor formation (Naito and DiGiovanni 1989). Dox (Harlan) was supplied ad libitum at 0.25–0.5g/L (Sigma-Aldrich) in 2.5%–5% (w/v) -sucrose-supplemented drinking water or in the food pellets. Double-transgenic mice are referred to as c-FosEp-tetON mice, and wild-type or single-transgenic littermates were used as controls. Mice were treated with Dox between 6 and 8 wk of age.
Study approval
Mice were kept in the animal facility in accordance with institutional policies and federal guidelines for animals used in biomedical research following the recommendations of the Ethics Committee and Animal Welfare of the Insituto de Salud Carlos III (Madrid, Spain), following the Royal Decree 10th of October 1201/2005. Animal experiments were approved by the Animal Experimental Ethics Committee of the Instituto de Salud Carlos III.
Chemical carcinogenesis
For DMBA/TPA-induced skin carcinogenesis, cohorts of c-FosEp-tetON and control mice at 5 wk of age received a single application of 100 μL of an acetone solution containing 0.5% DMBA (Sigma-Aldrich) applied to the dorsal surface. TPA (10−4 M; Sigma-Aldrich) in acetone was applied twice per week to the dorsal surface. Dox treatment was used either at 0.25 g/L in the drinking water or ad libitum in the food pellets. DMBA was combined with Dox or with Dox and TPA. A small cohort of mice also received 180 mg/L sulindac (Sigma-Aldrich) in the water at the same time as Dox in the food pellets.
MMP inhibitory treatment
Six-week-old to 8-wk-old control and c-FosEp-tetON mice were treated with Dox for 4 wk. TAPI-1 (MMP inhibitor; Peptides International) was injected i.p. every other day three times per week for 4 wk. TAPI was diluted in 5% DMSO in saline. The final concentration used was 10 mg/kg.
Histological analysis
Histology and IHC were performed according to Zenz et al. (2003). Briefly, tissues were fixed in PBS-buffered 3.7% formalin or frozen in OCT at −80°C. Formalin-fixed tissues were dehydrated, and 4-μm sections were used; 7-μm cryosections were cut from OCT-fixed and frozen tissues. Human tissue arrays from paraffin blocks were produced as 0.5-mm punches taken out of the paraffin-embedded material by parallel viewing of H&E-stained sections. H&E staining was performed according to standard procedures (Sigma-Aldrich). The following primary antibodies were used: rabbit monoclonal to Ki-67 (SP6 Master Diagnostica); c-Fos (Calbiochem); Flag-DYKDDDDK tag (Cell Signaling); K6, K1, K5, and loricrin (Covance); podoplanin (Developmental Studies Hybridoma Bank); S100a7a15 and S100A15 (donated by R. Wolf, Munich, Germany); MMP10 (Leica); c-FOS (Santa Cruz Biotechnology, D1); and CD4 (DAKO, 4B12). Alexa Fluor 488 dye-labeled secondary antibodies (Invitrogen) were used at concentrations from 1 μg/mL to 50 μg/mL. Sections were counterstained with Carazzi's hematoxylin (Panreac) and analyzed by light microscopy (Leica, DM2500).
Human skin SCC samples obtained after informed consent (ethics permission no. 125/10/2012 EK 405/2006) and approval were provided and evaluated by Dr. Peter Petzelbauer (Medical University of Vienna, Austria). Production of tissue arrays from paraffin blocks from BCC and SCC lesions (from the dermato-pathological data bank) was performed according to ethics committee permission 405/2006 and extension 125/10/2012 (Medical University of Vienna, Austria). In brief, by this technique, 0.5-mm punches are taken out of the paraffin-embedded material by parallel viewing of H&E-stained sections.
Flow cytometry
Total skin from control and c-FosEp-tetON mice was subjected to mild digestion with liberase and DNase and mechanically disrupted using GentleMACS Dissociator (Miltenyi). The isolated cells were then stained for 30 min with NK1.1 PerCPCy5.5 (1:200; BD Pharmingen), CD25 APC (1:200; BD Pharmingen), CD45.2 FITC (1:400; BD Pharmingen), g/d TCR Brilliant Violet 421 (1:200; Biolegend), CD3 (1:200; BD Pharmingen), CD4 PECy7 (1:100; BD Pharmingen), CD8 FITC (1:200; BD Pharmingen), Gr1 FITC (1:100; BD Pharmingen), CD45 (1:500; BD Pharmingen), CD11b PerCPCy5.5 (1:200; BD Pharmingen), B220 PE (1:200; BD Pharmingen), CD11c PE (1:100; BD Pharmingen), CD69 PE (1:200; BD Pharmingen), and CD44 and APC (1:200; BD Pharmingen). The cells were then fixed using BD Cytofix buffer. Samples were collected in a FACS CANTO II (BD Pharmingen) equipped with 488-nm, 640-nm, and 405-nm lines. We used pulse processing to exclude cell aggregates and live/dead fixable dye Aqua to exclude dead cells. At least 10,000 live single CD45+ events were collected; all data were analyzed using FlowJo 9.5.3 (Treestar). Skin draining lymph nodes were removed from the mouse and kept in 5% FBS RPMI. Lymph nodes were smashed with a plunger against a 40-mm cell strainer. The strainer was washed with 5% FBS RPMI. Cells were spun down at 300g for 10 min. Finally, leukocytes were resuspended in PB buffer with low BSA, and staining was carried out.
Cell culture and adenoviral infection
Isolation and culture of mouse primary tail keratinocytes was performed as described elsewhere (Zenz et al. 2003). Twenty-four hours after plating, keratinocyte medium was changed to KC-SFM (Gibco). Keratinocytes were cultured at 32°C. The medium was changed every other day and, if needed, supplemented with Dox at 1 mg/L. Ca2+ (0.05 mM or 2 mM CaCl2) stimulation of keratinocytes was performed in equally semiconfluent cultures for proper comparison. Cells were collected at different time points after Dox treatment. Primary c-Fosfl/fl keratinocytes were cultured in the same conditions and treated with adenoviruses. Adeno-Cre (AdCre) or Adeno-green fluorescent protein (AdGFP) adenoviruses were purchased from the University of Iowa. Primary keratinocytes were infected with 300 particles per cell in KC-SFM medium. Cells were treated with TPA for 3 h and collected at different times post-infection. The epidermal-derived SCC9 cell line was kindly provided by M. Sibilia (Institute for Cancer Research, Vienna, Austria) (Lichtenberger et al. 2010), and SCC12 and SCC22 were provided by G.P. Dotto (University of Lausanne, Lausanne, Switzerland) (Lefort et al. 2007). All of the SCC cells were grown in DMEM and 10% FBS. The SCC cells were cultured with vehicle or with a c-Fos/AP-1 inhibitor (Aikawa et al. 2008).
Packaging of retroviral constructs and retroviral transduction
For production of viruses, 293T cells were transfected with Lipofectamine 2000 (Gibco) and the indicated plasmids according to manufacturer's instructions. After 48 h, supernatants were filtered through a 0.45-μm sterile filter and used directly or stored at −80°C until further use. Plasmids used were LeGO-iG2 (Weber et al. 2008) and Precision LentiORF human FOS (Thermo Scientific, OHS5897-202616290). Viruses were pseudotyped with VSV-G. Human keratinocytes were infected with the different lentiviral particles.
Fos activity assay
TAPI- and vehicle-treated control and c-FosEp-tetON mouse skin extracts or keratinocyte lysates were used in c-Fos activity assays (Active Motif, TransAm AP-1 c-Fos) following the manufacturer's instructions.
CD4 T-cell chemotaxis assay
CD4 T cells were purified from wild-type mouse spleens by positive selection using MACS kits as recommended by the manufacturer (Miltenyi Biotec). CD4 T cells were activated with plate-bound anti-CD3 mouse antibody (3 μg/mL; BD Pharmingen, OKT3) and a soluble anti-CD28 mouse antibody (2 μg/mL; BD Pharmingen) for 5 h. SDF-1α (Vitro) was used at 10 ng/mL. Chemotaxis assay was performed using a plate with a transwell containing a 5-μm polycarbonate membrane. CD4 T cells were diluted in complete RPMI medium (5% fetal calf serum, 2.5 mg/L glucose, 2 mM glutamine, 10 μg/mL folate, 1 mM pyruvate, 50 μM β−mercaptoethanol). In the lower chamber, the condition medium was added (±SDF-1α). Two hours after plating CD4 T cells, the insert was lifted, and cells that migrated to the lower chamber were analyzed using a BD FACS Canto II flow cytometer (BD Biosciences).
MMP10 enzymatic assay
An enzymatic assay using back skin protein lysates was performed following the manufacturer's instructions (AnaSpec).
Protein isolation
Back skin from mice was shaved and snap-frozen in dry ice, or cells were scraped from the plate with a cell scraper. Protein isolation for Western blot was performed in RIPA buffer (50 mM Tris, 150 mM NaCl, 0.1% SDS, 0.5% deoxycholate, 1% NP-40) containing a protease inhibitor cocktail (0.1 mM Na3VO4, 40 mM b-glycerophospate, 40 mM NaPPi, 1 mM NaF) (Sigma-Aldrich), and tissues were homogenized using Precellys 24 (Bertin Technologies). For protein lysates from cultured keratinocytes, cells were scraped (100 μL for each well in a six-well plate) with a cell scraper and kept for 15 min in ice. The cell lysate was spun down at maximum speed, and the supernatant was collected. Protein lysates were quantified by using Pierce BCA protein assay reagent (Thermo Scientific). Western blot analysis was performed according to standard procedures using the following antibodies: c-Fos (BD-Pharmingen), Flag (Sigma-Aldrich), vinculin (Sigma-Aldrich), MMP10 (DAKO), S100a7a15 (provided by Dr. Ronald Wolf), b-Actin (Sigma-Aldrich), and GAPDH (Sigma-Aldrich).
The blots were incubated with the appropriate secondary horseradish peroxidase-coupled antibodies (GE Healthcare) and developed using the Amersham ECL Plus Western blotting detection reagent (GE Healthcare) and Amersham ECL Hyperfilms (GE Healthcare). Alternatively, blots were incubated with Alexa-Fluor 680-coupled secondary antibodies (Invitrogen) and visualized using the Odyssey imaging system (Li-Cor).
RT-qPCR
Total RNA was isolated using Trizol (Sigma-Aldrich) according to the manufacturer's instructions. Complementary DNA was synthesized using 1 μg of RNA, Ready-To-Go-You-Prime-First-Strand Beads (GE Healthcare), and random primers (Sigma-Aldrich) as described in the manufacturer's protocol. RT-qPCR reactions were performed using GoTaq RT-qPCR Master Mix (Promega) and Eppendorf fluorescence thermocyclers (Eppenddorf). The 2−ΔΔCT method was used to quantify amplified fragments. Expression levels were normalized using at least two housekeeping genes (gapdh and hprt). Primer sequences will be provided on request.
Genome-wide expression analyses
RNA of samples was hybridized to the Illumina MouseRef-8 V2 R3 BeadChip array according to the manufacturer's instructions (Illumina, Inc.). Microarray scanning was done using a Beadstation array scanner. Data preprocessing and quality control were conducted using packages of the Bioconductor project implemented at the DKFZ in-house-developed ChipYard framework for microarray data analysis (http://www.dkfz.de/genetics/ChipYard). In summary, microarray probes were annotated based on Ensembl (version 58) using an in-house BLAST-based pipeline. Before normalization with variance-stabilizing transformation (VST) and robust spline normalization (RSN) algorithms, beads with signals below the negative controls were removed, and positive, negative, and hybridization controls were excluded.
Normalized log2-transformed microarray data were used to calculate the gene fold change value (i.e., how many times the intensity of gene expression changed compared with the control). The fold change value has no measure of statistical significance but is biologically relevant and specifically usable for data that, other than the time series, have no other replication. Genes were ranked according to the Euclidian distance between the maximum expression within a probe set and the mean expression of a probe set over all time points (as shown in Busch et al. 2008), both scaled (normalized) to the maximum value measured in each data set.
ChIP
ChIP was performed using both the EZ-Zyme chromatin prep kit and EZ-Magna ChIP G ChIP kits (Millipore). Keratinocytes were isolated from the tails from adult c-Fosfl/fl or c-FostetON mice. Two confluent P100 plates were used per condition. c-FostetON keratinocytes were cultured as explained before and treated for 24 h with 1 μg/mL Dox. c-Fosfl/fl keratinocytes were infected with 300 particles of AdCre- or AdGFP-expressing viruses per cell. c-Fosfl/fl keratinocytes were cultured for 4 d after infection and serum-starved overnight. The next day, c-Fosfl/fl keratinocytes were treated for 3 h with 10 nM PMA (phorbol 12-myristate 13-acetate). Primers will be provided on request.
Flow cytometry and FACS sorting
Skin from c-FosEp-tetON mice and controls was shaved and removed from the mice. Dermal fat was scraped off using a scalpel. Tissues were finely minced with scissors and digested with 120 μL of 4 mg/mL DNase (Sigma-Aldrich) and 200 μL of 5 mg/mL liberase (Roche) in 4 mL of RPMI in MACs C tubes and incubated for 20 min at 37°C in a water bath. Tissue was homogenized using gentleMACS Dissociator. Enzymatic digestion was quenched by adding 10 mL of 10% Ch-FBS, 2 mM EDTA, and 250 μg/mL DNase RPMI. Released cells were filtered through a 70-μm cell strainer and spun down at 300g for 10 min at 4°C. Red blood cells were removed using 1 mL of Red Blood Cell Lysing buffer (Sigma-Aldrich) for 2 min at room temperature. Cells were washed in 5 mL of high PB buffer (2 %BSA, 2 mM EDTA, 250 μg/mL DNase in PBS) and resuspended in 1 mL of high PB buffer. FcBlock (1:200; BD Pharmingen) was used for 20 min at room temperature to avoid unspecific staining. Cells were washed and resuspended in low PB buffer (0.5 %BSA, 2 mM EDTA, 250 μg/mL DNase in PBS) and incubated with specific antibodies mixed previously in an antibody cocktail mix to avoid variability among samples (CD45-APC/PE [1:400], CD3-APC [1:200], Gr 1 FITC [1:100], B220 PE [1:200], CD44 APC [1:200], CD8 FITC [1:200], and F4/80 FITC [1:100] [BD-Pharmingen]). For FACS sorting, cells were used right after staining to avoid cell death. DAPI was used as a viability marker. Sorting was performed using a BD Influx cell sorter. For flow cytometry analyses, cells were fixed using BD cytofixed and Acqua live.
EdU proliferation
Sixty percent confluent c-FostetON keratinocyte cultures were treated ±Dox (1 mg/mL) for 1 d. An EdU proliferation assay was performed using the Click-iT EdU cell proliferation assay, and keratinocytes were treated following the manufacturer's instructions. Analyses were performed using a BD FACSCanto II flow cytometer. Analyses were done using FlowJo software.
Statistical analysis
Unless otherwise specified, results are presented as mean ± SD. P-value was calculated using Student's t-test.
Acknowledgments
We thank the members of the Wagner laboratory for critical reading of the manuscript and sharing reagents. We thank Marc Zapatka and Tobias Bauer and for normalization of expression profiling data and pathway analyses. We thank Guillermo Medrano and Gema Luque for taking care of the experimental mice. We thank Ultan Cronin for technical and analytical support in the cytometer and FACS sorter. E.F.W. is funded by the Banco Bilbao Vizcaya Argentaria (BBVA) Foundation and a European Research Council Advanced Grant (ERC FCK/2008/37). E.M.B. was funded by La Caixa International PhD Programme for 4 years. J.G.V. was funded by the Ramón y Cajal fellowship program (Spanish Ministry of Economy). P.A. and R.E. are funded by a grant from the German Ministry for Research and Education (BMBF) funding program AGENET.
Supplemental material is available for this article.
Article published online ahead of print. Article and publication date are online at http://www.genesdev.org/cgi/doi/10.1101/gad.223339.113.
Somatic mutations in the isocitrate dehydrogenase genes IDH1 and IDH2 occur frequently in acute myeloid leukemia (AML). Chen et al. find that IDH2 mutants cooperate with oncogenic Flt3 or NRas alleles to drive leukemia in mice by impairing the differentiation of myeloid cells. Inhibiting the bromodomain-containing protein Brd4 triggers rapid differentiation and death of IDH2 mutant AML. These results demonstrate a critical role for mutant IDH2 in leukemogenesis and identify an IDH-independent strategy to therapeutically target these cancers.
Somatic mutations in the isocitrate dehydrogenase (IDH) genes IDH1 and IDH2 occur frequently in acute myeloid leukemia (AML) and other cancers. These genes encode neomorphic proteins that produce the presumed oncometabolite 2-hydroxyglutarate (2-HG). Despite the prospect of treating AML and other cancers by targeting IDH mutant proteins, it remains unclear how these mutants affect tumor development and maintenance in vivo, and no cancer models exist to study the action of IDH2 mutants in vivo. We show that IDH2 mutants can cooperate with oncogenic Flt3 or Nras alleles to drive leukemia in mice by impairing the differentiation of cells of the myeloid lineage. Pharmacologic or genetic inhibition of IDH2 triggers the differentiation and death of AML cells, albeit only with prolonged IDH2 inhibition. In contrast, inhibition of the bromodomain-containing protein Brd4 triggers rapid differentiation and death of IDH2 mutant AML. Our results establish a critical role for mutant IDH2 in leukemogenesis and tumor maintenance and identify an IDH-independent strategy to target these cancers therapeutically.
Acute myeloid leukemia (AML) is a heterogeneous cancer involving the accumulation of immature cells of the myeloid lineage (Shih et al. 2012; The Cancer Genome Atlas Research Network 2013). Genomic and functional studies have identified two broad classes of mutations that cooperate during AML development (Kelly and Gilliland 2002; Takahashi 2011). Class I mutations confer a proliferative and/or survival advantage of hematopoietic stem and progenitor cells (HSPCs) and include activating mutations in NRAS or KRAS, loss of the Ras-GAP NF1, or upstream activation of RAS signaling through mutations affecting the FLT3 receptor tyrosine kinase. Class II mutations promote self-renewal and block the differentiation of HSPCs. Such mutations include translocations involving the MLL1 gene or the t(8:21) fusion involving AML1-ETO. Alone, class I mutations tend to trigger chronic myeloid leukemia-like diseases, whereas class II mutations lead to a myelodysplastic syndrome (MDS)-like state. Although an oversimplification, as a rule, class I and class II mutations cooperate to drive AML, leading to the aberrant proliferation and suppressed differentiation that are hallmarks of this disease.
Currently, most AML patients are treated with high-dose chemotherapy involving cytarabine (ara-C) and an anthracycline, a highly toxic drug combination to which >70% of patients develop resistance. However, the nature of certain genes mutated in AML suggests strategies for treating patients with molecularly targeted agents. For example, internal tandem duplication (ITD) mutations in Flt3 lead to constitutive activation of its tyrosine kinase activity, and drugs targeting these mutations are in clinical trials (Stirewalt and Radich 2003; Leung et al. 2013). Similarly, MEK inhibitors, which interfere with RAS effector mechanisms, show efficacy in certain preclinical models (Lauchle et al. 2009). Finally, inhibitors of DOT1L, a histone methyltransferase needed for the oncogenic activity of MLL fusion oncoproteins, are entering clinical trials (Bernt et al. 2011; Daigle et al. 2011). Such therapies require that leukemia remains dependent on the targeted oncogene or pathway, which is often difficult to predict a priori. Still, not all novel AML targets directly interfere with oncogenic drivers or their downstream effectors. For example, the BET family member Brd4, which is not mutated in AML, has been identified as a target in AML owing to its ability to sustain a Myc-dependent self-renewal program activated by certain AML drivers (Dawson et al. 2011; Zuber et al. 2011b). Regardless of their mode of action, genetically and pathologically accurate mouse models of AML have been important in informing target development (Bernt et al. 2011; Zuber et al. 2011b).
Recently, somatic mutations in the isocitrate dehydrogenase (IDH) genes IDH1 and IDH2 have been identified at high frequency in AML and other tumor types (Parsons et al. 2008; Mardis et al. 2009; Yan et al. 2009; Amary et al. 2011; The Cancer Genome Atlas Research Network 2013). These genes encode key metabolic enzymes that convert isocitrate to α-ketoglutarate (α-KG). IDH mutations mainly impact certain active site residues (e.g., IDH1R132, IDH2R140, or IDH2R172), resulting in loss of normal enzymatic function and the acquisition of a neomorphic activity that enables the mutant proteins to reduce α-KG to 2-hydroxyglutarate (2-HG) (Dang et al. 2009; Ward et al. 2010). The presumptive “oncometabolite” 2-HG can competitively inhibit multiple α-KG-dependent dioxygenases, including key epigenetic regulators such as histone demethylases and the DNA-demethylating TET proteins (Figueroa et al. 2010; Xu et al. 2011). Consequently, IDH mutants are associated with dramatic chromatin abnormalities, including globally altered histone and DNA methylation (Figueroa et al. 2010; Lu et al. 2012; Turcan et al. 2012). In the hematopoietic system and other cell types, these changes are associated with a differentiation block (Koivunen et al. 2012; Lu et al. 2012; Sasaki et al. 2012; Turcan et al. 2012).
The neomorphic action of IDH mutant proteins has created enthusiasm for targeting these enzymes with novel anti-cancer agents, and early studies using small molecules capable of inhibiting IDH1R132H and IDH2R140Q show some activity (Rohle et al. 2013; Wang et al. 2013). Still, there is a paucity of data documenting the oncogenic effects of IDH mutants on the development and maintenance of bona fide malignancies. On one hand, IDH2 mutants block the differentiation of cultured HSPCs (Figueroa et al. 2010), which accumulate in the hematopoietic compartment of mice expressing IDH1R132H (Sasaki et al. 2012). Also, enforced expression of this mutant can promote cytokine-independent growth and block the differentiation of an established erythroleukemic cell line in vitro (Losman et al. 2013). Still, in vivo models whereby IDH mutants drive a fully malignant disease have been lacking.
In this study, we describe a new mouse model in which IDH2 mutants cooperate with other lesions to drive an aggressive AML that accurately recapitulates features of the human disease. We further use this system to study the basis of IDH-mediated oncogenesis and as a preclinical model for testing novel therapies.
ResultsIDH2 mutants cooperate with Flt3-ITD or NrasG12D to promote leukemia
Considering evidence that IDH mutations can block the differentiation of HSPCs (Figueroa et al. 2010; Sasaki et al. 2012), we hypothesized that they might act as canonical class II mutations and thus could cooperate with class I mutations to promote AML. We chose mouse models incorporating two common class I mutations observed in human AML: FLT3-ITD (Nakao et al. 1996) and NrasG12D (Schubbert et al. 2007). Flt3-ITD knock-in mice develop a chronic myelomonocytic leukemia that never progresses to AML (Lee et al. 2007; Chu et al. 2012), whereas Mx-1-mediated activation of a latent “lox–stop–lox” NrasG12D allele (Haigis et al. 2008) in hematopoietic cells results in a myeloproliferative disorder (Li et al. 2011; Wang et al. 2011).
We applied a mosaic mouse modeling approach in which HSPCs are isolated from 5-fluorouracil (5-FU)-treated Flt3-ITD mice, transduced with retroviral vectors expressing IDH2 mutants or a vector control, and then assessed for tumorigenic potential following transplantation into sublethally irradiated syngeneic recipient mice (Schmitt et al. 2002). For experiments involving NrasG12D, mice were pretreated with polyinosinic:polycytidylic acid (pIpC) to trigger cre-mediated oncogene activation (Supplemental Fig. 1A). Immunoblotting of protein extracts obtained from sorted GFP+ HSPCs confirmed expression of IDH2 wild-type and mutant proteins (Supplemental Fig. 1B). Gas chromatography-mass spectrometry (GC-MS) analysis revealed that 2-HG levels were elevated in HSPCs expressing IDH2R140Q and IDH2R172K but not wild-type IDH2, confirming that the mutant alleles function as expected (Supplemental Fig. 1C).
Recipients of HSPCs expressing Flt3-ITD or NrasG12D together with either IDH2R140Q or IDH2R172K displayed significantly reduced survival compared with recipients of HSPCs expressing Flt3-ITD or NrasG12D transduced with empty vector or wild-type IDH2 (Fig. 1A,B). Recipients of Flt3-ITD;IDH2R140Q- and Flt3-ITD;IDH2R172K-expressing HSPCs died with a similarly short latency after transplantation (median leukemia-free survival = 97 d); similarly, both NrasG12D;IDH2R140Q- and NrasG12D;IDH2R172K-expressing HSPC recipients also showed accelerated disease onset, although in this instance, the effect of IDH2R140Q was less potent than IDH2R172K (median leukemia-free survival = 178 d for IDH2R140Q vs. 98 d for IDH2R172K; P < 0.0001) (Fig. 1B). Complete blood counts (CBCs) showed that all IDH2R140Q and IDH2R172K recipients displayed leukocytosis (Fig. 1C,D), anemia (Fig. 1E,F), and gross splenomegaly (Fig. 1G). Importantly, disease could be transferred to secondary recipients by transplanting bone marrow (BM) cells derived from moribund mice, indicating that neoplastic cells arising in the presence of IDH2R140Q or IDH2R172K were fully malignant (Fig. 1H).
IDH2 mutants cooperate with class I mutations to promote leukemia. (A) Kaplan-Meier survival curve of mice transplanted with Flt3-ITD HSPCs transduced with empty MSCV-IRES-GFP vector (pMIG), IDH2 wild type (WT), or mutants (IDH2R140Q and IDH2R172K). n = 9. (B) Kaplan-Meier survival curve of mice transplanted with NrasG12D HSPCs transduced with empty vector (pMIG), IDH2 wild type, or mutants (IDH2R140Q and IDH2R172K). n = 9. (C) White blood cell (WBC) counts of mice transplanted with Flt3-ITD HSPCs transduced with empty vector (pMIG), IDH2 wild type, or mutants (IDH2R140Q and IDH2R172K) at 12 wk after transplantation. n = 5. (D) WBC counts of mice transplanted with NrasG12D HSPCs transduced with empty vector (pMIG), IDH2 wild type, or mutants (IDH2R140Q and IDH2R172K) at 12 wk after transplantation. n = 5. (E) Red blood cell (RBC) counts of mice transplanted with Flt3-ITD HSPCs transduced with empty vector (pMIG), IDH2 wild type, or mutants (IDH2R140Q and IDH2R172K) at 12 wk after transplantation. n = 5. (F) RBC counts of mice transplanted with NrasG12D HSPCs transduced with empty vector (pMIG), IDH2 wild type, or mutants (IDH2R140Q and IDH2R172K) at 12 wk after transplantation. n = 5. (G) Representative pictures of the spleens of recipient mice transplanted with Flt3-ITD or NrasG12D HSPCs transduced with empty vector (pMIG), IDH2 wild type, or mutants (IDH2R140Q and IDH2R172K). (H) Kaplan-Meier survival curve of the secondary recipient mice transplanted with Flt3-ITD or NrasG12D HSPCs transduced with IDH2 wild type or mutants (IDH2R140Q and IDH2R172K). n = 5. (*) P < 0.05; (**) P < 0.01; (***) P < 0.001.
IDH2 mutations drive aggressive AML
To further characterize the hematopoietic malignancy induced by each IDH2 mutant, we subjected moribund animals to histopathological analyses and immunophenotyping. At the time of sacrifice, all recipients of HSPCs harboring Flt3-ITD or NrasG12D and an IDH2 mutant contained numerous circulating blasts in the peripheral blood (Fig. 2A). Leukemic cells replaced normal hematopoietic cells within the BM (Fig. 2B) and spleen and disseminated to normally nonhematologic organs such as the liver (Supplemental Fig. 2A,B). Flow cytometry revealed that all leukemic cells expressed mutant IDH2, indicated by GFP expression (Supplemental Fig. 3). Whether obtained from the spleen or BM, they expressed the myeloid marker Mac-1 and/or c-kit, a marker of more immature myeloid progenitors (MPs) (Fig. 2C,D,G).
IDH2 mutations result in AML. (A) Blood smear of recipient mice transplanted with Flt3-ITD or NrasG12D HSPCs transduced with IDH2 wild type (WT) or mutants (IDH2R140Q and IDH2R172K). (B) H&E staining of BM sections (400×) of recipient mice transplanted with IDH2 wild-type- or mutant-expressing cells. (C) Representative flow plots of BM cells from recipient mice transplanted with Flt3-ITD or NrasG12D HSPCs transduced with IDH2 wild type, or mutants (IDH2R140Q and IDH2R172K). (D) Representative flow plots of splenocytes stained with Mac-1 and c-kit of recipients transplanted with Flt3-ITD cells transduced with empty vector, IDH2 wild type, or mutants (IDH2R140Q and IDH2R172K). (E) Representative flow plots of peripheral blood stained with B220 and CD3 of recipients transplanted with Flt3-ITD cells transduced with empty vector, IDH2 wild type, or mutants (IDH2R140Q and IDH2R172K). (F) Representative flow plots of peripheral blood stained with Mac-1 and Gr-1 of recipients transplanted with Flt3-ITD cells transduced with empty vector, IDH2 wild type, or mutants (IDH2R140Q and IDH2R172K). (G) Representative flow plots of splenocytes stained with Mac-1 and CD19 of recipients transplanted with NrasG12D cells transduced with empty vector, IDH2 wild type, or mutants (IDH2R140Q and IDH2R172K). (H) Representative flow plots of peripheral blood stained with CD19 and Thy1 of recipients transplanted with NrasG12D cells transduced with empty vector, IDH2 wild type, or mutants (IDH2R140Q and IDH2R172K). (I) Representative flow plots of peripheral blood stained with Mac-1 and c-kit of recipients transplanted with NrasG12D cells transduced with empty vector, IDH2 wild type, or mutants (IDH2R140Q and IDH2R172K).
Peripheral blood from recipient mice transplanted with HSPCs expressing the control vector or one encoding wild-type IDH2 contained mostly CD3+ or Thy1+ T cells and B220+ or CD19+ B cells, which is consistent with the cellular composition of blood from nonleukemic mice (Fig. 2E,H). In contrast, the peripheral blood from mice bearing leukemia driven by IDH2R140Q or IDH2R172K was overrun with CD3− B220− Mac1+ or Thy1− CD19− Mac1+ cells (Fig. 2F,I). Taken together, these features indicate that IDH2 mutants in combination with Flt3-ITD or NrasG12D drive disease resembling human AML.
In vitro studies indicate that the high 2-HG levels produced by IDH2 mutants can impair the functions of TET proteins that convert 5-methylcytosine (5mC) to 5-hydroxymethylcytosine (5hmC), resulting in a decrease of 5hmC levels (Figueroa et al. 2010). Accordingly, murine AMLs triggered by either IDH2R140Q or IDH2R172K showed a marked increase in 2-HG production relative to nonleukemic BM, the immortalized myeloid cell line 32D, and AML induced by other oncogenes (Fig. 3A). Interestingly, although in vitro studies indicate that IDH2R172K produces more 2-HG than IDH2R140Q (Ward et al. 2013), the 2-HG levels present in Flt3-ITD;IDH2R140Q and Flt3-ITD;IDH2R172K AMLs were comparable, perhaps owing to the lower expression of IDH2R172K than that of IDH2R140Q (data not shown). Analysis of DNA methylation by dot blotting of normal and leukemia DNA showed no significant difference in the global levels of 5mC in AMLs expressing IDH2 mutants. However, consistent with their ability to inhibit TET proteins, IDH2 mutant leukemia displayed a massive reduction in 5hmC levels (Fig. 3B).
IDH2 mutant-induced AMLs display high 2-HG levels and altered DNA methylation and are chemoresistant. (A) 2-HG levels in BM, 32D, and MLL-AF9 or IDH2 mutant-induced AMLs. n = 3–5. (B) Representative dot blotting with antibodies against 5mC and 5hmC displaying DNA methylation of nonleukemic (BM and 32D) and leukemic cells. (C) Dose response of NrasG12D;MLL-AF9, NrasG12D;AML1-ETO, NrasG12D;IDH2R140Q, and NrasG12D;IDH2R172K leukemic cells to ara-C. Leukemic cells were treated with vehicle or 1–6000 nM ara-C for 3 d. n = 4. (D) Kaplan-Meier survival curve of mice transplanted with NrasG12D;MLL-AF9 or NrasG12D;IDH2R172K leukemic cells treated with ara-C or vehicle. n = 5.
IDH2R172 mutations are associated with poor prognosis in AML patients, while the clinical effects of IDH2R140 mutations are controversial (Marcucci et al. 2010; Paschka et al. 2010; Patel et al. 2012). As underlying differences in survival of patients with different AML genotypes can be linked to treatment response (Zuber et al. 2009; Patel et al. 2012), we asked whether IDH2 mutant AMLs were intrinsically sensitive or resistant to ara-C chemotherapy. Consistent with previous reports (Zuber et al. 2009), NrasG12D;AML1-ETO leukemic cells were much more sensitive to ara-C than NrasG12D;MLL-AF9 leukemic cells in a 3-d treatment assay (IC50 = 39 vs. 63 nM; P = 0.03). In contrast, both NrasG12D;IDH2R140Q and NrasG12D;IDH2R172K AML cells were highly resistant to ara-C in vitro (IC50 = 93 and 345 nM, respectively; P = 0.0001 for NrasG12D;IDH2R140Q vs. NrasG12D;AML1-ETO; P < 0.0001 for NrasG12D;IDH2R172K vs. NrasG12D;AML1-ETO). NrasG12D;IDH2R172K AML cells were even more resistant to ara-C than NrasG12D;IDH2R140Q AML cells (P = 0.0002), an intriguing observation in light of the poorer prognosis of patients with IDH2R172 mutations (Fig. 3C). This resistance was more striking in vivo, where some ara-C-treated mice harboring IDH2 mutant AMLs died even earlier than vehicle-treated controls (Fig. 3D). Therefore, IDH2 mutants can cooperate with class I mutations to promote AML with pathological, biological, and molecular features of the human disease.
IDH2 mutants are sufficient to suppress hematopoietic differentiation and alter DNA methylation in vivo
To determine whether the tumorigenic phenotype driven by mutant IDH2 might derive from their ability to impair differentiation in vivo, we examined the impact of IDH2 mutants on hematopoiesis using a competitive reconstitution assay. Here, CD45.1 HSPCs were transduced with a control vector expressing GFP alone or together with GFP vectors encoding either wild-type IDH2, IDH2R140Q, or IDH2R172K; the resulting populations were then mixed with CD45.2 BM cells as a competitive reference and transplanted into lethally irradiated CD45.1 mice (Fig. 4A). Peripheral blood, spleen, and BM from recipient animals were analyzed for cellular composition, 2-HG production, and changes in DNA methylation.
IDH2 mutations are sufficient to block differentiation of HSPCs and alter DNA methylation. (A) Schematic diagram of competitive transplant assay. Vector refers to transduction of cells with empty MSCV-IRES-GFP vector or one coexpressing wild-type IDH2 or an IDH2 mutant. CD45.1 c-kit+ HSPCs (105) transduced with vectors were mixed with 2 × 105 CD45.2 BM and transplanted into lethally irradiated CD45.1 recipient mice. (B) Percentage of lin−Sca-1+c-kit+ (LSK) cells in the whole BM of recipient mice 12 wk after transplantation. n = 3. (C) Percentage of lin−Sca-1−c-kit+ (MP) cells in the whole BM of recipient mice 12 wk after transplantation. n = 3. (D) Percentage of c-kit+, Mac-1+ or Gr-1+, B220+ or CD3+, and Ter119+ cells in the whole BM of recipient mice 12 wk after transplantation. n = 3. (E) Percentage of GFP+ cells in the whole BM of recipient mice 12 wk after transplantation. n = 3. (F) WBC counts of recipient mice with IDH2 wild-type- or mutant-expressing cells. n = 4–5. (G) BM cellularities of recipient mice with IDH2 wild-type- or mutant-expressing cells at 12 wk after transplantation. n = 3. (H) BrdU incorporation rate of LSK cells after 1 d of labeling. n = 3. (I) BrdU incorporation rate of MP cells after 1 d of labeling. n = 3. Of note, in B–G, vector, IDH2, or IDH2 mutants refers to mice transplanted with HSPCs transduced with the indicated vector but does not necessarily indicate that all of the cells analyzed expressed the constructs. Specifically, while BM from mice transplanted with IDH2 mutant cells almost completely consisted of GFP+ IDH mutant-expressing cells (shown in E), empty vector and wild-type IDH produced a competitive disadvantage, and BM consisted entirely of GFP− normal competitors at this time point.
As expected, white blood cell (WBC) counts in mice receiving HSPCs transduced with empty or wild-type IDH2-expressing vector increased with time after transplantation (Supplemental Fig. 4A). Under these conditions, transplanted cells were able to reconstitute a normal hematopoietic compartment, which displayed the expected distributions of Lin− Sca+ c-Kit+ (LSK) HSPCs, Lin− Sca− c-Kit+ (MP), and other hematopoietic cells (Fig. 4B,D). If anything, empty vector and wild-type IDH2 appeared to confer a competitive disadvantage during hematopoietic reconstitution, as virtually no GFP+ cells remained after 12 wk, with most cells being derived from untransduced or competitor cells (Fig. 4E).
Mice reconstituted with HSPCs expressing IDH2R140Q and IDH2R172K showed a markedly different phenotype. Although at 4 wk post-transduction these mice retained WBC counts comparable with controls, by 12 wk, they showed significantly reduced WBC and had developed an anemia (Fig. 4F; Supplemental Fig. 4A) associated with significantly reduced BM cellularity (Fig. 4G). Nevertheless, IDH2 mutant cells (GFP+) predominated over controls (GFP−), suggesting that these cells outcompeted and excluded the nontransduced and cotransplanted CD45.2 cells (Fig. 4E). These changes were associated with an increased fraction of LSK+ and MPs in the BM (Fig. 4B–D). Although LSK cells expressing both IDH2 mutants showed slightly increased BrdU incorporation when compared with controls, the opposite occurred in the MP compartment, where IDH2 mutant MP cells displayed reduced BrdU incorporation (Fig. 4H,I). The reduced proliferation and increased percentage of MPs in recipients of IDH mutants were associated with a skewing of BM differentiation toward the myeloid lineage (Fig. 4D) comparable with phenotypes observed in IDH1R132H knock-in mice (Sasaki et al. 2012). Thus, IDH2 mutant proteins block the differentiation of MP cells in vivo, leading to the accumulation of more immature cell populations.
As expected, BM cells derived from mice reconstituted with HSPCs expressing IDH2R140Q or IDH2R172K contained high levels of 2-HG (Fig. 5A). In contrast to what has been observed in other systems (Xu et al. 2011; Lu et al. 2012), no obvious changes in the global abundance of methylated histones (H3K4me3, H3K9me3, and H3K27me3) were observed in BM cells expressing IDH2 mutants (Supplemental Fig. 5) relative to vetor-only or wild-type IDH4-expressing cells, raising the possibility that IDH mutants can have tissue-specific effects on chromatin biology and cell fate. Nevertheless, these BM cells showed significantly decreased 5hmC levels without overt changes in global 5mC levels (Fig. 5B), an effect that was also observed in c-kit-positive cells subjected to flow cytometry (Fig. 5C,D; Supplemental Fig. 5B). Overexpression of wild-type IDH2 did not have significant effects on 5hmC levels (Fig. 5B,D; Supplemental Fig. 5B). Hence, IDH2 mutants are sufficient to drive 2-HG production and alter DNA methylation during leukemogenesis.
IDH2 mutants lead to abnormal DNA methylation in hematopoietic cells. (A) 2-HG levels in the whole BM cells of recipient mice. n = 3. (B) Representative dot blots showing 5mC and 5hmC levels of whole BM genomic DNA. (C) Relative mean fluorescence intensity (MFI) of 5hmC in GFP+ BM cells. n = 3. (D) Relative MFI of 5mC in GFP+ c-kit+ BM cells. n = 3.
IDH2 mutants are required for leukemia maintenance
The availability of IDH mutant-driven leukemia models enabled us to assess whether ongoing expression of IDH2 mutants and their product, 2-HG, were required for tumor maintenance. AGI-6780 is a commercially available small-molecule inhibitor that targets IDH2R140Q but not IDH2R172K and has been shown to have anti-leukemic activity in human AML (Wang et al. 2013). To assess its activity against IDH2 mutant-driven murine AML, control NrasG12D;MLL-AF9 leukemia or NrasG12D leukemias coexpressing IDH2R140Q or IDH2R172K were treated with AGI-6780 in vitro, and 2-HG production, cellular proliferation, and differentiation were monitored. Within 2 d, AGI-6780 inhibited 2-HG production in AML cells expressing IDH2R140Q but not control or IDH2R172K-expressing leukemia cells (Fig. 6A). Remarkably, this effect was associated with a 2-wk proliferative burst that was unique to IDH2R140Q-expressing AML (Fig. 6B), after which these cells ceased to proliferate and underwent differentiation (Fig. 6C,D). This proliferative burst and eventual differentiation have also been seen following AGI-6780 treatment of a human AML explant harboring the IDH2R140Q allele (Wang et al. 2013).
Inhibition of IDH2 mutants leads to loss of 2-HG and delayed differentiation. (A) 2-HG levels of leukemic cells treated with 5 μM AGI-6780 for 2 d. n = 3. (B) Growth of leukemic cells treated with 5 μM AGI-6780 or vehicle. n = 4. (C) Representative H&E staining of cytospin of NrasG12D;IDH2R140Q and NrasG12D;IDH2R172K leukemic cells treated with vehicle or 5 μM AGI-6780 at day 22. (D) Expression levels of Mac-1 by NrasG12D;IDH2R140Q and NrasG12D;IDH2R172K leukemic cells treated with vehicle or 5 μM AGI-6780 at day 22. (E) Transcript levels of hIDH2 in NrasG12D;IDH2R172K cells expressing shRen or shIDH2, normalized to actin mRNA. n = 4. (F) Western blot demonstrating hIDH2 knockdown in NrasG12D;IDH2R172K leukemic cells by shRNAs against mutant IDH2. (G) Relative growth of NrasG12D;MLL-AF9 and NrasG12D;IDH2R172K leukemic cells with shRen or shIDH2. n = 3. (H) Expression levels of Mac-1 by NrasG12D;IDH2R140Q and NrasG12D;IDH2R172K leukemic cells treated with shRen or shIDH2.
Small-molecule inhibitors of the IDH2R172K protein are not currently available. To test its role in leukemia maintenance, we designed shRNA specifically targeting human IDH2, thus inhibiting only the exogenously transduced IDH2 mutant used in this model (Fig. 6E,F). Control (shRen, a potent shRNA targeting Renilla luciferase) and IDH2-specific shRNAs were cloned downstream from a tetracycline-responsive element and dsRed and retrovirally transduced into IDH2 mutant leukemia cells also expressing a rtTA transgene, thereby making the shRNAs doxycycline-inducible (Zuber et al. 2011a). Cells were treated with doxycycline and analyzed for the prevalence of the dsRed marker (and thus the competitive fitness of shRNA-expressing cells) over time.
As expected, IDH2 shRNAs had no effect on MLL-AF9-driven AML. However, similar to the effects of AGI-6780, inhibition of IDH2R172K by RNAi led to an initial enhanced proliferation and subsequent depletion of shRNA-expressing cells from the population (Fig. 6H), an effect associated with myeloid differentiation (Fig. 6G). Thus, ongoing expression of IDH2 mutants is required to block differentiation, although long periods of IDH2 inhibition are required to reverse these effects. In principle, the delay in response may reflect the requirement for cell division to renormalize the epigenetic changes produced by mutant IDH2.
IDH mutant AMLs rapidly differentiate in response to Brd4 inhibition
The BET family protein BRD4 has been identified as a therapeutic target in AML produced by MLL fusion oncoproteins owing to its ability to sustain an aberrant self-renewal circuit controlled by Myc (Dawson et al. 2011; Zuber et al. 2011b). To test whether the aberrant self-renewal program produced by IDH2 mutants is also sensitive to Brd4 inhibition, we examined the impact of two independently validated Brd4 shRNAs (Zuber et al. 2011b) or the small-molecule Brd4 inhibitor JQ1 on leukemic cell proliferation, viability, differentiation, and Myc levels in IDH2 mutant AML. Cultured cells were infected with the doxycycline-inducible construct used above expressing either shRNAs targeting Brd4 or Renilla as a control. Brd4 suppression acutely reduced the competitive fitness of cultured NrasG12D;IDH2R172K cells compared with controls, suggesting that Brd4 is essential to maintain IDH2 mutant leukemia (Fig. 7A).
IDH mutant leukemias rapidly differentiate in response to Brd4 inhibition. (A) Relative growth of NrasG12D;IDH2R172K leukemic cells with Brd4 knockdown. n = 3. (B) Dose response of NrasG12D;MLL-AF9, NrasG12D;AML1-ETO, NrasG12D;IDH2R140Q, and NrasG12D;IDH2R172K to JQ1. Leukemic cells were treated with vehicle or 1–6000 nM JQ1 for 3 d. n = 4. (C) Representative flow plots showing Mac-1 expression levels in NrasG12D;IDH2R172K leukemic cells with shRen or shBrd4 3 d after doxycycline induction. (D) Representative flow plots showing Mac-1 expression levels in NrasG12D;IDH2R140Q and NrasG12D;IDH2R172K leukemic cells treated with vehicle or 50 nM JQ1 for 2 d. (E) Representative cytospin staining of NrasG12D;IDH2R140Q and NrasG12D;IDH2R172K leukemic cells treated with vehicle or 50 nM JQ1 for 2 d. (F) 2-HG levels of NrasG12D;MLL-AF9, NrasG12D;IDH2R140Q, and NrasG12D;IDH2R172K leukemic cells treated with vehicle or 50 nM JQ1 for 2 d. n = 3. (G) Western blotting of Myc levels in NrasG12D;MLL-AF9, NrasG12D;IDH2R140Q, and NrasG12D;IDH2R172K leukemic cells treated with vehicle or 50 nM JQ1 for 2 d. The numbers indicate normalized Myc levels by densitometry. (H) WBC counts of NrasG12D;IDH2R172K recipient mice treated with vehicle or JQ1 at day 20 after transplant. Inserts show the representative blood smear of vehicle- or JQ1-treated mice. Vehicle, n = 4; JQ1, n = 5. (I) RBC counts of NrasG12D;IDH2R172K recipient mice treated with vehicle or JQ1 20 d after transplant. Vehicle, n = 4; JQ1, n = 5. (J) Kaplan-Meier survival curve of NrasG12D;IDH2R172K recipient mice treated with vehicle or JQ1. The mice were treated with vehicle or 50 mg/kg per day JQ1 by gavage from day 5 to day 18 after transplant. n = 5.
In parallel, we tested response to the small-molecule JQ1 in vitro. As expected, NrasG12D;MLL-AF9 AML cells were highly sensitive to JQ1 (IC50, 58 nM) (Zuber et al. 2011b), while NrasG12D;AML1-ETO AML cells were more resistant (IC50, 1159 nM). IDH mutant AMLs were even more sensitive to JQ1 than NrasG12D;MLL-AF9-expressing cells (IC50, 34 nM and 19 nM for NrasG12D;IDH2R140Q and NrasG12D IDH2R172K, respectively; P = 0.01 for NrasG12D;IDH2R140Q vs. NrasG12D;MLL-AF9; P < 0.0001 for NrasG12D;IDH2R172K vs. NrasG12D;MLL-AF9) (Fig. 7B).
The hypersensitivity of IDH mutant AMLs to Brd4 inhibition was associated with rapid differentiation after either shRNA Brd4 knockdown or JQ1 treatment (Fig. 7C–E). Nevertheless, JQ1 treatment did not affect 2-HG levels in either IDH2R140Q or IDH2R172K cells (Fig. 7F), perhaps indicating that Brd4 inhibition acts downstream from mutant IDH2 proteins and mediates its effects without directly affecting its neomorphic activity. However, as has been reported in MLL-AF9-driven AML, immunoblotting of IDH2 mutant AML lysates treated for 48 h with JQ1 revealed reduced levels of Myc (Fig. 7G). Therefore, IDH2 mutant AMLs are addicted to a Brd4-driven, Myc-dependent, self-renewal program that can be inhibited by JQ1 treatment.
Although the pharmacologic properties of JQ1 are not ideal for in vivo studies (Matzuk et al. 2012), we tested its activity against IDH2 mutant AML in vivo. Mice were transplanted with IDH2R172K AML and, upon engraftment, treated (day 5) with vehicle or 50 mg/kg per day JQ1 by gavage for 2 wk (Filippakopoulos et al. 2010; Zuber et al. 2011b). Leukemic cells were apparent in the peripheral blood in vehicle-treated mice at day 19 but completely absent in JQ1-treated animals (Fig. 7H). JQ1-treated mice also displayed significantly improved erythropoiesis, leading to reduced anemia (Fig. 7I). While all vehicle-treated mice died within 3 d post-treatment, JQ1 treatment extended median survival to 18 d (Fig. 7J). Taken together, IDH mutant-driven AML is susceptible to Brd4 inhibition both in vitro and in vivo.
DiscussionA mouse AML model driven by IDH2 mutants suggests therapeutic approaches for AML treatment
In vivo cancer models are essential for understanding genetic alterations that contribute to oncogenesis and can provide preclinical systems for testing novel therapies. Here we demonstrate that the IDH2R140Q and IDH2R172K mutants observed in human cancers can be potent oncogenes in mice, acting as class II driver mutations that cooperate with the class I mutations Flt3-ITD and NrasG12D to promote aggressive AML. Murine AMLs expressing IDH2 mutants display the histopathological and molecular features that are characteristic of the human disease and show a marked chemoresistance phenotype that may underlie the association between IDH2R172 mutations and poor patient survival.
IDH2 mutant proteins promote AML by blocking the differentiation of HSPCs; in the absence of a cooperating event, this produces an MDS-like state, a disorder in which IDH2 mutations are frequent in humans (Patnaik et al. 2012). While these effects correlate with the ability of IDH2 mutants to produce 2-HG and alter DNA methylation, our experiments do not prove a causal role for these downstream changes and disease etiology or maintenance. Still, in a parallel study (Lu et al. 2013), IDH2R172K acts similarly to drive the malignant conversion of mesenchymal progenitor cells into sarcoma in vivo, which also correlates with its ability to produce 2-HG, block differentiation, and alter DNA methylation. Irrespective of the precise mechanism, these data imply a broad and potent action of IDH mutant oncogenes across diverse disease states.
Owing to their similarity to the human disease, the AML models described herein may be useful for evaluating therapies to target IDH2 mutant leukemia. Using genetic and pharmacological tools to manipulate IDH2 activity, we show that IDH2 mutants are required for sustained 2-HG production and leukemia maintenance. Suppression of IDH2 mutant levels and its neomorphic activity triggered myeloid differentiation, albeit requiring prolonged IDH2 inhibition. This delayed response may reflect the requirement for multiple rounds of proliferation to restore normal epigenetic states but might also present a confounding factor when treating patients with acute leukemia. In contrast, shRNA or small-molecule-mediated inhibition of Brd4 causes rapid terminal differentiation and elimination of IDH2 mutant leukemia cells even in the presence of sustained 2-HG levels. This effect is associated with loss of a Myc-dependent self-renewal circuit and triggers substantial anti-leukemic effects in vivo. Notably, IDH2R172K mutant AML, which is not inhibited by AGI-6780 (Wang et al. 2013) and is associated with poor prognosis (Marcucci et al. 2010; Paschka et al. 2010; Patel et al. 2012), displays the highest sensitivity to Brd4 inhibition.
While high-dose chemotherapy can be highly effective against AML, these agents produce substantial toxicity, and patients frequently relapse with resistant disease. One notable exception involves the treatment of acute promyelocytic leukemia (APL), which can be cured without substantial toxicity by a combination of all trans retinoic acid (ATRA) together with arsenic or other molecularly targeted agents (Shen et al. 1997). In contrast to conventional chemotherapy, which triggers leukemia cell death, ATRA acts more directly by reversing the differentiation block produced by the driving PML-RARα oncoprotein (Zhang et al. 2000). As shown here, similar effects can be achieved in murine IDH2 mutant AMLs following IDH2 or Brd4 inhibition, raising hope that improved agents will produce sustained anti-leukemic responses in patients. In any event, these observations validate IDH2 as a therapeutic target in AML and point to an alternative approach to IDH2 inhibition for treatment of IDH2 mutant cancers.
Materials and methodsMice
All mouse experiments were conducted in accordance with institutional guidelines at Memorial Sloan-Kettering Cancer Center (MSKCC). Recipient C57Bl/6 mice (National Cancer Institute) were irradiated at a dose of 4.5 Gy (Cs137) for sublethal irradiation or two doses of 5.5 Gy for lethal irradiation before transplantation. Mice were transplanted with 1 × 106 or the indicated number of cells by tail vein injection. Mice were monitored for leukemogenesis by complete blood cell count (Hemavet, Drew Scientific), spleen palpation, and blood smear.
Plasmid construction
Human IDH2 wild type and mutants R140Q and R172K were cloned into a pMIG vector (Lu et al. 2012). shRNAs against Renilla (TGCTGTTGACAGTGAGCGCAGGAATTATAATGCTTATCTATAGTGAAGCCACAGATGTATAGATAAGCATTATAATTCCTATGCCTACTGCCTCGGA), IDH2 (shIDH-1, TGCTGTTGACAGTGAGCGCCAAGCTGAAGAAGATGTGGAATAGTGAAGCCACAGATGTATTCCACATCTTCTTCAGCTTGATGCCTACTGCCTCGGA; shIDH2-2, TGCTGTTGACAGTGAGCGCTAAGACCGACTTCGACAAGAATAGTGAAGCCACAGATGTATTCTTGTCGAAGTCGGTCTTATTGCCTACTGCCTCGGA), and Brd4 (shBrd4-1, TGCTGTTGACAGTGAGCGCCCCATGGATATGGGAACAATATAGTGAAGCCACAGATGTATATTGTTCCCATATCCATGGGTTGCCTACTGCCTCGGA; shBrd4-2:,TGCTGTTGACAGTGAGCGACACAATCAAGTCTAAACTAGATAGTGAAGCCACAGATGTATCTAGTTTAGACTTGATTGTGCTGCCTACTGCCTCGGA) (Zuber et al. 2011b) were cloned into a TRIN (Tre-dsRed-mir30-PGK-Venus-IRES-Neo) vector (Zuber et al. 2011a).
Cell culture and transduction
BM cells were isolated from the indicated young adult donor mice, and c-kit-positive HSPCs were separated by autoMACS (Miltenyi Biotec, Inc.). HSPCs were cultured with stem cell medium as described (Schmitt et al. 2002; Zuber et al. 2009). IDH2 mutant AMLs were cultured with IMDM medium supplemented with 20% FBS and 0.2 ng/mL IL-3. NrasG12D;MLL-AF9 AML was cultured with RPMI-1640 plus 10% FBS. Retroviruses were made by calcium phosphate-mediated transfection of Plat-E (Morita et al. 2000) packaging cells. HSPCs and AML cells were transfected by spinoculation.
Measurement of 2-HG
Frozen cell pellets were extracted with 1 mL of ice-cold 80% methanol containing 20 μM deuterated 2-HG as an internal standard (D-hydroxyglutaric-2,3,3,4,4-d5). Methanol extracts were incubated at −80°C for 30 min and centrifuged at 21,000g for 20 min at 4°C to remove precipitated protein. Nine-hundred microliters was evaporated to dryness under a nitrogen gas stream. Dried organic acids were derivatized by the sequential addition of 50 μL of 40 mg/mL methoxyamine hydrochloride in pyridine with incubation for 90 min at 30°C followed by 80 μL of MSTFA + 1% TMCS (Thermo Scientific) and 70 μL of ethyl acetate with incubation for 30 min at 37°C using an automated sample preparation platform (Gerstel). One microliter of the trimethylsilyl-derivatized organic acids was analyzed by GC-MS using an Agilent 7890A gas chromatograph with an HP-5MS capillary column connected to an Agilent 5975 C mass selective detector operating in splitless mode with electron impact ionization. Relative quantitation of 2-HG was determined from extracted ion chromatograms (EICs) for 2-HG (m/z: 349 or 247) normalized to the EICs of intracellular citrate (m/z: 465).
Histology and pathology assay
Bone, spleen, and liver were fixed in 10% formalin. Embedding, sectioning, and H&E staining were performed by IDEXX RADIL. Blood smears were stained with Hema-Quik II stain solution (Fisher Scientific). Cytospin was performed on Shandon Cytospin 4 and then stained with JorVet DipQuick stain.
Flow cytometry
Flow cytometry was performed on LSR II or Fortessa machines (BD Bioscience). All of the cell surface marker antibodies were from eBioscience. 5-hmC antibodies were from Active Motif (no. 39769). Cell surface marker staining was performed with Hank's balanced salt solution plus 2% FBS. For 5-hmC staining, cells were fixed and permeabilized with BD fix/perm buffer and then treated with 0.3 mg/mL DNase I for 30 min at 37°C. For the BrdU incorporation assay, mice were intraperitoneally (i.p.) injected with 100 mg/kg BrdU and then fed with 1 mg/mL BrdU in water for 24 h (Chen et al. 2008). BrdU staining was performed with a BD BrdU kit (BD Bioscience).
Dot blotting
Dot blotting was performed as described (Ko et al. 2010). Briefly, purified genomic DNA was quantified on NanoDrop and denatured by 0.1 M NaOH. Serial-diluted DNA was spotted on a nitrocellulose membrane. 5mC and 5hmC antibodies (Active Motif) were diluted at 1:1000 and 1:5000, respectively.
Western blotting
For IDH2 and Myc detection, cells were lysed in RIPA buffer, and lysates were separated with SDS-PAGE gel electrophoresis. For histone acid extraction, cells were lysed in hypotonic lysis buffer. The primary antibodies were anti-IDH2 (Abcam, ab55271), anti-c-Myc (Santa Cruz Biotechnology, sc-764), anti-actin (Sigma, T6074), anti-tubulin (Sigma, T9026), anti-H3K4me3 (Active Motif, 39916), anti-H3K9me3 (Active Motif, 39765), anti-H3K27me3 (Millipore, 07-449), and anti-H3 (Cell Signaling Tech, 4499).
Quantitative real-time PCR
mRNA was extracted from FACS-sorted cells with Trizol (Invitrogen), and then cDNA was made with SuperScript III (Invitrogen) according to the manufacturer's manual. Quantitative PCR was performed on a 7900HT Fast Real-Time system (Applied Biosystems). The sequences of PCR primers were mouse Actin forward (TGGTGATAAGTGGCCTTGGAGTGT) and reverse (ATGCAAGGAGTGCAAGAACACAGC) and hIDH2 forward (AGACCGACTTCGACAAGAATAAG) and reverse (GACTGCACATCTCCGTCATAG).
Two-color shRNA competitive proliferation assay
Cells were transduced with shRNAs and then selected with G418 (US Biological). Infected cells were mixed with uninfected cells at a 1:1 ratio for the IDH2 knockdown experiment and 4:1 for the Brd4 knockdown experiment. Next, the shRNAs were induced by 1 mg/mL doxycycline (Sigma). The percentage of shRNA-expressing cells (Venus+dsRed+) was measured by a Guava easyCyte flow cytometer (Millipore) at the indicated time points.
Drug treatment
In vitro, cells were treated with vehicle, ara-C (Bedford Laboratories), AGI-6780 (Xcess Biosciences), or JQ1 for 3 d or the indicated times. Viable cells were counted by a Guava easyCyte flow cytometer. For in vivo treatment, JQ1 was suspended in vehicle (0.5% [w/v] hydroxy-propyl-methylcellulose [Sigma], 0.2% [v/v] Tween 80 [Sigma]) at a concentration of 10 mg/mL. The suspension was sonicated before use. Mice with AMLs were treated with vehicle, 100 mg/kg per day ara-C by i.p. injection for 5 d, or 50 mg/kg per day JQ1 by gavage for 2 wk.
Acknowledgments
We thank C.C. Sherr, Dr. Z. Zhao, Dr. C. Miething, Dr. C. Chen, and other members of the Lowe laboratory for suggestions and/or technical help; S. Kogan for histopathological analysis; C. Sherr, L. Dow, and S. Mayack for editorial assistance; and T. Jacks for the NrasG12D mice. C.C. was supported by a career development fellowship from the Leukemia and Lymphoma Society (LLS). This work was supported by a Specialized Center of Research (SCOR) grant from the LLS (to S.W.L.) and a U01 CTDD award from the National Cancer Institute (to S.W.L.). S.W.L. is the Goeffrey Beene Chair of Cancer Biology at MSKCC and an Investigator in the Howard Hughes Medical Institute.
Supplemental material is available for this article.
Article is online at http://www.genesdev.org/cgi/doi/10.1101/gad.226613.113.
ReferencesAmaryMF, BacsiK, MaggianiF, DamatoS, HalaiD, BerishaF, PollockR, O'DonnellP, GrigoriadisA, DissT, 2011IDH1 and IDH2 mutations are frequent events in central chondrosarcoma and central and periosteal chondromas but not in other mesenchymal tumours.
224: 334–34321598255BerntKM, ZhuN, SinhaAU, VempatiS, FaberJ, KrivtsovAV, FengZ, PuntN, DaigleA, BullingerL, 2011MLL-rearranged leukemia is dependent on aberrant H3K79 methylation by DOT1L.
20: 66–7821741597The Cancer Genome Atlas Research Network2013Genomic and epigenomic landscapes of adult de novo acute myeloid leukemia.
368: 2059–207423634996ChenC, LiuY, LiuR, IkenoueT, GuanKL, ZhengP2008TSC-mTOR maintains quiescence and function of hematopoietic stem cells by repressing mitochondrial biogenesis and reactive oxygen species.
205: 2397–240818809716ChuSH, HeiserD, LiL, KaplanI, CollectorM, HusoD, SharkisSJ, CivinC, SmallD2012FLT3-ITD knockin impairs hematopoietic stem cell quiescence/homeostasis, leading to myeloproliferative neoplasm.
11: 346–35822958930DaigleSR, OlhavaEJ, TherkelsenCA, MajerCR, SneeringerCJ, SongJ, JohnstonLD, ScottMP, SmithJJ, XiaoY, 2011Selective killing of mixed lineage leukemia cells by a potent small-molecule DOT1L inhibitor.
20: 53–6521741596DangL, WhiteDW, GrossS, BennettBD, BittingerMA, DriggersEM, FantinVR, JangHG, JinS, KeenanMC, 2009Cancer-associated IDH1 mutations produce 2-hydroxyglutarate.
462: 739–74419935646DawsonMA, PrinjhaRK, DittmannA, GiotopoulosG, BantscheffM, ChanWI, RobsonSC, ChungCW, HopfC, SavitskiMM, 2011Inhibition of BET recruitment to chromatin as an effective treatment for MLL-fusion leukaemia.
478: 529–53321964340FigueroaME, Abdel-WahabO, LuC, WardPS, PatelJ, ShihA, LiY, BhagwatN, VasanthakumarA, FernandezHF, 2010Leukemic IDH1 and IDH2 mutations result in a hypermethylation phenotype, disrupt TET2 function, and impair hematopoietic differentiation.
18: 553–56721130701FilippakopoulosP, QiJ, PicaudS, ShenY, SmithWB, FedorovO, MorseEM, KeatesT, HickmanTT, FelletarI, 2010Selective inhibition of BET bromodomains.
468: 1067–107320871596HaigisKM, KendallKR, WangY, CheungA, HaigisMC, GlickmanJN, Niwa-KawakitaM, Sweet-CorderoA, Sebolt-LeopoldJ, ShannonKM, 2008Differential effects of oncogenic K-Ras and N-Ras on proliferation, differentiation and tumor progression in the colon.
40: 600–60818372904KellyLM, GillilandDG2002Genetics of myeloid leukemias.
3: 179–19812194988KoM, HuangY, JankowskaAM, PapeUJ, TahilianiM, BandukwalaHS, AnJ, LampertiED, KohKP, GanetzkyR, 2010Impaired hydroxylation of 5-methylcytosine in myeloid cancers with mutant TET2.
468: 839–84321057493KoivunenP, LeeS, DuncanCG, LopezG, LuG, RamkissoonS, LosmanJA, JoensuuP, BergmannU, GrossS, 2012Transformation by the (R)-enantiomer of 2-hydroxyglutarate linked to EGLN activation.
483: 484–48822343896LauchleJO, KimD, LeDT, AkagiK, CroneM, KrismanK, WarnerK, BonifasJM, LiQ, CoakleyKM, 2009Response and resistance to MEK inhibition in leukaemias initiated by hyperactive Ras.
461: 411–41419727076LeeBH, TothovaZ, LevineRL, AndersonK, Buza-VidasN, CullenDE, McDowellEP, AdelspergerJ, FrohlingS, HuntlyBJ, 2007FLT3 mutations confer enhanced proliferation and survival properties to multipotent progenitors in a murine model of chronic myelomonocytic leukemia.
12: 367–38017936561LeungAY, ManCH, KwongYL2013FLT3 inhibition: A moving and evolving target in acute myeloid leukaemia.
27: 260–26822797419LiQ, HaigisKM, McDanielA, Harding-TheobaldE, KoganSC, AkagiK, WongJC, BraunBS, WolffL, JacksT, 2011Hematopoiesis and leukemogenesis in mice expressing oncogenic NrasG12D from the endogenous locus.
117: 2022–203221163920LosmanJA, LooperRE, KoivunenP, LeeS, SchneiderRK, McMahonC, CowleyGS, RootDE, EbertBL, KaelinWGJr2013(R)-2-hydroxyglutarate is sufficient to promote leukemogenesis and its effects are reversible.
339: 1621–162523393090LuC, WardPS, KapoorGS, RohleD, TurcanS, Abdel-WahabO, EdwardsCR, KhaninR, FigueroaME, MelnickA, 2012IDH mutation impairs histone demethylation and results in a block to cell differentiation.
483: 474–47822343901LuC, VennetiS, AkalinA, FangF, WardPS, DeMatteoRG, IntlekoferAM, ChenC, YeJ, HameedM, 2013Induction of sarcomas by mutant IDH2. Genes Dev (this issue). doi: 10.1101/gad.226753.113.MarcucciG, MaharryK, WuYZ, RadmacherMD, MrozekK, MargesonD, HollandKB, WhitmanSP, BeckerH, SchwindS, 2010IDH1 and IDH2 gene mutations identify novel molecular subsets within de novo cytogenetically normal acute myeloid leukemia: A Cancer and Leukemia Group B study.
28: 2348–235520368543MardisER, DingL, DoolingDJ, LarsonDE, McLellanMD, ChenK, KoboldtDC, FultonRS, DelehauntyKD, McGrathSD, 2009Recurring mutations found by sequencing an acute myeloid leukemia genome.
361: 1058–106619657110MatzukMM, McKeownMR, FilippakopoulosP, LiQ, MaL, AgnoJE, LemieuxME, PicaudS, YuRN, QiJ, 2012Small-molecule inhibition of BRDT for male contraception.
150: 673–68422901802MoritaS, KojimaT, KitamuraT2000Plat-E: An efficient and stable system for transient packaging of retroviruses.
7: 1063–106610871756NakaoM, YokotaS, IwaiT, KanekoH, HoriikeS, KashimaK, SonodaY, FujimotoT, MisawaS1996Internal tandem duplication of the flt3 gene found in acute myeloid leukemia.
10: 1911–19188946930ParsonsDW, JonesS, ZhangX, LinJC, LearyRJ, AngenendtP, MankooP, CarterH, SiuIM, GalliaGL, 2008An integrated genomic analysis of human glioblastoma multiforme.
321: 1807–181218772396PaschkaP, SchlenkRF, GaidzikVI, HabdankM, KronkeJ, BullingerL, SpathD, KayserS, ZucknickM, GotzeK, 2010IDH1 and IDH2 mutations are frequent genetic alterations in acute myeloid leukemia and confer adverse prognosis in cytogenetically normal acute myeloid leukemia with NPM1 mutation without FLT3 internal tandem duplication.
28: 3636–364320567020PatelJP, GonenM, FigueroaME, FernandezH, SunZ, RacevskisJ, Van VlierbergheP, DolgalevI, ThomasS, AminovaO, 2012Prognostic relevance of integrated genetic profiling in acute myeloid leukemia.
366: 1079–108922417203PatnaikMM, HansonCA, HodnefieldJM, LashoTL, FinkeCM, KnudsonRA, KetterlingRP, PardananiA, TefferiA2012Differential prognostic effect of IDH1 versus IDH2 mutations in myelodysplastic syndromes: A Mayo Clinic study of 277 patients.
26: 101–10522033490RohleD, Popovici-MullerJ, PalaskasN, TurcanS, GrommesC, CamposC, TsoiJ, ClarkO, OldriniB, KomisopoulouE, 2013An inhibitor of mutant IDH1 delays growth and promotes differentiation of glioma cells.
340: 626–63023558169SasakiM, KnobbeCB, MungerJC, LindEF, BrennerD, BrustleA, HarrisIS, HolmesR, WakehamA, HaightJ, 2012IDH1(R132H) mutation increases murine haematopoietic progenitors and alters epigenetics.
488: 656–65922763442SchmittCA, FridmanJS, YangM, BaranovE, HoffmanRM, LoweSW2002Dissecting p53 tumor suppressor functions in vivo.
1: 289–29812086865SchubbertS, ShannonK, BollagG2007Hyperactive Ras in developmental disorders and cancer.
7: 295–30817384584ShenZX, ChenGQ, NiJH, LiXS, XiongSM, QiuQY, ZhuJ, TangW, SunGL, YangKQ, 1997Use of arsenic trioxide (As2O3) in the treatment of acute promyelocytic leukemia (APL): II. Clinical efficacy and pharmacokinetics in relapsed patients.
89: 3354–33609129042ShihAH, Abdel-WahabO, PatelJP, LevineRL2012The role of mutations in epigenetic regulators in myeloid malignancies.
12: 599–61222898539StirewaltDL, RadichJP2003The role of FLT3 in haematopoietic malignancies.
3: 650–66512951584TakahashiS2011Current findings for recurring mutations in acute myeloid leukemia.
4: 3621917154TurcanS, RohleD, GoenkaA, WalshLA, FangF, YilmazE, CamposC, FabiusAW, LuC, WardPS, 2012IDH1 mutation is sufficient to establish the glioma hypermethylator phenotype.
483: 479–48322343889WangJ, LiuY, LiZ, WangZ, TanLX, RyuMJ, MelineB, DuJ, YoungKH, RanheimE, 2011Endogenous oncogenic Nras mutation initiates hematopoietic malignancies in a dose- and cell type-dependent manner.
118: 368–37921586752WangF, TravinsJ, DeLaBarreB, Penard-LacroniqueV, SchalmS, HansenE, StraleyK, KernytskyA, LiuW, GliserC, 2013Targeted inhibition of mutant IDH2 in leukemia cells induces cellular differentiation.
340: 622–62623558173WardPS, PatelJ, WiseDR, Abdel-WahabO, BennettBD, CollerHA, CrossJR, FantinVR, HedvatCV, PerlAE, 2010The common feature of leukemia-associated IDH1 and IDH2 mutations is a neomorphic enzyme activity converting α-ketoglutarate to 2-hydroxyglutarate.
17: 225–23420171147WardPS, LuC, CrossJR, Abdel-WahabO, LevineRL, SchwartzGK, ThompsonCB2013The potential for isocitrate dehydrogenase mutations to produce 2-hydroxyglutarate depends on allele specificity and subcellular compartmentalization.
288: 3804–381523264629XuW, YangH, LiuY, YangY, WangP, KimSH, ItoS, YangC, XiaoMT, LiuLX, 2011Oncometabolite 2-hydroxyglutarate is a competitive inhibitor of α-ketoglutarate-dependent dioxygenases.
19: 17–3021251613YanH, ParsonsDW, JinG, McLendonR, RasheedBA, YuanW, KosI, Batinic-HaberleI, JonesS, RigginsGJ, 2009IDH1 and IDH2 mutations in gliomas.
360: 765–77319228619ZhangJW, WangJY, ChenSJ, ChenZ2000Mechanisms of all-trans retinoic acid-induced differentiation of acute promyelocytic leukemia cells.
25: 275–28411022230ZuberJ, RadtkeI, PardeeTS, ZhaoZ, RappaportAR, LuoW, McCurrachME, YangMM, DolanME, KoganSC, 2009Mouse models of human AML accurately predict chemotherapy response.
23: 877–88919339691ZuberJ, McJunkinK, FellmannC, DowLE, TaylorMJ, HannonGJ, LoweSW2011aToolkit for evaluating genes required for proliferation and survival using tetracycline-regulated RNAi.
29: 79–8321131983ZuberJ, ShiJ, WangE, RappaportAR, HerrmannH, SisonEA, MagoonD, QiJ, BlattK, WunderlichM, 2011bRNAi screen identifies Brd4 as a therapeutic target in acute myeloid leukaemia.
478: 524–52821814200oai:pubmedcentral.nih.gov:37924752014-03-15genesdevpmc-openGenes DevGenes DevGADGenes & Development0890-93691549-5477Cold Spring Harbor Laboratory PressPMC3792475PMC379247537924752406576624065766871166010.1101/gad.226753.113Research PaperInduction of sarcomas by mutant IDH2Lu et al.Oncogenic mechanism of IDH mutationLuChao12VennetiSriram1AkalinAltuna3411FangFang4WardPatrick S.12DeMatteoRaymond G.1IntlekoferAndrew M.1ChenChong1YeJiangbin1HameedMeera5NafaKhedoudja5AgaramNarasimhan P.5CrossJustin R.6KhaninRaya7MasonChristopher E.3HealeyJohn H.8LoweScott W.19SchwartzGary K.10MelnickAri4ThompsonCraig B.112Cancer Biology and Genetics Program, Memorial Sloan-Kettering Cancer Center, New York, New York 10065, USA;Cell and Molecular Biology Graduate Group, Perelman School of Medicine, University of Pennsylvania, Philadelphia, Pennsylvania 19104, USA;Department of Physiology and Biophysics, Institute for Computational Biomedicine, Weill Cornell Medical College, New York, New York 10065, USA;Division of Hematology/Oncology, Weill Cornell Medical College, New York, New York 10065, USA;Department of Pathology, Memorial Sloan-Kettering Cancer Center, New York, New York 10065, USA;Donald B. and Catherine C. Marron Cancer Metabolism Center,Bioinformatics Core,Orthopaedic Surgery Service, Memorial Sloan-Kettering Cancer Center, New York, New York 10065, USA;Howard Hughes Medical Institute, New York, New York 10065, USA;Melanoma and Sarcoma Service, Department of Medicine, Memorial Sloan-Kettering Cancer Center, New York, New York 10065, USA
Present address: Friedrich Miescher Institute for Biomedical Research, Maulbeerstrasse 66, 4058 Basel, Switzerland.
Most chondrosarcoma patients exhibit gain-of-function mutations in either IDH1 or IDH2. Lu et al. found that IDH mutations were associated with DNA hypermethylation at CpG islands in chondrosarcoma biopsies. Regions of hypermethylation were enriched for genes implicated in stem cell maintenance/differentiation and lineage specification. In murine mesenchymal progenitor cells, mutant IDH2 led to DNA hypermethylation and impairment in differentiation, which could be reversed by DNA-hypomethylating agents. Mutant IDH2 also generated undifferentiated sarcomas in vivo. This work demonstrates that neomorphic IDH2 mutations can be oncogenic in mesenchymal cells.
More than 50% of patients with chondrosarcomas exhibit gain-of-function mutations in either isocitrate dehydrogenase 1 (IDH1) or IDH2. In this study, we performed genome-wide CpG methylation sequencing of chondrosarcoma biopsies and found that IDH mutations were associated with DNA hypermethylation at CpG islands but not other genomic regions. Regions of CpG island hypermethylation were enriched for genes implicated in stem cell maintenance/differentiation and lineage specification. In murine 10T1/2 mesenchymal progenitor cells, expression of mutant IDH2 led to DNA hypermethylation and an impairment in differentiation that could be reversed by treatment with DNA-hypomethylating agents. Introduction of mutant IDH2 also induced loss of contact inhibition and generated undifferentiated sarcomas in vivo. The oncogenic potential of mutant IDH2 correlated with the ability to produce 2-hydroxyglutarate. Together, these data demonstrate that neomorphic IDH2 mutations can be oncogenic in mesenchymal cells.
Nearly a century after Warburg (1956) observed that cancer cells metabolize glucose differently from quiescent tissues, the recent resurgence in cancer metabolism research has led to the increasing appreciation that metabolic reprogramming is a hallmark of cancer (Vander Heiden et al. 2009; Ward and Thompson 2012). However, it remains controversial whether metabolic reprogramming plays a significant role in the pathogenesis of cancer. An argument in support of this hypothesis is the identification of cancer-associated germline and somatic alterations of genes encoding for metabolic enzymes (Mullarky et al. 2011; Oermann et al. 2012), including the recent discovery of prevalent mutations in isocitrate dehydrogenase 1 (IDH1) and IDH2.
Cytosolic IDH1 and mitochondrial IDH2 are NADP+-dependent enzymes that metabolize isocitrate to α-ketoglutarate (αKG). Frequent somatic mutations of IDH1 and IDH2 were initially identified in ∼80% of intermediate-grade gliomas (Yan et al. 2009) and ∼20% of de novo acute myeloid leukemias (AMLs) (Mardis et al. 2009; Ward et al. 2010). More recently, they were also found in more than half of patients with chondrosarcomas (Amary et al. 2011a) and skeletal disorders characterized by cartilage tumors (Amary et al. 2011b; Pansuriya et al. 2011). Almost all mutations observed in IDH1 and IDH2 are monoallelic point mutations affecting only a few residues, suggesting that they are unlikely to be loss of function. Indeed, metabolomic and biochemical analysis revealed that mutant IDH enzymes gain a neomorphic activity of producing 2-hydroxyglutarate (2HG) from αKG (Dang et al. 2009; Ward et al. 2010). 2HG is normally present at very low levels in cells but exhibits a >100-fold increase in tumor samples with IDH mutations. It is believed that IDH mutations promote tumorigenesis through accumulating the putative “oncometabolite” 2HG.
At the molecular level, mounting evidence implicates a link between IDH mutation and epigenetic dysregulation. In hematologic tumors and gliomas, IDH mutations are associated with a DNA hypermethylation profile and a gene expression pattern associated with lineage-specific progenitors (Figueroa et al. 2010; Noushmehr et al. 2010; Turcan et al. 2012). Mechanistically, mutant IDH was shown to impair activities of αKG-dependent chromatin-modifying enzymes by producing 2HG as a competitive inhibitor. These include TET family enzymes, which convert 5-methylcytosine (5mC) to 5-hydroxymethylcytosine (5hmC), a novel epigenetic mark potentially involved in DNA demethylation (Figueroa et al. 2010). Similarly, 2HG-producing IDH mutation inhibits activities of αKG-dependent Jumonji-C histone demethylases and leads to increased methylation at certain histone lysine residues (Chowdhury et al. 2011; Xu et al. 2011; Lu et al. 2012).
Although the high frequencies of IDH mutations in cancer imply that they are potential “driver” mutations, in vitro and in vivo modeling of IDH mutations has yet to report overt cellular transformation. Introduction of a mutant IDH1 allele as either a transgene or knock-in into hematopoietic cells, neural progenitors, or immortalized cell lines has failed to produce tumors in vivo (Sasaki et al. 2012b; Turcan et al. 2012). To date, the properties of IDH mutant tumors of mesenchymal origin have not been characterized, and the role of IDH mutations in the tumorigenesis of such cells has not been examined. As in other tumor types, we now report that IDH mutation is associated with extensive changes in DNA methylation that are enriched at promoter CpG islands of genes implicated in the regulation of cellular differentiation. Similar changes were observed when an IDH2 mutant allele was introduced into immortalized mesenchymal cells. Furthermore, in such cells, expression of mutant IDH2 leads to impaired mesenchymal lineage differentiation and loss of contact inhibition in vitro and formation of mesenchymal tumors in vivo. Abolishing 2HG-producing activity from mutant IDH2 eliminates its transformation capacity. Collectively, the data presented here demonstrate that the 2HG-producing IDH2 mutation can directly induce tumor formation in vivo.
ResultsGenome-wide DNA methylation landscape of IDH mutant chondrosarcomas
A panel of snap-frozen surgical specimens from 21 patients with chondrosarcomas was collected through an institutional review board-approved protocol. Targeted sequencing results revealed an ∼50% frequency of IDH mutations in chondrosarcomas (seven of the samples had the R132 IDH1 mutation, three had the R172 IDH2 mutation, and 11 were wild type for IDH1/2), consistent with previous reports (Amary et al. 2011a; Pansuriya et al. 2011; Arai et al. 2012). Compared with samples that were wild type for IDH1/2, IDH1 or IDH2 mutant samples showed a significant increase in intratumoral 2HG levels (Fig. 1A).
ERRBS analysis of chondrosarcoma patient samples. (A) Twenty-one blinded chondrosarcoma samples were analyzed for 2HG levels by gas chromatography-mass spectrometry (GC-MS). Subsequently, samples were decoded and grouped according to IDH1/2 mutational status. (B) ERRBS was performed on genomic DNA extracted from chondrosarcoma patient samples to generate genome-wide base-pair resolution CpG methylation information. A heat map representing the hierarchical clustering of samples with wild-type (WT) or mutant IDH1/2 is shown, based on a supervised analysis for differentially methylated CpGs at CpG islands. Each row represents a sample, and each column represents a CpG. The level of methylation is represented using a color scale, as shown in the legend. (C) Bar graph showing the percentage of hypermethylated and hypomethylated CpGs comparing IDH1/2 mutant with wild-type chondrosarcoma samples. (*) P < 0.0001 by χ2 test.
Genome-wide DNA methylation levels were measured using enhanced reduced representation bisulfite sequencing (ERRBS), which was previously demonstrated to provide base-pair resolution DNA methylation information and extended genomic coverage beyond CpG island regions compared with traditional RRBS (Akalin et al. 2012a). A minimum cutoff of 40% methylation difference, in addition to statistical significance (Q < 0.01), was required to identify differentially methylated CpGs between IDH1/2 wild-type and mutant samples. A total of 12,236 CpGs were found to be differentially methylated.
Individual CpG sites were next annotated to CpG islands, shores, or regions beyond CpG shores. The results showed that IDH1/2 mutations were associated with DNA hypermethylation at CpG islands (77% of differentially methylated CpGs were hypermethylated in IDH1/2 mutant samples) (Fig. 1B,C). In contrast, a minority of CpG shores and other regions (29% and 3%, respectively) displayed increased methylation in IDH mutant samples.
Analysis of DNA methylation at gene promoters was performed by selecting differentially methylated CpGs at −1000 to +500 base pairs (bp) of each transcription start site. The group of genes that were promoter DNA-hypermethylated in IDH mutant chondrosarcomas was then subjected to Database for Annotation, Visualization, and Integrated Discovery (DAVID) analysis to examine their functional relevance. The results showed that the top enriched functional categories were involved in various organismal and cellular developmental processes (Supplemental Fig. S1; Supplemental Table S1). In addition, the most significantly hypermethylated genes in IDH mutant samples include lineage specification regulators such as retinoic acid receptor α (RARA), platelet-derived growth factor receptor α (PDGFRA), and BCL6 corepressor (BCOR) (Supplemental Table S2). Taken together, these data provide a comprehensive genome-wide DNA methylation landscape of IDH mutant chondrosarcomas and suggest that IDH mutations are associated with epigenetic dysregulation of genes implicated in the regulation of stem cell maintenance/differentiation and cell lineage specification.
IDH2 mutation induces DNA hypermethylation
To establish the causality between IDH mutation and aberrant DNA methylation, the DNA methylome between isogenic cell lines expressing wild-type or mutant IDH was compared. In contrast to others, we chose to examine the role of chondrosarcoma-associated mitochondrial IDH2 mutants. Unlike cytosolic IDH1 R132 mutants, mitochondrial IDH2 R172 mutants generate high levels of their neomorphic product, 2HG, independently of the action of wild-type enzymes (Ward et al. 2013). We retrovirally transduced 10T1/2 (10T) cells with vectors containing either wild-type or R172K mutant IDH2. 10T cells expressing R172K mutant IDH2 but not empty vector or wild-type IDH2 showed significant accumulation of 2HG (Supplemental Fig. S2).
Genomic DNA of 10T cells was extracted and processed for ERRBS to generate highly reproducible base-pair resolution DNA methylation profiles (Supplemental Fig. S3A). In agreement with results from chondrosarcoma biopsies, IDH2 R172K mutant cells compared with wild-type cells showed a profound DNA hypermethylation at CpG islands across all chromosomes (Fig. 2A). In contrast, significantly less differentially methylated CpGs were observed between wild-type IDH2 and vector cells, with even distribution of hypermethylated and hypomethylated sites (Fig. 2A). Increased methylation at several histone marks, such as H3K9me3, H3K9me2, and H3K4me3, was also found in IDH2 R172K mutant cells, which could reinforce with DNA methylation to modulate gene expression (Supplemental Fig. S3B).
Mutant IDH2 induces CpG island hypermethylation phenotype. (A) Stacking bar graph showing percentage of hypermethylated and hypomethylated CpGs of all CpGs located in CpG islands for each chromosome, comparing IDH2 R172K mutant with wild-type (WT) cells (left) or vector with IDH2 wild-type cells (right). Green represents proportion of hypomethylated cytosines, and magenta represents hypermethylated ones. Only CpGs with a Q-value <0.01 and a methylation difference of at least 25% are shown. (B) GSEA was performed on genes that were promoter DNA-hypermethylated in IDH2 R172K mutant cells. The table shows the top four most significantly enriched gene sets from the Broad Institute database and their P-values.
In total, there were 2400 genes with differentially methylated CpGs at their promoters, and a predominance in DNA hypermethylation was observed when comparing the IDH2 R172K mutant with wild-type cells (78.1% being hypermethylated, P-value < 0.0001) (Supplemental Table S3). To gain more insights into the mechanism of aberrant DNA methylation, the group of genes that were promoter DNA-hypermethylated in IDH2 R172K mutant cells was subjected to gene set enrichment analysis (GSEA) with the Broad Institute Molecular Signatures Database (http://www.broadinstitute.org/gsea/msigdb/index.jsp), which includes >3000 curated gene sets collected from various research sources. The top four most statistically significantly enriched gene sets were Polycomb-repressive complex 2 (PRC2) target genes or genes marked by H3K27me3 in embryonic stem cells (Fig. 2B; Supplemental Table S4). Notably, recent genome-wide mapping of Tet1-binding sites revealed that >95% of PRC2 target genes were bound by Tet1 (Wu and Zhang 2011), and Tet1 has been proposed as a guardian to protect these regions from accidental DNA methylation (Williams et al. 2012). Taken together, the DNA methylome analysis suggests that acquisition of an IDH2 R172 mutation is sufficient to establish DNA hypermethylation, the pattern of which matches the patterns of Tet1- and PRC2-binding activity.
10T cells were originally isolated from C3H mouse embryos. They were demonstrated to be multipotent, with the ability to differentiate into several mesenchymal lineages, including adipocytes, myoblasts, and chondrocytes (Taylor and Jones 1979). Since the DNA methylation signature associated with IDH mutant chondrosarcomas was enriched for stem cell maintenance and differentiation genes, the effects of IDH2 mutation on the mesenchymal differentiation potential of 10T cells were determined. Expression of R172K mutant IDH2 in 10T cells led to a profound impairment in either adipocyte or chondrocyte differentiation. Compared with vector and wild-type IDH2 cells, IDH2 R172K mutant cells showed no visible accumulation of lipid droplets and failed to express mature adipocyte markers (Adipoq and Fabp4) after adipocyte differentiation induction (Fig. 3A). Similarly, when cells were subjected to conditions that promote chondrocyte differentiation, morphological conversion to rounded shapes resembling mature chondrocytes and expression of mature chondrocyte markers (Acan and Col2a1) were only observed in vector and wild-type IDH2 cells, while IDH2 R172K mutant cells maintained a fibroblast-like undifferentiated state (Fig. 3B). Notably, overexpression of wild-type IDH2 enzyme led to enhanced expression of mature chondrocyte markers, suggesting that increased production of NADPH and αKG may promote cellular differentiation. Furthermore, while vector and wild-type IDH2 cells became proliferation-arrested after the induction of adipocyte or chondrocyte differentiation, the proliferation rate of IDH2 R172K mutant cells remained largely unaffected (Fig. 3C). IDH2 R172K mutant cells were also able to maintain cyclin D1 levels after differentiation induction, unlike vector and wild-type IDH2 cells (Fig. 3D).
IDH2 mutation inhibits mesenchymal differentiation. (A) Vector (Vec), wild-type (WT), or R172K mutant IDH2 cells were treated with adipocyte differentiation cocktail. After 7 d of differentiation induction, representative microscopic images of cell morphology were recorded, and mRNA expression of Adipoq and Fabp4 was measured by quantitative real-time PCR (qRT-PCR). (B) Vector, wild-type, or R172K mutant IDH2 cells were treated with chondrocyte differentiation cocktail. After 10 d of differentiation induction, representative microscopic images of cell morphology were recorded (arrowheads point to mature chondrocyte-resembling cells), and mRNA expression of Acan and Col2a1 was measured by qRT-PCR. (C) 10T vector, wild-type, or R172K mutant IDH2 cells were treated with adipocyte or chondrocyte differentiation cocktails. Cell numbers were counted at days 0, 3, and 6 after differentiation induction. (D) Six days after adipocyte or chondrocyte differentiation induction, 10T vector, wild-type, or R172K mutant IDH2 cells were lysed, and protein levels of cyclin D1 were measured by Western blot. Tubulin was used as loading control. For all experiments, the average ± SD from three biological replicates are shown.
Differentiation impairment by mutant IDH2 correlates with high levels of 2HG accumulation
We previously demonstrated that a rare mutation in IDH1 (A134D) identified in thyroid cancers abolished the catalytic activity of the enzyme (Ward et al. 2012). To determine whether the 2HG-producing activity is required for R172K mutant IDH2 to inhibit differentiation, a similar mutation was introduced to the analogous residue in IDH2 (A174D), which is predicted to interfere with the substrate binding of R172 and eliminate the enzymatic activity of R172K mutant IDH2 (Fig. 4A). Indeed, 2HG levels in 10T cells stably expressing R172K/A174D mutant IDH2 were comparable with vector and wild-type IDH2 cells (Fig. 4B). 10T cells stably expressing R140Q mutant IDH2 were also generated. R140Q mutation in IDH2 is only found in hematological malignancies and was shown to be a weak 2HG-producing mutant (Ward et al. 2013). In 10T cells, expression of R140Q mutant IDH2 increased 2HG levels by fivefold, while R172K mutant IDH2 increased 2HG levels by >100-fold (Fig. 4B). When cells were tested for their differentiation potentials, the A174D mutation was found to abolish R172K mutant IDH2's ability to block adipocyte or chondrocyte differentiation (Fig. 4C). Similarly, weak 2HG-producing R140Q mutant IDH2 did not result in any differentiation impairment.
Differentiation impairment by mutant IDH2 correlates with 2HG production. (A) Structural modeling of IDH2 catalytic site showing Arg 172 and Ala 174. Isocitrate carbons are in yellow except carbon 6 containing the β-carboxyl, which is highlighted in cyan. Carbon atoms of amino acids (green), amines (blue), and oxygens (red) are also depicted. Hydrogen atoms are omitted for clarity. Dashed lines show <3.1 Å interactions corresponding to hydrogen and ionic bonds. The prime (′) denotes that the residue comes from the other monomer of the IDH dimer. (B) 10T cells expressing vector (Vec), wild-type (WT), R172K, R172K/A174D, or R140Q mutant IDH2 were lysed, and IDH2 expression was measured by Western blot. 2HG levels were measured by GC-MS and normalized to internal standard (D5-2HG) and cell number. (C) 10T cells expressing wild-type or various mutant IDH2 were treated with adipocyte or chondrocyte differentiation cocktails. mRNA expression of Adipoq, Fabp4, Acan, and Col2a1 was measured by qRT-PCR after 8 d of differentiation induction. For all experiments, the average ± SD from three biological replicates are shown.
DNA-hypomethylating treatment reverses differentiation block by mutant IDH2
To functionally link DNA hypermethylation to differentiation impairment, IDH2 R172K mutant cells were treated with 5-azacytidine (5-aza) to test whether inhibiting DNA methyltransferase had any effect on reversing the differentiation inhibition by mutant IDH2. After treatment with low doses of 5-aza for 48 h to induce global DNA demethylation (Supplemental Fig. S4), IDH2 R172K mutant cells were allowed to proliferate in the absence of 5-aza until reaching confluence before being subjected to adipocyte differentiation induction. Compared with untreated cells, a transient exposure to 5-aza led to the accumulation of lipid droplets and a dose-dependent increase in the expression of mature adipocyte markers (fabp4 and adipoq) after 8 d of differentiation induction (Fig. 5A,B), suggesting that IDH2 R172K mutant cells' differentiation potential can be restored upon treatment with a DNA-hypomethylating agent.
DNA-hypomethylating agent reverses the differentiation defect in mutant IDH2 cells. (A) After 8 d of adipocyte differentiation induction, microscopic images of cell morphology were recorded in IDH2 R172K mutant cells with or without transient 5-aza treatment. (B) After 8 d of adipocyte differentiation induction, mRNA expression of Adipoq and Fabp4 in IDH2 R172K mutant cells with or without transient 5-aza treatment was measured by qRT-PCR. The average ± SD from three biological replicates are shown.
IDH2 mutant cells escape contact inhibition
In addition to mesenchymal multipotency, 10T cells are sensitive to confluence-induced proliferation arrest and show no tumorigenicity in xenograft studies (Reznikoff et al. 1973). However, the differentiation experiments described above require cells to be seeded at confluence before induction. Therefore, the findings that IDH2 R172K mutant cells were able to continue proliferation at post-confluence (Fig. 3C,D) prompted us to investigate whether these cells, in addition to differentiation impairment, have acquired resistance to contact inhibition. When cultured in normal growth medium, vector, wild-type IDH2, or IDH2 R172K mutant cells showed no difference in proliferation rate when cells were sparse (days 0–2) (Fig. 6A). However, while vector and wild-type IDH2 cells stopped proliferating shortly after reaching confluence (days 2–6), the accumulation of IDH2 R172K mutant cells continued even after confluence was achieved. Measurement of cell mitotic activity by detection of the incorporation of nucleoside analog 5-ethynyl-2′-deoxyuridine (EdU) also revealed that IDH2 R172K mutant cells were more mitotically active at post-confluence (Fig. 6B). Moreover, compared with vector and wild-type IDH2 cells, the induction of the cell cycle inhibitor p27 and the decrease in cyclin D1 levels after contact inhibition were less pronounced in IDH2 R172K mutant cells (Fig. 6C). These features were not observed in IDH2 R172K/A174D or R140Q mutant cells (Fig. 6D), suggesting that loss of contact inhibition induced by mutant IDH2 requires high levels of 2HG production.
IDH2 mutant cells show loss of contact inhibition. (A) 10T vector (Vec), wild-type (WT), or R172K mutant IDH2 cells were cultured in Dulbecco's modified Eagle's medium (DMEM), and cell numbers were counted at days 0, 2, 4, and 6. Cells reached confluence between days 2 and 4. (B) Vector, wild-type, or R172K mutant IDH2 cells at sparse or post-confluence day 2 were incubated with EdU for 4 h. Percentage of EdU-positive cells was measured by flow cytometry. Histograms from a representative experiment from a total of two experiments are shown. (C) Vector, wild-type, or R172K mutant IDH2 cells were lysed at sparse, confluence, or 2 d post-confluence. Protein levels of cyclin D1 and p27 were measured by Western blot. Tubulin was used as loading control. (D) 10T cells expressing wild-type or various mutant IDH2s at post-confluence day 2 were incubated with EdU for 4 h. Percentage of EdU-positive cells was measured using flow cytometry. Cells were also lysed, and cyclin D1 levels were measured by Western blot. Tubulin was used as loading control. (E) Vector, wild-type, or R172K mutant IDH2 cells were lysed at sparse or 2 d post-confluence. Levels of N-cadherin protein expression were measured by Western blot, and mRNA expression was measured by qRT-PCR. For all experiments, the average ± SD from three biological replicates are shown.
IDH2 R172K mutant cells underwent cell cycle arrest normally after serum deprivation (Supplemental Fig. S5A), suggesting that their escape from contact inhibition is unlikely to be due to defects in cell cycle regulation. The cadherins are critical components of cell–cell adhesion and have been proposed as the main upstream mediator of contact inhibition (Gumbiner 2005). Unlike vector and wild-type IDH2 cells, IDH2 R172K mutant cells failed to up-regulate N-cadherin protein and mRNA expression after reaching confluence (Fig. 6E), suggesting that the post-confluence growth of these cells results from the inability to sense cell–cell contact at the surface membrane. This may be at least in part due to epigenetic gene silencing, as two cell adhesion genes (Itga9 and cntn1) that were DNA-hypermethylated in IDH2 R172K mutant cells (Supplemental Table S3) failed to increase their mRNA expression upon cell–cell contact (Supplemental Fig. S5B).
IDH2 mutant cells generate poorly differentiated sarcomas in vivo
Given the malignant phenotypes of differentiation blockade and loss of contact inhibition observed in vitro, xenograft studies were performed to test whether 10T cells with IDH2 mutation could generate tumors in vivo. IDH2 R172K mutant cells formed palpable tumors 20 d after subcutaneous injection and continued to grow up to 1000 mm3, while vector and wild-type IDH2 cells showed no tumorigenicity over the course of study (Fig. 7A). Immunohistochemistry staining revealed that these tumors resembled poorly differentiated sarcomas with a high Ki67 index (Fig. 7B). Consistent with in vitro studies, these tumors also had high levels of cyclin D1, low p27 expression (Fig. 7B), and no detectable staining for mature mesenchymal lineage markers (data not shown). Tumors from IDH2 R172K mutant cells showed 2HG levels that were comparable with cells cultured in vitro (Fig. 7C), and IDH2 R172K/A174D or R140Q mutant cells failed to induce xenograft growth (Fig. 7D).
IDH2 mutant cells generate mesenchymal tumors in vivo. (A) We injected 1 × 107 10T vector (Vec), wild-type (WT), or R172K mutant IDH2 cells subcutaneously into nude mice. Tumor growth was monitored and measured. The insert image is shown for mice implanted with wild-type cells (left) or mutant cells (right) at the time of sacrifice. (B) Immunohistochemical staining was performed on R172K mutant IDH2 tumors using specific antibodies, and representative images are shown for sections stained with hematoxylin and eosin (H&E), Ki67, cyclin D1, and p27. (C) 2HG levels in R172K mutant IDH2 tumors and parental or R172K mutant IDH2 10T cells cultured in vitro were measured by GC-MS. The ratio of 2HG to citrate is shown. (D) We injected 1 × 107 10T R172K, R172K/A174D, or R140Q mutant IDH2 cells subcutaneously into nude mice. Tumor growth was monitored and measured. For all experiments, the average tumor volumes ± SD of five mice per group are shown.
Discussion
The study of IDH mutations in carcinogenesis has been hampered by a lack of robust model systems. We previously found that IDH mutation blocked adipocyte differentiation from 3T3-L1 murine fibroblasts (Lu et al. 2012). This finding has been extended to the hematopoietic system with the observations that mutant IDH and 2HG could impair EPO-induced erythrocyte differentiation in an erythroleukemic cell line (Losman et al. 2013) and that hematopoietic conditional knock-in IDH1 mutant mice had an expansion in early progenitor/stem cell population (Sasaki et al. 2012b). Nevertheless, none of the previous studies reported in vivo tumorigenicity as a result of the presence of mutant IDH. We report here that, using a nontransformed mesenchymal multipotent cell line, expression of an IDH2 mutant enzyme producing high levels of 2HG not only arrested cells from differentiating into adipocytic and chondrocytic lineages, but also resulted in loss of contact inhibition and tumor formation in vivo. It is noteworthy that 10T cells have been frequently used as an in vitro model to test the carcinogenic potential of chemicals in mesenchymal cells (for review, see Schechtman 2012). While it is possible that pre-existing genetic and epigenetic alterations in 10T cells render them more susceptible to 2HG-induced transformation, the in vivo tumorigenicity established here depends on the introduction of a 2HG-producing IDH2 R172K mutant transgene. The morphology of the mutant IDH2-induced 10T tumors in vivo resembles that of human chondrosarcomas in which the block to cartilage formation becomes more pronounced as the tumor progresses. This is consistent with the in vitro impairment in cell differentiation exhibited by 10T cells expressing a chondrosarcoma-associated IDH2 R172K mutant enzyme.
As a common product of recurrent cancer-associated IDH1 and IDH2 mutations, 2HG is believed to mediate mutant IDH's tumor-promoting potential. We and others have shown that exogenous 2HG could recapitulate mutant IDH's effect on blocking cell differentiation (Lu et al. 2012; Losman et al. 2013). In the present study, we further demonstrate that high levels of 2HG production were required for mutant IDH2's transformation capacity. Therefore, it appears that enzymatic activity is necessary for oncogenesis by mutant IDH. It should be noted that the truncated reverse reaction catalyzed by mutant IDH also consumes NADPH. Since the R172K/A174D mutant IDH2 was enzymatically inactive, our experiments do not rule out the possibility that NADPH consumption and the resulting generation of reactive oxygen species (ROS) are also important for mutant IDH's oncogenic mechanism. However, since IDH mutations in cancer are monoallelic and the reverse reaction by mutant IDH occurs at a much slower rate compared with the forward reaction by the wild-type enzyme (Dang et al. 2009), the mutation's impact on the total cellular NADPH pool is likely to be minimal. In agreement, studies have shown that the intracellular ROS levels actually decrease as a result of mutant IDH expression (Sasaki et al. 2012a; Li et al. 2013).
In agreement with previous findings in other tumor types, quantitative profiling of genome-wide DNA methylation using ERRBS revealed a significant increase in CpG island methylation in IDH mutant chondrosarcoma samples. Such association is unlikely to result from the disparate cells of origin affected by IDH mutations, as the expression of the IDH2 R172K mutant in 10T cells was sufficient to cause CpG island DNA hypermethylation. GSEA showed a remarkable overlap between R172K mutant IDH2-associated promoter-hypermethylated genes and PRC2 target/Tet1-bound genes in embryonic stem cells. Since other cell types, including mesenchymal progenitor cells, show a similar PRC2 target preference (Bracken et al. 2006; Squazzo et al. 2006), our results suggest that the establishment of DNA hypermethylation by mutant IDH2 in 10T cells may be the result of TET inhibition. Importantly, functional analysis and validation experiments implicate that aberrant promoter DNA methylation may underlie the inability to regulate the expression of genes important for differentiation and cell–cell contact. Consistent with this notion, we showed that low-dose pulsing of 5-aza, which has been shown to exert durable anti-tumor activities (Tsai et al. 2012), was sufficient to reverse the differentiation defect of IDH2 mutant cells. Therefore, IDH mutation may initiate and promote tumorigenesis by epigenetically “locking” a proliferative progenitor cell in that state.
Recently, inhibitors that specifically abrogate mutant IDH's ability to produce 2HG have been developed (Wang et al. 2013). It was shown that the inhibitors could induce differentiation, delay xenograft growth, and reverse histone hypermethylation in tumor cells containing ectopic or endogenous mutant IDH (Rohle et al. 2013; Wang et al. 2013), suggesting that some of the epigenetic abnormalities are potentially correctable. To date, no inhibitors of 2HG production by the chondrosarcoma-associated IDH2 R172 mutation studied here have been reported. However, given 2HG's remarkable impact on the epigenome, it would be interesting to test whether compounds targeting chromatin modifiers are selectively toxic to tumors with an IDH mutation.
Materials and methodsChondrosarcoma patient samples and ERRBS analysis
Acquisition of snap-frozen surgical tumor specimens from patients with chondrosarcomas was carried out through a Memorial Sloan-Kettering Cancer Center institutional review board-approved protocol. MALDI-TOF mass spectrometry (MS) (Sequenom) platform was used to identify IDH1/2 mutations. ERRBS was performed with genomic DNA extracted from 10T cells or chondrosarcoma biopsies as previously described (Akalin et al. 2012a). The ERRBS raw data were aligned to the whole genome, and the methylation was called using the epigenetic core pipeline at Weill Cornell Medical School. The differential methylated CpGs were identified with the methylKit R package (Akalin et al. 2012b). The data were processed and plotted using R script. The two-dimensional (2D) hierarchical clustering of 21 chondrosarcoma samples was performed using the heat map function in R 2.15.1 (http://www.r-project.org) with 1-Pearson correlation distance and Ward's agglomeration method. The ERRBS data have been deposited in the NCBI's Gene Expression Omnibus (accession no. GSE50539).
GSEA
Gene sets from the Broad Institute Molecular Signatures Database (http://www.broadinstitute.org/gsea/msigdb/index.jsp) were used. The overrepresentation of gene sets in the gene list was done using hypergeometric calculation implemented in the GOstats Bioconductor package (http://www.bioconductor.org/packages/release/bioc/html/GOstats.html) in R statistical language (http://www.r-project.org). Gene ontology analysis was performed with the online software DAVID on functional analysis of gene lists (Huang et al. 2009).
Plasmid construction
The cDNA clone of human IDH2 (BC009244) was purchased from Invitrogen in pOTB7. Standard site-directed mutagenesis techniques were used to make IDH2 R172K by introducing a g515a change in the IDH2 ORF. Wild-type and mutant sequences were then subcloned into the LPC vector, and IDH2 R172K/A174D was made by introducing a c521a change in the ORF of the LPC-IDH2 R172K construct. All sequences were confirmed by direct sequencing before retrovirus generation.
Cell culture, transfection/transduction, and generation of stable cell lines
C3H10T1/2 cells were cultured in Dulbecco's modified Eagle's medium (DMEM; Invitrogen) with 10% fetal bovine serum (FBS; CellGro). To generate 10T cell lines with stable expression of wild-type or mutant IDH2, supernatant from 293T cells transfected with pCL-Eco helper virus and plasmids was collected after 72 h, filtered, and applied to 10T parental cells overnight. Pooled populations of puromycin-resistant cells were obtained by growing cells in 2.5 μg/mL puromycin for 7 d following retroviral transduction and then continuously culturing in puromycin.
10T cell differentiation
Adipocyte differentiation of 10T cells was done following the same procedure as 3T3-L1 cell differentiation previously described (Wellen et al. 2009). In brief, confluent 10T cells were stimulated with a cocktail containing 0.5 mM isobutylmethylxanthine, 1 μM dexamethasone, 5 μg/mL insulin, and 5 μM troglitazone (all from Sigma) in DMEM with 10% FBS to induce differentiation. Cells were maintained in DMEM with insulin after 2 d of differentiation until they were ready to be harvested. Chondrocyte differentiation of 10T cells was performed as previously described (Mikami et al. 2011). In brief, confluent 10T cells were stimulated with a cocktail containing DMEM with 1% FBS, 10 μg/mL human insulin (Sigma), 3 × 10−8 M sodium selenite (Sigma), 10 μg/mL human transferrin (Sigma), 10−8 M dexamethasone, and 100 ng/mL rhBMP-2 (R&D Systems). Medium was replaced with fresh cocktail every 2–3 d until cells were ready to be imaged or harvested.
Histone extraction and Western blotting
For histone acid extraction, cells were lysed in hypotonic lysis buffer (10 mM HEPES, 10 mM KCl, 1.5 mM MgCl2, 0.5 mM DTT, protease inhibitors) for 1 h. H2SO4 was added to 0.2 N followed by overnight rotation at 4°C. After centrifugation, supernatants were collected, and proteins were precipitated in 33% TCA, washed with acetone, and resuspended in deionized water. For whole-cell lysates, cells were lysed in standard RIPA buffer (1% NaDOC, 0.1% SDS, 1% Triton X-100, 0.01 M Tris at pH 8.0, 0.14 M NaCl), and lysates were then sonicated and centrifuged at 14,000g at 4°C. Supernatants were collected and normalized for total protein concentration, separated by SDS-PAGE, transferred to nitrocellulose membrane, blocked in 5% nonfat milk in PBS plus 0.5% Tween 20, probed with primary antibodies, and detected with horseradish peroxidase-conjugated anti-rabbit or anti-mouse secondary antibodies (GE Healthcare, NA934V and NA931V). The primary antibodies used include anti-IDH2 (Abcam, ab55271), anti-H3K9me2 (Cell Signaling Technology, 9753), anti-H3K9me3 (Active Motif, 39765), anti-H3K36me3 (Abcam, ab9050), anti-H3K27me3 (Millipore, 07-449), anti-H3K4me3 (Active Motif, 39916), anti-H3K79me2 (Cell Signaling Technology, 9757), anti-acetyl H3 (Upstate Biotechnology, 06-599), anti-H3 (Cell Signaling Technology, 4499), anti-tubulin (Sigma, T9026), anti-cyclin D1 (EMD Millipore, CC12), anti-p27 (Santa Cruz Biotechnology, sc-1641), anti-N-cadherin (Cell Signaling Tech, 4061).
Metabolite extraction and gas chromatography (GC)-MS analysis
Frozen chondrosarcoma tissue (30–60 mg) or 10T cells were homogenized in 1.2 mL of ice-cold 80% methanol containing 20 μM deuterated 2HG as an internal standard (D-hydroxyglutaric-2,3,3,4,4-d5) using a tissue homogenizer (Omni International). Methanol extracts were incubated for 30 min at −80°C and centrifuged at 21,000g for 20 min at 4°C to remove precipitated protein. Five-hundred microliters of extracts was evaporated to dryness under a nitrogen gas stream. Dried organic acids were derivatized by the addition of 20 μL of 40 mg/mL methoxyamine hydrochloride in pyridine with incubation for 90 min at 30°C followed by 80 μL of MSTFA + 1% TMCS (Thermo Scientific) with incubation for 30 min at 37°C. One microliter of the trimethylsilyl-derivatized organic acids was analyzed by GC-MS using an Agilent 7890A GC equipped with an HP-5MS capillary column and connected to an Agilent 5975 C mass selective detector operating in splitless mode with electron impact ionization. Relative quantification of 2HG was determined from extracted ion chromatograms for 2HG (m/z: 349) normalized to either the d5-2HG internal standard (m/z: 354) or citrate (m/z: 465) and corrected by wet weight of tissue sample or cell numbers.
Cell proliferation and cell cycle analysis
EdU incorporation assay was performed using the Click-iT EdU flow cytometry assay kit (Invitrogen, C10425) according to the manufacturer's instructions. In brief, 10T cells at sparse or post-confluence were incubated with 10 μM EdU for 4 h. Cells were then trypsinized, washed with PBS, and fixed for 15 min at room temperature. After fixation, cells were washed with PBS and incubated in permeabilization buffer containing CuSO4 and Alexa Fluor 448 azide for 30 min at room temperature. Levels of EdU incorporation were measured using a Becton Dickinson Calibur flow cytometer.
For FACS analysis of cell cycle distribution, adherent cells in a six-well plate were trypsinized and collected in 15-mL centrifuge tubes. The collected cells were fixed by ethanol (final concentration of 70%) overnight at −20°C. Cells were then stained with 50 μg/mL propidium iodide (Sigma) with 0.1 mg/mL RNase A for 40 min at 37°C. After washing with PBS, the DNA content of the stained cells was then analyzed by a Becton Dickinson Calibur flow cytometer.
Quantitative real-time PCR
RNA was isolated using TRIzol (Invitrogen). After incubating with DNase, cDNA was synthesized using SuperScript II reverse transcriptase (Invitrogen). Quantitative PCR was performed on a 7900HT sequence detection system (Applied Biosystems) using TaqMan gene expression assays (Applied Biosystems). Gene expression data were normalized to 18S rRNA.
Immunohistochemical staining
Immunohistochemical detection was performed using the Discovery XT processor (Ventana Medical Systems). Paraffin-embedded tissue sections were blocked for 30 min in 10% normal goat serum plus 2% BSA in PBS. Sections were incubated for 5 h with 0.4 μg/mL rabbit polyclonal anti-Ki67 (Vector Laboratories, VP-K451), 1 μg/mL mouse monoclonal anti-p27 (BD Biosciences, 55890), or 6 μg/mL rabbit polyclonal anti-Cyclin D1 (Santa Cruz Biotechnology, H-295) antibodies. Tissue sections were then incubated for 60 min with biotinylated goat anti-rabbit IgG (Vector Laboratories, PK6101) or goat anti-mouse IgG (Vector Laboratories, BA9200) at 1:200 dilutions. Blocker D, streptavidin-HRP, and the DAB detection kit (Ventana Medical Systems) were used according to the manufacturer's instructions.
Tumor xenograft study
10T cells were harvested with trypsin and resuspended in PBS. Approximately 107 cells were injected subcutaneously into both flanks of 6- to 8-wk-old athymic female nude mice (Harlan Laboratories). Formation of tumors was monitored. After tumors became palpable, tumor volumes were estimated by caliper measurements, assuming spherical geometry (volume = d3 × π/6). Mice were cared for in accordance with the institutional animal care and use committee at Memorial Sloan-Kettering Cancer Center.
Structural modeling
The active site of human IDH2 with isocitrate was modeled based on the highly homologous porcine IDH2 crystal structure, as previously described (Ward et al. 2010, 2012).
Dot blot
Genomic DNA was phenol–chloroform-extracted and quantified using NanoDrop. DNA was denatured with 0.1 M NaOH, serial-diluted, and spotted on a nitrocellulose membrane. 5mC antibody (Millipore, MABE146) was used at 1:2000 dilution.
Statistical analysis
Statistical analysis was performed using a Student's t-test (two-sample equal variance; two-tailed distribution) unless stated otherwise.
Acknowledgments
We thank members of the Thompson laboratory for technical help and critical reading of the manuscript. We thank the Memorial Sloan-Kettering Cancer Center (MSKCC) Molecular Cytology Core Facility for technical help with the immunohistochemical study, the MSKCC Anti-tumor Assessment Core Facility for technical help with the mouse xenograft study, and the Epigenomics Core Facility of Weill Cornell Medical College for help with ERRBS. This work was supported by the Starr Cancer Consortium, grants from NCI and NIH (to C.B.T.), a Specialized Center of Research (SCOR) grant from the Leukemia and Lymphoma Society (to S.W.L.), and a U01 CTDD award from the National Cancer Institute (to S.W.L.). S.W.L. is the Goeffrey Beene Chair of Cancer Biology at MSKCC and an Investigator in the Howard Hughes Medical Institute. C.C. receives a career development fellowship from the Leukemia and Lymphoma Society. P.S.W. was supported in part by the University of Pennsylvania Medical Scientist Training Program. J.H.H. is funded by the Stephen McDermott Chair in Surgery.
Supplemental material is available for this article.
Article is online at http://www.genesdev.org/cgi/doi/10.1101/gad.226753.113.
Double-strand break repair by recombination requires a homology search, and its efficiency is affected by the mobility of both the break site and the homologous template. In yeast, induced breaks move significantly more than undamaged loci. Seeber et al. find that the yeast checkpoint factors Mec1, Rad9, and Rad53 are required for genome-wide increases in chromatin mobility. Mec1 activation enhances chromatin mobility even in the absence of damage. The INO80 chromatin remodeler acts downstream from Mec1 to increase chromatin mobility.
Double-strand break repair by recombination requires a homology search. In yeast, induced breaks move significantly more than undamaged loci. To examine whether DNA damage provokes an increase in chromatin mobility generally, we tracked undamaged loci under DNA-damaging conditions. We found that the yeast checkpoint factors Mec1, Rad9, and Rad53 are required for genome-wide increases in chromatin mobility, but not the repair protein Rad51. Mec1 activation by targeted Ddc1/Ddc2 enhances chromatin mobility even in the absence of damage. Finally, the INO80 chromatin remodeler is shown to act downstream from Mec1 to increase chromatin mobility, highlighting an additional damage-related role of this nucleosome remodeling complex.
DNA damage responseMec1chromatin mobilitydouble-strand breaksINO80
DNA double-strand breaks (DSBs) can be repaired by either homologous recombination (HR) or nonhomologous end-joining (NHEJ). Whereas NHEJ is the dominant pathway in G1-phase cells and in mammals, the more efficient and preferred pathway of repair in budding yeast is by HR. In the S and G2 phases of the cell cycle, recombinational repair makes use of the undamaged sister chromatid. However, if a sister chromatid is not present or if both sisters have been damaged, a search for an alternative homologous template ensues. This search appears to be rate-limiting for HR (Wilson et al. 1994; Agmon et al. 2013), and its efficiency was predicted to be affected by the mobility of both the break site and the homologous template (Gehlen et al. 2011). This notion is supported by computer simulations, which show that two randomly moving spots confined within a sphere collide more frequently than they would if one were immobile (Gehlen et al. 2011). New data on recombination rates in yeast support these predictions experimentally (Agmon et al. 2013).
Recent studies have also examined the mechanisms that drive chromatin movement of damaged sites (Dion et al. 2012; Krawczyk et al. 2012; Mine-Hattab and Rothstein 2012; Neumann et al. 2012; for review, see Dion and Gasser 2013). The recruitment of repair proteins, such as the strand exchange protein Rad51, enhances the movement of the broken DNA locus, tagged by Rad52-YFP (Dion et al. 2012). By analogy to the bacterial RecA, which contributes to the sequence search in three-dimensional (3D) space in vitro (Forget and Kowalczykowski 2012), Rad51 has also been suggested to mechanistically drive homology search (Renkawitz et al. 2013). In eukaryotes, Rad51 recruits the Snf2-type ATPase Rad54, which also contributes to the enhanced mobility of a DSB through an unknown mechanism (Dion et al. 2012).
Alongside these repair proteins, the DNA damage response (DDR) pathway appears to regulate the movement of the DSB. The resection of DNA at a cut site leads to activation of the ATR kinase complex Mec1/Ddc2 and, in turn, the downstream checkpoint kinase Rad53. These kinases regulate a number of processes, including cell cycle transitions, transcriptional programs, and DNA repair. Importantly, both Mec1 and its target protein, Rad9, were needed to increase the mobility of a DSB, whereas Rad53 was not (Dion et al. 2012).
It has remained unresolved whether undamaged chromatin also becomes more mobile in a nucleus that contains DNA damage. One report showed that the induction of a DSB at the MAT locus on chromosome (Chr) III in diploid yeast cells led to increased mobility of an undamaged site on the short arm of Chr V (Mine-Hattab and Rothstein 2012). In contrast, Dion et al. (2012) showed that an I-SceI endonuclease-induced DSB in haploid cells did not increase the mobility of an undamaged Chr VI locus, nor did treating cells with the DNA-damaging agent Zeocin at 50 μg/mL. The source of this discrepancy was unclear. Factors likely to influence the outcome include (1) the differential regulation of HR in haploid versus diploid yeast cells, (2) the types and levels of damage induced, or (3) the specific genomic context of the locus monitored; e.g., its proximity to a telomere or centromere. Indeed, the nuclear organization of chromosomes and chromosome territories do seem to affect the efficiency of repair by HR (Agmon et al. 2013). In this study, we set out to resolve this discrepancy. We found that checkpoint kinases in yeast induce a genome-wide alteration in chromatin structure, which is manifested as enhanced locus mobility, even in the absence of DNA damage. This increase in mobility appears to be driven by the INO80 nucleosome remodeling complex.
ResultsDNA damage increases global chromatin mobility independently of Rad51
Here we exploited single-particle tracking of fluorescently tagged genomic loci in Saccharomyces cerevisiae to quantify the mobility of the chromatin fiber in vivo. To this end, we recorded 3D image stacks on a spinning-disc confocal microscope every 1.5 sec during 5 min. The images were then deconvolved (Ponti et al. 2007) and projected onto a two-dimensional (2D) plane (Fig. 1A). Using the ImageJ plug-in, spots were tracked with respect to the center of the nucleus (SpotTracker) (Sage et al. 2005), and the X and Y coordinates of the spot as well as the center of the nucleus were determined in each of the 200 images of a typical time-lapse movie. From these values, we calculated the mean-squared displacement (MSD = 〈Xt − Xt + Δt〉2, where X is the position of a spot at time t). From the MSD plot, we derived the radius of constraint (Rc; the square root of the plateau of the MSD curve multiplied by 5/4) (Meister et al. 2010; Neumann et al. 2012), which is a robust measurement of locus confinement given that thousands of data points were averaged in each graph (Meister et al. 2010).
DNA damage causes a global increase in chromatin mobility. (A, top) Schematic of strain GA-6879 showing the tracked met10∷TetO Locus. (Bottom) S-phase cell (GA-6879) after 1 h of Zeocin or undamaged. Shown is nuclear pore (GFP-Nup49), TetR-mCherry locus, and Rad52-YFP damage focus. Bar, 2 μm. (Right) Image stack is projected onto a 2D plane for tracking. (B) Western blot of Rad53 and H2A phosphorylation after 1-h Zeocin treatment of GA-6879. Actin and Mcm2 were used as loading controls. (C) Rad52-YFP foci accumulation after 1-h treatment with Zeocin or induction of pGAL-ECORI. Numbers of nuclei scored: 510 (undamaged), 311 (50 μg/mL Zeocin), 197 (100 μg/mL Zeocin), 395 (250 μg/mL Zeocin), 403 (375 μg/mL Zeocin), 102 (0 h pGAL-ECORI), and 124 (1-h pGAL-ECORI). (D–H) MSD plots of TetO-tagged met10 in S-phase cells. (D) Without damage (blue) or incubated with Zeocin at 50, 100, or 250 μg/mL. The inset graph represents the first five time intervals to generate the initial slope. The percentage nuclear volume (nvol) explored is indicated. (E) MSD plot of LacO-tagged ZWF1 (brown) and Rad52-YFP (green) compared with met10∷TetO either with 1 h of Zeocin (red) or without (blue). ZWF1 and I-SceI-induced Rad52-YFP from Dion et al. (2012). (F) MSD plots of met10∷TetO after growth in 2% galactose for 1 h either with pGAL-ECORI (red) or without (blue). (G) MSD plots of met10∷TetO during S phase in a pseudodiploid strain (GA-7591) with an extra copy of MATα either without damage (blue) or after 1 h of 250 μg/mL Zeocin (red); MSD of met10∷TetO in haploid GA-6879 (gray). (H) MSD of met10∷TetO in rad51Δ cells (GA-7550) without damage (blue) or after 1 h of 250 μg/mL Zeocin (red) or in wild-type cells (GA-6879) in 250 μg/mL Zeocin (light gray). All error bars show the SEM. The numbers of movies tracked and parameters are given in Table 1.
We tracked genomic loci that were tagged with mCherry-TetR (e.g., the met10 locus on Chr VI) (Fig. 1A) or GFP-LacI (e.g., ATG2 on the long arm of Chr XIV or PES4 and HXK1, both on Chr VI). These haploid cells also express yellow fluorescent protein (YFP) fused to Rad52, which is fully functional for HR (Supplemental Table S1; Lisby et al. 2004). To ensure that we tracked an undamaged locus, time-lapse movie data was used only if the genomic tagged locus did not colocalize with Rad52-YFP. Moreover, we confirmed that our imaging regime itself does not induce damage by showing that cells divide with normal kinetics after imaging (Supplemental Fig. S1; Dion et al. 2012).
In line with earlier studies, we measured an Rc of 0.41 μm for the met10 locus in haploid S-phase cells grown on glucose in the absence of damage (Dion et al. 2012; Mine-Hattab and Rothstein 2012). This value indicates that the locus can sample ∼9.5% of the nuclear volume (nvol) (Fig. 1D). We note that movement is higher in G1-phase cells as compared with S-phase cells (Heun et al. 2001) and that the Rc of undamaged loci can range from 0.4 to 0.6 μm, depending on the chromosomal location of the tagged locus (Table 1; Supplemental Fig. S2; Heun et al. 2001; Gartenberg et al. 2004; Bystricky et al. 2009).
Summary of MSD results presented in this study
To examine the movement of the undamaged locus in a cell responding to DNA damage, we next exposed cells bearing the tagged met10 locus to increasing doses of Zeocin, a copper-chelated glycopeptide antibiotic that induces both DSBs and single-strand nicks, in the ratio of ∼1:9 (Povirk 1996; Burger 1998). After 1 h of treatment with Zeocin, we assessed DDR activation by scoring for phosphorylated forms of H2A and Rad53 (Fig. 1B) and the frequency of Rad52-YFP foci formed (Fig. 1C). Surprisingly, the tagged and undamaged met10 locus exhibited an increase in mobility after Zeocin treatment, increasing with the concentration of Zeocin used and reaching a maximum of nvol searched of ∼34% at 250 μg/mL Zeocin (Fig. 1D). The tracked locus did not colocalize with Rad52-YFP, although each nucleus analyzed had a Rad52-YFP focus elsewhere at both the beginning and end of the movie. In agreement with our previous report, low-level Zeocin (50 μg/mL) did not increase mobility at an undamaged site (Dion et al. 2012), but inducing more damage with higher levels of Zeocin did (Fig. 1C,D).
We then examined whether the increase in chromatin mobility was locus-dependent by scoring the movement of three LacO-tagged loci (PES4, ATG2, and HXK1) on different chromosomes. Each showed an equivalent increase in mobility upon Zeocin treatment, although their chromosomal locations varied significantly, with one being subtelomeric (HXK1) (Supplemental Fig. S2A,B). The undamaged loci tested all showed an increase in mobility, albeit slightly less than that scored at the site of damage (Fig. 1E; Table 1).
To rule out that the increase was due to secondary effects unique to Zeocin, we expressed the restriction enzyme EcoRI in yeast cells bearing the tagged met10 locus. EcoRI induces DSBs that can be monitored through Rad52-YFP foci formation (Fig. 1C). We found that induction of EcoRI also increased the Rc of the undamaged met10 locus from 0.39 μm to 0.51 μm (from ∼9% to ∼18% nvol) (Fig. 1F). This increase is more modest than that scored at 250 μg/mL Zeocin and correlates with the reduced number of Rad52 foci formed upon EcoRI induction (Fig. 1C).
Given that specialized chromosomal domains (e.g., telomeres or centromeres) have been implicated in both constraining movement (Heun et al. 2001) and limiting recombination (Agmon et al. 2013), we speculated that the general increase in chromatin movement might arise from the release of perinuclear chromosomal anchorage points. To test this, we scored whether the subtelomeric, LacO-tagged locus HXK1 loses its anchorage and moves away from the nuclear envelope after DNA damage. Although HXK1 shows increased mobility, it does not lose its perinuclear localization upon incubation with Zeocin, suggesting that movement along the nuclear envelope increases (Supplemental Fig. S2D; Supplemental Table S2). This agrees with earlier results showing that yeast telomeres remained associated with the nuclear envelope after single DSB induction (Martin et al. 1999). The observed increase in chromatin mobility in the presence of damage is therefore not a passive event resulting from a loss of telomere anchoring.
Chromatin mobility in general depends on ATP (Heun et al. 2001; Weber et al. 2012), and the ionophore carbonyl cyanide m-chlorophenyl hydrazine (CCCP) was shown to reduce the mobility of chromosomal loci as well as of an excised chromatin ring (Heun et al. 2001; Gartenberg et al. 2004). Consistently, we found that preincubation of cells with 40 μM CCCP suppressed the general chromatin mobility induced by DNA damage (Supplemental Fig. S3A; Supplemental Table S2), suggesting a role for an active mechanism in this process.
The HR repair factors Rad51 and Rad54 are needed to increase mobility at the site of damage itself (Dion et al. 2012; Mine-Hattab and Rothstein 2012). We asked whether repair factors are similarly required for the increased mobility scored at undamaged loci, provoked by DNA lesions elsewhere in the genome. This is not the case; the Rc of the met10 locus started at the same point in rad51Δ and RAD51+ cells and increased after incubation with 250 μg/mL Zeocin to equal degrees, reaching ∼34% of nvol in both of these otherwise isogenic haploid strains (Fig. 1H).
In diploid yeast, Rad51 is hyperactive (Morgan et al. 2002), and it was suggested (Ira and Hastings 2012) that this hyperactivity might be responsible for the enhanced mobility observed at unbroken sites in diploid cells. To test this hypothesis, we integrated an extra copy of MAT bearing the opposite mating type information (MATα) into our wild-type haploid strain (MATa). Rad51 is hyperactivated in this pseudodiploid strain, as in diploid yeast (Morgan et al. 2002; Haber 2012). Tracking of the met10 locus yielded an increase in Rc upon treatment with Zeocin (from 0.43 μm to 0.65 μm) in the pseudodiploid that was indistinguishable from that in the haploid strain (Fig. 1G). Thus, the increase in chromatin mobility in response to damage is independent of the Rad51 hyperactivity associated with diploid cells and occurs equally in haploid and diploid strains.
The stage of the cell cycle has a clear effect on chromatin mobility, since in S-phase, movement is much decreased compared with G1 (Heun et al. 2001) due to constraint from sister–sister cohesion (Dion et al. 2013). We were unable to test G1-phase cells, since we used Rad52 focus formation as a marker for damage, and G2-phase nuclei often have distorted nuclear shapes that interfere with accurate tracking. We did score for differences in mobility between early and mid-S phase at met10 and found a similar increase in response to damage at both stages (Supplemental Fig. S3B,C; Supplemental Table S2). Given that met10 is early-replicating, it is likely that the tracked loci are replicated and have cohesin loaded (Dion et al. 2013).
The DDR increases global chromatin mobility
An earlier study showed that loss of Mec1, the homolog of human ATR kinase, compromised the increased mobility scored for an induced DSB, while loss of Rad53 (CHK2) did not (Dion et al. 2012). It was therefore of interest to test whether checkpoint kinase activation was necessary for the increase in chromatin mobility genome-wide. This was scored in a sml1Δ background because ablation of Sml1 up-regulates dNTP synthesis and suppresses the lethality of mec1Δ or rad53Δ strains (Zhao et al. 1998). Upon tracking met10, we found that the background chromatin mobility (Rc = 0.32 μm) in mec1Δsml1Δ cells was nearly unchanged after 1 h of exposure to Zeocin (0.36 μm) (Fig. 2A), and loss of the DDR effector kinase Rad53 completely ablated the damage-induced increase in global chromatin mobility (Fig. 2B). Consistent with earlier observations, sml1 deletion itself has a partial phenotype: The increase in general chromatin mobility after damage in a sml1Δ strain reaches 0.47 μm instead of the 0.65 μm of wild-type cells (Fig. 2C). This lower increase in mobility may reflect the fact that sml1Δ strains have altered dNTP levels, although overexpression of the factor Sml1, which down-regulates Rnr1, did not have a similar effect (data not shown). Nonetheless, mec1Δsml1Δ or rad53Δsml1Δ cells fail to show a damage-induced, global increase in chromatin mobility, while sml1Δ cells do.
Checkpoint proteins Mec1, Rad9, and Rad53 are essential for damage-induced increases in global chromatin mobility. (A–D) MSD of met10∷TetO in S-phase wild-type GA-6879 after 1 h of 250 μg/mL Zeocin (gray). MSD plots of the same locus without damage (blue) and Zeocin-treated (red) as above in the following backgrounds: mec1Δsml1Δ (GA-7556) (A), rad53Δsml1Δ (GA-7552) (B), sml1Δ (GA-7553) (C), and rad9Δ (GA-7555) (D). The error bars show the SEM. The numbers of movies tracked and parameters are given in Table 1. (E) Western blots showing checkpoint activation after 1 h of Zeocin (250μg/mL) as in Figure 1B, in wild-type (wt) (GA-6879), sml1Δ (GA-7553), mec1Δsml1Δ (GA-7556), hta1/2 S129* (GA-4188), rad9Δ (GA-7555), and rad53Δsml1Δ (GA-7552) cells. (F) Scheme of key kinases and regulators in the DNA damage checkpoint.
Next, we tested the role of factors downstream from Mec1. One key target of Mec1 that contributes to both Rad53 activation and the downstream response is the BRCT-containing protein Rad9 (Vialard et al. 1998). Whereas RAD9 deletion alone increased chromatin mobility in the undamaged state, it completely blocked the general chromatin mobility increase induced by Zeocin (Fig. 2D). By Western blot analysis, we confirmed that checkpoint activation was compromised in all of the DDR mutants tested. Interestingly, although mobility did not increase in the rad9Δ mutant, H2A phosphorylation was comparable with that in wild-type cells, indicating that this Mec1 phosphorylation target (γH2AX in mammals) does not induce chromatin movement (Fig. 2E), while the DNA checkpoint signaling cascade, which ends with Rad53 activation, does (Fig. 2F). Interestingly, Rad53, unlike Mec1/Ddc2, has a dispersed nuclear localization, allowing it to propagate changes throughout the nucleoplasm (Melo et al. 2001).
Checkpoint activation without DNA damage increases global chromatin mobility
The question remained whether checkpoint activation is sufficient to enhance chromatin movement in the absence of damage. To examine this, we constructed a strain containing GFP- and LacI-tagged versions of both Ddc1 and Ddc2, whose artificial juxtaposition is sufficient to initiate a damage-independent checkpoint response (Bonilla et al. 2008). Upon expression from a galactose-inducible promoter, these two proteins bound the integrated LacO array at PES4 near MET10 and created a GFP focus that could be tracked (Fig. 3A,B). Upon induction of Ddc1/Ddc2-GFP-LacI, the checkpoint was activated as indicated by H2A phosphorylation (Fig. 3C), and the Rc of the tagged PES4 locus increased from 0.38 μm to 0.5 μm (Fig. 3B,D).
Checkpoint activation is sufficient to enhance both local and global chromatin mobility. (A) Schematic of strain GA-7676 showing the tracked locus at PES4. (B) Cartoon illustrating in cis tracking of Ddc1/Ddc2-GFP-LacI foci at PES4 or in trans tracking of mCh-TetO at LEU2. (C) Western blot for H2A S129 phosphorylation after galactose induction of pGAL-Ddc1/2-GFP-LacI in GA-7676 or a wild-type (wt) strain, GA-1461. (D) MSD plots of LacO-tagged PES4 during S phase after 1 h on galactose in cells expressing Ddc1/Ddc2-GFP-LacI (green; GA-7676) or GFP-LacI (brown; GA-1461). (E) TetR-mCherry at LEU2 in a strain containing GFP-LacI (blue; GA-8088), expressing Ddc1/Ddc2-GFP-LacI in the presence (red; GA-8023) or absence (purple; GA-8158) of PES4∷lacO. The numbers of movies tracked and parameters are given in Table 1.
To see whether checkpoint activation in the absence of DNA damage suffices to trigger the general increase in chromatin mobility, we monitored the LEU2∷TetO locus in these same conditions. Whereas overexpression of Ddc1/2-GFP-LacI without a LacO array at PES4 did not cause an increase in mobility at LEU2, by targeting Ddc1/Ddc2 to PES4∷LacO, not only did the PES4 locus increase mobility, but also LEU2, albeit to a lower extent (Fig. 3B,E). Given that we saw a dose-dependent increase in mobility with Zeocin, we assume that this more modest increase reflects the efficiency of checkpoint activation. We conclude that DNA damage checkpoint activation is sufficient, even in the absence of DNA damage, to increase both local and global chromatin mobility.
Intact INO80 remodeler complex is required to increase chromatin mobility in trans
Several chromatin remodeling enzymes are targets of Mec1 and the checkpoint response (Morrison et al. 2007; Smolka et al. 2007). Since INO80 is known to increase the mobility of a locus to which it is targeted (Neumann et al. 2012) and has recently been implicated in various DSB repair events (Agmon et al. 2013), we hypothesized that it may have a role in increasing global chromatin mobility. We found that strains lacking either Arp5 or Arp8 abolish increased chromatin mobility after Zeocin treatment (Fig. 4A,B). Complementation of arp8Δ with wild-type Arp8 under its endogenous promoter restores chromatin mobility after Zeocin-induced damage just as it restores growth on Zeocin-containing plates (Fig. 4B,C). Importantly, we show that arp8Δ does not impair checkpoint activation on Zeocin (Fig. 4D), confirming previously published results on hydroxyurea (van Attikum et al. 2004). In contrast to INO80, we found that the Chd1 and Swr1 chromatin remodelers do not have an effect on global chromatin mobility, nor does the sister chromatid cohesion-promoting factor Tof1 (Supplemental Fig. S4; Supplemental Table S2).
The INO80 complex is required to increase global chromatin mobility. MSD plots of met10∷TetO during S phase in wild-type cells after 1 h in 250 μg/mL Zeocin (gray) and of the same locus without damage (blue) and treated with Zeocin 250 μg/mL (red) in the following mutant backgrounds: arp5Δ (GA-8202) (A) and arp8Δ (GA-8132) (B). (C) Serial dilution (10×) showing complementation of arp8Δ with p416-ARP8-URA3 (Shen et al. 2003). (D) Western blot of γH2A accumulation after Zeocin treatment. (E) MSD plot of LacO-tagged PES4 during S phase after 1 h on galactose in cells expressing Ddc1/Ddc2-GFP-LacI (red; GA-8203) or GFP-LacI (blue; GA-8204) in arp8Δ versus wild-type (green) backgrounds. (F) TetR-mCherry at LEU2 in a strain expressing Ddc1/Ddc2-GFP-LacI (red; GA-8203) or expressing Ddc1/Ddc2-GFP-LacI in the absence of PES4∷lacO (blue; GA-8204) versus wild-type (gray). The error bars in all panels show the SEM. The numbers of movies tracked and parameters are given in Table 1.
To confirm that the INO80 complex is needed for global chromatin mobility in direct response to Mec1 activation, we targeted Ddc1/Ddc2 to activate movement in the absence of damage. Under these conditions, Arp8 was partially required to increase the mobility of a locus in cis (analogous to the partial effect of arp8Δ on DSB mobility) (Neumann et al. 2012), and its loss completely compromised the increased mobility of an undamaged locus (LEU2) (Fig. 4E,F). We conclude that the INO80 complex acts downstream from checkpoint activation and is needed to increase global chromatin mobility even when the checkpoint is activated artificially, without widespread DNA damage (Bonilla et al. 2008). The model in Figure 5 illustrates the two pathways that lead to enhanced chromatin mobility—one acting locally, and the second affecting chromatin mobility globally—both showing dependence on INO80.
Model of the factors influencing local and global chromatin mobility in response to a DSB. Resection at a DSB leads to accumulation of ssDNA and binding of RPA. This signals the recruitment of repair proteins such as Rad51, Rad52, and finally, Rad54 and activates the DDR. At a DSB, Rad51, Rad54, Rad9, and Mec1 are required to increase DSB mobility (Dion et al. 2012). Global chromatin movement is enhanced by the checkpoint kinase activation and requires Rad53. Downstream from checkpoint activation, the INO80 subunits Arp5 and Arp8 are required to increase mobility globally and have a partial effect on the increase scored at a DSB. Increased mobility of both damaged and undamaged loci would facilitate contact between the DSB and ectopic sites of homology for HR.
Discussion
This study resolves the discrepancy between previous studies (Dion et al. 2012; Mine-Hattab and Rothstein 2012) with respect to chromatin mobility at undamaged loci in cells exposed to damage. Chromatin mobility does indeed increase globally, yet the increase appears to require a threshold level of damage, which correlates with the induction of the DDR through Mec1 and Rad53 kinases. Low-level damage (e.g., after incubation with 50μg/mL Zeocin) does not provoke a detectable increase in general chromatin movement. We rule out other explanations for the discrepancy, such as cell ploidy or the type of damage induced. Moreover, we can exclude that the global increase in movement arises from chromosome fragmentation given that we scored a checkpoint kinase-dependent increase in mobility in the absence of damage. Finally, by scoring multiple loci, including telomeres, we rule out that the effect depends on the chromosomal context of the locus monitored.
Our studies suggest that changes in chromatin structure that lead to increased mobility in response to DNA damage are different at the site of damage as compared with an undamaged locus. Enhanced mobility in cis requires the repair factors Rad51 and Rad54 but is independent of Rad53 kinase activation. Global chromatin mobility increases require the downstream checkpoint kinase Rad53 but not the repair protein Rad51. We note that loss of Arp8 has only a partial effect on the increased mobility of a DSB, while it is essential for the global increase (Fig. 4B,D,F).
We speculate that differential control of chromatin mobility at damaged and undamaged sites may be advantageous to the cell. The enhanced movement of a DSB enhances the probability of harmful translocations or deletions even as it promotes a homology search for HR-mediated repair. The movement, like the checkpoint activation itself, is dependent on the level of damage, consistent with the two being linked. Mec1–Ddc2 activation requires both a threshold and a specific processing event at damage, which may be used to determine in which circumstances global chromatin movement should be enhanced to maximize recombinational repair. The genomic tradeoff for movement is likely to be an elevated risk of deleterious recombination (Dion and Gasser 2013; Seeber et al. 2013).
Several observations suggest that our insights are likely to be relevant to mammalian genomes. For one, recurrent translocations in B lymphocytes occur proportionally to DNA damage (Hakim et al. 2012), and down-regulation of 53BP1 (Rad9) reduces both chromatin mobility and chromosomal end-to-end fusions (Dimitrova et al. 2008). Furthermore, the mobility of arrays that generate translocations in mammalian cells is significantly higher than that of arrays not producing translocations (Roukos et al. 2013). We propose that lesions that do not require a long-range search for a homologous donor and those that do not activate a checkpoint response also fail to trigger a general increase in chromatin mobility.
In summary, we show here that a DNA damage-triggered kinase response controls chromatin organization, with the likely effector being the INO80 complex (Fig. 5). We monitor this as expanded Rc values for undamaged fluorescently tagged chromatin loci. This may reflect local changes in chromatin structure (Neumann et al. 2012) or alterations in the long-range folding of chromatin genome-wide. When S-phase damage is repaired from the sister chromatid, damage movement is constrained and not enhanced, which can be overcome by destruction of cohesin (Dion et al. 2013). Thus, we speculate that the checkpoint kinases Mec1–Ddc2 and Rad53 modify INO80 and possibly cohesin in response to damage to regulate global chromatin mobility differentially during the DDR.
Materials and methodsYeast growth conditions and plasmids
Yeast strains used in this study were W303-derived (see Supplemental Table S1). Yeast growth was at 30°C, and imaging was at 25°C. Zeocin exposure experiments were done in synthetic complete (SC) medium with 1-h incubations with drugs prior to microscopy or other assays, which were performed in fresh SC medium. Precise conditions are in the Supplemental Material. pGAL-ECOR1 plasmid was a gift of Dr. P. Schär, and the pseudodiploid strain was constructed by integrating a MATα plasmid at URA3 (gift of Dr. S. Marcand).
Microscopy, movie analysis, and zoning assay
Live microscopy used an Olympus IX81 microscope equipped with a Yokogawa CSU-X1 scan head, an EM-CCD Cascade II (Photometrics), an ASI MS-2000 Z-piezo stage, and a PlanApo ×100, NA 1.45 total internal reflection fluorescence microscope oil objective. For excitation and exposure times, see the Supplemental Material.
Time-lapse image stacks were analyzed as in Dion et al. (2012) using a custom-made Fiji plug-in (Sage et al. 2005). Analysis of locus position was performed with the zoning assay described in Meister et al. (2010), and phototoxicity was tested by exposing wild-type cells (GA-6879) to standard imaging conditions and then following outgrowth for 5 h by morphological analysis, comparing them with unexposed cells.
Western blotting
DDR activation was scored by Western blotting TCA-precipitated proteins separated on a SDS-PAGE gel (Invitrogen). The antibodies used are noted in the Supplemental Material.
Acknowledgments
We thank C. Horigome, U. Rass, I. Marcomini, P. Zeller, M. Hauer, and M. Di Pietro for critical reading and preparation of the manuscript and figures; P. Schär, S. Marcand, and D.P. Toczyski for reagents; and the Friedrich Miescher Institute Facility for Advanced Imaging and Microscopy for technical assistance. The Gasser laboratory is supported by FP7 Marie Curie Network Nucleosome 4D, the Swiss National Science Foundation National Centre for Competence in Research, “Frontiers in Genetics,” and the Friedrich Miescher Institute for Biomedical Research. A.S., V.D., and S.M.G. planned experiments and wrote the paper, and A.S. and V.D. performed experiments.
Supplemental material is available for this article.
Article published online ahead of print. Article and publication date are online at http://www.genesdev.org/cgi/doi/10.1101/gad.222992.113.
ReferencesAgmonN, LiefshitzB, ZimmerC, FabreE, KupiecM2013Effect of nuclear architecture on the efficiency of double-strand break repair.
15: 694–69923644470BonillaCY, MeloJA, ToczyskiDP2008Colocalization of sensors is sufficient to activate the DNA damage checkpoint in the absence of damage.
30: 267–27618471973BurgerRM1998Cleavage of nucleic acids by bleomycin.
98: 1153–117011848928BystrickyK, Van AttikumH, MontielMD, DionV, GehlenL, GasserSM2009Regulation of nuclear positioning and dynamics of the silent mating type loci by the yeast Ku70/Ku80 complex.
29: 835–84819047366DimitrovaN, ChenYC, SpectorDL, de LangeT200853BP1 promotes non-homologous end joining of telomeres by increasing chromatin mobility.
456: 524–52818931659DionV, GasserSM2013Chromatin movement in the maintenance of genome stability.
152: 1355–136423498942DionV, KalckV, HorigomeC, TowbinBD, GasserSM2012Increased mobility of double-strand breaks requires Mec1, Rad9 and the homologous recombination machinery.
14: 502–50922484486DionV, SeeberA, GasserSM2013Cohesin and the nucleolus constrain the mobility of spontaneous repair foci.
(in press)ForgetAL, KowalczykowskiSC2012Single-molecule imaging of DNA pairing by RecA reveals a three-dimensional homology search.
482: 423–42722318518GartenbergMR, NeumannFR, LarocheT, BlaszczykM, GasserSM2004Sir-mediated repression can occur independently of chromosomal and subnuclear contexts.
119: 955–96715620354GehlenL, GasserSM, DionV2011How broken DNA finds its template for repair: A computational approach.
191: 20–29HaberJE2012Mating-type genes and MAT switching in Saccharomyces cerevisiae.
191: 33–6422555442HakimO, ReschW, YamaneA, KleinI, Kieffer-KwonKR, JankovicM, OliveiraT, BothmerA, VossTC, Ansarah-SobrinhoC, 2012DNA damage defines sites of recurrent chromosomal translocations in B lymphocytes.
484: 69–7422314321HeunP, LarocheT, ShimadaK, FurrerP, GasserSM2001Chromosome dynamics in the yeast interphase nucleus.
294: 2181–218611739961IraG, HastingsPJ2012DNA breakage drives nuclear search.
14: 448–45022552144KrawczykPM, BorovskiT, StapJ, CijsouwT, CateRT, MedemaJP, KanaarR, FrankenNA, AtenJA2012Chromatin mobility is increased at sites of DNA double-strand breaks.
125: 2127–213322328517LisbyM, BarlowJH, BurgessRC, RothsteinR2004Choreography of the DNA damage response: Spatiotemporal relationships among checkpoint and repair proteins.
118: 699–71315369670MartinSG, LarocheT, SukaN, GrunsteinM, GasserSM1999Relocalization of telomeric Ku and SIR proteins in response to DNA strand breaks in yeast.
97: 621–63310367891MeisterP, GehlenLR, VarelaE, KalckV, GasserSM2010Visualizing yeast chromosomes and nuclear architecture.
470: 535–56720946824MeloJA, CohenJ, ToczyskiDP2001Two checkpoint complexes are independently recruited to sites of DNA damage in vivo.
15: 2809–282111691833Mine-HattabJ, RothsteinR2012Increased chromosome mobility facilitates homology search during recombination.
14: 510–51722484485MorganEA, ShahN, SymingtonLS2002The requirement for ATP hydrolysis by S. cerevisiae Rad51 is bypassed by mating-type heterozygosity or RAD54 in high copy.
22: 6336–634312192033MorrisonAJ, KimJA, PersonMD, HighlandJ, XiaoJ, WehrTS, HensleyS, BaoY, ShenJ, CollinsSR, 2007Mec1/Tel1 phosphorylation of the INO80 chromatin remodeling complex influences DNA damage checkpoint responses.
130: 499–51117693258NeumannFR, DionV, GehlenLR, Tsai-PflugfelderM, SchmidR, TaddeiA, GasserSM2012Targeted INO80 enhances subnuclear chromatin movement and ectopic homologous recombination.
26: 369–38322345518PontiA, SchwarbP, GulatiA, BäckerV2007Huygens remote manager.
9: 57–58PovirkLF1996DNA damage and mutagenesis by radiomimetic DNA-cleaving agents: Bleomycin, neocarzinostatin and other enediynes.
355: 71–898781578RenkawitzJ, LademannCA, KalocsayM, JentschS2013Monitoring homology search during DNA double-strand break repair in vivo. . 50: 261–27223523370RoukosV, VossTC, SchmidtCK, LeeS, WangsaD, MisteliT2013Spatial dynamics of chromosome translocations in living cells.
341: 660–66423929981SageD, NeumannFR, HedigerF, GasserSM, UnserM2005Automatic tracking of individual fluorescence particles: Application to the study of chromosome dynamics.
14: 1372–138316190472SeeberA, HauerM, GasserSM2013Nucleosome remodelers in double-strand break repair.
23: 174–18423352131ShenX, RanalloR, ChoiE, WuC2003Involvement of actin-related proteins in ATP-dependent chromatin remodeling.
12: 147–15512887900SmolkaMB, AlbuquerqueCP, ChenSH, ZhouH2007Proteome-wide identification of in vivo targets of DNA damage checkpoint kinases.
104: 10364–1036917563356van AttikumH, FritschO, HohnB, GasserSM2004Recruitment of the INO80 complex by H2A phosphorylation links ATP-dependent chromatin remodeling with DNA double-strand break repair.
119: 777–78815607975VialardJE, GilbertCS, GreenCM, LowndesNF1998The budding yeast Rad9 checkpoint protein is subjected to Mec1/Tel1-dependent hyperphosphorylation and interacts with Rad53 after DNA damage.
17: 5679–56889755168WeberSC, SpakowitzAJ, TheriotJA2012Nonthermal ATP-dependent fluctuations contribute to the in vivo motion of chromosomal loci.
109: 7338–734322517744WilsonJH, LeungWY, BoscoG, DieuD, HaberJE1994The frequency of gene targeting in yeast depends on the number of target copies.
91: 177–1818278360ZhaoX, MullerEG, RothsteinR1998A suppressor of two essential checkpoint genes identifies a novel protein that negatively affects dNTP pools.
2: 329–3409774971oai:pubmedcentral.nih.gov:37924772014-03-15genesdevpmc-openGenes DevGenes DevGADGenes & Development0890-93691549-5477Cold Spring Harbor Laboratory PressPMC3792477PMC379247737924772406576724065767871166010.1101/gad.223396.113Research PaperExchange of associated factors directs a switch in HBO1 acetyltransferase histone tail specificityLalonde et al.Histone tail selection by MYST-associated factorsLalondeMarie-Eve1AvvakumovNikita1GlassKaren C.27JoncasFrance-Hélène1SaksoukNehmé1HollidayMichael3PaquetEric1YanKezhi5TongQiong2KleinBrianna J.2TanSong4YangXiang-Jiao56KutateladzeTatiana G.23CôtéJacques18Laval University Cancer Research Center, Hôtel-Dieu de Québec (CHUQ), Quebec City, Québec G1R 2J6, Canada;Department of Pharmacology,Molecular Biology Program, University of Colorado School of Medicine, Aurora, Colorado 80045, USA;Center for Gene Regulation, Department of Biochemistry and Molecular Biology, The Pennsylvania University, University Park, Pennsylvania 16802, USA;The Rosalind and Morris Goodman Cancer Research Center, Department of Biochemistry, McGill University, Montreal, Québec H3A 1A1, Canada;Department of Medicine, McGill University Health Center, Montreal, Québec H3A 1A1, Canada
Present address: Department of Pharmaceutical Sciences, Albany College of Pharmacy and Health Sciences, Colchester, VT 05446, USA.
Histone acetyltransferases (HATs) assemble into multisubunit complexes that target distinct lysine residues on nucleosomal histones. Here, Lalonde et al. characterize native HAT complexes assembled by the BRPF family of scaffold proteins and show that the specificity of the HBO1 HAT changes from the H3 to the H4 tail upon association with either JADE or BRPF1 protein. A region within scaffold proteins responsible for histone tail selection on chromatin is also identified. This study provides critical new insight into how histone-modifying enzyme substrate specificity is determined.
Histone acetyltransferases (HATs) assemble into multisubunit complexes in order to target distinct lysine residues on nucleosomal histones. Here, we characterize native HAT complexes assembled by the BRPF family of scaffold proteins. Their plant homeodomain (PHD)–Zn knuckle–PHD domain is essential for binding chromatin and is restricted to unmethylated H3K4, a specificity that is reversed by the associated ING subunit. Native BRPF1 complexes can contain either MOZ/MORF or HBO1 as catalytic acetyltransferase subunit. Interestingly, while the previously reported HBO1 complexes containing JADE scaffold proteins target histone H4, the HBO1–BRPF1 complex acetylates only H3 in chromatin. We mapped a small region to the N terminus of scaffold proteins responsible for histone tail selection on chromatin. Thus, alternate choice of subunits associated with HBO1 can switch its specificity between H4 and H3 tails. These results uncover a crucial new role for associated proteins within HAT complexes, previously thought to be intrinsic to the catalytic subunit.
In eukaryotic cells, DNA wraps around histone octamers to form nucleosomes, the basic units of chromatin. This structural organization not only allows compaction of the DNA in the nuclei, but also regulates diverse cellular processes such as DNA repair, transcription, and replication. This regulation is mainly exerted via the different post-translational modifications of the protruding N-terminal tail residues of histones, such as acetylation, methylation, phosphorylation, ubiquitination, and sumoylation. These modifications act as docking sites for different protein effectors involved in many cell pathways (Musselman et al. 2012). Acetylation of histone lysines is one of the best-characterized functional modifications. It is deposited by histone acetyltransferases (HATs) by transferring the acetyl group from acetyl-CoA on the ε-amino group of lysine residues. Although histone acetylation has mostly been associated with transcriptional activators and regulators, it has also been implicated in other processes, such as DNA repair and replication and mRNA splicing (Shahbazian and Grunstein 2007; van Attikum and Gasser 2009; de Almeida and Carmo-Fonseca 2012).
The MYST (MOZ, Ybf2/Sas3, Sas2, and Tip60) family of acetyltransferases is composed of evolutionarily conserved enzymes that are assembled into multisubunit protein complexes. They acetylate histone tails within chromatin but also target nonhistone substrates in both humans and yeast (Sapountzi and Cote 2011). We previously purified several native MYST complexes and found that they are based on a tetrameric core structure associated with an ING tumor suppressor subunit, the Eaf6 subunit, and an EPC (enhancer of polycomb)-related scaffold subunit (Doyon et al. 2006; Saksouk et al. 2009; Avvakumov et al. 2012). Four human MYST complexes assemble in such a manner; namely, Tip60 (KAT5), HBO1 (MYST2 and KAT7), MOZ (MYST3 and KAT6A), and MORF (MYST4 and KAT6B) (Fig. 1A). These enzymes have been shown to play crucial roles in DNA repair, recombination, and replication as well as in transcription activation (Avvakumov and Cote 2007b), which in turn regulates developmental processes (Voss and Thomas 2009) and is involved in leukemia and several genetic diseases (Avvakumov and Cote 2007a; Yang and Ullah 2007; Voss and Thomas 2009).
Characterization of the two PHD fingers in the PZP domain of BRPF1. (A) Subunit organization of human MYST acetyltransferase complexes used in this study. The core subunits have a scaffold protein (JADE, BRPF, or EPC), an ING tumor suppressor protein (ING3, ING4, or ING5), and a catalytic enzyme protein (Tip60, HBO1, or MOZ/MORF). (B) Schematic representation of the conserved protein domains found in the scaffold subunits of MYST–ING HAT complexes (yeast and human). (C) The PHD1 finger of the PZP domain of BRPF1 recognizes unmethylated H3 in vitro, while the PHD2 finger shows interaction with H3 peptides independently of methylation status. Peptide pull-down assays with different biotinylated peptides and recombinant PHD fingers fused to GST were analyzed by Western blotting with anti-GST antibody (Western blott: α-GST). (D) Superimposed 1H,15N heteronuclear single quantum coherence (HSQC) spectra of BRPF1 PHD1, collected as unmodified H3 peptide, was titrated in. The spectrum is color-coded according to the protein–peptide ratio. (E) Binding affinities of the BRPF1 PHD1 finger for histone H3 peptides with different K4 methylation statuses were measured by tryptophan fluorescence. Numbers in parentheses represent the amino acid positions of histone H3 included in the peptides. (F) Superimposed 1H,15N HSQC spectra of BRPF1 PHD2 (collected as indicated) peptides or unlabeled BRPF1 PHD1 were added stepwise. The spectra are color-coded according to the protein–peptide ratio. (G) The PZP domain of BRPF1 binds to unmethylated histone H3K4 in vitro. Peptide pull-down assays with different biotinylated peptides and recombinant PZP domain fused to GST were analyzed by Western blot with α-GST antibody. The PHD1 dictates the specificity of the entire domain toward unmethylated H3K4. (H) Binding affinities of the PZP domain for the indicated H3 peptides were measured by tryptophan fluorescence. Numbers in parentheses represent the amino acid positions included in the peptides. (NB) No binding was detected.
As for several chromatin-related proteins, the MYST–ING complexes comprise diverse subunits that carry various histone recognition modules that bind to different post-translationally modified residues. Indeed, they contain chromodomains, bromodomains, and PWWP domains, but the most abundant histone-binding domain found within these proteins is the PHD (plant homeodomain) finger. PHD fingers form a versatile recognition motif family that binds to different histone modifications of the N-terminal tail of histone H3. Most of the PHD fingers can read the histone methylation state of H3K4 (me0 vs. me2/3), but some can recognize other histone H3 residues and/or modifications (Musselman et al. 2012). We recently characterized the binding properties of the different PHD fingers found in the MYST acetyltransferase complex HBO1 (Saksouk et al. 2009; Avvakumov et al. 2012). Even though the various domains found within the complex have unique recognition motifs, we showed that they cooperate to bind chromatin on H3K4me3, a mark found near the transcription start sites (TSSs) of actively transcribed genes (Shilatifard 2006). The two PHD fingers of the JADE1L scaffold subunit (Fig. 1B) work together to recognize unmethylated H3K4, while the PHD finger of the ING4 subunit directs the binding of the entire complex toward H3K4me3. Indeed, ING tumor suppressor proteins all contain a PHD finger that recognizes H3K4me3 (Pena et al. 2006; Shi et al. 2006; Champagne et al. 2008; Saksouk et al. 2009). The integration of the different binding properties of PHD fingers within the HBO1 complex allows for its tumor suppressor activity through its regulation of transcriptional activity, leading to the control of cell proliferation (Avvakumov et al. 2012).
The BRPF1 protein is the scaffold subunit of the MYST acetyltransferase complex MOZ/MORF (Doyon et al. 2006; Ullah et al. 2008). It has been shown to play a role in maintaining anterior HOX gene expression during zebrafish development and consequently in determining segmental identity (Laue et al. 2008). It is also thought to be part of the TrxG family of genes, which are important for maintaining active genes during development (Laue et al. 2008). The mutually exclusive catalytic subunits of the complex (MOZ and MORF) are also known to play a role in HOX gene expression and development in both mice and zebrafish (Miller et al. 2004; Voss et al. 2009; Qiu et al. 2012). Moreover, the MOZ acetyltransferase is frequently translocated in acute myeloid leukemia and is required for proper hematopoietic stem cell (HSC) proliferation (Katsumoto et al. 2006; Thomas et al. 2006; Perez-Campo et al. 2009; Aikawa et al. 2010). As for many chromatin-related proteins, BRPF1 contains a variety of histone recognition modules that can bind to different modifications (Fig. 1B). Its N-terminal region has two PHD domains linked by a Zn knuckle (PZP [PHD–Zn knuckle–PHD] domain), while the C terminus has both a bromodomain and a PWWP domain. This PWWP domain can bind to the H3K36me3 mark found on the coding regions of active genes (Vezzoli et al. 2010). As for other scaffold subunits within MYST complexes, BRPF1 also contains at its N terminus a region of homology with the EPcA domain found in EPC proteins (Stankunas et al. 1998; Avvakumov et al. 2012). We showed that two homology subdomains within EPcA serve as docking sites—one for the HAT subunit (domain I) and one for the hEaf6 and ING proteins (domain II) (Fig. 1B; Avvakumov et al. 2012).
In this study, we first dissected the molecular interactions of the PZP domain in BRPF1 for different chromatin modifications, showing that the first PHD domain acts dominantly over the second one in targeting unmethylated H3K4, but they function together to drive binding to chromatin. Moreover, we found that the different subunits of the complex are distributed genome-wide over H3K4me3-rich regions, targeted by the ING5 subunit. Finally, we discovered that the different scaffold subunits of MYST–ING complexes not only play an essential role in enabling chromatin acetylation, but also select which histone tail is modified. We propose a model in which the HBO1 acetyltransferase is competing for binding to either BRPF or JADE scaffold subunits in cells and where this differential association determines which histone tail is acetylated by the HAT on chromatin. While JADE directs the acetylation toward the H4 tail, BRPF1 targets H3 acetylation. Our results thus provide novel insights into the mechanism by which MYST acetyltransferases target chromatin acetylation from yeast to human cells and thus help us to understand how these enzymes regulate their different cellular functions.
ResultsThe PZP domain of BRPF1 binds to unmethylated histone H3K4
BRPF1 contains many potential chromatin recognition domains, as depicted in Figure 1B. We previously demonstrated that the JADE1L scaffold subunit found within the HBO1 HAT complex contains a PZP domain that recognizes unmethylated H3K4 (Saksouk et al. 2009; Avvakumov et al. 2012). Importantly, the first PHD domain of BRPF2 (also known as Brd1, a paralog of BRPF1) has recently been shown to bind the N-terminal tail of histone H3 (Qin et al. 2011; Liu et al. 2012). As BRPF1 also contains a similar tandem PHD finger module, we characterized its affinity for histone marks. We first examined the affinity of the two separated PHD fingers using peptide pull-down experiments with recombinant domains fused to GST and biotinylated peptides (Fig. 1C). The first PHD (PHD1) domain shows specific binding to unmethylated H3K4 peptides, while the second PHD (PHD2) domain shows some interaction with H3 peptides irrespective of their methylation status. These interactions were also analyzed using nuclear magnetic resonance (NMR). Substantial chemical shift changes observed in the spectra upon addition of the peptide indicated that the PHD1 finger recognizes the unmodified histone H3 tail (Fig. 1D). Moreover, methylation or acetylation of H3K9 did not affect binding of PHD1 to H3 (see Supplemental Fig. 1a,b). Likewise, binding of the PHD1 finger to the histone peptides mono-, di-, and trimethylated at Lys4 was examined by tryptophan fluorescence (Fig. 1E). A single methyl group attached to Lys4 reduced binding of the PHD1 finger by ∼10-fold, whereas affinity for H3K4me2 and H3K4me3 was dropped by ∼100-fold. Thus, methylation of Lys4 disrupts the association of BRPF1 PHD1 with histone H3. Additionally, PHD1 requires the first N-terminal amino acids of the H3 tail for proper binding (see Supplemental Fig. 1b). These results indicate that PHD1 associates with the extreme N terminus of histone H3, likely through hydrogen bonds and ionic interactions with Ala1, Arg2, and unmodified Lys4, as reported for other PHD fingers (Musselman et al. 2012), and that this binding is disrupted by methylation of Lys4. To demonstrate that PHD1 has a mechanism of interaction similar to those of other H3K4me0-binding PHD fingers, we mutated two key conserved residues and measured significantly lower binding in peptide pull-down assays (Supplemental Fig. 1c,d).
Conversely, neither the N-terminal part of the histone tail (residues 1–12) nor the downstream sequence (residues 9–19) was able to induce chemical shift changes in the 15N-labeled PHD2 finger of BRPF1, illustrating that this module alone is not capable on its own of specific binding to these sections of the histone H3 tail (Fig. 1F). We also found that PHD2 and PHD1 do not interact with each other, as no resonance perturbations were observed in the PHD2 finger during gradual addition of unlabeled PHD1 (Fig. 1F). On the other hand, nice chemical shift changes were obtained when incubating PHD2 with increasing amounts of deoxyribonucleoside monophosphates (dNMPs), suggesting that PHD2 could in fact interact with DNA (Supplemental Fig. 1e).
The two PHD fingers were then tested together as the full PZP module. Peptide pull-downs demonstrate that binding of the PZP to the N-terminal tail of H3 is inhibited by methylation of H3K4 (Fig. 1G), revealing PHD1 as a dominant recognition module over PHD2 within the PZP domain. To examine whether the Zn knuckle and/or the second PHD2 finger cooperate with PHD1 in the association with H3 by recognizing the histone sequence downstream from Lys9, binding affinity of BRPF1 PZP for a longer peptide was measured by tryptophan fluorescence. We found that PZP binds to the peptide containing residues 1–20 of histone H3un with a Kd of 12 μM (Fig. 1H; Supplemental Fig. 1f). This value was comparable with the Kd value measured for the interaction of a single PHD1 module with the short H3un peptide (6 μM) (Fig. 1E). Furthermore, the PZP domain and PHD1 alone exhibited similar affinities toward the short H3un peptide (Kd = 2 μM and 6 μM, respectively). Methylation or acetylation of Lys9 or Lys14 had very little to no effect on the interaction of PZP with the short H3un peptide (Fig. 1H; Supplemental Fig. 1f). Together, these data demonstrate that the PZP domain of BRPF1 recognizes the histone H3 N terminus that is unmethylated on Lys4 and that this in vitro interaction is driven by the first PHD.
Each PHD finger of the PZP domain is critical for chromatin binding and acetylation
To determine the functional relevance of the second PHD finger, we immunopurified wild-type and ΔPHD2 BRPF1 complexes from cotransfected 293T cells. Western analysis of the wild-type complex indicates the copurification of endogenous histone H3 (Fig. 2A). This cofractionation is completely lost after removal of the PHD2 domain of BRPF1, implying a crucial role in binding histone H3 in vivo. Furthermore, when the purified complexes were used in HAT assays, acetylation of chromatin was abolished by the deletion of PHD2, while acetylation of free histones was not affected (Fig. 2B). These data indicate that, while the NMR studies did not support a role in binding to the H3 tail, the second PHD of the PZP domain is essential for binding to chromatin and its subsequent acetylation. In order to compare these observations with the deletion of the first PHD domain, which drives the in vitro interaction, we then immunopurified both ΔPHD1 and ΔPHD2 complexes (Fig. 2C) and compared their acetyltransferase activity on chromatin and free histones. Similar to PHD2, the PHD1 finger is essential for acetylation of chromatin by the complex while not affecting acetylation of free histones (Fig. 2D,E). Altogether, these data demonstrate that both PHD fingers of BRPF1 are necessary for the complex to bind and acetylate chromatin, suggesting that the PZP domain functions as a single module binding to nucleosomes.
BRPF1 PHD1 and PHD2 fingers are required for binding to chromatin and its acetylation. (A) BRPF1 associates with histone H3 in vivo, and this binding is lost when the PHD2 finger is deleted. 293T cells were cotransfected with HA-BRPF1 (wild-type [WT] or ΔPHD2) plasmids, and Flag-MOZ (MYST domain), Flag-ING5, and Flag-hEaf6 plasmids and whole-cell extracts (WCEs) were used for HA immunoprecipitation (IP). As the MYST domain of the MOZ catalytic subunit is sufficient for HAT activity and complex assembly (Ullah et al. 2008), it was used to avoid the high degradation sensitivity of the full-length protein (225 kDa). Mock control was cotransfected with both HA empty and Flag empty plasmids. Histone H3 binding was analyzed by Western blot α-H3. (B) PHD2 is required for proper in vitro chromatin acetylation. HAT assays with the same purified complexes as in A were performed on chromatin or free histones. Reactions were spotted on membranes and counted by liquid scintillation. Values are based on three independent experiments with standard error. (C) Western blot on purified wild-type, ΔPHD1, and ΔPHD2 complexes in 293T cells showing equal expression of the different subunits. WCE was used after cotransfection of HA-BRPF1 plasmids (wild type, ΔPHD1, or ΔPHD2), and Flag-MOZ, Flag-ING5, and Flag-hEaf6 plasmids were used for HA immunoprecipitation. HA beads were eluted with HA peptide, and fractions were loaded on gel. (D,E) PHD1 is also important for acetylation of chromatin in vitro. HAT assays on free histones (D) or chromatin (E) using the complexes purified in C were spotted on membranes for liquid scintillation counting. Values are based on three independent experiments with standard error.
ING5 directs BRPF1 localization to H3K4me3-enriched chromatin at the 5′ end of active genes
We previously identified ING5 as the ING tumor suppressor subunit of the MOZ/MORF complexes (Doyon et al. 2006; Ullah et al. 2008). This association occurs via the conserved domain II of BRPF proteins (Ullah et al. 2008; Avvakumov et al. 2012). As ING proteins contain a PHD domain in their C termini that has been shown to recognize H3K4me3 (Pena et al. 2006; Champagne et al. 2008; Musselman et al. 2012), we asked whether the presence of the ING5 protein within the complex is targeting BRPF1 to the H3K4me3 mark in chromatin. First, we used recombinant complexes purified from SF9 cells ± ING5 to perform HAT assays on histone peptides. We observed that the HAT activity of the ING5-containing complex is greatly stimulated on H3K4me3 peptides when compared with unmodified or H3K9me peptides (Fig. 3A). This is reminiscent of our previous observations for ING4 and the JADE1/HBO1 complex, where the presence of ING4 stimulated acetylation of H3K4me3 peptides (Saksouk et al. 2009; Avvakumov et al. 2012). On the other hand, we observed a clear inhibition of HAT activity when peptides carry methylated Lys4 and ING5 is absent from the BRPF1 complex (Fig. 3B). This result corroborates what we observed in Figure 1G, where the PZP domain of BRPF1 is unable to bind H3 peptides that are methylated on Lys4. Together, these data suggest that there is an interplay between the different PHD domains found within the complex and that the PHD of ING5 is prevailing over the others in driving interaction with H3K4me3, while the BRPF1 PZP domain is required for binding to chromatin per se.
BRPF1 localizes to H3K4me3-enriched regions through its association with ING5. (A,B) ING5 directs the binding of the entire MORF complex toward H3K4me3 in vitro. Complexes ± ING5 were purified from SF9 cells and used in HAT assays on diverse modified histone peptides. Values are based on two independent experiments with standard error. (C–E) ING5, ING2, BRPF1/2, and HBO1 colocalize with H3K4me3 near the TSSs of genes. Heat maps of each protein ChIP-seq signal on ±5 kb surrounding the TSSs of genes are shown. Genes were sorted by gene expression level (see the Materials and Methods) from high (top) to low (bottom), and the signals were corrected over reads per million (RPM). (C) Enrichment of H3K4me3 signal compared with H3 signal. ING5 and ING2 (D) and BRPF1/2 and HBO1 (E) are bound at H3K4me3-enriched regions near the TSSs of genes. (F,G) BRPF1/2 and ING5 colocalize at the p21/CDKN1A gene and the HOXA cluster. The different signals for each protein are shown to illustrate their binding at H3K4me3-enriched region near the p21/CDKN1A TSS (F) and each TSS of the HOXA cluster (G).
To test how this interplay occurs in vivo, we then performed chromatin immunoprecipitation (ChIP) combined with deep sequencing (ChIP-seq) experiments in human RKO cells using H3K4me3, ING, and BRPF antibodies. This p53-positive human colon carcinoma cell line shows the expected features when mapping the H3K4me3 chromatin mark; i.e., enrichment near the TSSs of active genes, with signals increasing with the expression levels (Fig. 3C; Supplemental Figs. 2a, 3). Genome-wide localization analysis of the ING5 protein showed enrichment around the TSSs of active genes, on the same regions where H3K4me3 is located (Fig. 3D; Supplemental Fig. 2b). In comparison, the ING2 protein, which also binds H3K4me3 through its PHD domain (Pena et al. 2006; Shi et al. 2006), was found within the same genomic regions (Fig. 3D; Supplemental Fig. 2c). ING2 is a subunit of the histone deacetylase (HDAC) complex mSin3a (Doyon et al. 2006; Shi et al. 2006). Our data indicate that ING5 (HAT) and ING2 (HDAC) complexes colocalize at the TSSs of actively transcribed genes, arguing for a primary role of their H3K4me3-binding PHD finger in this targeting. Previous genome-wide mapping of several HAT and HDAC proteins showed that they are similarly found near TSSs of actively transcribed regions in a dynamic process allowing rapid resetting of chromatin (Wang et al. 2009). We then used an antibody raised against the BRPF2 paralog to evaluate genomic distribution by ChIP-seq. This antibody is equally efficient at recognizing the BRPF1 protein (but not BRPF3) because of the highly conserved region used as an antigen (Supplemental Fig. 4). The BRPF1/2 distribution was also localized near the TSSs of active genes at H3K4me3-enriched regions, similar to what was observed for ING5 (Fig. 3E; Supplemental Fig. 3d). This is also almost identical to our recently published genome-wide profile of the HBO1 HAT in the same cell line (cf. heat maps in Fig. 3E; Avvakumov et al. 2012). Upon closer examination at specific loci, we clearly observe colocalization of BRPF1/2 and ING5 proteins over H3K4me3-rich regions near the p21(CDKN1A) TSS and at the HOXA cluster (Fig. 3F,G).
Overall, these results show that BRPF1/2, ING5, and HBO1 proteins colocalize on H3K4me3-rich regions surrounding the TSSs of actively transcribed genes and that their signals are proportional to the expression level of these genes. In addition, it indicates that the ING5 protein is a primary determinant for BRPF1/2 complex association with specific genomic loci, correlating with its binding to the H3K4me3 histone mark both in vitro and in vivo.
Two distinct forms of HBO1 complexes exist with different scaffold subunits
We showed previously that the ING5 protein is associated with two different MYST acetyltransferase complexes in HeLa cells: the HBO1–JADE1/2/3 and MOZ/MORF–BRPF1/2/3 complexes (Fig. 1A; Doyon et al. 2006). A recent study in K562 cells argued that the BRPF2 scaffold protein could be associated with the HBO1 HAT (Mishima et al. 2011). To investigate this possible distinct interaction, we purified BRPF1 from stably transduced HeLa cells (Fig. 4A). Mass spectrometry and Western blot analyses confirmed association of MOZ/MORF HATs, hEaf6, and ING5, as we reported previously (Fig. 4A,B; Doyon et al. 2006; Ullah et al. 2008). Significant signals were also obtained for HBO1. Thus, both MOZ/MORF and HBO1 catalytic HAT subunits can be associated with BRPF1 in vivo. In addition, ING4 was also identified as a BRPF1-associated factor, in contrast to our previous results that suggested its restriction to HBO1–JADE1/2/3 complexes (Doyon et al. 2006). Moreover, using transduced HBO1 as bait for purification from HeLa cells, we were able to identify by Western blot and mass spectrometry both JADE1L and BRPF2/3 as interacting partners (Fig. 4C; data not shown). We also confirmed the interaction in 293T cells, where immunoprecipitation of Flag-HBO1 brings down endogenous BRPF1/2 proteins (Supplemental Fig. 5). These results demonstrate the presence of a new MYST–ING acetyltransferase complex within HeLa cells, formed by HBO1 and a BRPF paralog.
The BRPF1 scaffold subunit copurifies with both MOZ/MORF and HBO1 catalytic subunits. (A) A transduced stable HeLa cell line expressing 3xFlag-BRPF1 was used to purify associated proteins by anti-Flag immunoprecipitation/elution. Immunopurified proteins were analyzed on gel by silver staining. Nonspecific contaminants are shown by the asterisk. (B) The purified complex in A was analyzed by Western blot with the indicated antibodies. The two bands appearing with the MORF antibody are both MOZ and MORF, as the MORF antibody recognizes both proteins (Ullah et al. 2008). (C) Western blot analysis of a tandem affinity purification (TAP) from a transduced HeLa cell line stably expressing HA-HBO1-TAP (protein A and calmodulin-binding protein). (C) BRPF1 complexes acetylate histone H3 on chromatin in vitro. HAT assays were performed on free histones or native chromatin with the purified BRPF1 fraction in A. Acetylated histones were separated by SDS-PAGE and revealed by fluorography. (E–G) BRPF1 complexes acetylate histone H3 on chromatin in vivo. ChIP analysis of H3K14/K23 and H4 acetylation in transduced HeLa cells stably expressing BRPF1. Acetylation levels corrected for nucleosome occupancy (total H3 signal) were measured at the p21 TSS, 2 kb downstream, and at an intergenic locus. BRPF1 significantly increases H3K14ac and H3K23ac at the p21 TSS while showing no effect at the intergenic locus. (H) HBO1 occupancy on p21 gene does not change upon increased BRPF1 expression. ChIP analysis of HBO1 in BRPF1 and control transduced cell lines. All ChIP values are based on two independent experiments with standard error.
BRPF1–MYST complexes acetylate only histone H3 in chromatin
We used the purified BRPF1 fraction for in vitro HAT assays and observed a striking specificity for histone H3 in chromatin, while both H3 and H4 are targeted as free histones (Fig. 4D). While we previously found MOZ/MORF–BRPF–ING5 complexes to have similar specificity (Doyon et al. 2006; Ullah et al. 2008), the presence of HBO1 in the fraction suggested otherwise. Indeed, HBO1 is known to acetylate mainly the N-terminal tail of histone H4 on chromatin both in vitro and in vivo, specifically on Lys5, Lys8, and Lys12 (Doyon et al. 2006). We then used ChIP assays to determine whether the presence of additional BRPF1 protein in the transduced HeLa cells influences the level of H3 and H4 acetylation at specific loci in vivo. When compared with cells transduced with an empty vector, the Flag-BRPF1 cell line shows a significant increase of H3 acetylation on both Lys14 and Lys23 at the p21 TSS (Fig. 4E,F). Moreover, a slight increase can also be observed 2 kb downstream from the p21 TSS, where the BRPF1 complex is still likely bound (see Fig. 3F). We also noticed some increase of H4 acetylation at the same loci, but the effect seems much smaller and could be indirect/subsequent to H3 acetylation (Fig. 4G). Indeed, H3K14ac was recently shown to inhibit demethylation of H3K4 (Maltby et al. 2012), which would favor recruitment of other HATs through increased H3K4me3. Importantly, no change in H3 and H4 acetylation levels was observed at an intergenic control locus (Fig. 4E–G). These data indicate that purified BRPF1 complexes target mainly H3K14 and H3K23 in chromatin in vitro and in vivo despite the fact that a previously characterized H4-specific MYST enzyme, HBO1, is present in the fraction. However, HBO1 occupancy measured by ChIP on the same regions did not show any increase in BRPF1 transduced cells compared with the control cell line (Fig. 4H). This indicates that while increased BRPF1 level leads to higher H3 acetylation on p21, it does not seem to target additional HBO1 protein.
The scaffold subunits of the MYST–ING complexes direct histone tail specificity during acetylation of chromatin
Since no H4 acetylation was observed in vitro with the purified BRPF1 complexes containing either MOZ, MORF, or HBO1 catalytic subunits, we sought to determine the histone lysine residues targeted by the newly identified HBO1–BRPF1–ING5–hEaf6 complex. For this purpose, we overexpressed in 293T cells the different combinations of the desired subunits and purified the resulting complexes using the Flag-tagged scaffold subunit (Fig. 5A). These overexpressed complexes were then used in HAT assays on both free histones and chromatin. As expected, the BRPF1–MOZ complex acetylates H3 on chromatin, whereas the JADE1–HBO1 complex acetylates H4 (Fig. 5B). This is in sharp contrast to the lack of histone specificity when using free histones as the substrate, where histones H3/H4 are acetylated equally by both complexes. Remarkably, when the HBO1 acetyltransferase is associated with BRPF1, acetylation of chromatin by this complex is restricted to histone H3 (Fig. 5B). To more precisely identify the lysine residues acetylated on H3, we performed HAT assays with unlabeled acetyl-CoA followed by Western blot analysis with specific histone mark antibodies. We observed an increase in H3K23ac and H3K14ac with the HBO1–BRPF1 complex compared with the mock fraction, while no change in H3K9ac was detected (Fig. 5C). This H3K14/23 specificity was also observed using peptides in HAT assays (Supplemental Fig. 6). These results indicate that HBO1 can acetylate both H3 and H4 lysine residues on chromatin but that its specificity is determined by the associated scaffold subunit. While HBO1–JADE–ING–hEaf6 targets H4K5/8/12 on chromatin, HBO1–BRPF–ING–hEaf6 targets H3K14/23.
Association of different scaffold subunits with the HBO1 HAT is responsible for switching its histone acetylation specificity between H4 and H3. (A) Purification of the MOZ–BRPF1, HBO1–BRPF1, and HBO1–JADE1L complexes. 293T cells were transfected with flagged scaffold subunit plasmids (BRPF1 and JADE1L), while other cotransfected subunits were HA-tagged. Mock control was cotransfected with Flag empty and HA empty plasmids. Complexes were purified from WCEs by Flag immunoprecipitation (IP) and eluted with 3xFlag peptides. (B) The scaffold subunit changes chromatin acetylation specificity. The purified complexes in A were used in HAT assays on both chromatin and free histones. Acetylated histones were separated by SDS-PAGE and revealed by fluorography. Coomassie-stained gels show equivalent amounts of histones between samples. (C) The HBO1–BRPF1 complex acetylates H3K14 and H3K23. HAT assays with the purified HBO1–BRPF1 complex were performed on chromatin followed by Western blot analysis of different histone marks. (D) Purification of various deletions of Brpf1. Cotransfected 293T cells with either wild-type (WT), ΔBromo, or ΔPWWP Flag-BRPF1 plasmids combined with HA-HBO1, HA-ING5, and HA-hEaf6 plasmids were used for Flag immunoprecipitation purification, and complexes were analyzed by Western blot with the indicated antibodies. (E) The PWWP domain and bromodomain of BRPF1 are not essential for acetylation specificity. The purified complexes in D were used in HAT assays on both chromatin and free histones. The amount of complex used for HAT assays was normalized to free histone activity. Acetylated histones were separated by SDS-PAGE and revealed by fluorography.
Since JADE and BRPF PZP domains behave similarly in histone/chromatin-binding functions (Figs. 1, 2; Saksouk et al. 2009; Avvakumov et al. 2012) and their domain II associates with the same set of ING proteins (ING4/5), the drastic change of nucleosomal histone specificity put on the MYST acetyltransferase must originate from other parts of these scaffold proteins. Obvious candidate features are present on BRPF proteins, since they also contain a Kac-binding bromodomain and a H3K36me3-binding PWWP domain at their C termini (Fig. 1B; Vezzoli et al. 2010; Filippakopoulos et al. 2012). We constructed C-terminal deletions of the PWWP domain and the bromodomain in BRPF1. We purified HBO1 complexes containing either wild-type BRPF1 or BRPF1 lacking these domains (Fig. 5D) and used them in HAT assays (Fig. 5E). Neither the deletion of the PWWP domain nor the deletion of the bromodomain of BRPF1 affected the specificity of HBO1 for histone H3 on chromatin substrate. These data indicate that the two histone mark reader modules at the C terminus of BRPF1 are not involved in selecting the histone tail specificity of the HAT complex.
A short N-terminal region within scaffold subunits directs which histone tail is acetylated by MYST complexes on chromatin
When comparing sequence homologies between scaffold subunits of human MYST–ING HAT complexes, it became apparent that some features were conserved at the N-terminal region, just before the domain I, responsible for binding the MYST subunit (Fig. 6A). This region is considered part of the larger EPcA domain in EPC proteins, scaffold subunits of the NuA4/Tip60 HAT complex (Fig. 1A,B). We showed previously that this small region at the beginning of EPcA is important for chromatin binding and nucleosomal HAT activity of the yeast NuA4 complex (Selleck et al. 2005; Chittuluru et al. 2011). To investigate whether the corresponding N-terminal region in BRPF1 or JADE1 scaffold subunits is implicated in nucleosomal HAT activity and, perhaps, histone tail selection, we produced N-terminal deletions that removed the first 20 amino acids of the EPcA-related region. The wild-type or truncated JADE1 and BRPF1 scaffold subunits were purified as tetrameric complexes from cotransfections with either HBO1 or MOZ as the catalytic subunit. HAT assays with the purified complexes were performed on chromatin (Fig. 6B). Strikingly, association of MOZ with JADE1 instead of BRPF1 also shifts its histone tail specificity from H3 to H4, as we observed for HBO1 (Fig. 6B, cf. lanes 4 and 8 and lanes 2 and 6). Since MOZ is mostly known for acetylating histone H3, this result clearly supports our previous conclusion about HBO1 and expands it to other MYST HATs; i.e., that it is the scaffold subunit that is responsible for directing the histone tail specificity during acetylation of chromatin, not the acetyltransferase subunit.
A small N-terminal domain in the scaffold subunits of the MYST–ING complexes is responsible for directing specific histone tail acetylation. (A) Sequence alignment of scaffold subunits of the MYST–ING complexes with the N-terminal region of the EPcA domain found in the EPC proteins (EPC1/2 and Epl1). An arrow indicates the location of the N-terminal truncation in BRPF1 and JADE1 mutants. Domain I is the region of association with the MYST HAT. (B) Deletion of the N-terminal part of scaffold subunits BRPF1 and JADE1L alters chromatin acetylation specificity. 293T cells were cotransfected with the indicated expression plasmids, HA-tagged catalytic subunits, and Flag-tagged scaffold subunits. HA-ING5 and HA-hEAF6 expression plasmids were also cotransfected for each purification. Flag immunoprecipitations (IPs) were performed on WCE and were eluted with 3xFlag peptides. Purified complexes were used in HAT assays, and acetylated histones were separated by SDS-PAGE and revealed by fluorography. The complexes were normalized to the same HAT activity on free histones (by liquid assays). (C) Deletion of the N-terminal part of scaffold subunit EPC1 also modifies chromatin acetylation specificity. EPC1 wild-type and Δ1–12 complexes were purified and used for HAT assays by coexpressing Flag-EPC1, HA-ING3, HA-hEaf6, and HA-Tip60. (D) The N-terminal domain of the yeast homolog Epl1 acts similarly in directing histone specificity. Recombinant yeast piccolo NuA4 complexes with wild-type or Δ1–71 Epl1 were used for HAT assays on yeast chromatin, and acetylated histones were treated as in A.
When associated with HBO1, removal of the small N-terminal EPcA-related region of JADE1 protein resulted in not only a loss of H4 acetylation on chromatin, but also a clear appearance of H3 acetylation (Fig. 6B, lanes 2,3). This was also the case when truncated JADE1 was associated with MOZ (Fig. 6B, lanes 4,5). These results indicate that this region of JADE1 is important for not only acetylation of chromatin, but also histone tail selection. However, removal of the same region in BRPF1, while leading again to a loss of HAT activity on chromatin for both HBO1 and MOZ, did not seem to significantly change histone tail specificity, as only H3 acetylation could be observed (Fig. 6B, cf. lanes 6 and 7 and lanes 8 and 9). However, repeating the assay with equivalent amounts of nucleosomal HAT activity between the wild-type and mutant complexes showed a significant loss of histone tail specificity, as H4 acetylation is now detected in the mutants (Supplemental Fig. 7). Thus, deletion of the EPcA-related region leads to reduced nucleosomal HAT activity, as we showed for yeast Epl1 in NuA4 (Selleck et al. 2005; Chittuluru et al. 2011). However, our results also indicate that the same region of JADE1 is indeed responsible for selecting the H4 tail versus H3 for acetylation on chromatin. Since the loss of tail specificity detected with the BRPF1 mutant is more subtle, it is possible that H3 tail acetylation is the default target driven by the PZP domain in JADE and BRPF proteins.
To further investigate the molecular mechanisms of histone tail selectivity, we analyzed other MYST–ING HAT complexes that contain scaffold subunits naturally lacking a PZP domain. We purified the human Tip60–EPC1–ING3–hEaf6 complex from cotransfected cells and the recombinant yeast piccolo NuA4 complex from bacteria. Both of these complexes selectively acetylate histone H4 and H2A tails on chromatin substrates (Fig. 6C,D), while they can target H3 in free histones (Boudreault et al. 2003; Doyon et al. 2004). Removal of only the first 12 amino acids of human EPC1 and its EPcA domain completely abolishes acetylation of the histone H4 tail by Tip60 on chromatin (Fig. 6C). In a clear contrast, acetylation of the histone H2A tail is preserved. Similar results were obtained with yeast piccolo NuA4, as a complex containing the equivalent deletion of the yeast scaffold protein Epl1 also lost its activity toward nucleosomal histone H4 tail but retained its activity toward H2A (Fig. 6D). Thus, deletion of the first portion of the EPcA-related domain resulted in loss of H4 acetylation in Tip60/EPC1, NuA4/Epl1, JADE1/MOZ, and JADE1/HBO1 complexes, supporting a role for this domain in orienting the MYST HAT to acetylate the H4 tail in chromatin. However, since there was no loss of H2A tail-specific acetylation by Tip60 and NuA4, in comparison with gain of H3 tail acetylation in the case of JADE1, these results suggest that other histone tail specificity determinants are at play, likely within the same region of the scaffold subunits. Altogether, these data indicate that scaffold subunits in MYST acetyltransferase complexes are not only essential to enable acetylation of chromatin, but also required to direct which histone tail gets acetylated.
Discussion
Post-translational modifications of histone residues can directly alter chromatin structure by modulating the interactions between histones and DNA. They can also serve as docking platforms for the binding of chromatin-associated proteins and thus in activating nuclear signaling pathways (Musselman et al. 2012). A tight regulation of the deposited modifications and the related enzymes is thus necessary to ensure proper chromatin dynamics. The MYST family of acetyltransferases assembles in different multiprotein complexes. Several subunits of these complexes contain such histone recognition motifs (Avvakumov and Cote 2007b; Avvakumov et al. 2012). In this study, we dissected the binding properties of the PZP domain located in the BRPF1 protein, a scaffold subunit of MOZ/MORF HAT complexes. We found that, as for JADE1 and BRPF2 PZP domains (Saksouk et al. 2009; Qin et al. 2011; Avvakumov et al. 2012), the first PHD finger of BRPF1 has strong affinity for the histone H3 N-terminal domain but only when H3K4 is not methylated. Interestingly, PHD1 acts dominantly over PHD2 within the PZP in blocking interaction with methylated forms of H3K4 (Fig. 1). Nevertheless, both PHD1 and PHD2 are required for the MOZ–BRPF1–ING5–hEaf6 complex to bind histone H3 in vivo and acetylate chromatin in vitro (Fig. 2). Thus, although PHD2 does not show any structured binding to histone peptides in vitro, it is still required for proper binding of the BRPF1 complexes to chromatin. This is reminiscent of the JADE1 PHD2 finger, which is also essential for binding chromatin in vivo and for the tumor suppressor activity of the HBO1 complex (Saksouk et al. 2009). However, our results with BRPF1 PHD fingers suggest that they act together as a single functional module, the PZP domain, to bind chromatin and allow its acetylation by the MYST HAT. This is supported by similar results obtained when only the Zn knuckle region between PHD1 and PHD2 is deleted (data not shown). Interestingly, it was recently suggested that the BRPF2 PHD2 finger could in fact bind DNA (Liu et al. 2012). Our NMR data with dNMPs and BRPF1 PHD2 also support this model (Supplemental Fig. 1e). Thus, within the PZP domain, PHD2 could assist PHD1 by allowing binding to nucleosomal DNA, while PHD1 locks in the histone H3 N-terminal domain.
As multiple chromatin-binding domains are found within the different subunits of MYST–ING HAT complexes, further study is still required to understand the interplay that exists between them. We showed that the ING5 PHD domain directs binding of the associated complexes to H3K4me3-rich regions and stimulates acetylation both in vitro and in vivo (Figs. 3, 4). Since both MOZ and BRPF1 have been linked to HOX gene activation (Laue et al. 2008; Voss et al. 2009; Qiu et al. 2012), it is thus very likely that their transcriptional regulation occurs via their binding to the TSSs of these genes, which are highly enriched in H3K4me3 (Fig. 3G). Moreover, in the absence of ING5, we clearly demonstrate that the PZP domain-binding features inhibited the acetylation on H3K4 methylated peptides (Fig. 3B). Thus, the PHD of the ING5 subunit prevails over the PZP domain of BRPF1, redirecting the binding of the complex to H3K4me3-containing chromatin. Nevertheless, even when ING5 is present, the PZP domain is required for binding to chromatin and acetylation. It is possible that BRPF–ING HAT complexes target asymmetric nucleosomes in which only one H3 tail is methylated on Lys4. These complexes may also favor spreading of the H3K4 methylation mark by simultaneously binding a methylated nucleosome through ING5 and an adjacent unmethylated one through the PZP, leading to acetylation of H3K14, which stimulates methylation of H3K4 (Nakanishi et al. 2008; Maltby et al. 2012). The MOZ/MORF HATs found associated with BRPF1 also contain a tandem PHD domain that has recently been shown to bind unmodified H3R2 and acetylated H3K14 (Ali et al. 2012; Qiu et al. 2012). It will be interesting to determine how the five PHD fingers found in different subunits of the MOZ/MORF–BRPF1–ING5–hEaf6 complex functionally interact with each other and other histone reader modules during binding to chromatin.
Some apparent contradictions are present in the literature regarding the HBO1 acetyltransferase. We and others have shown that the HBO1 enzyme is purified with JADE scaffold proteins and is responsible for histone H4 tail acetylation (Doyon et al. 2006; Iizuka et al. 2006, 2008; Foy et al. 2008; Miotto and Struhl 2010). We even showed that HBO1 siRNA-mediated knockdown in HeLa cells leads to a global loss of H4 acetylation on Lys5, Lys8, and Lys12, matching in vitro specificity on chromatin and arguing that HBO1 was the main H4-specific HAT in mammals (Doyon et al. 2006). On the other hand, it was later shown that HBO1 gene knockout in mouse embryos leads instead to a loss of bulk H3K14ac in primary embryonic fibroblasts at embryonic day 9.5 (E9.5), while H4 acetylation persisted (Kueh et al. 2011). In addition, a HBO1–BRPF2 complex was reported in K562 leukemic cells and shown to target global H3K14 acetylation and erythroid regulators (Mishima et al. 2011). These contradicting results are quite adequately explained in the present study with the finding that HBO1–JADE and HBO1–BRPF HAT complexes coexist within HeLa cells (Fig. 4). Moreover, this differential association with distinct scaffold subunits is responsible for switching HBO1 specificity on chromatin toward different histone tails (Fig. 5). The HBO1–JADE1 complex targets mainly H4 residues, whereas the HBO1–BRPF1 complex acetylates only H3 in the context of chromatin. The varying protein expression levels between different tissues and/or during different developmental stages thus allow for fine-tuned regulation, leading to differential patterns of acetylated histones across the genome. Such functionally important tissue-specific variability of paralog subunits in chromatin regulators has been well documented for the BAF(SWI/SNF) remodeling complex (Hargreaves and Crabtree 2011). It is important to point out that HBO1 is nevertheless confirmed as a major mammalian HAT, since its depletion leads to global loss of histone acetylation on H3 in mouse erythroblasts/embryonic fibroblasts or on H4 in HeLa cells (Doyon et al. 2006; Kueh et al. 2011; Mishima et al. 2011). It will be very interesting to determine what is responsible for bulk H4 acetylation in Hbo1−/− mouse embryonic fibroblasts (Kueh et al. 2011). Is it Tip60? Is it another MYST HAT now associated with a JADE protein?
We identified a short EPcA-related N-terminal domain in BRPF1 and JADE1 as the region responsible for histone tail specificity of the associated MYST acetyltransferase on chromatin substrates (Figs. 6, 7). Interestingly, truncation of this domain in human EPC1 and yeast Epl1 protein cripples Tip60/NuA4's ability to acetylate nucleosomal H4, but histone H2A acetylation persists. This basic region of Epl1 was recently shown in cross-linking experiments to bind the histone H2A N-terminal tail in nucleosomes (Huang and Tan 2013). Thus, it is tempting to speculate that this binding to H2A is orienting the NuA4 complex on the nucleosome to target the H4 tail for acetylation. The corresponding regions in BRPF and JADE proteins would bind nucleosomes in distinct manners, leading to different histone tail selection for acetylation. The residual acetylation of H3 tail detected in BRPF1 and JADE1 truncations may be driven by the H3-binding function of the PZP domain and/or the ING subunit. Indeed, we showed that H3K4me3-binding ING4/5 subunits favor acetylation of histone H3K14 even from within the HBO1–JADE1 complex (Hung et al. 2009; Saksouk et al. 2009). It is important to point out that ING4/5 in this case allows H3K14 acetylation on top of the main H4 acetylation performed by HBO1–JADE1, not a complete switch of histone tail specificity, as demonstrated here between the HBO1–JADE1 and HBO1–BRPF1 complexes.
Model for MYST acetyltransferase assembly in alternate complexes, leading to different histone tail specificities. (A) Schematic representation of protein domains in BRPF paralogs and their demonstrated specific interactions/roles. (B) The HBO1 and MOZ/MORF catalytic subunits can be associated with different scaffold proteins, leading to a switch in histone tail specificity for acetylation of chromatin substrates. Thus, protein complexes associated with HAT proteins not only enable them to acetylate chromatin substrates, but also select which histone tail is targeted, a specificity previously thought to reside in the acetyltransferase itself. The arrow between MOZ/MORF and JADE1/2/3 is gray, since this interaction has only been reported in cotransfection experiments.
It is well established that acetylation neutralizes the charge of lysine residues to modulate the interactions with nucleosomal DNA and neighboring nucleosomes, regulating higher-order chromatin structure (Tse et al. 1998). Indeed, H3 acetylation shows distinctive effects on modulating the tertiary structure compared with H2A or H4 acetylation (Siino et al. 2003; Wang and Hayes 2008), underlying the importance of specific histone acetylation in regulating different cellular processes. H3 acetylation essentially appears to affect DNA accessibility in individual nucleosomes, while H4 acetylation has more long-range effects on chromatin compaction (Wang and Hayes 2008). Moreover, acetylated lysines on histones H3 and H4 can recruit distinct nuclear effector proteins, such as transcriptional coactivators like Rsc4 by H3K14ac, TRIM24 by H3K23ac, and Brd2 by H4ac (Agalioti et al. 2002; Kasten et al. 2004; Agricola et al. 2006; Tsai et al. 2010; Draker et al. 2012). Such differential recruitment of specific bromodomain-containing proteins could thus help regulate the expression levels of specific genes (Filippakopoulos et al. 2012). It remains to be determined how the specific diacetylation of H3K14/23 by BRPF1–MYST complexes is interpreted compared with the different combinations deposited on H3 by PCAF/GCN5 and CBP/p300.
In conclusion, this study uncovers a new crucial role of factors associated with HAT proteins in multisubunit complexes. We and others demonstrated previously that complex assembly is required to enable HAT enzymes to acetylate their targets in native chromatin substrates (Carrozza et al. 2003). We now show that scaffold subunits associated with MYST family HATs not only allow chromatin binding and acetylation, but also select which histone tail becomes acetylated. Until now, histone tail specificity has been thought to reside in the acetyltransferase protein. The alternate association of the HBO1 catalytic subunit with BRPF and JADE proteins induces a striking shift of acetylation specificity between H3 and H4 tails. These results highlight the new role of the associated scaffold subunits within MYST–ING acetyltransferase complexes in directing the acetylation of specific histone tails. These findings add a new mechanism to the regulation of chromatin dynamics and call for caution when interpreting and comparing studies in which the function of HAT proteins is analyzed outside their physiological context.
Materials and methodsPurification of MYST HAT complexes
The native BRPF1 complex was purified from a retrovirus transduced HeLa cell line expressing 3xFlag-BRPF1 from a CMV promoter (pRCF vector). Nuclear extract were prepared following standard procedures (Abmayr 1993), and immunoprecipitation with anti-Flag agarose beads (Sigma) was done before eluting with Flag peptide buffer (100 mM KCl, 20 mM HEPES at pH 7.5, 20% glycerol, 0.1% Triton X-100, 400 μg/mL 3xFlag peptide, 1 mM DTT, 0.1 mM ZnCl2, 1 mM PMSF). Tandem affinity purification of the native HBO1 complexes was done as previously described (Doyon et al. 2006). Purification of MYST complexes from transient transfections was performed in 293T cells. Cells were transfected near confluency with 6 μg of each plasmid (MYST, ING, BRPF/JADE/EPC, and hEAF6 with the indicated tags) per 150-mm plate. Cells were harvested 48 h post-transfection, and whole-cell extracts (WCEs) were prepared followed by anti-Flag immunoprecipitation/elution with Flag peptide or anti-HA immunoprecipitation/elution with HA peptide as previously described (Saksouk et al. 2009; Avvakumov et al. 2012). Tandem mass spectrometry analysis was done after in-gel digestion with trypsin at the Taplin Mass Spectrometry Facility. Expression vectors were constructed and, when indicated, mutated following standard procedures (details are available on request). PHD1 and PHD2 deletions correspond to amino acids 265–355 and 359–450.
Recombinant protein purifications and peptide pull-downs
The BRPF1 PHD1 (amino acids 275–329), PHD2 (amino acids 385–456), and PZP (amino acids 256–543) domains were expressed in Escherichia coli Rosetta pDEST15 or BL21 pGEX4T3 cells grown in LB or 15NH4Cl minimal medium supplemented with 1.5 mM ZnCl2. After induction with 1.0 mM IPTG for 16 h at 20°C, bacteria were harvested by centrifugation and lysed with lysozyme and/or by sonication. The unlabeled and 15N-labeled GST fusion proteins were purified on glutathione Sepharose 4B beads (GE Healthcare). The GST tag was either cleaved with PreScission protease or kept for Western blot analysis/peptide pull-downs, in which case the GST fusion protein was eluted off the glutathione Sepharose beads using 50 mM reduced L-glutathione (Sigma Aldrich). For NMR analysis, the proteins were concentrated into 20 mM Tris-HCl (pH 6.8) in the presence of 150 mM NaCl, 10 mM dithiothreitol, and 10% D2O. Protein complex purification from bacteria and SF9 cells was done as previously described (Selleck et al. 2005; Ullah et al. 2008). Peptide pull-downs with GST fusion proteins were performed as previously described using biotinylated peptides and streptavidin magnetic beads (Saksouk et al. 2009).
NMR spectroscopy
NMR experiments were performed at 25°C on Varian INOVA 600- and 500-MHz spectrometers using pulse field gradients to suppress potential artifact and eliminate water signal. 1H,15N heteronuclear single quantum coherence (HSQC) spectra of uniformly 15N-labeled PHD1, PHD2, and PZP (0.1–0.2 mM) were recorded as histone tail peptides (synthesized by the University of California at Davis Biophysics Core Facility), dNMPs (a mixture of dAMP, dTMP, dCMP, and dGMP, 1:1:1:1), or unlabeled PHD1 were added stepwise.
Fluorescence spectroscopy
Tryptophan fluorescence measurements were carried out at 25°C on a Fluoromax-3 spectrofluorometer. The samples of 1–10 μM PHD1 or PZP containing progressively increasing concentrations of histone peptides (up to 1 mM) were excited at 295 nm. Emission spectra were recorded between 305 and 405 nm with a 0.5-nm step size and a 1-sec integration time and were averaged over three scans. Kd values were determined by a nonlinear least-squares analysis using the equationwhere [L] is the concentration of the peptide, [P] is the concentration of the protein, ΔI is the observed change of signal intensity, and ΔImax is the difference in signal intensity of the free and fully bound states of the protein. The Kd values were averaged over three separate experiments, with error calculated as the standard deviation between the runs.
Antibodies and peptides
The following antibodies were used for Western blotting with the indicated dilutions: anti-Flag M2 HRP (1:10000; Sigma), anti-HA HRP (1:1000; Roche), anti-HA.11 (1:1000; Babco), anti-HBO1 (1:2000; Abcam), anti-hEaf6 (1:1000; Abcam), anti-MORF (1:1000) (Ullah et al. 2008), anti-ING5 (1:1000; Abcam), and anti-H3 (1:5000; Abcam). For ChIP and ChIP-seq, the following antibodies were used: anti-H3K4me3 (Abcam), anti-H3 (Abcam), anti-H4ac (Millipore), anti-H3K14ac (Millipore), anti-H3K23ac (Millipore), anti-ING5 (Abcam), anti-HBO1 (Abcam), anti-ING2 (Epitomics), anti-MRG15, and anti-BRPF2 (Bethyl Laboratories). Biotinylated histone peptides were purchased from Millipore.
HAT assays
Native human chromatin and free histone were purified as previously described (Utley et al. 1996). HAT assays with 300 ng of histone peptides (Millipore), 500 ng of core histones, or 500 ng of H1-depleted oligonucleosomes prepared from HeLa S3 cells were performed in a 15-μL reaction containing 50 mM Tris-HCl (pH 8.0), 10% glycerol, 1 mM DTT, 0.1 mM EDTA, 1 mM PMSF, 10 mM sodium butyrate (Sigma), and 1.25 nCi 3H-labeled (Perkin Elmer Life Sciences) or unlabeled acetyl-CoA (Sigma). Samples were either spotted on P81 membranes (GE Healthcare) for counting or loaded on 18% SDS-PAGE gels. For gel assays, Coomassie staining was followed by EN3HANCE (Perkin Elmer) treatment and fluorography.
ChIP assays
Chromatin preparation from RKO cells was done as previously described (Avvakumov et al. 2012). For immunoprecipitation of chromatin, we used 200 μg of chromatin with 1–3 μg of specific antibodies incubated overnight at 4°C. Next, 40 μL of Protein A Dynabeads (Invitrogen) was added to each sample and incubated for 4 h at 4°C. The beads were washed extensively and eluted with 1% SDS and 0.1 M NaHCO3. Cross-linked samples were reversed by heating overnight at 65°C in the presence of 0.2 M NaCl. Samples were then treated with RNase A and proteinase K for 2 h, and DNA was recovered by phenol-chloroform and ethanol precipitation. Quantitative real-time PCR corrected for primer efficiencies in the linear range was performed using SYBR Green I (Roche) on a LightCycler 480 (Roche). The error bars represent standard errors based on two independent experiments. The primers used for quantitative PCR amplified the following genomic regions (build hg18): p21 TSS chr6, 36,754,421–36,754,542; p21 + 2 kb chr6, 36,756,871–36,756,972; and intergenic chr12, 65,815,182–65,815,318.
ChIP-seq analysis
ChIP and library preparation for sequencing were done as previously described (Avvakumov et al. 2012). Samples were sequenced by 50-base-pair (bp) single reads on either a Genome Analyzer platform (HBO1, BRPF1/2, and input) or a HiSeq 2000 platform (H3K4me3, H3, IgG, ING2, and ING5) (Illumina). Raw sequences were mapped using Bowtie (PubMed identification [PMID]: 19261174) on build hg18 of the human genome and deposited in the Gene Expression Omnibus (GEO) database under accession number GSE47190. HBO1 and input data were previously deposited under accession number GSE33221. Uniquely mapped sequences were kept for downstream analysis. The global profiles at TSSs presented in Supplemental Figure 3 were produced using the University of California at Santa Cruz genome browser gene definitions and the Python package HTseq. We extended the reads to 200 bp to be in line with our sonication protocol. In the case of multiple TSSs associated with the same gene, we selected the one with the highest number of H3K4me3 mapped reads within 5000 bp around the TSS. For gene expression level in RKO cells, we used publicly available data from an Affymetrix U133 plus 2.0 chip (PMID: 16300372). The binning of genes by their level of expression was performed by first sorting the log2 expression level and then subdividing genes in four equal categories (quartiles). The heat maps presented in Figure 3 were generated using a custom script in R (http://www.r-project.org). Briefly, we computed the profile for every ChIP-seq experiment for every gene, considering 5000 bp on both sides of the TSS. We took the profile around the TSS and binned it at every 50 bp. We then generated the heat maps using the binned values. For all of the heat maps, genes were sorted by function of their expression values in the RKO cell line, with the most expressed genes at the top of the heat map (PMID: 16300372).
Acknowledgments
We thank Céline Roques for experimental help during revisions, and Calcul Québec/Compute Canada for the use of their supercomputers. This research is supported by grants from the Canadian Institutes of Health Research (CIHR) (MOP-64289 to J.C., and MOP-97957 to X.J.Y.) and the National Institutes of Health (NIH) (GM096863 and GM101664 to T.G.K., and GM60489 to S.T.). M.-E.L. is supported by a Fonds de Recherche du Québec–Santé studentship, and J.C. is a Canada Research Chair.
Supplemental material is available for this article.
Article is online at http://www.genesdev.org/cgi/doi/10.1101/gad.223396.113.
ReferencesAbmayrSM, YaoT, ParmelyT, WorkmanJL1993Preparation of nuclear and cytoplasmic extracts from mammalian cells.
12: 12.1.1–12.1.9AgaliotiT, ChenG, ThanosD2002Deciphering the transcriptional histone acetylation code for a human gene.
111: 381–39212419248AgricolaE, VerdoneL, Di MauroE, CasertaM2006H4 acetylation does not replace H3 acetylation in chromatin remodelling and transcription activation of Adr1-dependent genes.
62: 1433–144617121596AikawaY, KatsumotoT, ZhangP, ShimaH, ShinoM, TeruiK, ItoE, OhnoH, StanleyER, SinghH2010PU.1-mediated upregulation of CSF1R is crucial for leukemia stem cell potential induced by MOZ-TIF2.
16: 580–58520418886AliM, YanK, LalondeME, DegernyC, RothbartSB, StrahlBD, CoteJ, YangXJ, KutateladzeTG2012Tandem PHD fingers of MORF/MOZ acetyltransferases display selectivity for acetylated histone H3 and are required for the association with chromatin.
424: 328–33823063713AvvakumovN, CoteJ2007aFunctions of myst family histone acetyltransferases and their link to disease.
41: 295–31717484133AvvakumovN, CoteJ2007bThe MYST family of histone acetyltransferases and their intimate links to cancer.
26: 5395–540717694081AvvakumovN, LalondeME, SaksoukN, PaquetE, GlassKC, LandryAJ, DoyonY, CayrouC, RobitailleGA, RichardDE, 2012Conserved molecular interactions within the HBO1 acetyltransferase complexes regulate cell proliferation.
32: 689–70322144582BoudreaultAA, CronierD, SelleckW, LacosteN, UtleyRT, AllardS, SavardJ, LaneWS, TanS, CoteJ2003Yeast enhancer of polycomb defines global Esa1-dependent acetylation of chromatin.
17: 1415–142812782659CarrozzaMJ, UtleyRT, WorkmanJL, CoteJ2003The diverse functions of histone acetyltransferase complexes.
19: 321–32912801725ChampagneKS, SaksoukN, PenaPV, JohnsonK, UllahM, YangXJ, CoteJ, KutateladzeTG2008The crystal structure of the ING5 PHD finger in complex with an H3K4me3 histone peptide.
72: 1371–137618623064ChittuluruJR, ChabanY, Monnet-SaksoukJ, CarrozzaMJ, SapountziV, SelleckW, HuangJ, UtleyRT, CrametM, AllardS, 2011Structure and nucleosome interaction of the yeast NuA4 and piccolo-NuA4 histone acetyltransferase complexes.
18: 1196–120321984211de AlmeidaSF, Carmo-FonsecaM2012Design principles of interconnections between chromatin and pre-mRNA splicing.
37: 248–25322398209DoyonY, SelleckW, LaneWS, TanS, CoteJ2004Structural and functional conservation of the NuA4 histone acetyltransferase complex from yeast to humans.
24: 1884–189614966270DoyonY, CayrouC, UllahM, LandryAJ, CoteV, SelleckW, LaneWS, TanS, YangXJ, CoteJ2006ING tumor suppressor proteins are critical regulators of chromatin acetylation required for genome expression and perpetuation.
21: 51–6416387653DrakerR, NgMK, SarcinellaE, IgnatchenkoV, KislingerT, CheungP2012A combination of H2A.Z and H4 acetylation recruits Brd2 to chromatin during transcriptional activation.
8: e100304723144632FilippakopoulosP, PicaudS, MangosM, KeatesT, LambertJP, Barsyte-LovejoyD, FelletarI, VolkmerR, MullerS, PawsonT, 2012Histone recognition and large-scale structural analysis of the human bromodomain family.
149: 214–23122464331FoyRL, SongIY, ChitaliaVC, CohenHT, SaksoukN, CayrouC, VaziriC, CoteJ, PanchenkoMV2008Role of Jade-1 in the histone acetyltransferase (HAT) HBO1 complex.
283: 28817–2882618684714HargreavesDC, CrabtreeGR2011ATP-dependent chromatin remodeling: Genetics, genomics and mechanisms.
21: 396–42021358755HuangJ, TanS2013Piccolo NuA4-catalyzed acetylation of nucleosomal histones: Critical roles of an Esa1 Tudor/chromo barrel loop and an Epl1 enhancer of polycomb A (EPcA) basic region.
33: 159–16923109429HungT, BindaO, ChampagneKS, KuoAJ, JohnsonK, ChangHY, SimonMD, KutateladzeTG, GozaniO2009ING4 mediates crosstalk between histone H3 K4 trimethylation and H3 acetylation to attenuate cellular transformation.
33: 248–25619187765IizukaM, MatsuiT, TakisawaH, SmithMM2006Regulation of replication licensing by acetyltransferase Hbo1.
26: 1098–110816428461IizukaM, SarmentoOF, SekiyaT, ScrableH, AllisCD, SmithMM2008Hbo1 Links p53-dependent stress signaling to DNA replication licensing.
28: 140–15317954561KastenM, SzerlongH, Erdjument-BromageH, TempstP, WernerM, CairnsBR2004Tandem bromodomains in the chromatin remodeler RSC recognize acetylated histone H3 Lys14.
23: 1348–135915014446KatsumotoT, AikawaY, IwamaA, UedaS, IchikawaH, OchiyaT, KitabayashiI2006MOZ is essential for maintenance of hematopoietic stem cells.
20: 1321–133016702405KuehAJ, DixonMP, VossAK, ThomasT2011HBO1 is required for H3K14 acetylation and normal transcriptional activity during embryonic development.
31: 845–86021149574LaueK, DaujatS, CrumpJG, PlasterN, RoehlHH, KimmelCB, SchneiderR, HammerschmidtM2008The multidomain protein Brpf1 binds histones and is required for Hox gene expression and segmental identity.
135: 1935–194618469222LiuL, QinS, ZhangJ, JiP, ShiY, WuJ2012Solution structure of an atypical PHD finger in BRPF2 and its interaction with DNA.
180: 165–17322820306MaltbyVE, MartinBJ, Brind'AmourJ, ChruscickiAT, McBurneyKL, SchulzeJM, JohnsonIJ, HillsM, HentrichT, KoborMS, 2012Histone H3K4 demethylation is negatively regulated by histone H3 acetylation in Saccharomyces cerevisiae.
109: 18505–1851023091032MillerCT, MavesL, KimmelCB2004moz regulates Hox expression and pharyngeal segmental identity in zebrafish.
131: 2443–246115128673MiottoB, StruhlK2010HBO1 histone acetylase activity is essential for DNA replication licensing and inhibited by Geminin.
37: 57–6620129055MishimaY, MiyagiS, SarayaA, NegishiM, EndohM, EndoTA, ToyodaT, ShingaJ, KatsumotoT, ChibaT, 2011The Hbo1-Brd1/Brpf2 complex is responsible for global acetylation of H3K14 and required for fetal liver erythropoiesis.
118: 2443–245321753189MusselmanCA, LalondeME, CoteJ, KutateladzeTG2012Perceiving the epigenetic landscape through histone readers.
19: 1218–122723211769NakanishiS, SandersonBW, DelventhalKM, BradfordWD, Staehling-HamptonK, ShilatifardA2008A comprehensive library of histone mutants identifies nucleosomal residues required for H3K4 methylation.
15: 881–88818622391PenaPV, DavrazouF, ShiX, WalterKL, VerkhushaVV, GozaniO, ZhaoR, KutateladzeTG2006Molecular mechanism of histone H3K4me3 recognition by plant homeodomain of ING2.
442: 100–10316728977Perez-CampoFM, BorrowJ, KouskoffV, LacaudG2009The histone acetyl transferase activity of monocytic leukemia zinc finger is critical for the proliferation of hematopoietic precursors.
113: 4866–487419264921QinS, JinL, ZhangJ, LiuL, JiP, WuM, WuJ, ShiY2011Recognition of unmodified histone H3 by the first PHD finger of bromodomain-PHD finger protein 2 provides insights into the regulation of histone acetyltransferases monocytic leukemic zinc-finger protein (MOZ) and MOZ-related factor (MORF).
286: 36944–3695521880731QiuY, LiuL, ZhaoC, HanC, LiF, ZhangJ, WangY, LiG, MeiY, WuM, 2012Combinatorial readout of unmodified H3R2 and acetylated H3K14 by the tandem PHD finger of MOZ reveals a regulatory mechanism for HOXA9 transcription.
26: 1376–139122713874SaksoukN, AvvakumovN, ChampagneKS, HungT, DoyonY, CayrouC, PaquetE, UllahM, LandryAJ, CoteV, 2009HBO1 HAT complexes target chromatin throughout gene coding regions via multiple PHD finger interactions with histone H3 tail.
33: 257–26519187766SapountziV, CoteJ2011MYST-family histone acetyltransferases: Beyond chromatin.
68: 1147–115621132344SelleckW, FortinI, SermwittayawongD, CoteJ, TanS2005The Saccharomyces cerevisiae piccolo NuA4 histone acetyltransferase complex requires the enhancer of polycomb A domain and chromodomain to acetylate nucleosomes.
25: 5535–554215964809ShahbazianMD, GrunsteinM2007Functions of site-specific histone acetylation and deacetylation.
76: 75–10017362198ShiX, HongT, WalterKL, EwaltM, MichishitaE, HungT, CarneyD, PenaP, LanF, KaadigeMR, 2006ING2 PHD domain links histone H3 lysine 4 methylation to active gene repression.
442: 96–9916728974ShilatifardA2006Chromatin modifications by methylation and ubiquitination: Implications in the regulation of gene expression.
75: 243–26916756492SiinoJS, YauPM, ImaiBS, GatewoodJM, BradburyEM2003Effect of DNA length and H4 acetylation on the thermal stability of reconstituted nucleosome particles.
302: 885–89112646255StankunasK, BergerJ, RuseC, SinclairDA, RandazzoF, BrockHW1998The enhancer of polycomb gene of Drosophila encodes a chromatin protein conserved in yeast and mammals.
125: 4055–40669735366ThomasT, CorcoranLM, GugasyanR, DixonMP, BrodnickiT, NuttSL, MetcalfD, VossAK2006Monocytic leukemia zinc finger protein is essential for the development of long-term reconstituting hematopoietic stem cells.
20: 1175–118616651658TsaiWW, WangZ, YiuTT, AkdemirKC, XiaW, WinterS, TsaiCY, ShiX, SchwarzerD, PlunkettW, 2010TRIM24 links a non-canonical histone signature to breast cancer.
468: 927–93221164480TseC, SeraT, WolffeAP, HansenJC1998Disruption of higher-order folding by core histone acetylation dramatically enhances transcription of nucleosomal arrays by RNA polymerase III.
18: 4629–46389671473UllahM, PelletierN, XiaoL, ZhaoSP, WangK, DegernyC, TahmasebiS, CayrouC, DoyonY, GohSL, 2008Molecular architecture of quartet MOZ/MORF histone acetyltransferase complexes.
28: 6828–684318794358UtleyRT, Owen-HughesTA, JuanLJ, CoteJ, AdamsCC, WorkmanJL1996In vitro analysis of transcription factor binding to nucleosomes and nucleosome disruption/displacement.
274: 276–2918902812van AttikumH, GasserSM2009Crosstalk between histone modifications during the DNA damage response.
19: 207–21719342239VezzoliA, BonadiesN, AllenMD, FreundSM, SantiveriCM, KvinlaugBT, HuntlyBJ, GottgensB, BycroftM2010Molecular basis of histone H3K36me3 recognition by the PWWP domain of Brpf1.
17: 617–61920400950VossAK, ThomasT2009MYST family histone acetyltransferases take center stage in stem cells and development.
31: 1050–106119722182VossAK, CollinC, DixonMP, ThomasT2009Moz and retinoic acid coordinately regulate H3K9 acetylation, Hox gene expression, and segment identity.
17: 674–68619922872WangX, HayesJJ2008Acetylation mimics within individual core histone tail domains indicate distinct roles in regulating the stability of higher-order chromatin structure.
28: 227–23617938198WangZ, ZangC, CuiK, SchonesDE, BarskiA, PengW, ZhaoK2009Genome-wide mapping of HATs and HDACs reveals distinct functions in active and inactive genes.
138: 1019–103119698979YangXJ, UllahM2007MOZ and MORF, two large MYSTic HATs in normal and cancer stem cells.
26: 5408–541917694082oai:pubmedcentral.nih.gov:37924782013-10-21genesdevpmc-openGenes DevGenes DevGADGenes & Development0890-93691549-5477Cold Spring Harbor Laboratory PressPMC3792478PMC379247837924782406576824065768871166010.1101/gad.221960.113Research PaperSpliceosome-mediated decay (SMD) regulates expression of nonintronic genes in budding yeastVolanakis et al.SMD regulates gene expressionVolanakisAdam13PassoniMonica13HectorRalph D.2ShahSneha1KilchertCornelia1GrannemanSander24VasiljevaLidia14Department of Biochemistry, University of Oxford, Oxford OX1 3QU, United Kingdom;Institute for Structural and Molecular Biology, Centre for Synthetic and Systems Biology (SynthSys), University of Edinburgh, Edinburgh EH9 3JD, United Kingdom
This study uncovers a role for the spliceosome in regulating mRNA expression levels. Transcriptome-wide studies reveal splice junctions in transcripts that are not known to have introns in budding yeast. Volanakis et al. show that spliceosomal cleavage of bromodomain factor 2 (BDF2) mRNA generates unstable products degraded by the nuclear surveillance machinery, and BDF2 regulation requires its paralog, Bdf1. The authors thus propose a mechanism—termed spliceosome-mediated decay (SMD)—for the regulation of gene expression involving splicing coupled to RNA decay.
We uncovered a novel role for the spliceosome in regulating mRNA expression levels that involves splicing coupled to RNA decay, which we refer to as spliceosome-mediated decay (SMD). Our transcriptome-wide studies identified numerous transcripts that are not known to have introns but are spliced by the spliceosome at canonical splice sites in Saccharomyces cerevisiae. Products of SMD are primarily degraded by the nuclear RNA surveillance machinery. We demonstrate that SMD can significantly down-regulate mRNA levels; splicing at canonical splice sites in the bromodomain factor 2 (BDF2) transcript reduced transcript levels roughly threefold by generating unstable products that are rapidly degraded by the nuclear surveillance machinery. Regulation of BDF2 mRNA levels by SMD requires Bdf1, a functionally redundant Bdf2 paralog that plays a role in recruiting the spliceosome to the BDF2 mRNA. Interestingly, mutating BDF2 5′ splice site and branch point consensus sequences partially suppresses the bdf1Δ temperature-sensitive phenotype, suggesting that maintaining proper levels of Bdf2 via SMD is biologically important. We propose that the spliceosome can also repress protein-coding gene expression by promoting nuclear turnover of spliced RNA products and provide an insight for coordinated regulation of Bdf1 and Bdf2 levels in the cell.
The biogenesis of pre-mRNA involves multiple processing reactions, including 5′ capping, splicing, cleavage, and polyadenylation at the 3′ end, leading to production of translationally competent mRNA (Moore and Proudfoot 2009). In nuclear pre-mRNA splicing, the excision of introns is catalyzed by the spliceosome, a ribonucleoprotein machine comprising five small nuclear RNAs (snRNAs) and >100 conserved proteins (Wahl et al. 2009). Spliceosomal snRNAs contain a conserved PuAU4–6GPu sequence called the Sm site, which provides a platform for the assembly of the heteroheptameric Sm complex, comprised of seven Sm proteins: SmB/B′, SmD1, SmD2, SmD3, SmE, SmF, and SmG (Beggs 2005). These form a ring, referred to as the Sm complex, around the Sm site of the spliceosomal U1, U2, U4, and U5 snRNAs, and play roles in multiple aspects of small nuclear ribonucleoprotein (snRNP) biogenesis. The U6 snRNA associates with a structurally related set of seven Lsm (like Sm) proteins. Being a part of several snRNPs, the Sm complex was shown to facilitate assembly of the spliceosome on pre-mRNA (Zhang et al. 2001) and play a role in multiple aspects of snRNA biogenesis, such as cellular localization, processing, and stability.
The spliceosomal snRNPs and multiple non-snRNP proteins assemble cotranscriptionally on pre-mRNAs through recognition of the 5′ splice site (5′ss), the branch point (BP), and the 3′ss to form the spliceosome (Carrillo Oesterreich et al. 2011). This involves interactions of the spliceosomal components with the 5′ cap-binding complex and the C-terminal domain (CTD) of RNA polymerase II (RNA Pol II). Splicing factors such as yeast Prp40 and several human SR proteins were implicated in mediating interactions with the phosphorylated CTD (Morris and Greenleaf 2000; Kotovic et al. 2003). Previous analyses of spliceosomal recruitment to RNA Pol II-transcribed genes were performed by studying the genome-wide distribution of the individual components of the spliceosome using chromatin immunoprecipitation (ChIP) (Kotovic et al. 2003; Moore et al. 2006; Tardiff et al. 2006). Surprisingly, these studies revealed that spliceosomal components are also recruited to some genes whose transcripts are not known to be spliced. It was recently shown that several nonintronic transcripts can be spliced; however, it remained unclear whether this had any regulatory role or was the result of stochastic splicing events that occur due to the lack of specificity in the spliceosome recruitment (Harigaya and Parker 2012).
Unexpectedly, by sequencing SmD1-associated RNAs, we identified many RNA Pol II transcripts that are not known to be spliced, indicating that spliceosome assembly on nonintronic mRNAs may be a common phenomenon. To understand the functional significance of this discovery, we analyzed the effect of splicing mutants and strains defective in RNA surveillance on the protein-coding transcriptome. Remarkably, these mutants displayed a marked up-regulation of a number of intronless mRNAs, including bromodomain factor 2 (BDF2) mRNA, which encodes a bromodomain factor, and OYE3 mRNA, which encodes an NADPH oxidoreductase. Interestingly, it was previously shown that BDF2 overexpression is toxic, implying that it could be important to down-regulate levels of this transcript in the cell (Yoshikawa et al. 2011; Fu et al. 2013). We demonstrate that splicing of these mRNAs at consensus splice sites generates unstable products that are primarily degraded by the nuclear RNA surveillance machinery. This spliceosome-mediated decay (hereafter referred to as SMD) of BDF2 mRNA is dependent on Bdf1, another bromodomain-containing protein, providing a plausible explanation of how expression of these paralogous genes is regulated. Mutating BDF2 5′ss and BP consensus sequences suppressed bdf1Δ growth phenotypes, suggesting that maintaining proper levels of Bdf2 via SMD is biologically important. We propose a model in which SMD regulates coordinated expression of Bdf1 and Bdf2. Collectively, our results have revealed a new role for the spliceosome in the regulation of mRNA expression.
ResultsPurification and analyses of the Sm complex-bound RNAs
To identify RNA targets of the spliceosome, we purified spliceosome-associated RNAs by affinity chromatography using a strain expressing tandem affinity purification (TAP)-tagged SmD1, a protein required for stabilizing the U1 snRNA–pre-mRNA interaction (Zhang et al. 2001). As a negative control, we used a nontagged parental strain. SmD1-coprecipitated RNA was extracted and analyzed by Northern blotting and high-throughput RNA sequencing (RNA-seq). Methylene blue staining of total RNA showed a significant depletion of ribosomal RNAs (rRNAs) in the SmD1 pull-down compared with input (Supplemental Fig. 1A). Conversely, Northern analysis revealed substantial enrichment of the Sm site-containing RNAs U1 (Supplemental Fig. 1A) and telomerase (TLC1) (data not shown), validating the biological relevance of this approach. U1 was undetectable in the mock samples, demonstrating the specificity of the results (Supplemental Fig. 1A, lanes 5,7). Mass spectrometric analysis of coprecipitated proteins identified other Sm complex members (SmB, SmD1, SmD2, SmE, and SmG) and many spliceosome components (Supplemental Fig. 1B).
Bioinformatics analysis of the RNA-seq data (Supplemental Table 1) revealed that 35%–40% of uniquely mapped cDNAs corresponded to snRNAs directly bound by the Sm complex, with the highly abundant U2 snRNA being the most highly represented RNA. U6 snRNA was also present at low levels, indicating that the U4/U6 particle as well as late spliceosome assembly intermediates also coprecipitated (Will and Luhrmann 2011). Consistent with our Northern blotting data, the low-abundant TLC1 RNA that contains an Sm-binding site (Seto et al. 1999) was substantially enriched relative to other noncoding RNAs (ncRNAs), such as SCR1, which coprecipitated nonspecifically at low levels compared with control samples (Supplemental Table 1). Notably, the snR190 box C/D small nucleolar RNA (snoRNA) was also enriched, consistent with previously published data (Zagorski et al. 1988). Collectively, these data demonstrate the specificity and sensitivity of our analysis.
Of the top 500 enriched mRNAs, 412 were reproducibly identified in both data sets (Fig. 1A). These included 198 intron-containing mRNAs, representing ∼60% of the known spliced mRNAs in Saccharomyces cerevisiae (341) (Fig. 1B). It is possible that the fraction corresponding to the remaining 40% included pre-mRNAs that are rapidly spliced and therefore underrepresented in the sequencing data. Strikingly, more than half of the 412 mRNAs identified were annotated as nonintronic (Fig. 1B). We envisage two possible explanations for this observation. First, the Sm complex could assemble directly on mRNAs independently of the spliceosome. Indeed, in silico analyses revealed 2558 potential Sm-binding sites in intronless genes and 94 in the 213 Sm-associated mRNAs (data not shown). Alternatively, this interaction could be mediated via the spliceosome, which is known to be recruited to a subset of intronless mRNAs (Kotovic et al. 2003; Tardiff et al. 2006). The relevance of these interactions, however, was not addressed, begging the question of whether these mRNAs are spliced by the spliceosome. Our in silico analyses revealed that 44 of the 213 intronless mRNAs contained canonical 5′ss (GUA[U/C/A]GU) and BP (ACUAAC[G/A/U]) sequences in the correct orientation (Fig. 1C). Three were previously described (FBA1, UTH1, and PGI1) (Kotovic et al. 2003), and six had a putative Sm site ([A/G]AU4,6G[A/G]) within 50 base pairs (bp) upstream of the 5′ss (KAR2, KEM1, PSK1, STE6, and YPR045C). To validate our findings, we initially focused on the BDF2 mRNA, as it was most likely to be targeted by the spliceosome for the following reasons: (1) BDF2 RNA was highly enriched in both Sm-IP data sets comparable with intron-containing mRNAs (Supplemental Table 1) and has predicted 5′ss, BP, and potential 3′ss (Zhang et al. 2007) in the correct orientation (Fig. 1E). (2) U1, U2, and U5 components were found to chromatin-immunoprecipitate over the BDF2 gene at levels comparable with a typical intron-containing gene, RPL28A (Fig. 1D; Tardiff et al. 2006). (3) Intron tiling array data from the dbr1Δ strain, where the lariat intermediate is stabilized due to the loss of debranching enzyme, have indicated the presence of a putative intron in BDF2 mRNA (Zhang et al. 2007).
Sm proteins associate with nonintronic mRNAs. (A) Venn diagram showing the overlap between the top 500 mRNAs enriched in two separate SmD1-TAP experiments. (B) Venn diagram showing that of the 412 genes that reproducibly coprecipitated with the Sm complex, 198 contained introns, and 213 were annotated as intronless genes. The pink circle shows the 412 common genes from Sm-TAP immunoprecipitation; of these, 213 genes are nonintronic. The green circle shows 341 annotated intronic genes in S. cerevisiae; of these, 198 were found in Sm-TAP immunoprecipitation. (C) Venn diagram showing that out of all intronless genes, 835 are predicted to contain BP and 5′ss sequences in the correct order and have a potential to recruit the spliceosome. Forty-four genes with the splice site consensus were present in Sm-IP. (D,E) University of California at Santa Cruz Genome Browser outputs displaying the Sm RNA-seq results for RPL28A, BDF2, and neighboring genes. Included are the positions of the predicted 5′ss and BP splice signals. Spliceosome ChIP tiling array data (U1, U2, and U5) (Tardiff et al. 2006) and dbr1Δ tiling array data are indicated as “Arrays,” and log2 changes between mutant and wild type are shown (Tardiff et al. 2006; ENCODE Project Consortium et al. 2007).
The spliceosome endonucleolytically cleaves BDF2 mRNA
To determine whether expression of the intronless BDF2 mRNA is regulated by the spliceosome, we analyzed its steady-state levels in the prp40-1 temperature-sensitive strain. Prp40 is a subunit of the U1 snRNP involved in cotranscriptional recruitment of early splicing factors to pre-mRNAs and later steps of spliceosome assembly. Previous analysis has demonstrated global down-regulation of mRNA levels in this mutant as a result of splicing defects (Albulescu et al. 2012). As expected, at the nonpermissive temperature, mRNA levels of the intron-containing gene RPL28A were substantially reduced (Fig. 2A). In contrast, in this strain, BDF2 mRNA levels increased roughly threefold (Fig. 2B). Similarly, BDF2 mRNA levels were also elevated in U1-depleted cells and prp2-1 (Fig. 2C, lanes 1–4) and prp42-1 (data not shown) mutants, where the first catalytic step of the splicing reaction is affected. A mutant defective in catalysis of the second transesterification reaction (prp17-1) (Jones et al. 1995; Noble and Guthrie 1996; Pleiss et al. 2007) also accumulated BDF2 mRNA (Fig. 2C, lanes 5,6). Finally, a 2-base substitution introduced into the predicted 5′ss (Fig. 2D) also increased BDF2 mRNA levels (Fig. 2E). We conclude that the spliceosome plays a direct role in down-regulating BDF2 expression. To substantiate our results, we also performed Northern blots to measure levels of splicing intermediates in strains defective in splicing (prp40-1), nuclear RNA surveillance (rrp6Δ), and/or debranching of the intron lariat (dbr1Δ). Using a probe complementary to a region downstream from the predicted 5′ exon (probe 2) (Fig. 3A), we could readily detect a lariat in the dbr1Δ strain (Fig. 3B [lane 5], C [lane 2]). Cells lacking both Dbr1 and Rrp6 also accumulated a 3′ extended lariat intermediate (Fig. 3C [lanes 3,6], D [lane 3]). Excised exon 1, a product of the first splicing reaction, could also be detected in the rrp6Δ strain (Fig. 3E, lane 4), an exonuclease-defective dis3 mutant, and a rat1-1 temperature-sensitive mutant (Fig. 3E, lanes 6,8). We note that exon 1 fragments were heterogeneous in length, which could be the result of exosome-mediated degradation and/or low-fidelity splicing events. Deletion of the cytoplasmic exonuclease Xrn1 and the nonsense-mediated decay (NMD) factor Upf1 had no detectable effect on the stability of the exon 1 splicing products (Fig. 3E, lanes 5,7). Altogether, our data indicate that spliceosomal BDF2 cleavage products are mainly degraded in the nucleus by both 5′–3′ (Rat1), and 3′–5′ (Rrp6 and Dis3) exonucleases. In contrast to BDF2, we could not detect accumulation of spliced RPL28A 5′exons (Fig. 2A), which is consistent with published observations for typical intronic genes (Egecioglu and Chanfreau 2011). This is most likely due to the fast kinetics of the second splicing reaction.
BDF2 mRNA levels are up-regulated in the splicing mutants or upon 5′ss mutation. (A,B) Northern blot analysis on total RNA isolated from the parental strain (wild type [wt]) and prp40-1, prp40-1 rrp6Δ, and rrp6Δ strains. RNA was extracted 2 h after the shift to 37°C, resolved on agarose (A) or polyacrylamide gels (B), transferred to positively charged membrane, and probed for RPL28A (using 5′ exon-specific probe) and BDF2 mRNAs, respectively. The positions of pre-mRNA, mature mRNA, and 5′ exon (for RPL28A) species are indicated. ADH1 (A) or methylene blue-stained 25S and 18S rRNAs (B) are shown as loading controls. (C) Splicing mutants accumulate BDF2 mRNA. To analyze BDF2 mRNA levels, total RNA was extracted from U1 snRNA-depleted cells (lane 2) and prp2-1 (lanes 3,4) and prp17-1 (lanes 5,6) mutants. Cultures were shifted to nonpermissive temperature (37°C) for 1 h (prp2-1 and prp17-1) or grown in the presence of glucose for 8 h to allow for U1 depletion. (D) Schematic representation of the predicted splicing signals in BDF2. The purple nucleotides indicate the substitutions that were made in the 5′ss (red) to generate the 5′ss* construct. (E) Northern blot analysis of BDF2 mRNA levels in wild-type (BDF2) or 5′ss mutant (5′ss*) cells.
The spliceosome performs a one-step splicing reaction on BDF2 mRNA. (A) Schematic representation of the BDF2 gene. The approximate positions of probes used to detect BDF2 mRNA and spliced products are indicated. (B,C) A BDF2 intron–lariat accumulates in the exosome and debranching mutants. Northern blot analysis of the BDF2 transcript was performed on RNA extracted from wild-type (wt), prp40-1, prp40-1 rrp6Δ, rrp6Δ, dbr1Δ, and dbr1Δrrp6Δ strains. Probes 2 and 3 were used to detect the BDF2 mRNA and lariat–intron splicing intermediates. rRNAs are shown as a loading control. A schematic representation of splicing intermediates is indicated at the side of each panel. (D) Primer extension analysis of the BDF2 splicing intermediate in YF336, dbr1Δ, rrp6Δ dbr1Δ, and rrp6Δ strains. The position of the primer used in this experiment is indicated in the diagram. Primer extension products were separated on a 10% urea-PAGE. The arrow indicates the position of the primer extension product corresponding to the 3′ extended intron–lariat. Primer extension was performed on PMA1 to control for the RNA levels. (E) Released BDF2 exon 1 is degraded in the nucleus. Northern blotting was performed on RNA extracted from the wild-type, prp40-1, prp40-1 rrp6Δ, rrp6Δ, upf1Δ, xrn1Δ, dis3exo-, and rat1-1 strains as described above. Probe 1 was used to detect the exon 1 fragment. (F) Mutations in the 5′ss and BP abolish the first step of BDF2 splicing. Shown is a schematic representation of the mutations introduced to BDF2 splice sites and Northern blot analysis of BDF2 exon 1 in the splice site mutants. (G) The first step of splicing is affected in the prp17-1 mutant. Exon 1 levels were analyzed by Northern blot. RNA was extracted from the wild-type (DIS3), prp17-1, and prp17-1 dis3 exo- strains.
Accumulation of BDF2 5′ exon was also dependent on intact 5′ss and the BP sequences (Fig. 3F, lanes 3–6). In contrast, mutations in three predicted 3′ss had no noticeable effect on exon 1 levels or BDF2 mRNA levels in general (Supplemental Fig. 2A; data not shown), indicating that splicing at these sites does not occur or is very inefficient. We conclude that splicing products generated using these splice elements are products of a single splicing step and are degraded by the nuclear surveillance system. This indicates that the second step of splicing in the BDF2 mRNA is inefficient, as it was reported for the telomerase RNA (Box et al. 2008; Kannan et al. 2013). BDF2 mRNA levels also increased in the prp17-1 mutant, suggesting that BDF2 can be fully spliced (Fig. 2C); however, we also noticed a reduction in exon 1 levels in the prp17-1 strain. Consistent with previous reports, this suggests that first-step splicing is also delayed in this mutant (Sapra et al. 2008), making it difficult to determine what fraction of BDF2 is fully spliced.
Shortening BP–3′ss distance partially restores the second step of splicing in BDF2
The distance between the BP and 3′ss was reported to be a factor affecting efficient 3′ss selection during the second catalytic step of splicing (Luukkonen and Seraphin 1997; Box et al. 2008; Meyer et al. 2011; Kannan et al. 2013). Hence, we reasoned that the unusual long distance between these two elements in the BDF2 mRNA might be responsible for the accumulation of the RNA species that are produced after the first step of splicing. To test whether BDF2 mRNA has a potential to be fully spliced, we generated several BDF2 constructs in which the distance between the BP and previously predicted 3′ss sites (Zhang et al. 2007) was shortened (Supplemental Fig. 2B, Δ1–Δ3). Most of the tested constructs failed to allow complete splicing at detectable levels even using a sensitive technique such as RT–PCR (Supplemental Fig. 2C, lanes 1,3,4,6). Intriguingly, low levels of spliced BDF2 Δ2 product could be detected in an Rrp6 deletion strain (Supplemental Fig. 2C, lane 5), indicating that this product is degraded by the nuclear exosome. Sequencing of the PCR product revealed that this indeed represented a spliced BDF2 mRNA in which the distal 3′ss was used (Supplemental Fig. 2D). We conclude that the previously proposed splicing signals are suboptimally positioned in BDF2 mRNA to promote an efficient second step of splicing.
Both a single splicing event and complete splicing contribute to regulation of BDF2 levels via SMD
While this manuscript was in preparation, the Parker laboratory (Harigaya and Parker 2012) reported splicing of BDF2 in cells lacking Xrn1 and a cytoplasmic decapping factor, Dcp2. However, the biological relevance of this splicing event had not been addressed (Harigaya and Parker 2012). Sequencing of the spliced product revealed that the 5′ss reported here and a more downstream 3′ss at nucleotide +1672 were used (Fig. 4A; Harigaya and Parker 2012). A BP consensus located 12 nucleotides (nt) upstream of this 3′ss was proposed; however, usage of this element was not experimentally validated.
Two-step splicing also contributes to regulation of BDF2 levels. (A) Shown is a schematic representation of the BDF2 splice sites, with the position relative to ATG (+1) indicated. RT product is indicated in red, and primers used for PCR are shown as black arrows. (B) RT–PCR results performed on oligo[d(T)]-enriched RNA extracted from the wild-type (wt), rrp6Δ, upf1Δ, and xrn1Δ dcp2Δ strains using BDF2 and ADH1 primers. RT–PCR reactions without enzyme (RT−) are included as a negative control. The spliced product is indicated. (C) RT–PCR results performed on poly(A)-enriched RNA extracted from strains expressing the wild-type (BDF2) and BDF2 5′ss, BP (+254), BP (+1660) (changing AAC to ACC), and 3′ss (+1672) (changing AG to AA) mutants in bdf2Δ rrp6Δ. (D) Northern blot analysis of exon 1 levels in the strains indicated.
We were able to detect splicing at these sites in the wild-type strain; however, no significant additional accumulation of the spliced product was in the xrn1Δ dcp2Δ double mutant (Fig. 4B, lanes 1,4). Deleting UPF1 also did not noticeably stabilize the spliced product (Fig. 4B, lane 3), indicating that the cytoplasmic RNA surveillance machinery plays a minor role in the degradation of the spliced product. In sharp contrast, deleting RRP6 led to an ∼4.5-fold increase in the level of the spliced product (Fig. 4B, lane 2). We conclude that SMD triggers the degradation of partially and completely spliced BDF2 products by the nuclear surveillance machinery.
To address the importance of this two-step splicing event in the regulation of BDF2 expression in more detail, we introduced mutations in the splice sites proposed by Harigaya and Parker (2012). BDF2 mRNAs with mutations in the 5′ss or the BP at +1660 (Fig. 4A) were not detectably spliced, indicating that these sites are indeed used by the spliceosome. Interestingly, although the BP sequence at +254 (Fig. 4A) is clearly important for the single-step splicing event that leads to the accumulation of the exon 1 fragment in the rrp6Δ strain (Fig. 3F), it is not used for the two-step splicing reaction (Fig. 4C, lane 3), presumably because the distance between this BP and the downstream 3′ss is far from optimal to support splicing (>1400 nt). Surprisingly, mutating the 3′ss at +1672 did not block splicing of BDF2, as an alternative 3′ss was used at +1702 (Fig. 4C, lane 5). Mutations in the 5′ss or the two BPs led to a twofold to threefold increase in the BDF2 mRNA levels (Fig. 4C, lanes 2–4). Mutation in the 3′ss at +1672 had only a modest effect on BDF2 steady-state levels (Fig. 4C, lane 5), presumably because alternative 3′ss could still be used. In contrast, the BP at +1660 and the 3′ss at +1672 are not required for the single-step splicing event, as we could still detect accumulation of exon 1 in the rrp6Δ strain in cells expressing these BDF2 mutations (Fig. 4D).
We conclude that SMD of BDF2 involves (at least) two separate splicing events: a single cleavage event, which requires the BP at +254, and a two-step splicing event involving the BP at +1660. Both pathways use the same 5′ss, and spliced products generated by these pathways are (primarily) degraded by the nuclear surveillance machinery.
Spliceosome recruitment to BDF2 mRNA is compromised in bdf1Δ
Bdf2 and the closely related Bdf1 both bind to acetylated histones, albeit with different specificities (Matangkasombut and Buratowski 2003). These genes are genetically redundant, as only one of the two genes is necessary and sufficient for cell viability. Deletion of both genes is lethal (Matangkasombut et al. 2000). Interestingly, deletion of BDF1 resulted in an approximately threefold increase in the levels of BDF2 mRNA (Fig. 5A; Fu et al. 2013), similar to what we observed in the splicing mutants (Fig. 2), and a twofold increase in Bdf2 protein levels (Fig. 5B). Although Bdf1 binds to a BDF2 promoter element (Durant and Pugh 2007; Fu et al. 2013), nuclear run-on experiments demonstrated that deletion of BDF1 did not noticeably affect BDF2 transcription (Fig. 5C), ruling out that Bdf1 regulates Bdf2 at the transcriptional level. Deletion of BDF1 is known to reduce splicing of a subset of intron-containing transcripts (Albulescu et al. 2012), but the effect of BDF1 deletion on intronless mRNA levels was not reported. This prompted us to investigate whether Bdf1 is required for spliceosome recruitment to BDF2. ChIP assays revealed a substantial U1 snRNP enrichment over the BDF2 gene, in agreement with previously published data (Fig. 5D,E [lanes 1–4], F; Tardiff et al. 2006). Strikingly, deleting BDF1 dramatically reduced the U1 binding to BDF2 (Fig. 5E [lanes 5–8], F). Similarly, deletion of BDF1 was reported to result in compromised U1 recruitment to several intron-containing genes, implying that Bdf1 might play a general role in splicing (Albulescu et al. 2012). We conclude that BDF1 down-regulates BDF2 mRNA levels by stimulating splicing of BDF2.
Bdf1 is required for the recruitment of the spliceosome at the BDF2 locus. (A) Northern blot analysis of BDF2 mRNA levels in the wild type (wt) and bdf1Δ. (B) Western blot analysis of Bdf2-TAP and Rpb1 (3E1 antibody from Millipore) protein levels in the wild type and bdf1Δ. (C) Deletion of Bdf1 does not noticeably affect BDF2 transcription. Transcriptional run-on experiments were performed to measure transcription on BDF2 in wild-type and bdf1Δ strains. RNA probes corresponding to BDF2 [probe 1 (exon 1) and probe 2], ADH1 and a nontranscribed region on chromosome V (nc) were used for detection. (D) Schematic representations of the BDF2 gene and BDF2 mRNA with the predicted 5′ exon (exon 1) and BP. Black bars show the locations of the PCR products generated in the ChIP experiments. (E,F) Deletion of BDF1 reduces spliceosome recruitment to BDF2. ChIP experiments were performed in wild-type and bdf1Δ strains expressing TAP-tagged Yhc1 (U1C), a U1 snRNP component. U1 enrichment was measured by radioactive PCR using primers that generate the PCR fragments shown in D. PCR products were resolved on polyacrylamide gels and detected by autoradiography. Quantification results from four independent experiments are shown in F; the error bars indicate the standard error. Stars indicate position of the PCR product from a nontranscribed region on the chromosome. (G–I) Mutating BDF2 5′ss and BPs suppresses bdf1Δ salt- and temperature-sensitive phenotypes. The bdf1Δ strains expressing either BDF2 or bdf2 5′ss, 3′ss, and BP mutants were grown in liquid medium until OD(A600) reached 0.5. To assay for growth, 5-μL aliquots of 10-fold serial dilutions were spotted onto YPD and −LEU medium in the presence or absence of 0.6 M NaCl.
Disrupting BDF2 splicing suppresses bdf1Δ temperature- and salt-sensitive phenotypes
To address the biological significance of regulation of BDF2 mRNA levels via SMD, we used a genetic approach that relied on growth defects observed in the bdf1Δ strain. Overexpression of BDF2 was recently shown to suppress bdf1Δ temperature- and salt-sensitive phenotypes in a dosage-dependent manner (Matangkasombut et al. 2000; Fu et al. 2013). We asked whether completely blocking splicing of BDF2 by mutating the 5′ss or blocking two-step splicing by mutating the BP at +1660 (see Fig. 4A) could at least partially suppress bdf1Δ growth defects. Judging from the number of colonies in each dilution and the colony size, it can be concluded that at the nonpermissive temperature (37°C), bdf1Δ strains expressing BDF2 mRNAs with mutated 5′ss grew better compared with the bdf1Δ cells expressing wild-type BDF2 (approximately eightfold to 10-fold) (Fig. 5G). Moreover, completely blocking BDF2 splicing substantially enhanced the cells' resistance (∼30-fold to 50-fold compared with bdf1Δ BDF2) to a combination of high temperature and high salt concentrations (0.6 M NaCl) (Fig. 5H). Finally, mutating the BPs also improved growth in high salt conditions (Fig. 5I). Collectively, these results demonstrate that even a relatively small change in BDF2 mRNA and protein levels can dramatically affect a cell's resistance to certain stress conditions, underscoring the impact that SMD can have on cell fitness.
SMD may regulate expression of ∼1% of the intronless genes
Several recent studies have described interactions between the spliceosome and mRNAs that were not known to contain introns (Kotovic et al. 2003; Moore et al. 2006; Tardiff et al. 2006; Harigaya and Parker 2012). Our bioinformatics analysis indicates that >800 genes encoding intronless mRNA have canonical 5′ss and BP sequences in the correct orientation (Fig. 1C), suggesting that the spliceosome could potentially target and regulate expression of hundreds of intronless mRNAs. To test this possibility, we performed RNA-seq analyses on rRNA-depleted RNA extracted from a wild-type strain and the temperature-sensitive prp40-1 and rrp6Δ mutants.
Based on our analysis of BDF2 mRNA, we reasoned that splicing products derived from intronless genes would be stabilized upon deleting Rrp6, whereas unspliced or full-length transcripts would accumulate in the prp40-1 mutant. RNA was extracted from these strains 2 h after the shift to the nonpermissive temperature. This was sufficient to detect accumulation of pre-mRNAs (prp40-1) and 3′ extended RNA species (rrp6Δ) (Supplemental Fig. 3). As expected, compared with the parental strain, the levels of intron-containing mRNAs—in particular, ribosomal protein mRNAs—were substantially reduced in the prp40-1 strain (Supplemental Fig. 4A). This global reduction of spliced mRNAs in the prp40-1 mutant was likely the result of degradation of pre-mRNAs that accumulated in this mutant (Bousquet-Antonelli et al. 2000; Hilleren and Parker 2003). Despite this, the prp40-1 data set contained a higher number of cDNAs that mapped to introns (Supplemental Fig. 4B), indicative of pre-mRNA accumulation due to defective splicing in this strain.
Remarkably, in sharp contrast to intron-containing mRNAs, a large number of intronless mRNAs were up-regulated in the prp40-1 mutant, indicative of a potential regulation by SMD (Supplemental Fig. 4A). However, we cannot rule out a possibility that increase in the levels of some of these intronless mRNAs in the prp40-1 mutant is the result of pleiotropic effects. Compared with the parental strain, levels of 197 intronless transcripts increased at least threefold in the prp40-1 mutant, and these were significantly enriched for genes involved in the unfolded protein response (P-value < 0.001). Twenty-four of 197 (∼12%) contained predicted 5′ss and BPs in the right orientation, and 16 (∼8%) were also enriched in the Sm-IP data (Supplemental Fig. 4C). Our analyses identified four transcripts (OYE3, SSA4, KAR2, and BDF2) that we predicted were most likely to be SMD targets: They reproducibly coprecipitated with the Sm proteins and contained splice signals in the correct orientation, and their steady-state levels increased at least threefold in the prp40-1 mutant (RNA-seq and quantitative RT–PCR [qRT–PCR]) (Fig. 6A; Supplemental Fig. 4C).
Other intronless mRNAs might be targeted by SMD. (A) KAR2, SSA4, and OYE3 transcript levels increase at least threefold in the prp40-1 mutant. To validate the RNA-seq data, mRNA levels of the transcripts predicted to be targeted by SMD was measured by qRT–PCR. Error bars indicate standard deviations. (B) End point RT–PCR using oligonucleotides that span exon junctions confirms splicing of the KAR2, SSA4, and OYE3 transcripts. (C) OYE3 spliced products accumulate in the rrp6Δ strain. Spliced and unspliced OYE3 products were detected by Northern blot using a probe spanning the predicted intron. The BDF2 exon 1 probe was used as a positive control. (D) Spliced products are stabilized in exosome (rrp6Δ) and (upf1Δ) mutants. RNA from the RNA-seq experiments was subjected to qRT–PCR using oligonucleotides that span exon junctions. The data were normalized to the PPM2 reference gene. The Y-axis indicates the fold increase in the spliced RNA levels over the levels in the parental strain. Error bars indicate standard deviations.
Because SMD requires the activity of the nuclear RNA surveillance machinery, we looked for canonical splice junctions (GU/AG) in the sequencing data. The TopHat program (Kim and Salzberg 2011) identified splice junctions in the vast majority of known intronic mRNAs (79% [wild type] to 88% [rrp6Δ]) (Supplemental Fig. 5A). Splice junctions were also identified in 0.45% (wild type) to 1.1% (rrp6Δ) of mRNAs that are not known to contain introns. Based on these data, we predict that at least 1% of the intronless mRNAs might be subjected to SMD. For the vast majority of these transcripts, splicing would generate premature stop codons, and any spliced products would presumably be degraded. Indeed, in agreement with the unstable nature of SMD products, more splice junctions were identified in the rrp6Δ data set in addition to a higher number of reads mapped to protein-coding genes (Supplemental Fig. 5A). This indicates at least a partial stabilization of spliced products in the absence of Rrp6. Fewer splice junctions were recovered from the prp40-1 data set, indicating that many of the identified junctions originated from a splicing event (Supplemental Fig. 5A).
A total of 227 canonical splice junctions were found in 49 transcripts, including SSA4, KAR2, and OYE3 (Supplemental Table 2; Supplemental Fig. 5B), and 22 transcripts contained a BP consensus sequence within the predicted intron (Supplemental Table 3). Notably, splicing of OYE3 at the identified splice sites would generate an in-frame deletion (Supplemental Fig. 6). However, as with BDF2, splicing could also occur at an alternative downstream 3′ss, generating an out-of-frame product (Supplemental Fig. 6). Unfortunately, we did not find any BDF2 splice junctions in the RNA-seq data, but this could be due to the relatively low read coverage over BDF2. We did observe a modest increase in reads covering the predicted first exon in the rrp6Δ sequencing data (Supplemental Fig. 7), consistent with at least a partial stabilization of this product in the absence of Rrp6 (Fig. 2B).
A subset of the genes contained putative introns in the 5′ untranslated region (UTR) (CPA2, GCY1, PIR3, SBH2, TBC3, YGR210C, and YRO2) or 3′ UTR (MSL5, RPL41A, QDR2, and YOR097C). Splicing of these mRNAs is therefore not expected to alter the reading frame (Supplemental Table 2) but could potentially affect mRNA stability and/or translation efficiency.
Splicing of SSA4, OYE3, and KAR2 was confirmed by RT–PCR and sequencing of PCR products (Fig. 6B,D; data not shown), and a strong accumulation of OYE3 spliced products could be detected in the rrp6Δ strain by Northern blot (Fig. 6C). Interestingly, spliced mRNA products could be (partially) stabilized by deleting Rrp6 or Upf1, indicating that, in these cases, both the nuclear exosome and cytoplasmic surveillance machineries contribute to their degradation (Fig. 6D). Because not all of the factors involved in cytoplasmic mRNA decay were analyzed here, it is possible that the actual percentage of intronless genes regulated by SMD is higher than we predicted (1%).
We conclude that, in addition to BDF2, SMD also regulates expression of several other intronless genes.
Discussion
The Sm proteins play crucial roles in pre-mRNA splicing and exert their function by binding several spliceosomal snRNAs. We tandem affinity-purified RNAs associated with the Sm complex to identify novel Sm/splicesome targets. To our surprise, more than half of the top 500 recovered mRNAs were not known to contain introns. Interactions between spliceosome components and nonintronic RNAs have been reported by other groups (Kotovic et al. 2003; Moore et al. 2006; Tardiff et al. 2006; Harigaya and Parker 2012). However, the biological relevance of these interactions remained unclear. Intriguingly, recent studies have shown that the spliceosome is also used for purposes other than removing intron sequences. In particular, it has been reported that fission yeast uses (L)Sm proteins and the spliceosome to mature the 3′ end of the telomerase RNA, involving a single spliceosome-dependent cleavage step (Box et al. 2008; Tang et al. 2012). This encouraged us to investigate the association of the spliceosome with intronless transcripts in more detail. Our bioinformatics analyses show that many intronless mRNAs contain canonical splice signals in the correct orientation and therefore have the potential to recruit the splicing machinery.
To investigate the potential function of the spliceosome on intronless genes, we focused on BDF2, a gene encoding a bromodomain transcription factor, where we and others (Tardiff et al. 2006) observed recruitment of the spliceosome. Using yeast strains defective in nuclear RNA surveillance and pre-mRNA splicing, we discovered that splicing of the BDF2 mRNA led to an approximately threefold reduction in BDF2 mRNA levels, and the resulting splicing products were rapidly degraded. This phenomenon, which we refer to as SMD, down-regulates mRNA expression as opposed to a well-documented positive outcome of conventional splicing and could potentially play an important role in down-regulating transcript levels. Indeed, the 5′ss and BP consensus sequences are conserved in genes homologous to BDF2 in related yeast species (Supplemental Fig. 8). Transcriptome sequencing of splicing and exosome mutants identified several splicing events on transcripts that were previously regarded not to have introns. We furthermore demonstrate that KAR2, SSA4, and OYE3 mRNA levels are also regulated via SMD and identify dozens of other potential targets. Finally, we show that Bdf1, a Bdf2 paralog, is required for the recruitment of the spliceosome to BDF2, revealing a novel mechanism by which expression of paralogous genes can be regulated.
A growing body of evidence suggests that defective or inefficient pre-mRNA splicing leads to the production of nonfunctional and potentially toxic RNAs such as unspliced mRNA precursors as well as various splicing intermediates. These are targeted by the cellular RNA surveillance systems for destruction. The current model implicates spliceosomal components in the initial identification of defective splicing intermediates, which are discarded by the spliceosome and released to be degraded by the cellular RNA decay machineries (Houseley et al. 2006; Egecioglu and Chanfreau 2011; Parker 2012; Schmid and Jensen 2013). Unspliced RNAs are degraded by the exosome (from 3′ to 5′), Rat1/Xrn2, and Xrn1 (from 5′ to 3′) in both nuclear and cytoplasmic compartments. Our results show that BDF2 splicing intermediates that accumulate due to delayed second step of splicing are also degraded. It was previously proposed that splicing intermediates are degraded in the cytoplasm (Hilleren and Parker 2003; Sayani et al. 2008; Mayas et al. 2010; Harigaya and Parker 2012). However, here we demonstrate that splicing products are also degraded by the nuclear exosome, as they accumulate in cells lacking Rrp6. How are these RNAs recognized by the nuclear exosome? We speculate that recognition by the nuclear surveillance machinery is coupled to the nuclear retention times of these molecules. Growing evidence suggests that splicing and surveillance are tightly connected and that the fate of pre-mRNAs is dictated by competition between degradation, splicing, and export machineries (Gudipati et al. 2012; Sayani and Chanfreau 2012). Unspliced transcripts and splicing intermediates that are efficiently exported to the cytoplasm are presumably more likely to be degraded by the cytoplasmic RNA decay machinery, whereas transcripts with relatively slow export kinetics or long nuclear retention times are (also) targeted by the nuclear degradation pathway (Sayani and Chanfreau 2012). This model also implies that slow transition kinetics from the first to the second step of splicing could render RNA more susceptible to nuclear degradation. The efficiency of the 3′ss selection depends on several factors, including the distance of 3′ss from the BP, secondary structure of the intron, and interactions with the specific protein factors (Box et al. 2008; Meyer et al. 2011). Here we demonstrate that regulation of BDF2 by SMD can involve (at least) two different splicing events in which the same 5′ss is used but two different BP sequences (see Fig. 7). We speculate that these two BPs are competing with each other for splicing factors and that selection of the BP is stochastic. If the spliceosome uses the upstream BP (+254), then it is very unlikely that the second splicing step will take place, as the BP is more than several hundred nucleotides away from 3′ss elements (Zhang et al. 2007; Harigaya and Parker 2012). When the downstream BP is used, complete splicing can occur. Surprisingly, products generated by either pathway appear to be primarily degraded by the nuclear RNA surveillance machinery, as deletion of Rrp6 resulted in a substantial increase of partially and completely spliced products. Mutating either BP results in a twofold to threefold increase in BDF2 steady-state levels, suggesting that both pathways contribute significantly to SMD. The selection of BP and 3′ss signals is likely the rate-limiting step in splicing of BDF2, and any delay presumably leads to the activation of the nuclear surveillance machinery. This model is supported by a recent study from the Baumann laboratory (Kannan et al. 2013) that focused on dissecting the mechanism of a one-step splicing reaction required for 3′ end formation of the telomerase RNA in fission yeast. This study reports that slow kinetics of the second splicing reaction due to the suboptimal positioning of the 3′ss could lead to the structural rearrangements of the spliceosome complex and subsequent discard of the intermediates. We propose that, depending on the rate of the second step of splicing, the spliceosome could either direct RNA degradation (BDF2) or splice pre-mRNA (conventionally spliced mRNA-encoding genes). Importantly, each of these two scenarios leads to a completely different outcome: Production of unstable aberrant RNAs negatively impacts gene expression (Fig. 7), whereas conventional splicing produces functional protein-coding mRNA.
Model for the regulation of BDF2 expression via SMD. In the presence of Bdf1, the spliceosome assembles on the BDF2 mRNA. This leads to cleavage at the 5′ss and (1) subsequent release of exon 1 and (2) complete splicing. Splicing products are degraded by the nuclear exosome (and 5′–3′ exonuclease Rat1 may contribute as well), leading to down-regulation of Bdf2 levels. In the absence of Bdf1, the BDF2 mRNA is not efficiently recognized by the spliceosome, and mainly nonspliced full-length BDF2 mRNA is generated.
In S. cerevisiae, only a small proportion of genes contains introns and encodes transcripts that are spliced. As a possible explanation for this phenomenon, it was previously proposed that introns in many genes of the budding yeast were “lost” during the course of evolution (Fink 1987). However, recent studies as well as our data show recruitment of the spliceosome coinciding with the presence of conserved splice signals on genes that are not known to be spliced. It seems unlikely that these conserved sequences were retained through evolution without any functional implication. Indeed, we observed that a number of RNAs undergo splicing, creating fully spliced but nonfunctional RNAs that are subsequently degraded via SMD. While spliced BDF2 products are mainly targeted in the nucleus by SMD, some products of unproductive splicing can also be degraded by NMD in the cytoplasm.
It is not clear exactly how many other genes are regulated by SMD to the same degree as BDF2. Under the experimental conditions used, for some intronless transcripts in which we could detect splice junctions in the RNA-seq data, we could not detect a significant increase in pre-mRNA levels in splicing mutants, indicating that splicing of these mRNAs is inefficient. It is conceivable that the presence of (canonical) splice sites is sufficient to recruit the spliceosome and perhaps drive splicing to some degree, but for most transcripts, this is not sufficient to dramatically alter steady-state RNA levels. It is also possible that significant changes will only be detectable under specific growth or stress conditions. Examples of this regulation were reported in higher eukaryotes. Thus, the well-documented unproductive outcome of alternative splicing can be used to regulate the expression of specific genes by suppressing the production of a protein in the absence of the proper biological context (Lareau et al. 2007; Ni et al. 2007). For example, the expression of PTBP1 and PTBP2 factors is regulated via this mechanism in order to control their function during neuronal differentiation (Wollerton et al. 2004). In this scenario, alternative splicing results in production of a nonfunctional RNA degraded by NMD. Experiments in the laboratory are most frequently performed when cells are growing exponentially in high concentrations of glucose or galactose, conditions rarely found in nature. Under these conditions, cells devote the vast majority of their resources to the synthesis and splicing of pre-mRNAs encoding ribosomal proteins, masking the detection of less abundant or less efficient splicing events. Indeed, differential recruitment profiles of the spliceosome and degradation machineries to target substrates were reported in response to different carbon sources (Bousquet-Antonelli et al. 2000), environmental stress (Moore et al. 2006; Pleiss et al. 2007), the cell cycle (Reis and Campbell 2007; Mullen and Marzluff 2008), and meiosis (Moore et al. 2006; McPheeters et al. 2009; Cremona et al. 2011). BDF2 expression has also been reported to be significantly affected under different stress conditions (Causton et al. 2001; Tkach et al. 2012).
Consistent with recently published work, we also observed an increase in BDF2 mRNA levels upon deletion of another bromodomain-containing protein, BDF1, whose function in transcription initiation is redundant with Bdf2 (Matangkasombut et al. 2000; Durant and Pugh 2007). We reasoned that the spliceosome could potentially regulate Bdf2 function in the cell by modulating its levels and therefore analyzed recruitment of U1 snRNP in bdf1Δ. Indeed, our data suggest that the approximately threefold increase in BDF2 mRNA levels (and approximately twofold increase in protein level) in bdf1Δ is due to the reduced recruitment of the spliceosome to BDF2 (Fig. 7). The function of Bdf1 in regulating gene expression via splicing, in addition to its role in transcription initiation, may explain the multiple phenotypes associated with the lack of Bdf1. For example, BDF1 deletion causes slow growth, temperature sensitivity, salt hypersensitivity, flocculence in liquid culture, sensitivity to the DNA-damaging agent MMS, poor sporulation, and inability to grow on nonfermentable carbon sources (Lygerou et al. 1994; Chua and Roeder 1995; Liu et al. 2007).
It remains to be determined whether Bdf1 is directly involved in the recruitment of the splicing machinery or whether this is a secondary effect of its function in transcription. Deletion of Bdf1 was reported to affect snRNA levels, indicating a role in transcription of these genes (Lygerou et al. 1994); however, a more recent study has questioned these findings (Albulescu et al. 2012). We favor a model in which Bdf1 plays a direct role in regulating BDF2 expression via modulating spliceosome recruitment. The latter could be inferred from the fact that the presence of Bdf1 on the BDF2 gene (Venters et al. 2011) has no noticeable impact on transcription (Fig. 4C). Moreover, loss of Bdf1 was previously linked with genome-wide defective splicing and impaired recruitment of the spliceosome to spliced genes, implying a general role in splicing (Albulescu et al. 2012). As regulation at the level of chromatin has recently emerged as an additional regulatory mechanism affecting splicing (de Almeida and Carmo-Fonseca 2012), it would be interesting to examine whether Bdf1 function in splicing is mediated via recognition of the acetylated histones.
In conclusion, we propose that the cooperative contribution of multiple mechanisms during RNA processing—such as recruitment of the spliceosome, the checkpoint allowing the second step of splicing, and finally, recruitment of the exosome to the defective RNAs—is important to ensure dynamic regulation of gene expression in eukaryotes.
Materials and methodsYeast strains
The S. cerevisiae strains used in this study are listed in Supplemental Table 4. A list of the oligonucleotides used is given in Supplemental Table 5.
Northern blotting
RNA extractions and Northern blot experiments were performed as previously described (Vasiljeva and Buratowski 2006). Gene-specific PCR-generated fragments were used as probes using the oligonucleotides listed in Supplemental Table 5.
Purification of the Sm complex
TAP-tagged Sm complex was purified from 16 L of yeast culture essentially as described in Vasiljeva and Buratowski (2006). RNA was extracted from calmodulin-bound material and subjected to RNA-seq.
Transcriptional run-on
Transcriptional run-on was performed as described in Birse et al. (1997). Single-stranded probes for transcriptional run-on analysis were generated by in vitro transcription using T7 or SP6 polymerase from plasmids harboring regions corresponding to different regions of BDF2. BDF2- and ADH1-derived fragments were generated using the primers listed in Supplemental Table 5 and were cloned into the pCR-Blunt II-TOPO vector (Invitrogen).
ChIP experiments
ChIP procedures and quantification were performed as described (Keogh and Buratowski 2004; Kim et al. 2004). For temperature shift experiments, cells were incubated at 23°C until OD(A600) = 0.5, and cells were further incubated for 120 min at 37°C. The primers are listed in Supplemental Table 5.
High-throughput RNA-seq
Procedures for cDNA library preparation and RNA-seq are described in the Supplemental Material. Raw (fastq) and processed (bedgraph and GTF files) RNA-seq data can be downloaded from the NCBI Gene Expression Omnibus repository (http://www.ncbi.nlm.nih.gov/geo, accession number GSE49966).
Acknowledgments
We thank our colleagues at the University of Oxford—in particular, Shona Murphy and Nick Proudfoot—for their valuable comments on the manuscript, helpful discussions, and encouragement. We thank Jean Beggs for insightful suggestions and discussion. We are very grateful to Jeffrey Pleiss, Sherif Abou Elela, Roy Parker, and Steve Buratowski for strains and constructs. We thank the GenePool facility in Edinburgh for high-throughput sequencing. This work was supported by a research fellowship from the German Research Foundation (Deutsche Forschungsgemeinschaft, KI1657/1-1 for C.K.) and Wellcome Trust Research and Career fellowships (for S.G. and L.V.).
Supplemental material is available for this article.
Article is online at http://www.genesdev.org/cgi/doi/10.1101/gad.221960.113.
Freely available online through the Genes & Development Open Access option.
ReferencesAlbulescuLO, SabetN, GudipatiM, StepankiwN, BergmanZJ, HuffakerTC, PleissJA2012A quantitative, high-throughput reverse genetic screen reveals novel connections between Pre-mRNA splicing and 5′ and 3′ end transcript determinants.
8: e100253022479188BeggsJD2005Lsm proteins and RNA processing.
33: 433–43815916535BirseCE, LeeBA, HansenK, ProudfootNJ1997Transcriptional termination signals for RNA polymerase II in fission yeast.
16: 3633–36439218804Bousquet-AntonelliC, PresuttiC, TollerveyD2000Identification of a regulated pathway for nuclear pre-mRNA turnover.
102: 765–77511030620BoxJA, BunchJT, TangW, BaumannP2008Spliceosomal cleavage generates the 3′ end of telomerase RNA.
456: 910–91419052544Carrillo OesterreichF, BiebersteinN, NeugebauerKM2011Pause locally, splice globally.
21: 328–33521530266CaustonHC, RenB, KohSS, HarbisonCT, KaninE, JenningsEG, LeeTI, TrueHL, LanderES, YoungRA2001Remodeling of yeast genome expression in response to environmental changes.
12: 323–33711179418ChuaP, RoederGS1995Bdf1, a yeast chromosomal protein required for sporulation.
15: 3685–36967791775CremonaN, PotterK, WiseJA2011A meiotic gene regulatory cascade driven by alternative fates for newly synthesized transcripts.
22: 66–7721148298de AlmeidaSF, Carmo-FonsecaM2012Design principles of interconnections between chromatin and pre-mRNA splicing.
37: 248–25322398209DurantM, PughBF2007NuA4-directed chromatin transactions throughout the Saccharomyces cerevisiae genome.
27: 5327–533517526728EgeciogluDE, ChanfreauG2011Proofreading and spellchecking: A two-tier strategy for pre-mRNA splicing quality control.
17: 383–38921205840ENCODE Project Consortium, BirneyE, StamatoyannopoulosJA, DuttaA, GuigoR, GingerasTR, MarguliesEH, WengZ, SnyderM, DermitzakisET, 2007Identification and analysis of functional elements in 1% of the human genome by the ENCODE pilot project.
447: 799–81617571346FinkGR1987Pseudogenes in yeast?49: 5–63549000FuJ, HouJ, LiuL, ChenL, WangM, ShenY, ZhangZ, BaoX2013Interplay between BDF1 and BDF2 and their roles in regulating the yeast salt stress response.
280: 1991–200123452060GudipatiRK, XuZ, LebretonA, SeraphinB, SteinmetzLM, JacquierA, LibriD2012Extensive degradation of RNA precursors by the exosome in wild-type cells.
48: 409–42123000176HarigayaY, ParkerR2012Global analysis of mRNA decay intermediates in Saccharomyces cerevisiae.
109: 11764–1176922752303HillerenPJ, ParkerR2003Cytoplasmic degradation of splice-defective pre-mRNAs and intermediates.
12: 1453–146514690599HouseleyJ, LaCavaJ, TollerveyD2006RNA-quality control by the exosome.
7: 529–53916829983JonesMH, FrankDN, GuthrieC1995Characterization and functional ordering of Slu7p and Prp17p during the second step of pre-mRNA splicing in yeast.
92: 9687–96917568198KannanR, HartnettS, VoelkerRB, BerglundJA, StaleyJP, BaumannP2013Intronic sequence elements impede exon ligation and trigger a discard pathway that yields functional telomerase RNA in fission yeast.
27: 627–63823468430KeoghMC, BuratowskiS2004Using chromatin immunoprecipitation to map cotranscriptional mRNA processing in Saccharomyces cerevisiae.
257: 1–1614769992KimD, SalzbergSL2011TopHat-Fusion: An algorithm for discovery of novel fusion transcripts.
12: R7221835007KimM, AhnSH, KroganNJ, GreenblattJF, BuratowskiS2004Transitions in RNA polymerase II elongation complexes at the 3′ ends of genes.
23: 354–36414739930KotovicKM, LockshonD, BoricL, NeugebauerKM2003Cotranscriptional recruitment of the U1 snRNP to intron-containing genes in yeast.
23: 5768–577912897147LareauLF, InadaM, GreenRE, WengrodJC, BrennerSE2007Unproductive splicing of SR genes associated with highly conserved and ultraconserved DNA elements.
446: 926–92917361132LiuX, ZhangX, WangC, LiuL, LeiM, BaoX2007Genetic and comparative transcriptome analysis of bromodomain factor 1 in the salt stress response of Saccharomyces cerevisiae.
54: 325–33017334841LuukkonenBG, SeraphinB1997The role of branchpoint–3′ splice site spacing and interaction between intron terminal nucleotides in 3′ splice site selection in Saccharomyces cerevisiae.
16: 779–7929049307LygerouZ, ConesaC, LesageP, SwansonRN, RuetA, CarlsonM, SentenacA, SeraphinB1994The yeast BDF1 gene encodes a transcription factor involved in the expression of a broad class of genes including snRNAs.
22: 5332–53407816623MatangkasombutO, BuratowskiS2003Different sensitivities of bromodomain factors 1 and 2 to histone H4 acetylation.
11: 353–36312620224MatangkasombutO, BuratowskiRM, SwillingNW, BuratowskiS2000Bromodomain factor 1 corresponds to a missing piece of yeast TFIID.
14: 951–96210783167MayasRM, MaitaH, SemlowDR, StaleyJP2010Spliceosome discards intermediates via the DEAH box ATPase Prp43p.
107: 10020–1002520463285McPheetersDS, CremonaN, SunderS, ChenHM, AverbeckN, LeatherwoodJ, WiseJA2009A complex gene regulatory mechanism that operates at the nexus of multiple RNA processing decisions.
16: 255–26419198588MeyerM, PlassM, Perez-ValleJ, EyrasE, VilardellJ2011Deciphering 3′ss selection in the yeast genome reveals an RNA thermosensor that mediates alternative splicing.
43: 1033–103921925391MooreMJ, ProudfootNJ2009Pre-mRNA processing reaches back to transcription and ahead to translation.
136: 688–70019239889MooreMJ, SchwartzfarbEM, SilverPA, YuMC2006Differential recruitment of the splicing machinery during transcription predicts genome-wide patterns of mRNA splicing.
24: 903–91517189192MorrisDP, GreenleafAL2000The splicing factor, Prp40, binds the phosphorylated carboxyl-terminal domain of RNA polymerase II.
275: 39935–3994310978320MullenTE, MarzluffWF2008Degradation of histone mRNA requires oligouridylation followed by decapping and simultaneous degradation of the mRNA both 5′ to 3′ and 3′ to 5′.
22: 50–6518172165NiJZ, GrateL, DonohueJP, PrestonC, NobidaN, O'BrienG, ShiueL, ClarkTA, BlumeJE, AresMJr2007Ultraconserved elements are associated with homeostatic control of splicing regulators by alternative splicing and nonsense-mediated decay.
21: 708–71817369403NobleSM, GuthrieC1996Identification of novel genes required for yeast pre-mRNA splicing by means of cold-sensitive mutations.
143: 67–808722763ParkerR2012RNA degradation in Saccharomyces cerevisae.
191: 671–70222785621PleissJA, WhitworthGB, BergkesselM, GuthrieC2007Rapid, transcript-specific changes in splicing in response to environmental stress.
27: 928–93717889666ReisCC, CampbellJL2007Contribution of Trf4/5 and the nuclear exosome to genome stability through regulation of histone mRNA levels in Saccharomyces cerevisiae.
175: 993–101017179095SapraAK, KhandeliaP, VijayraghavanU2008The splicing factor Prp17 interacts with the U2, U5 and U6 snRNPs and associates with the spliceosome pre- and post-catalysis.
416: 365–37418691155SayaniS, ChanfreauGF2012Sequential RNA degradation pathways provide a fail-safe mechanism to limit the accumulation of unspliced transcripts in Saccharomyces cerevisiae.
18: 1563–157222753783SayaniS, JanisM, LeeCY, ToescaI, ChanfreauGF2008Widespread impact of nonsense-mediated mRNA decay on the yeast intronome.
31: 360–37018691968SchmidM, JensenTH2013Transcription-associated quality control of mRNP.
1829: 158–16822982197SetoAG, ZaugAJ, SobelSG, WolinSL, CechTR1999Saccharomyces cerevisiae telomerase is an Sm small nuclear ribonucleoprotein particle.
401: 177–18010490028TangW, KannanR, BlanchetteM, BaumannP2012Telomerase RNA biogenesis involves sequential binding by Sm and Lsm complexes.
484: 260–26422446625TardiffDF, LacadieSA, RosbashM2006A genome-wide analysis indicates that yeast pre-mRNA splicing is predominantly posttranscriptional.
24: 917–92917189193TkachJM, YimitA, LeeAY, RiffleM, CostanzoM, JaschobD, HendryJA, OuJ, MoffatJ, BooneC, 2012Dissecting DNA damage response pathways by analysing protein localization and abundance changes during DNA replication stress.
14: 966–97622842922VasiljevaL, BuratowskiS2006Nrd1 interacts with the nuclear exosome for 3′ processing of RNA polymerase II transcripts.
21: 239–24816427013VentersBJ, WachiS, MavrichTN, AndersenBE, JenaP, SinnamonAJ, JainP, RolleriNS, JiangC, Hemeryck-WalshC, 2011A comprehensive genomic binding map of gene and chromatin regulatory proteins in Saccharomyces.
41: 480–49221329885WahlMC, WillCL, LuhrmannR2009The spliceosome: Design principles of a dynamic RNP machine.
136: 701–71819239890WillCL, LuhrmannR2011Spliceosome structure and function.
3: a00370721441581WollertonMC, GoodingC, WagnerEJ, Garcia-BlancoMA, SmithCW2004Autoregulation of polypyrimidine tract binding protein by alternative splicing leading to nonsense-mediated decay.
13: 91–10014731397YoshikawaK, TanakaT, IdaY, FurusawaC, HirasawaT, ShimizuH2011Comprehensive phenotypic analysis of single-gene deletion and overexpression strains of Saccharomyces cerevisiae.
28: 349–36121341307ZagorskiJ, TollerveyD, FournierMJ1988Characterization of an SNR gene locus in Saccharomyces cerevisiae that specifies both dispensible and essential small nuclear RNAs.
8: 3282–32902850487ZhangD, AbovichN, RosbashM2001A biochemical function for the Sm complex.
7: 319–32911239461ZhangZ, HesselberthJR, FieldsS2007Genome-wide identification of spliced introns using a tiling microarray.
17: 503–50917351133oai%3Apubmedcentral.nih.gov%3A3792479!!!pmc!genesdev