Initiation of stem cell differentiation involves cell cycle-dependent regulation of developmental genes by Cyclin D

1Embryonic development, organogenesis, and tissue repair are controlled through the proliferation and differentiation of stem cells and progenitors. The importance of these mechanisms has been demonstrated in adult stem cells and also in pluripotent stem cells. Human embryonic stem cells (hESCs) represent an advantageous model system for investigating the cross-talk between cell cycle and cell fate choice. hESCs can proliferate indefinitely while maintaining the capacity to differentiate into the three primary germ layers neuroectoderm, mesoderm, and endoderm in vitro. Furthermore, hESCs are characterized by a specific cell cycle profile with a short G1 phase, while their differentiation is associated with an increase in G1 phase length. A recent study has shown that Cyclin D1–3/CDK4/6 complexes, which control G1 length and progression, also limit the transcriptional activity associated with Activin/Nodal signaling during the late G1 phase. This mechanism results in a cell cycle-dependent capacity of differentiation in which hESCs can only differentiate into endoderm in their early G1 phase and into neuroectoderm in their late G1. Finally, S- and G2/M-related mechanisms appear to control the inhibition of pluripotency upon differentiation, suggesting the existence of differentiation checkpoints during cell cycle progression in stem cells. These reports collectively demonstrate that cell cycle regulators have a direct function in early events of differentiation by controlling the activity of extracellular signals. Interestingly, cell cycle regulators are also known to function at the chromatin level and control gene transcription. For instance, Cyclin D proteins have been found on DNA in mouse cancer cell lines but also in mouse retinal tissue. However, the function of Cyclin D-mediated transcriptional mechanisms has never been studied in the context of stem cells and cell fate choice.

2Here we demonstrate that the cell cycle machinery is directly involved in the transcriptional initiation of developmental genes controlling primary germ layer specification. Cyclin D1, which is highly expressed in late G1, recruits transcriptional corepressors to endoderm genes and coactivators to neuroectoderm genes, thereby blocking or promoting the induction of the corresponding germ layers. Thus, Cyclin D1 acts as a cell cycle-dependent transcriptional regulator, which can direct lineage specification of stem cells. These results show how cell cycle regulators can orchestrate cell fate decisions by organising transcriptional networks in human stem cells.

3We recently showed that hESC differentiation is regulated by the cell cycle through mechanisms involving control of the Activin/Nodal signaling pathway via Smad2/3 phosphorylation by Cyclin D–CDK4/6. We also observed that constitutive expression of Cyclin D1 and, to a lesser extent, Cyclin D2 and Cyclin D3 can rapidly increase the expression of neuronal markers independently of Smad2/3 inhibition. These results suggested that Cyclin Ds might prime the hESCs toward neuronal differentiation independently of Smad2/3–CDK4/6 cross-talk. To explore this hypothesis further, we decided to perform teratoma assays as an unbiased approach to evaluate pluripotency of hESCs overexpressing GFP or Cyclin D1 (Fig. 1A–D). Histological analyses of the resulting tumors were performed to define the proportion of germ layer derivatives generated. These analyses revealed that teratomas derived from control GFP-hESCs contained similar proportions of derivatives from the three germ layers, while Cyclin D1-hESC-derived teratomas contained 77% of neuroectodermal tissues (Fig. 1A–D; Supplemental Fig. S1A–C). In addition, statistical analyses showed that neuroectoderm was the main germ layer affected by Cyclin D1 overexpression (X2= 152.5, Bonferroni corrected; P < 6.6 × 10−16, χ2 test). Thus, Cyclin D1 appears to trigger differentiation of hESCs toward the neuroectodermal lineage independently of the surrounding environment. Next, we investigated whether Cyclin D1 could promote neuroectoderm specification in the absence of CDK4/6 activity by taking advantage of a highly specific CDK inhibitor, PD0332991 (Supplemental Fig. S1D; ). The addition of this small molecule in culture medium and thus the absence of Smad2/3 inhibition by CDK4/6 were not sufficient to block Cyclin D1 overexpression from inducing neuroectoderm and repressing endoderm differentiation, and this was confirmed by CDK4/6 knockdown (Fig. 1E; Supplemental Fig. S1E–H). Similar effects were obtained by overexpressing in hESCs a Cyclin D1 K112E mutant (CycD1-K112E) (Fig. 1F,G) that does not bind and activate CDK4/6 (Supplemental Fig. S1I; ). Considered together, these findings confirm that Cyclin D1 can direct cell fate decisions of hESCs independently of CDK4/6 activity.

4The above results suggest the existence of cell-autonomous mechanisms allowing Cyclin D1 to direct cell fate choice. Interestingly, studies in mouse retinal tissue and mouse cancer lines have shown that Cyclin D1 can participate in transcriptional regulation. However, whether this cell cycle regulator could also have a similar role in pluripotency exit and stem cell differentiation is unknown. Hence, we decided to explore whether similar mechanisms could occur in hESCs and could help to explain the CDK4/6-independent function of Cyclin D1 in neuroectoderm specification. For that, we performed Western blot analyses to determine the subcellular localization of Cyclin D proteins in hESCs and during their differentiation. These analyses revealed that Cyclin D1–3 not only localize to cytoplasm but also reside on chromatin in pluripotent cells (Fig. 2A,B). Cyclin D1–3 could also be found on the chromatin of neuroectodermal derivatives (Fig. 2C) and, to a lesser extent, in endoderm/mesoderm cells (Fig. 2C; Supplemental Fig. S2A,B). Together, these observations suggest that Cyclin Ds could indeed have a function on chromatin in hESCs.

5To further uncover the mechanisms involving Cyclin Ds in the nucleus, we decided to perform a series of functional experiments focusing on Cyclin D1, since this is the most abundant Cyclin D on the chromatin while having the strongest effect on neuroectoderm differentiation (; data not shown). We first overexpressed in hESCs a mutant form of Cyclin D1 (Cyclin D1 T286A [CycD1-T286A]) that accumulates in the nucleus due to the absence of phosphorylation of the residue T286 (Fig. 2D,E; ). CycD1-T286A hESCs showed a pronounced loss in pluripotency and mesoderm/endoderm gene expression and an increased neuroectoderm gene expression when compared with GFP or wild-type Cyclin D1 hESCs (Fig. 2F,G; Supplemental Figs. S2C–G, S3A–F). Furthermore, the effects of CycD1-T286A were maintained even with CDK4/6 inhibition (Supplemental Fig. S3G). Collectively, these results confirm that Cyclin D1 can influence cell fate decisions through nuclear mechanisms independently of CDK4/6.

6Since Cyclin D1 controls accumulation of Smad2/3 into the cytoplasm during the late G1 phase of the cell cycle, we decided to determine whether the nuclear function of Cyclin D1 could be independent of the Activin/Nodal signaling pathway. FUCCI hESCs transitorily transfected with Flag-NLS-Smad2/3 could not overcome the inhibition of endoderm differentiation in late G1 (Fig. 2H), suggesting that forced nuclear import of Smad2/3 is not sufficient to bypass the mechanisms preventing this differentiation. Furthermore, inhibition of CDK4/6 results in only a limited increase in Sox17-expressing cells in endoderm differentiation induced in late G1, confirming the existence of Smad2/3 phosphorylation-independent mechanisms (Fig. 2H; Supplemental Fig. S3H–J). Finally, CycD1-K112E was able to significantly reduce the activity of a constitutive form of Activin receptor, ActRIB (ActRIB-T206A; P ≤ 0.05) (Fig. 2I), without affecting the binding of Smad2/3 to its target loci (Fig. 2J). Taken together, these data demonstrate that Cyclin D1 has a nuclear function in hESCs that promotes neuroectoderm and blocks endoderm specification independently of CDK4/6-mediated inhibition of Activin/Nodal–Smad2/3 signaling (Fig. 2K).

7The functional studies described above support the hypothesis that Cyclin D1 functions via chromatin binding for controlling cell fate choice in hESCs. Previous studies on Flag-tagged Cyclin D1 in mice have shown that this protein can regulate gene expression by interacting with transcription factors and epigenetic modifiers in cancer cells and mouse embryos. Consequently, we hypothesized that Cyclin D1's role in the nucleus could be to control transcription of genes that are involved in cell fate decisions of hESCs.

8To investigate this potential function, we performed genome-wide identification of endogenous Cyclin D1-binding sites in hESCs using chromatin immunoprecipitation (ChIP) followed by high-throughput sequencing (ChIP-seq). We identified 400 significantly read-enriched genomic regions (or peaks) to be reproducible between two independent biological replicates (P ≤ 1 × 10−5; false discovery rate [FDR] ≤0.01). Peaks were associated with 1519 target genes within regions spanning 50 kb upstream of and 50 kb downstream from the Cyclin D1 peaks, and about half of the genes are protein-coding (Fig. 3A; Supplemental Table S1). The majority of the peaks were overlapping with or located close to transcription start sites (Fig. 3B; Supplemental Tables S2, S3), and >30% of the Cyclin D1-binding sites in hESCs were located in proximal promoter regions within 5 kb of genes (Fig. 3C). Interestingly, Cyclin D1 also binds a large number of intergenic regions >10 kb upstream of or downstream from genes, which could represent enhancers and regulatory regions, since a proportion of these overlap with H3K4me1 and H3K27ac regions. Thus, Cyclin D1 could also control gene expression via long-distance interactions of enhancer–promoter regions. Together, these observations reinforce our hypothesis concerning the involvement of Cyclin D1 in transcriptional regulation of genes in hESCs.

9To investigate the putative functional pathways of the Cyclin D1 target genes, we performed gene ontology (GO) analysis. GO terms associated with the 740 protein-coding genes were significantly enriched for terms related to enzyme binding (P = 1.63 × 10−6) and chromosome organization (P = 3.32 × 10−5) (Supplemental Table S4). Representative Cyclin D1 target genes were chromosome segregation factors such as AURKB and CENPM (Fig. 3D) that have previously been identified as targets of Cyclin D1 in mouse embryonic fibroblasts and have been implicated in chromosome instability in cancers. Among the target genes were also replication protein RPA1, cell cycle regulator RGCC (Fig. 3D; Supplemental Table S4), and DNA topoisomerase TOP1. Importantly, the GO terms enriched for Cyclin D1 target loci also included cell morphogenesis (P = 1.1 × 10−5), anatomical development (P = 2.2 × 10−4), cell differentiation (P = 5.63 × 10−4), and DNA binding (P = 7.64 × 10−3) (Fig. 3E). These results suggest that Cyclin D1 could act as a transcriptional regulator during cell fate choice of stem cells.

10To further determine whether Cyclin D1 target genes included regulators of cell fate decisions, we compared their expression levels in endoderm, mesoderm, and neuroectoderm using published RNA sequencing (RNA-seq) data (Fig. 3F; ). We did not observe an overall uniform change in expression of all Cyclin D1 target genes, suggesting that Cyclin D1 does not act as a genome-wide repressor or activator of transcription (Supplemental Fig. S4B). However, Cyclin D1 could act as a locus-specific regulator of developmental genes, as, among its target loci, we found important inducers of germ layer specification that showed up-regulation primarily in neuroectoderm (cluster 3: OTX1, LMO1, and TIAM1) (Fig. 3F; Supplemental Table S5), endoderm (cluster 2: Wnt3, Smad2, and VEGFA) (Fig. 3F), or mesoderm (cluster 1: RARB, DKK1, and Smad6) (Fig. 3F) as well as genes that indicate a down-regulation upon stem cell differentiation to all three germ layers (cluster 4: Wnt6, GLI2, and LIN28A) (Fig. 3F).

11Since Cyclin D1 target genes included regulators of germ layer specification such as transcription factors and components of signaling pathways (Wnt, Bmp4, and Activin/Nodal/TGFβ), we decided to focus on a subset of these genes that showed a strong enrichment in Cyclin D1 binding in ChIP-seq analysis (Fig. 4A; Supplemental Fig. S4A–C). First, we confirmed the binding of Cyclin D1 to these genomic regions (Fig. 4B; Supplemental Fig. S4D). Of note, we also showed the absence CDK6 on the same genome regions by ChIP and quantitative PCR (ChIP-qPCR) (Supplemental Fig. S5A), thereby confirming that Cyclin D control of transcription is independent of its phosphorylation activity. Furthermore, these developmental target genes showed distinct expression patterns in hESCs differentiated into three germ layers using defined culture condition (Supplemental Fig. S5B,C). Importantly, the endoderm-specific genes were repressed by Cyclin D1 overexpression, whereas the neuroectoderm-specific genes were induced (Fig. 4C). Furthermore, Cyclin D1/Cyclin D2 double knockdown resulted in opposite changes of expression for these genes when compared wth overexpression (Fig. 4C), suggesting that Cyclin D1 could control stem cell fate choice by transcriptional induction of neuroectoderm genes (Pax2, Sox3, PBX1, DACH1, and Otx1) and transcriptional repression of endoderm genes (Sox18, Wnt3, and Smad2). Finally, we tested the binding of Cyclin D1 to its target loci in FUCCI hESCs sorted to distinct cell cycle phases. Since Cyclin Ds are specifically expressed in late G1, we hypothesized that Cyclin D1 would bind those genomic regions in specific phases of the cell cycle. Accordingly, ChIP-qPCR analyses demonstrated that Cyclin D1 binds promoters of neuroectoderm (Fig. 4D) and endoderm (Fig. 4E) factors specifically during the late G1 phase and G1/S transition. These data collectively suggested that Cyclin D1 could control the transcriptional activity of developmental genes by binding proximal regulatory regions in a cell cycle-dependent manner.

12To understand the mechanisms involved in Cyclin D1-dependent transcriptional regulation, we analyzed the impact of Cyclin D1 on the epigenetic landscape of its target genes. The knockdown of Cyclin D1 resulted in the loss of the histone modification H3K4me3 on neuroectoderm loci but not on endoderm loci (Fig. 5A). Thus, Cyclin D1 could be necessary for the deposition of this positive epigenetic mark. Interestingly, Cyclin D1 has previously been shown to interact with transcriptional coactivator and histone modifier p300 on the PPARγ-responsive element of the lipoprotein lipase promoter during adipocyte formation. We therefore performed coimmunoprecipitations (co-IPs) between Cyclin Ds and p300 from the chromatin fraction of hESCs, which indicated that these proteins can also form a complex in stem cells (Fig. 5B). Protein truncation analyses further showed that these interactions involved the domain of Cyclin D1 spanning residues 91–179, thereby confirming a direct interaction (Supplemental Fig. S6A). Finally, ChIP analyses showed that p300 recruitment to neuroectodermal loci was impaired in hESCs knocked down for the expression of Cyclin Ds, whereas overexpression of Cyclin D1 increases the presence of p300 on these loci (Fig. 5C; Supplemental Fig. S6B–D).

13In contrast to H3K4me3, the knockdown of Cyclin Ds led to the loss of the repressive histone mark H3K27me3 in endoderm loci (Fig. 5A). In agreement with this and with a previous study, we also found that Cyclin Ds can interact with the transcriptional repressor HDAC1 (Fig. 5B), and Cyclin D knockdown causes its loss on endoderm loci (Fig. 5D). We performed sequential ChIP of Cyclin D1 and HDAC1 or p300, validating the simultaneous binding of these transcriptional and cell cycle regulators on developmental gene promoters (Fig. 5E,F), while Cyclin D1 overexpression in hESCs increases HDAC1 binding to endoderm loci (Supplemental Fig. S6E,F). Altogether, these results imply that Cyclin Ds bind the promoters of developmental genes in stem cells, influencing their epigenetic status by interacting with epigenetic modifiers.

14Finally, we validated the functional importance of these interplays by generating a luciferase reporter system with regulatory regions bound by Cyclin D1 on its corresponding target loci. We cotransfected these luciferase constructs into hESCs with Cyclin D1–3 and analyzed promoter activities after 48 h. These experiments showed that Cyclin Ds could activate reporters for neuroectoderm genes (Fig. 5G), indicating the capacity of these cell cycle regulators to induce transcription. On the other hand, we observed that overexpression of Cyclin Ds limits their transcriptional activity of endoderm genes, confirming that these cell cycle regulators could also act as transcriptional repressors (Fig. 5G).

15These data collectively demonstrate that Cyclin Ds control the expression of regulators of neuroectoderm and endoderm differentiation in hESCs by recruiting transcriptional coactivators and corepressors onto their promoters regions.

16Cyclin D1 is likely to recruit epigenetic regulators onto distinct genomic regions by cooperating with sequence-specific transcription factors. To identify its potential coregulators, we performed de novo motif discovery analysis followed by a comparison of detected motifs against databases using MEME-ChIP on the DNA sequences associated with Cyclin D1-binding sites. The most significantly overrepresented motifs found corresponded to SP1, E2Fs, EGR1, Zfp281, FoxP1, CTCF, and Zfn263 (Fig. 6A; Supplemental Fig. S7A; full list in Supplemental Table S7). Next, we decided to characterize the functional cooperation between Cyclin D1 and the identified transcription factors in more detail, focusing on SP1 and E2F1/4/6 as being among the top hits identified by the transcription factor motif comparison. Co-IP experiments verified that Cyclin D1 and SP1 could be found in the same protein complexes on chromatin in hESCs (Fig. 6B). Transient overexpression of SP1 in hESCs resulted in the induction of SOX3, OTX1, and PAX2 reporter genes, suggesting that the binding of SP1 on these neuroectodermal gene promoters could be functional. Of note, sequential ChIP with Cyclin D1 and SP1 confirmed the interactions between these two proteins on DNA (Fig. 6D). Finally, Cyclin D1 overexpression did not alter the presence of SP1 (Supplemental Fig. S7B) on neuroectodermal promoters, while SP1 binding to neuroectoderm loci did not change significantly during the cell cycle (Fig. 6E).

17Concerning endoderm genes, co-IP (Fig. 6B; Supplemental Fig. S7D,E) and sequential ChIP (Fig. 6F) experiments confirmed that Cyclin D1 and E2Fs could be found in the same protein complex on Wnt3, Sox18, and INHBA regulatory regions. Transient overexpression of E2F1 was able to reduce the activity of endoderm reporter genes, while cotransfection of Cyclin D1 further strengthened this repressive effect (Fig. 6G), thereby implying that E2F1 and Cyclin D1 binding could be functional. The increase or decrease in Cyclin D1 expression did not affect the presence of E2F1 on Wnt3, Sox18, and INHBA promoters (Supplemental Fig. S7C), and E2F1 binding to DNA was not cell cycle-dependent (Fig. 6H). Therefore, E2F1 could bind endodermal promoters independently of cell cycle regulations. These results suggest that E2F1 is involved in transcriptionally repressive interaction with Cyclin D1 protein specifically in late G1.

18Taken together, these results reveal that the activity of transcription factors can be cell cycle-regulated through their interactions with Cyclin D1 (Fig. 7) and that this regulation establishes the epigenetic modifications necessary for transcriptional initiation of developmental master regulators during primary germ layer specification.

19Here we uncovered that, in addition to their cross-talk with the signaling pathway Activin/Nodal–Smad2/3, the mitogenic sensor Cyclin D1 also orchestrates cell fate decisions of stem cells via a complementary mechanism that involves a direct transcriptional regulation of developmental master regulators (Fig. 7). These results reveal the mechanisms by which cell fate decisions are initiated during the narrow opportunity window of G1 phase when Cyclin Ds are dynamically expressed. Accordingly, endoderm induction occurs in early G1 when the level of Cyclin Ds is low, allowing Smad2/3 to bind and activate endoderm genes. In late G1, Cyclin D1 is induced and forms a complex with locus-specific transcription factors to recruit transcriptional coactivators onto neuroectoderm genes and corepressors onto endoderm genes. This model explains the molecular mechanisms allowing cell cycle regulators to directly control differentiation by regulating transcription factor activity during cell cycle progression and provides further support for cell cycle-coordinated cell fate decisions in stem cells.

20Importantly, similar cell cycle-dependent transcriptional mechanisms converging on developmental factors could apply to other cell cycle regulators in pluripotent cells. For instance, RB genes (pRB, Rbl1, and Rbl2) affect differentiation in mouse ESCs (mESCs) and hESCs. Furthermore, the CDK inhibitor p27 cooperates with Rbl2 to repress Sox2 expression during differentiation of mESCs while promoting the expression of mesoderm markers brachyury and twist in hESCs. Transcriptional regulation of developmental genes by cell cycle regulators could therefore be a common mechanism, and a complex cross-talk is likely to occur between these proteins.

21Our results also uncover the mechanisms by which cell cycle regulators are recruited to specific genomic locations in hESCs. Indeed, we performed for the first time a genome-wide survey of Cyclin D1 location on the stem cell chromatin, and this analysis proved that Cyclin Ds have the capacity to interact with transcription factors SP1 and E2F1. Interestingly, SP1 knockout mice have reduced embryo size and a diversity of phenotypes that are lethal before 9.5 d post-coitum (dpc). These observations suggest that SP1 is involved in controlling cell proliferation and tissue differentiation in the early embryo and thus that the mechanisms described in our model could be relevant for normal development. The importance of E2F1 in vivo is more complicated to establish, since E2F genes are known to be functionally redundant during embryonic development. However, pRB and E2F family members can control neuronal and pancreatic development by direct regulation of genes such as Ngn3. Thus, E2F factors could orchestrate stem cell and progenitor differentiation in a diversity of tissues, underlining that cell cycle control of cell fate choice could be a common mechanism between a diversity of progenitors.

22The absence of Cyclin Ds or CDK4/6 has only modest effects on early mouse development despite their universal function in primary and cancer cells. Functional redundancy between cell cycle regulators is frequently mentioned to explain this relative lack of role in vivo. Nevertheless, Cyclin Ds have a tissue-specific expression during gastrulation that resembles the pattern observed during differentiation of hESCs in vitro, suggesting that their role in the mechanisms controlling early cell fate decisions could be conserved in vivo. Cyclin D activity also promotes neuroectoderm formation during induced pluripotent stem cell generation through up-regulating Pax6, and, importantly, Cyclin D maintains the self-renewal of a broad number of adult stem cells, including mammary stem and progenitor cells, neuronal stem cells, and hematopoietic stem cells. Therefore, the basic mechanisms uncovered by these studies could also be relevant for adult stem cells and represent an important step toward understanding the balance between differentiation and self-renewal during organ development and repair.

23hESCs (H9 from WiCell) and mEpiSCs were grown in defined culture conditions as described previously. H9 cells were passaged weekly using collagenase IV and maintained in chemically defined medium (CDM) supplemented with 10 ng/mL Activin A and 12 ng/mL FGF2.

24hESCs were differentiated into neuroectoderm, endoderm, and mesoderm as described previously. Briefly, cells were cultured in CDM supplemented with 10 µM SB-431542 (Tocris) and 12 ng/mL FGF2 for neuroectoderm; in CDM + PVA supplemented with 100 ng/mL Activin A, 20 ng/mL FGF2, 10 ng/mL BMP4, 10 µM Ly294002 (Promega), and 3 µM CHIR99021 (Selleck) for mesoderm; and in CDM − PVA supplemented with 100 ng/mL Activin A, 20 ng/mL FGF2, 10 ng/mL BMP4, and 10 µM Ly294002 (Promega) for endoderm. hESCs were differentiated as described before.

25ChIP samples were sequenced at the Cambridge Institute Next-Generation Sequencing service of the University of Cambridge using Illumina HiSeq, and the raw data have been deposited on ArrayExpress under accession number E-MTAB-3807. Peak calling analysis was performed using PeakRanger, and only the peaks that were reproducible at an FDR of ≤0.01 in two biological replicates were selected for further processing.

26Extended Materials and Methods with the associated references are in the Supplemental Material.