|Home | About | Journals | Submit | Contact Us | Français|
Embryonic development is characterized by dynamic changes in gene expression, yet the role of chromatin remodeling in these cellular transitions remains elusive. To address this question, we profiled the transcriptome and select chromatin modifications at defined stages during pancreatic endocrine differentiation of human embryonic stem cells. We identify removal of Polycomb group (PcG)-mediated repression on stage-specific genes as a key mechanism for the induction of developmental regulators. Furthermore, we discover that silencing of transitory genes during lineage progression associates with reinstatement of PcG-dependent repression. Significantly, in vivo-, but not in vitro-differentiated endocrine cells exhibit close similarity to primary human islets in regard to transcriptome and chromatin structure. We further demonstrate that endocrine cells produced in vitro, do not fully eliminate PcG-mediated repression on endocrine-specific genes, likely contributing to their malfunction. These studies reveal dynamic chromatin remodeling during developmental lineage progression and identify possible strategies for improving cell differentiation in culture.
Embryonic development is a process characterized by rapid alternations in cellular states, resulting from dynamic changes in gene expression programs. These changes in gene expression are initiated by rapidly changing environmental cues, which affect cellular states by altering chromatin structure and gene transcription. Increasing evidence suggests that changes in specific histone modifications, particularly the trimethylation of histone H3 at lysine 4 and 27, may play a role in coordinating this highly regulated process. While H3K4me3 is found at virtually all active transcriptional start sites (TSS) (Schneider et al., 2004), H3K27me3 enrichment occurs in genes that are transcriptionally repressed by Polycomb group (PcG) proteins (Cao et al., 2002). Genome-wide studies have shown that PcG-dependent H3K27 trimethylation is not a universal repression mechanism, but represses a restricted subset of genes that typically encode for developmental regulators (Boyer et al., 2006; Lee et al., 2006). Notably, in undifferentiated embryonic stem cells (ESCs), a major portion of developmental genes that are trimethylated at H3K27 are also enriched in H3K4me3, creating a bivalent state in which genes are poised for future activation, but remain inactive until the repressive mark is removed (Bernstein et al., 2006). When ESCs are differentiated into neurons or fibroblasts, many of these genes resolve their bivalent state to either H3K27me3 or H3K4me3 (Mikkelsen et al., 2007). However, critical questions about the significance of H3K27me3 and H3K4me3 modifications in tissue and organ development remain unanswered. For example, it is still unclear whether H3K27me3- and H3K4me3-dependent changes in chromatin structure are associated with the rapid cellular transitions of development, whether bivalency can be reinstated once resolved, and whether this mechanism is relevant during terminal differentiation when cells acquire their functional properties.
The major limitation for investigating these important questions has been that transitory lineage intermediates cannot be isolated in sufficient numbers from embryos to conduct global chromatin profiling. This limitation can be overcome by recapitulating essential steps of in vivo development in vitro through directed differentiation of ESCs. Recently, a pancreatic differentiation protocol was developed that sequentially exposes human ESCs (hESCs) to different signaling factors, thereby moving cells stepwise through several precursor populations towards the pancreatic fate (D’Amour et al., 2006; Kroon et al., 2008). First, hESCs are induced to develop into definitive endoderm (DE), which is a transitory developmental cell population that gives rise to intestine, lungs, liver, and pancreas. DE is then converted into cells of the primitive gut tube (GT). Extrinsic cues, known to be involved in anterior-posterior regionalization of the GT, are subsequently applied to generate posterior foregut (FG), followed by the formation of pancreatic endoderm (PE). Upon implantation into mice, hESC-derived PE from late-stage cultures differentiates into glucose-responsive, insulin-secreting cells capable of reversing diabetes (Kelly et al., 2011; Kroon et al., 2008). Thus, this differentiation protocol enables us to explore how cells transition through intermediary developmental stages and to define chromatin remodeling mechanisms associated with these transitions.
In contrast to insulin-producing cells produced after engraftment, insulin+ cells generated in vitro are devoid of mature beta-cell characteristics. These cells produce little insulin, are not glucose-responsive, and frequently co-express other pancreatic hormones (D’Amour et al., 2006). Therefore, while PE generated by directed differentiation of hESCs is competent to differentiate into functional beta-cells when implanted into mice, at present, mature beta-cells cannot be produced in vitro. What remains to be determined is how closely the molecular features of hESC-derived beta-cells resemble those of their primary human counterparts and which of these features are insufficiently induced in vitro.
In this study, we examined changes in chromatin architecture associated with the rapid cellular transitions of development by conducting RNA-seq profiling as well as ChIP-seq for H3K27me3 and H3K4me3 of lineage intermediates during the stepwise pancreatic differentiation of hESCs in vitro, and after in vivo differentiation into functional endocrine (FE) cells in mice. We demonstrate that bivalency is highly dynamic and tightly associated with activation and silencing of developmental regulators during lineage progression. Moreover, we reveal that the chromatin of critical beta-cell genes is aberrantly remodeled during endocrine cell differentiation in vitro.
To characterize global gene expression and concomitant changes in chromatin structure during pancreatic lineage progression, we generated highly pure populations of PE and intermediate lineage precursors from hESCs in vitro. In addition, FE cells were produced by further differentiation in mice. To understand how endocrine differentiation is aberrantly regulated in vitro, we further analyzed in vitro-differentiated polyhormonal cells (PH) from late-stage cultures and compared PH to in vivo-generated FE cells (Fig. 1A). To assess purity of each intermediate cell population, we examined known markers for each stage by immunofluorescence staining, flow cytometry, and reverse transcription quantitative PCR (RT-qPCR). As expected, the transition from hESCs to the DE stage was associated with induction of the DE marker SOX17 in 97% of cells along with efficient repression of the pluripotency marker OCT4 (Fig. 1B,C; Fig. S1A,B,H). At the GT stage, SOX17 was drastically reduced, while the GT marker HNF4A became detectable in 99% of cells (Fig. 1D; Fig. S1B,C,H). Progression to the FG stage was accompanied by the appearance of the pancreatic progenitor marker PDX1 in 90% of cells (Fig. 1E; Fig. S1D,H). At the PE stage, additional pancreatic markers, including SOX9 and NKX6.1, became expressed (98% SOX9+ and 62% NKX6.1+ cells in PE; Fig. 1F,G; Fig. S1E,F,H).
Further in vitro differentiation led to induction of the pan-endocrine marker chromogranin A (CHGA) and pancreatic hormones in almost half of the cells (Fig. 1H–K; Fig. S1G,H). As reported (D’Amour et al., 2006; Kroon et al., 2008), endocrine cells co-expressed multiple hormones (Fig. 1H) and were mostly devoid of the beta-cell transcription factors (TFs) PDX1 and NKX6.1 (Fig. 1I,J). Hormone-negative cells in late-stage cultures continued to exhibit features of PE, expressing SOX9, PDX1, and NKX6.1 (Fig. 1I–K).
Implantation of cell aggregates into Scid-Beige mice resulted in glucose-dependent release of human C-peptide after 16–18 weeks (Table S1). In grafts retrieved 20 weeks post implantation, 90% of cells expressed CHGA (Fig. S1I,J). Similar to endocrine cells in the adult human pancreas, insulin+ cells in grafts were devoid of other hormones and strongly expressed NKX6.1, PDX1, and MAFB, while ARX was expressed in glucagon+ cells (Fig. 1L–N; Fig. S1K–M). RT-qPCR analysis of grafts further revealed induction of the mature beta-cell marker MAFA (Fig. S1H). The acinar marker trypsin was not detected (Fig. S1N) and small clusters of SOX9+ and SPP1+ cells with duct-like morphology (Fig. S1O) were rare, confirming that the majority of engrafted cells adopted endocrine characteristics. Together, our analysis demonstrates that this protocol allows for highly efficient generation of endocrine cells by synchronously moving cells through distinct lineage intermediates.
To identify global changes in gene expression associated with lineage progression of hESCs to functional endocrine cells, we performed RNA-seq analysis at six defined stages (ES, DE, GT, FG, PE, and FE). Bayesian clustering of mRNAs was performed to identify stage-specific signature genes (Fig. 2A; Table S2). As expected, the hESC-specific cluster included genes involved in maintenance of cellular pluripotency, such as OCT4 and SOX2 (Fig. 2A). Similarly, the DE cluster contained known regulators of endoderm formation, including EOMES, MIXL1, and SOX17 (Fig. 2A,B; Table S2A). Gene ontology (GO) analysis of DE signature genes revealed the expected enrichment for genes involved in endoderm development, gastrulation, and pattern specification (Fig. 2B). Moreover, members of the Wnt signaling pathway were enriched in the DE cluster, consistent with previously recognized roles for Wnt in DE formation (Liu et al., 1999).
The PE-specific cluster was enriched for genes involved in pancreas development and cell fate commitment (Fig. 2A,C; Table S2D), and included pancreatic progenitor markers, such as PDX1, SOX9, HNF6, NKX6.1, and PTF1A. GO term enrichment for neuron differentiation in the PE cluster is consistent with the previously observed similarity in gene expression profiles between neuronal and pancreatic precursors (van Arensbergen et al., 2010).
Genes in the FE-specific cluster were functionally associated with hormone transport, regulation of hormone levels, and maturity onset diabetes of the young (MODY) (Fig. 2A,D; Table S2E). These GO categories encompassed genes involved in glycolysis (e.g. GCK, G6PC2), insulin processing (e.g. PCSK1, PCSK2, SLC30A8), and insulin secretion (e.g. GLP1R, KCNJ11, ABCC8, FFAR1, UCN3), revealing that engraftment induces a complement of genes critical for beta-cell function. We also observed induction of TFs with still unknown roles in beta-cells, including NFIC, the beta-cell TFs HDAC9 and HOPX (Dorrell et al., 2011), and HIC1, which exhibits reduced expression in islets of diabetic rats (Liu et al., 2011). Overall, our transcriptome analysis shows that key developmental programs are stage-specifically regulated, suggesting that further examination of the gene signatures could enable discovery of novel lineage determinants.
In ESCs, genes encoding key lineage-specific regulators are bivalent, i.e. marked by both H3K4me3 and H3K27me3 (Bernstein et al., 2006; Pan et al., 2007). In striking contrast to ESCs, neural progenitor cells and adult stem cells of the hair follicle display very few genes in a bivalent state (Lien et al., 2011; Mikkelsen et al., 2007), suggesting that lineage commitment is associated with resolution of bivalency to either a fully active or repressed state. To study bivalency during pancreatic differentiation, we globally mapped H3K4me3 and H3K27me3 modifications in hESCs, in vivo-differentiated FE cells, and each lineage intermediate (Table S3). In hESCs, 17% of known transcripts (3182/18,822) displayed bivalency at their TSSs and the majority of these genes (78%) resolved bivalency during hESC differentiation to the FE stage (Fig. S2). In addition to loss of bivalency, we also observed that genes acquired a bivalent state during the differentiation process (Fig. 3A), resulting in a relatively constant overall number of bivalent genes during progression to PE. However, this percentage dropped after engraftment, with only 4.9% (918/18,822) of genes exhibiting bivalency at the FE stage. In total, the largest number of genes resolved the bivalent state during terminal endocrine differentiation.
To determine which classes of genes resolve bivalency during pancreatic differentiation, we performed GO analysis. Genes bivalent in hESCs that resolve to solely H3K4me3 at any time point during differentiation (n=1717) included many with roles in endocrine development and function (Fig. 3B). Consistent with the large representation of genes involved in endocrine cell function, a high percentage of these genes (46%) did not lose the repressive H3K27me3 mark until progression to the FE stage (Fig. 3B). Conversely, genes bivalent in hESCs undergoing loss of the active H3K4me3 mark during progression to the FE stage (n=147) were functionally associated with the development of non-endodermal lineages (Fig 3B). Surprisingly, the majority of these bivalently marked genes (61%) retained H3K4me3 until engraftment (Fig. 3B), suggesting that a large proportion of genes involved in the development of alternative lineages remain bivalent at the PE stage. These findings indicate that PE maintains a high degree of plasticity to adopt other lineage choices.
Of the promoters resolving or gaining bivalency, the majority changed their H3K27me3 state, while H3K4me3 modifications were far less dynamically regulated (Fig. 3C). This observation implies a significant role for PcG proteins in chromatin remodeling during pancreatic endocrine differentiation. To glean additional insight into the stage-specific dynamics of H3K27me3 modifications during differentiation, we employed hierarchical clustering of genes which exhibit a change in H3K27me3 levels at any time point during differentiation. From this analysis, the ES, DE, and FE stages surfaced as most distinct, while GT, FG, and PE stages had similar patterns (Fig. 3D). Therefore, we focused our subsequent analysis of stage-specific chromatin modifications on cellular transitions to the DE, PE, and FE stages.
To understand how chromatin is modified as DE-specific genes are rapidly induced and silenced during development, we examined histone profiles of DE signature genes. Consistent with the notion that lineage-specific developmental regulators display bivalent modifications in ESCs (Bernstein et al., 2006), many DE signature genes were bivalent in hESCs (37% of DE signature genes vs. 17% of all genes; P<1E-36, Hypergeometric Distribution; Fig. S3A). Sequential ChIP-qPCR on DE signature genes demonstrated a simultaneous presence of H3K4me3 and H3K27me3 at the ES stage (Fig. S4A,B), indicating concurrent enrichment of H3K4me3 and H3K27me3 at the same locus. Most (54%) DE signature genes bivalent in hESCs resolved to H3K4me3 at the DE stage, while genes solely marked by H3K4me3 or unmarked in hESCs showed little change in their histone profile (Fig. 4A). Strikingly, the list of bivalent DE signature genes exhibiting H3K27me3 removal featured functional association with endoderm formation and included most known transcriptional regulators of DE specification (e.g. GSC, EOMES, MIXL1, SOX17, CER1, and GATA6; Fig. 4A; Table S4). Reflecting the requirement for Wnt signaling in DE formation (Liu et al., 1999), the subset of DE signature genes resolving bivalency was also enriched for Wnt receptor signaling pathway components (Fig. 4A). These results suggest that removal of PcG-mediated H3K27me3 accumulation at DE regulators could be the predominant mechanism associated with endoderm induction.
To examine the functional relevance of PcG-dependent chromatin remodeling in DE induction, we prevented H3K27me3 removal during the transition from ES to DE by inhibiting Jumanji C domain-containing protein (JMJD3). JMJD3 antagonizes PcG-mediated gene silencing by demethylating H3K27me3 (Hong et al., 2007). We designed two independent shRNAs against JMJD3, which each led to an 80% reduction in JMJD3 levels (Fig. 4B). Analysis of JMJD3 knockdown cells at the DE stage revealed significantly reduced expression of the DE signature genes EOMES, LHX1, NRP2, and SOX17 (Fig. 4B). Furthermore, compared to cells transduced with control shRNA, JMJD3 knockdown cells retained higher H3K27me3 levels at the promoter regions of these DE signature genes (Fig. 4B). Our findings demonstrate that induction of DE fate determinants requires removal of H3K27me3 and provides direct evidence that modulation of PcG-mediated repression is necessary to initiate endoderm differentiation.
To gain insight into the chromatin modifications associated with the silencing of DE-specific programs during lineage progression, we next analyzed histone profiles of DE signature genes subsequent to the DE stage. We found that a subset of DE signature genes became bivalent again by progressively acquiring H3K27me3 at the GT and FG stages (Fig. 4C,D; Fig. S4A,B). The reacquisition of H3K27me3 after the DE stage raises the possibility that DE-specific programs are repressed by PcG-dependent mechanisms. Cluster analysis of histone methylation patterns revealed three distinct groups of DE signature genes: genes acquiring the H3K27me3 modification either at the GT or FG stage, and genes not acquiring this modification (Fig. 4D). While all three groups of DE genes exhibited decreased expression at the GT stage, expression of PcG-modified genes declined earlier than of genes silenced in a PcG-independent manner (Fig. 4D; Fig. S3B). Remarkably, TFs involved in DE formation, including the most upstream DE inducer EOMES (Teo et al., 2011), belonged to the group of early silenced genes acquiring H3K27me3 at the GT stage (Fig. 4D; Fig. S3B), suggesting a role for PcG proteins in the silencing DE-specific developmental regulators.
We next examined whether activation of genes during pancreatic lineage commitment coincides with the loss of PcG-dependent repression. Analysis of histone profiles of GT, FG, and PE signature genes (Fig. 2A; Table S2B–D) showed that all stage-specific signature genes exhibit a significantly higher propensity than total genes to be bivalent in hESCs (17% of all genes vs. 46% of GT (P<1E-17) vs. 31% of FG (P<1E-15) vs. 50% of PE (P<1E-31) signature genes, Hypergeometric Distribution; Fig. 5A). PE signature genes lost bivalency gradually during progression from the GT to PE stage (Fig. 5B), a finding we confirmed by sequential ChIP-qPCR (Fig. S4A,B). However, 26% of PE signature genes remained bivalent at the PE stage (Fig. 5B), which we reasoned could be because later pancreatic differentiation factors are just beginning to be expressed. For example, similar to its expression in the early pancreatic epithelium of mouse embryos (Henseleit et al., 2005), NKX6.1 was detected in only a subset of PDX1+ cells at the PE stage (Fig. 1F; Fig. S1F), but became uniformly expressed in non-endocrine cells at later stages (Fig. 1J). Therefore, we also analyzed the histone profiles of PE signature genes in further differentiated pancreatic endoderm isolated by fluorescence-activated cell sorting (FACS) from late-stage cultures (Late PE; Fig. S5A). As predicted, almost all PE signature genes (87%) had resolved bivalency to H3K4me3 in this Late PE population (Fig. 5B,C). Thus, similar to DE-specific genes, activation of PE-specific genes was mostly associated with removal of PcG-mediated repression (Fig. 5C). Strikingly, the short list of 104 genes displaying this pattern contained many known TFs necessary for early pancreas development, encompassing SOX9, PDX1, PTF1A, HNF6, NKX6.1, and NKX6.2 (Fig. 5C; Table S5). This list also contained several components of the Notch (HEY1, DLK1, HES4, and JAG1) and Netrin (DCC) signaling pathways, as well as the TFs MEIS1 and NFKB1 with still unknown functions in pancreas development. Immunohistochemical analysis of some of these Notch components and TFs in human fetal pancreas revealed expression in native pancreatic progenitors (Fig. S5B), suggesting that stage-specific gene activation and H3K27me3 loss is a strong predictor for putative regulators of human pancreatic development.
Analysis of the histone modification pattern of PE signature genes during lineage progression showed distinct groups of genes, exhibiting early, intermediate, or late reversal of PcG-mediated repression (Fig. 5D). Genes with early H3K27me3 loss displayed earlier onset of expression than genes maintaining H3K27me3 until later stages (Fig. 5D). Among the genes derepressed early were SOX9 and PDX1, while derepression of NKX6.1, NKX6.2, and PTF1A occurred later. The same order of gene activation is seen during mouse development (Seymour and Sander, 2011), indicating that this in vitro differentiation system mimics normal pancreas specification.
We next sought to determine which chromatin remodeling events occur during the in vivo differentiation of PE after engraftment. Surprisingly, we found that bivalency in hESCs was not a unique feature of developmental regulators, but that FE signature genes also displayed a higher propensity to be bivalent in hESCs than total genes (17% of all genes vs. 44% of FE signature genes (P<1E-57), Hypergeometric Distribution; Fig. 6A). Contrasting findings at earlier stages, only 8.7% of bivalently modified FE signature genes encoded TFs (compared to 14% of DE and 22% of PE signature genes bivalently modified in hESCs). 43% of these genes encoded transmembrane proteins, including KCNJ11, ABCC8, and GPR120, which are critical for beta-cell function. This was unexpected, because previous studies suggested that bivalent modifications in hESCs are predominantly relevant for future activation of TFs involved in lineage specification (Pan et al., 2007). A stage-wise heatmap of chromatin modifications revealed that the majority of FE signature genes bivalently marked at the PE stage were also bivalent in hESCs (Fig. 6A). Sequential ChIP-qPCR verified the presence of H3K4me3 and H3K27me3 on FE signature genes at all stages prior to the FE stage (Fig. S4A,B). Interestingly, almost all (87%) FE signature genes bivalently marked at the PE stage resolved to an active H3K4me3 mark after engraftment (Fig. 6B; Table S6A). This subset was enriched for genes encoding proteins regulating hormone levels (Fig. 6B). In contrast to earlier stages, induction of a substantial number of FE signature genes coincided with the acquisition of H3K4me3 (18% of DE (Fig. 4A) vs. 33% of PE (Fig. 5C) vs. 52% of FE signature genes (Fig. 6B)). These H3K4me3 “induced” genes mainly comprised genes encoding secreted proteins, such as INS, GCG, and IAPP (Fig. 6A,B; Table S6B). Together, these findings demonstrate that induction of endocrine-specific genes during pancreas differentiation is associated with both removal of repressive H3K27me3, as well as de novo acquisition of active H3K4me3 modifications.
To examine how closely hESC-derived FE cells resemble their in vivo counterparts, we directly compared endocrine cell composition, transcriptome, and H3K4/K27 trimethylation profiles of FE to primary human islets (HI). Intracellular flow cytometry and morphometric analysis for the relative abundance of different endocrine cell types revealed high similarity between FE and human islets (40% insulin+ cells in FE vs. 55% in HI; 27% glucagon+ cells in FE vs. 38% in HI; and 20% somatostatin+ cells in FE vs. 7% in HI; Fig. S6A; Cabrera et al., 2006). Notably, the cell type composition of human islets varies significantly between individuals and the percentage of insulin+ cells has been reported to range between 28% and 75% (Brissova et al., 2005). Comparison of mRNAs for all genes or FE signature genes demonstrated highly similar expression profiles of FE and human islets (Fig. S6B). Echoing the findings at the mRNA level, cluster analysis of histone modifications at FE signature genes (Fig. S6C) confirmed close proximity of hESC-derived in vivo-matured cells to human islets.
Previous studies have shown that insulin+ cells generated from hESCs in vitro lack important beta-cell characteristics, such as the ability to secrete insulin in a glucose-dependent manner (D’Amour et al., 2006). It has been further demonstrated that implantation of these cells into mice does not induce their maturation (Kelly et al., 2011). To define the distinguishing features between in vitro-generated endocrine cells (PH) and FE, we compared the expression profile of FACS-purified PH (Fig. S5A) with FE retrieved from grafts. Although PH and FE expressed a significant number of genes at comparable levels (n=9111, P<1E-15, Fisher’s Exact Test), 1438 genes displayed expression levels ≥ 2-fold higher in FE than PH (FE high genes; Fig. 7A; Table S7A). FE high genes associated with the GO categories of regulation of hormone secretion, hormone metabolic process, blood vessel development, and ECM-receptor interaction (Fig. 7B). Significantly, FE high genes were found similarly expressed in human islets (Fig. 7C). Comparison of H3K4me3 and H3K27me3 profiles of FE high genes in PE, PH, FE, and primary islets further revealed high similarity between PH and PE, but not between PH and FE or islets (Fig. 7D), suggesting that chromatin architecture is inappropriately remodeled during in vitro differentiation.
To determine whether aberrant chromatin modifications are associated with insufficient induction of beta-cell genes in vitro, we further analyzed the specific changes in histone modifications during the transition from PE to either the FE or PH stage. As expected, 82% (368/451) of FE high genes with bivalent marks at the PE stage lost the repressive H3K27me3 mark after engraftment (Fig. 7E). Strikingly, 48% (175/368) of these genes aberrantly retained H3K27me3 in PH (Fig. 7E; Table S7B). Another group of FE high genes was marked solely by H3K27me3 at the PE stage (n=36). While 75% of these genes lost the repressive H3K27me3 mark during in vivo differentiation to FE, only 44% underwent the same change during in vitro differentiation (Fig. 7E; Table S7B). Moreover, 50% of FE high genes with neither H3K4me3 nor K3K27me3 modifications in PE (n=496) acquired the active H3K4me3 mark after engraftment, but only 5.8% gained H3K4me3 in vitro (Fig. 7E; Table S7C). Together, these findings suggest that the insufficient induction of endocrine genes during in vitro differentiation is associated with inadequate changes in H3K27me3 and H3K4me3 modification patterns.
GO analysis of FE high genes remaining inappropriately PcG-repressed (n=186) or failing to acquire H3K4me3 (n=222) in PH revealed enrichment for genes with functions related to hormone activity, cell proliferation, and ion homeostasis (Fig. 7E). In addition to insulin, the list of genes showing aberrant histone modifications in PH included GLP1R, FFAR1, and UCN3 (Fig. 7E), which are required for glucose-stimulated insulin secretion (Itoh et al., 2003; Li et al., 2007; Preitner et al., 2004), Given that in vitro-generated insulin-expressing cells exhibit insufficient insulin production and lack glucose-regulated secretion (D’Amour et al., 2006), our data suggest that inappropriate chromatin remodeling during in vitro differentiation could contribute to the malfunction of these cells.
Due to the limited accessibility and quantity of transitory embryonic cell populations, bivalent domains have only been studied in select cellular contexts (Bernstein et al., 2006; Lien et al., 2011; Mikkelsen et al., 2007; Pan et al., 2007). Therefore, the coexistence of repressive H3K27me3 and activating H3K4me3 histone marks at the same locus has remained controversial, as it is not easy to determine whether this dual histone mark represents different cells within a heterogeneous tissue. In this study, we experimentally demonstrate the concurrent enrichment of H3K4me3 and H3K27me3 at the same locus and show that at each step of lineage progression, critical developmental regulators resolve their bivalent state concomitant with gene activation. Thus, our study supports the existence of bivalent domains and lends clear evidence to their biological relevance. For example, the group of genes featuring bivalency in hESCs, loss of H3K27me3, and induction of expression at the DE stage included most of the known cell fate determinants for DE formation. Likewise the major regulators of early pancreas development followed this pattern during PE formation. Therefore, removal of PcG-mediated repression appears to be the preferred induction mechanism for genes involved in developmental lineage progression. Consequently, genes with still unknown functions exhibiting stage-specific induction and loss of bivalency are likely to have developmental roles. Particularly interesting was the presence of several components of the Notch and Netrin signaling pathways in this short list of genes associated with pancreatic lineage commitment. While Notch signaling has a well-established role in terminal differentiation of pancreatic cells (Apelqvist et al., 1999), it is still unknown whether Notch or Netrin signaling is required for cell fate commitment to the pancreatic lineage. Notably, Netrins are expressed in embryonic pancreatic endothelial cells (Yebra et al., 2011) and endothelial cues are required for early pancreas development (Lammert et al., 2001), suggesting that Netrins could relay signals between the endothelium and foregut progenitors during pancreatic fate induction.
We found that critical transcriptional regulators of DE formation first exhibited removal of PcG repression at DE entry and then quickly reacquired PcG repression upon exiting the DE stage. Based on the observation that SMAD2/3 recruits the H3K27 demethylase JMJD3 to EOMES, SOX17, and GSC (Kim et al., 2011; Teo et al., 2011), it has been suggested that endodermal gene activation depends on active removal of H3K27me3 repressive marks. By directly demonstrating that JMJD3-mediated removal of H3K27me3 is necessary for the activation of DE-specific genes, our study establishes a causative role for H3K27me3 modification in gene regulation. Moreover, we show that during lineage progression, key regulators of DE formation quickly regain PcG-dependent H3K27me3 modifications resulting in rapid loss of gene expression upon exiting the DE stage. Many of the genes regaining the H3K27me3 repressive mark at DE exit (i.e. ROR2 and SOX17) need to be reactivated when DE differentiates into other DE-derived tissues and organs, such as the lung, intestine, or gallbladder (Spence et al., 2009; Yamada et al., 2010). Thus, PcG-mediated gene repression appears to provide a particularly rapid and flexible mode of gene regulation during differentiation and therefore specifically associates with genes important for transitory lineage intermediates. This mechanism might also be relevant for similar gene regulation during the development of other germ layers because it allows cells to maintain the molecular flexibility of quickly reinitiating expression at a later time point.
Our data further revealed that resolving bivalency to an active H3K4me3 state also associated with the induction of genes important for endocrine cell function during terminal differentiation. This contrasts findings in neuronal development, during which the majority of initially bivalently modified genes resolve bivalency in neuronal progenitors and not during terminal differentiation into neurons (Mikkelsen et al., 2007). Thus, our study reveals a previously unrecognized role for bivalent domains after early lineage commitment and demonstrates differences in how chromatin architecture is remodeled during development of different tissues.
A major limitation of significant health relevance is that culture conditions supporting efficient production of beta-cells capable of reversing diabetes remain to be identified. At present, effects of a large number of different culture conditions on beta-cell maturation are difficult to examine in vitro, because analysis tools are largely limited to functional studies of glucose-dependent insulin secretion. Our global gene expression analysis shows that engraftment of PE into mice induces a genes expression program with remarkable similarity to primary human islets. The list of genes that distinguishes in vivo- from in vitro- differentiated cells will be a valuable molecular read out for assessing the extent to which modified culture conditions can produce cells in vitro with closer similarity to functional beta-cells.
We observed that the insufficient induction of endocrine genes in vitro was associated with aberrant histone modification patterns during terminal differentiation into endocrine cells. This aberrant pattern comprised important beta-cell genes, such as insulin, the recently identified beta-cell maturation marker UCN3 (Blum et al., 2012), and the FFAR1 and GLP1 receptor, which modulate glucose-dependent insulin secretion (Itoh et al., 2003; Preitner et al., 2004). Thus, incorrect chromatin remodeling during terminal endocrine differentiation in vitro might be an important factor that contributes to the malfunction of these cells. Notably, TFs known to regulate functional properties of beta-cells, such as the MODY3 gene HNF1A (Horikawa et al., 1997) and the type 2 diabetes-associated genes TCF7 and HHEX (Sladek et al., 2007) are also insufficiently induced during in vitro differentiation. It is tempting to speculate that the absence of these TFs could be responsible for insufficient activation and inappropriate chromatin remodeling of beta-cell genes, such as insulin, UCN3, FFAR1, and GLP1R during in vitro differentiation. Supporting this idea, HNF1A has been shown to recruit histone-modifying complexes that lead to site-specific H3K4 methylation, while preventing H3K27 trimethylation (Luco et al., 2008). Future experiments will determine whether manipulations of these TFs or chromatin modifying enzymes could help produce fully differentiated beta-cells in vitro. Protocols to generate such cells will have important applications for cell replacement therapies and research into the pathogenesis of diabetes.
Chromatin immunoprecipitations were performed as previously described (Bhandare et al., 2010). Briefly, samples were crosslinked in 1.1% formaldehyde/PBS for 10 min at room temperature and then quenched with 0.125M glycine/PBS. Samples were subsequently washed twice with PBS and then lysed in 1% SDS. For sonication, lysates were sonicated with a Bioruptor Sonicator (Diagenode) three times for 5 min each with a 30 sec on and off cycle, resulting in 200–500bp chromatin fragments. Sheared chromatin was incubated overnight at 4 °C w ith 50 μl of Dynabeads sheep anti-rabbit IgG (Life Technologies), preincubated with 5 μg rabbit anti-H3K4me3 (Millipore, 04-745) or rabbit anti-H3K27me3 (Millipore, 07-499) antibodies. Immunoprecipitated complexes were further eluted, reverse crosslinked, and subjected to library preparation. For aggregated hESC-derived and FACS-sorted samples, a total of 2 to 5 × 106 cells were used for each ChIP-seq run. For transplanted samples, three grafts from each transplantation cohort were pooled and used for both ChIP-seq and RNA-seq analysis.
ChIP-seq libraries were prepared as per Illumina’s instructions (http://www.illumina.com). For input library preparation, 10 ng of input DNA from each sample was used. After adaptor ligation, DNA fragments were size-fractionated by gel electrophoresis and excised at 200±25 bp. Following gel purification, DNA fragments were amplified with 18 PCR cycles and purified using a MiniElute PCR Purification kit (Qiagen). 10 nM purified DNA was loaded on the flow cell, and sequencing was performed on an Illumina/Solexa Genome Analyzer II in accordance with the manufacturer’s protocols.
The Homer package was used to define regions enriched for H3K4me3 and H3K27me3 binding (Heinz et al., 2011), overlapping gene TSS regions defined by RefSeq. See Supplemental Experimental Procedures for additional details.
Strand-specific RNA-seq libraries were prepared as previously described (Parkhomchuk et al., 2009), with minor modifications. Briefly, cells were lysed in Trizol (Life Technologies) for extraction of total RNA. Residual contaminating genomic DNA was removed using the Turbo DNase kit (Ambion). mRNA was isolated from 2 μg of DNA-free total RNA using the Dynabeads mRNA Purification kit (Life Technologies). Following purification, mRNA was primed with Olig(dT)s and random hexamers and reverse-transcribed to first-strand cDNA. Residual dNTPs were removed using Illustra MicroSpin G-25 columns (GE Healthcare). In the second-strand synthesis reaction, dUTPs were used instead of dTTPs. The double-strand cDNA was fragmented using a Bioruptor Sonicator for 60 cycles of 30 sec on and off. After end-repair and adenine base addition, the cleaved double-strand cDNA fragments were ligated to Pair-end Adaptor Oligo Mix (Illumina) and size-fractionated on a 2% agarose gel (Sigma). cDNA fragments of 200±25 bp were recovered, and incubated with uracil-N-glycosylase (UNG) to digest the second-strand cDNA. Purified single-strand cDNA was then used as template for 15 cycles of amplification using pair-end PCR primers (Illumina). The amplified products were separated on a 2% agarose gel and a band between 225–275 bp excised.
For each sample, sequence reads were aligned to the transcriptome using RUM, and a “Feature Quantification” (FQ) value was computed for each Refseq mRNA transcript, where each FQ value = the number of reads overlapping each transcript per million reads sequenced, per kb of transcript length. In accordance with recommendations from ENCODE and the BCBC, these experiments were performed on two independent biological replicates. The FQ values for each pair of sample replicates showed high correlation, and were therefore averaged together before subsequent analysis. For each unique gene symbol, a single representative transcript was chosen, having the highest single FQ value for any one sample (see Supplemental Experimental Procedures).
ChIP-seq and RNA-seq data sets have been deposited into Array Express under accession numbers E-MTAB-1086.
We thank C. Wright (Vanderbilt University) and K. Miyabayashi (Kyushu University) for antibodies; T. Harper for lentivirus preparation; N. Shah and E. Lee for technical assistance; E. Baetge and members of the Sander laboratory for helpful discussions. We acknowledge use of human fetal pancreas provided by the Birth Defects Research Laboratory of the University of Washington. We are grateful to E. O’Connor for his help with FACS. This work was supported by National Institutes of Health grants U01-DK089567, R01-DK07243 to M.S., U01-DK089529, R01-DK088383 to K.H.K., and P30-DK19525 to K.J.W..
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.