Epigenome rewiring in human pluripotent stem cells

Abstract

The epigenome plays a crucial role in modulating the activity of regulatory elements, thereby orchestrating diverse transcriptional programs during embryonic development. Human pluripotent stem cell (hPSC) stepwise differentiation provides an excellent platform for capturing dynamic epigenomic events during lineage transition in human development. Here we discuss how recent technological advances, from epigenomic mapping to targeted perturbation, are providing a more comprehensive appreciation of remodeling of the chromatin landscape during human development with implications for aberrant rewiring in disease. We predict that the continuous innovation of hPSC differentiation methods, epigenome mapping and CRISPR perturbation technologies will allow researchers to build towards not only a comprehensive understanding of the epigenomic mechanisms governing development, but also a highly flexible way to model diseases with opportunities for translation.

Human pluripotent stem cells for investigating epigenome rewiring in human development

Human development is a dynamic process that depends on the faithful execution of lineage-specific transcriptional programs. Essential to the correct establishment of cell identity during lineage transition is the epigenome, a collection of information that is on top of (epi) that propagated by the DNA sequence, involving chemical modifications as well as physical interactions of the chromatin. The dynamic changes of the epigenome, or epigenome rewiring, underlie both the exquisite regulation of normal development and dysregulation in the context of diseases. The rapid development of sequencing technologies and (epi)genomic editing tools is making it possible to not only map epigenomic rewiring in developmental processes but also to examine their functional consequences, creating opportunities to dissect the molecular mechanisms underlying development and disease. However, it is challenging to obtain enough cells of the same developmental state from a rapidly developing mammalian embryo for high-resolution epigenomic mapping. Furthermore, access to human embryos is extremely limited and experimental manipulation is largely out of reach. Therefore, the impact of the technological advances can be greatly enhanced if they are applied to model systems that are both accurate as a representation of in vivo human development and amenable to genomic characterization as well as genetic and epigenomic perturbations.

Human pluripotent stem cells (hPSCs), including human embryonic stem cells (hESCs) and induced pluripotent stem cells (iPSCs), are an attractive model system given their ability to self-renew indefinitely and to differentiate into diverse cell types that resemble their in vivo counterparts [1,2]. HPSC differentiation platforms typically utilize defined chemical compounds or growth factors in each step of differentiation, making it feasible to generate sequential developmental intermediates for reproducible and temporally resolved molecular characterizations. In addition to modeling development, hPSC-derivatives are therapeutically valuable for their ability to model disease phenotypes and their potential as a source for transplantation, such as in the case of stem cell derived β-cells for the treatment of diabetes. This review focuses on epigenomic rewiring in hPSC models while highlighting relevant studies in mouse PSCs and other differentiation platforms. Due to space limitations, we refer readers interested in epigenomic rewiring during pluripotency reprogramming to recent reviews [3,4] that have provided excellent summaries on this topic. Here, we first describe how recent technological advances are revealing epigenomic features during hPSC differentiation, from one-dimensional (1D) chromatin state to three-dimensional (3D) spatial architecture. Together with genetic manipulation, they are providing a mechanistic understanding of epigenomic rewiring in human development. We further discuss how epigenomic information can be used to identify functional regulatory elements in cell fate transition, and to interrogate the consequences of epigenomic rewiring on gene expression. In addition, we highlight the relevance of epigenomic rewiring in diseases and the utilization of epigenomic information in inferring the targets of risk variants. Finally, we discuss challenges and future opportunities towards gaining a deeper understanding of how epigenetic rewiring shapes normal development, and how dysregulation leads to disease.

Mapping chromatin modification dynamics in hPSC differentiation

Cell state transition is orchestrated by dynamic interplay between epigenomic features such as chromatin accessibility, histone post-translational modifications, transcription factor (TF) binding, and DNA methylation (Table 1; Fig. 1). For instance, pancreatic differentiation has been extensively characterized and used in epigenomic studies that provided insight into the regulation of enhancer activity by epigenetic rewiring at each step of development [5–8]. The activation of lineage-specific gene expression is accompanied by chromatin remodeling, including increased chromatin accessibility and the gaining of mono-methylation at the 4th lysine residue of the histone H3 protein (H3K4me1) and acetylation at the 27th lysine residue of H3 (H3K27ac) marks at lineage-specific enhancers, as well as the loss of Polycomb-mediated tri-methylation at the 27th lysine residue H3 (H3K27me3) at developmental promoters [5,8]. Furthermore, transition of chromatin state is also associated with the binding of pioneer TFs, which could access closed chromatin and induce its opening. For example, the binding of archetypal pioneer TFs FOXA1 and FOXA2 coincides with chromatin opening during pancreatic differentiation [5,6]. An additional layer of epigenetic regulation is DNA methylation, a more stable mechanism for transcriptional silencing compared to repressive histone modifications. Large hypomethylated genomic regions (≥ 5 kb) termed DNA methylation valleys (DMVs) are enriched for developmental genes [9], and aberrant hypermethylation at DMVs is associated with failure to upregulate pancreatic genes during differentiation [10]. DMVs are reported to be protected from hypermethylation in mouse embryonic stem cells by Polycomb Repressive Complex 2 (PRC2), pointing to an interplay between DNA methylation and histone modifications [11]. Despite their importance, epigenomic features and TF binding alone do not provide the complete picture. For instance, while genomic data implicate both FOXA1 and FOXA2 in enhancer priming during hPSC pancreatic differentiation, genetic knockout studies show that FOXA2 plays a more critical role [6,7]. Similarly, genetic perturbations have been used to discover how regulators of DNA demethylation affect development in broad hPSC differentiation contexts with implications for pancreatic, cardiac and neural development [10,12–14]. In addition to providing mechanistic insight into chromatin remodeling during development, findings from such epigenomic studies could also be harnessed to improve guided differentiation. Profiling chromatin accessibility and histone modification during hPSC β cell differentiation unveiled that circadian clock regulators gained regulatory activity during in vitro islet maturation and circadian entrainment improved glucose responsiveness, demonstrating the potential for application of epigenomic findings to improve differentiation protocols [15].

Table 1.

Technologies for mapping and functionally characterizing the epigenome

	Relevant technologies in this review	Notable applications	Notable single-cell applications
Mapping
1D epigenome	DNase-seq ATAC-seq ChlP-seq MethylC-seq Whole Genome Bisulfite Sequencing 5hmc-Seal	Ref 5–10 and 14–15 cover the characterization of dynamic histone modification, transcription factor binding, and DNA methylation during in vitro pancreatic differentiation; ref 12–13 study the role of safeguarding DNA methylation during hPSC to neural differentiation and embryonic heart development. hPSC pancreatic differentiation (ref 5)	Single cell ATAC-seq have been performed in human hematopoietic progenitors (ref 97–98), human early embryos (ref 98), mouse skin (ref 99), and human pancreatic islet (ref 100). Single-cell ChIP-seq has been done on mouse PSCs, fibroblasts, and human hematopoietic progenitors (ref 103) Single cell methylome sequencing have been performed in human early embryos (ref 101) and mouse oocytes and PSCs (ref 102).
3D interactions	3C:ref 21 4C: ref 22–23 Hi-C:ref 24 ChlA-PET: ref 25 Capture Hi-C: ref 26 HiChlP:ref 27 PLAC-seq: ref 28 HiCAR:ref 29 Micro-C: ref 30 Micro-Capture-C: ref 31	Ref 33 applied Hi-C to investigate compartment and TAD dynamics in hPSC-derived lineages; Ref 34–35 utilized Promoter-Capture Hi-C (PCHi-C) to study enhancer-promoter (E-P) interactions in hPSC to neuroectoderm differentiation and primary human keratinocytes; Ref 36 uncovers the both pre-formed and dynamic chromatin looping during human macrophage development with in situ Hi-C; Ref 37 identified enhancer hubs during mouse muscle progenitor specification with Hi-C and PCHi-C; Ref 38 mapped TADs and E-P contacts during mouse neural development using in situ Hi-C	Single-cell Hi-C have been applied to mouse immune cells (ref 104), mouse fibroblasts and immortalized human cell lines (ref 105).
Functional characterization
Genetic editing	Cas9	Ref 45, 46, 55, 60 interrogated enhancer activity by using CRISPR-Cas9 based editing to mutate or delete enhancer sequences.
Epigenetic editing	Repressive domains fused with catalytically dead Cas9 (dCas9): KRAB (ref 64–65) LSD1(ref 66), and DNA methyltransferase DNMT3A (ref. 67–68) Active domains: p300 (ref 69), p65 (ref 70), HSF1 (ref 70), PRDM (ref 71), and TET1 (ref 72–73).	Ref 59 is a proof-of-concept demonstration of CRISPRi/a’s ability to repress pluripotency genes or activate silenced developmental genes in hPSCs. Ref 12 utilized dCas9 fused with the catalytic domain of TET1 to establish a causal link between promoter hypomethylation and transcriptional activation during hPSC neuroectodermal differentiation.	Ref 61–62 applied CRISPRi to perturb regulatory elements in immortalized cell line using scRNA-seq as readout.
Inducible looping	Heterodimerizing dCas9 constructs (refs 85–87) inducible by ligand (ref 86) or light (ref 87)	Ref 86 induced modest expression of pluripotency gene Oct4 in 293T cells through induced looping between its promoter and an enhancer. Ref 87 forced looping between a pluripotency enhancer and a lowly expressed gene in mouse PSCs and observed modest increase in gene expression.
Editing structural variants	CRISPR-Cas based editing with a pair of sgRNAs targeted at two distal genomic sites (ref 96)	Ref 96 engineered chromatin structural variants associated with human limb malformation in mice and demonstrated a link between genome topology and developmental disease.

Figure 1. Layers of epigenomic rewiring during differentiation.

hPSC differentiation is accompanied by reorganization of 3D chromatin topology and chromatin modifications. On the larger scale of 3D genome organization, compartment switching between A compartment associated with active transcription and B compartment associated with repressed transcription correlates with lineage-specific gene expression. On finer scales, the formation and dissolution of TADs and enhancer-promoter loops accompany the dynamic gain and loss of cell-type-specific regulatory relationships. Chromatin and DNA modifications at regulatory elements also impact their activity. For example, chromatin mark H3K27Ac and DNA hypomethylation are correlated with active regulatory elements, whereas H3K27me3 and DNA hypermethylation are associated with repressed regulatory elements.

Mapping chromatin architecture changes in hPSC differentiation

While profiling biochemical modifications can map putative enhancers and promoters in a given cell type, it cannot establish the link between these regulatory elements and their target genes. With the development of technology for mapping chromatin interactions in the 3D space, researchers have started to appreciate the significance of genome topology in gene regulation and cell state transition. Genome organization is characterized at several scales. At the multi-megabase (Mb) level, chromatin interactions are segregated into A/B compartments, with compartment A typically enriched for transcriptionally active regions and compartment B for inactive regions [16] (Fig. 1). Sequences within either type of compartment preferentially interact with other sequences within the same type of compartment [16]. At the Mb to sub-Mb level, topologically associated domains (TADs) are insulated regions with a tendency to interact within themselves [3,16] (Fig. 1). Chromatin loops anchored by CCCTC-binding factor (CTCF), an insulator-binding protein, form insulated neighborhoods that restrict enhancer-promoter interactions and confine the range of gene regulation [17]. Finally, enhancer-promoter contacts usually ranging from tens to hundreds of kilobases (kb) participate in transcriptional regulation by bringing enhancers into physical proximity to their target genes [18] (Fig. 1).

How chromatin organization regulates gene expression and cell identity has received increasing scrutiny and is addressed with an array of methods. Imaging techniques such as fluorescence in situ hybridization (FISH) have made important contributions to understanding interactions between specific loci as comprehensively reviewed elsewhere [19]. Here we focus on chromosome conformation capture (3C) based technologies that are central to mapping global chromatin interactions [20] (Table 1). 3C-based methods commonly crosslink chromatin with their interacting regions, followed by restriction enzyme digestion and re-ligation to linearly connect these spatially proximal regions in preparation for PCR- (3C [21]) or sequencing- (4C [22,23], Hi-C [24], ChIA-PET [25]) based detection of interacting sequences. 3C and 4C methods can provide targeted view of local interactions including enhancer-promoter contacts, whereas the genome-wide Hi-C and ChIA-PET methods are better suited for capturing higher-order organization features such as compartments and TADs. Some newer technologies, such as Capture Hi-C [26], HiChIP [27], and PLAC-seq [28], HiCAR [29], Micro-C [30], and Micro-Capture-C [31], seek to achieve high-resolution global mapping with cost-effective sequencing depth. Here we highlight some of these emerging methods for fine-mapping the 3D genome. HiCAR is developed to capture open chromatin-anchored interactions through Tn5 transposition prior to restriction enzyme digestion, thus enriching for contacts involving regulatory elements [29]. By combining Hi-C protocol with micrococcal nuclease (MNase) digestion, Micro-C profiles chromatin interactions at nucleosome resolution and uncovers loops previously missed by Hi-C in hPSCs [30]. Micro-Capture-C further leverages the higher resolution afforded by MNase digestion with target enrichment through oligonucleotide capture to resolve promoter-enhancer interactions at close range that have eluded conventional 3C methods [31]. The rapid development of 3C-based approaches presents new opportunities to map regulatory loops globally and to link architectural rewiring with gene expression changes in development.

Equipped with these technologies, researchers have investigated the cell type specificity of chromatin organization at different scales and its rewiring during human development. At the compartment level, Hi-C experiments using hPSCs and hPSC-derived mesendoderm, mesenchymal stem cells, neural progenitor cells, and trophoblasts-like cells have demonstrated compartmental plasticity of the 3D genome during differentiation, with gene expression showing a subtle yet lineage-specific correlation with compartment switching [32] (Fig. 1). On the scale of TADs, Hi-C mapping showed that while topological boundaries were largely unchanged, some interactions within TADs were rewired during hPSC differentiation and associated with differential gene expression [33] (Fig. 1). Furthermore, CTCF-CTCF loops were shown to be largely cell-type invariant between hPSCs in two different states of pluripotency, naïve and primed [17]. Mapping enhancer-promoter loops with promoter capture Hi-C in hPSCs and neuroectodermal cells identified correlated changes in chromatin interactions and chromatin modifications during neural differentiation [34] (Fig. 1). The loss or gain of interactions between promoters and active chromatin (characterized by the chromatin accessibility and/or the presence of histone marks such as H3K27ac, H3K4me1, or H3K4me3) was significantly associated with changes in gene expression, suggesting a role of chromatin connectivity in gene regulation [34]. Beyond these early lineage transitions modeled by guided hPSC differentiation, correlations between lineage-specific gene expression and interactions of promoters with active chromatin have also been observed in epidermal keratinocytes differentiation [35], macrophage development [36], mouse muscle progenitor development [37], and mouse neural differentiation [38]. In conclusion, while different chromatin features have different dynamics during development, rewired interactions are often implicated with functional consequences on gene expression. The next step is to move beyond the correlation, and to assign specific functions to the rewired interactions as discussed in the next section.

The need for functional characterization

While mapping epigenomic features has led to the identification of candidate regulatory elements in various contexts of human development, it does not directly assess the contribution of such elements to the regulation of gene expression [39–43]. Biochemically predicted regulatory elements are not always required for gene expression, whereas some regions lacking canonical epigenetic marks of active enhancers have also been shown to drive gene expression [44–46]. Furthermore, a single gene could contact multiple biochemically active regulatory elements, each of which may contribute to gene expression in varying degrees [36,47]. Conversely, a single candidate regulatory element can also form contacts with multiple genes [48,49], further confounding the assignment of regulatory relationship. Therefore, functional assays using transcriptional output as a readout are required to validate whether a candidate regulatory element is a bona fide enhancer. In addition to confirming candidate enhancers’ ability to regulate transcription, how they acquire such regulatory activity is also a point of investigation. While an enhancer’s activity can be considered as a function of the collective input from chromatin modifications and interactions, the functional consequences of specific chromatin features on gene expression levels have only been incompletely determined. We will discuss how the development of genetic and epigenetic perturbation technologies enable the field to address these questions in the following sections.

Discover functional enhancers

To identify functional enhancers, reporter assays have been used to test enhancer activity by measuring transcriptional output. This methodology could be applied to examine the activity of individual regulatory elements under desired cellular contexts in vitro or in vivo [50], or to screen a large pool of enhancer candidates by reading out enhancer-driven expression of barcoded transcripts through next-generation sequencing, as in the case of massively parallel reporter assays (MPRA) and self-transcribing active regulatory region sequencing (STARR-seq) [51,52]. These episome-based reporter assays have generated rich datasets of active regulatory sequences in a number of hPSC differentiation contexts [53,54]. On the other hand, the assessment of regulatory elements away from their native chromatin contexts could also complicate the interpretation of results and the assignment of endogenous transcriptional targets. To interrogate enhancers’ functions in their native chromatin context, CRISPR-based systems are increasingly used to alter enhancers’ DNA sequences or epigenomic features. Coupled with the nuclease activity of Cas9, CRISPR-based tools are able to either delete an entire enhancer region or generate small insertions or deletions (indels), thus enabling researchers to both assess the function of an enhancer and dissect its functional motifs at close to single-nucleotide resolution [46,55–58] (Table 1; Fig. 2). In addition, epigenetic modifiers are fused with catalytically dead Cas9 (dCas9) for locus-specific epigenetic CRISPR interference (CRISPRi) or CRISPR activation (CRISPRa) [57–59] (Table 1; Fig.2). These tools have been applied to individual loci but also in large-scale screening efforts, where guide RNAs (gRNAs) are designed to target regions encompassing genes of interest either unbiasedly or focused on putative enhancers based on chromatin features such as chromatin accessibility or H3K27ac occupancy [57,58]. Such CRISPR enhancer screens have been carried out in both human [46,55] and mouse PSCs [45,60], interrogating regions harboring key pluripotency genes such as POU5F1 and NANOG and uncovering both known and novel enhancers. We envision future enhancer screens will go beyond the pluripotent state and extend to diverse hPSC differentiation contexts.

Figure 2. Functional interrogation of epigenetic regulation.

CRISPR-based tools have been employed to delete enhancer sequences or mutate TF binding motifs, epigenetically silence or activate regulatory elements, and force enhancer-promoter looping to examine the consequences of rewiring the epigenome. Deletion of an enhancer or core TF motifs can reduce gene expression, and the loss of TF binding can also impact both local chromatin state and 3D interactions. Epigenetic editing can activate or repress transcription, which can be used to examine whether changes in chromatin marks can interfere with chromatin contact. Forced looping by heterodimerizing CRISPR constructs can induce ectopic gene expression and cause remodeling of chromatin modifications at regions gaining interactions with active chromatin.

Discovery of developmental enhancers can be further accelerated through the incorporation of single cell genomics. In particular, CRISPR screens coupled with scRNA-seq do not rely on the expression of one or a few genes as a readout and are therefore able to interrogate enhancer candidates of a large number of target genes, as shown in their successful applications in immortalized cell lines (Table 1; [61,62]). The expanded usage of hPSC differentiation system for large-scale enhancer discovery can lead to a comprehensive atlas of functional enhancers involved in lineage transitions throughout human development. Furthermore, data from CRISPR screens have been used to develop or validate computational models, as exemplified by the activity-by-contact and GraphReg models [48,63]. These studies suggest the possibility of using the rich datasets from functional screens with single-cell readouts, together with the epigenome mapping data, to further refine computational models to inform the chromatin features of functional enhancers and to predict enhancers in cell types not easily accessible for experimental manipulations with improved sensitivity and precision.

Interrogating the functional consequences of chromatin modifications

Various epigenetic marks are associated transcriptional activation or repression, yet the direct contribution of each mark to the activity of regulatory elements is not completely understood. Genetic knockout studies have unveiled the epigenomic and transcriptional impact of many chromatin-remodeling enzymes in organismal development as well as in hPSC self-renewal and differentiation. However, secondary effects due to changes in the global epigenetic landscape have made it challenging to establish a direct link between a specific chromatin modification on a regulatory element and changes in the expression of the target gene. While locus-specific genetic editing can be conducted to disrupt the sequence required for the targeting of DNA-binding factors, chromatin remodelers generally lack a specific consensus binding motif that can be altered in this way. Therefore, epigenome perturbation represents a uniquely powerful approach for targeted interrogation. Chromatin-modifying proteins or their catalytic domains are fused with DNA-binding proteins, notably dCas9, to introduce or remove specific chromatin modifications at a target locus (Table 1; Fig. 2). Domains that have been engineered as repressive epigenetic editors include the scaffold protein KRAB [64,65], histone H3K4 demethylase LSD1 [66], and DNA methyltransferase DNMT3A [67,68], and those adapted as activating epigenetic editors include histone acetyltransferase p300 [69], the NF-κB transactivation subunit p65 [70], the activation domain from human heat-shock factor 1 (HSF1) [70], histone methylase PRDM9 [71], and methylcytosine dioxygenase TET1 [72,73]. Such epigenetic editors have been applied at developmental loci to establish the causality between a specific modification and gene expression or developmental phenotypes. For example, dCas9 fused with the catalytic domain of TET1 was able to partially reverse DNA hypermethylation at the PAX6 promoter in TET1/2/3 triple knockout hPSCs, leading to a rescue of PAX6 induction in neuroectoderm differentiation [12]. The targeted demethylation strategy helped establish a direct connection between promoter hypomethylation in hPSCs prior to differentiation and gene activation during differentiation. With improvement in the versatility and efficiency of epigenetic editing tools, it would be possible to dissect layers of epigenetic modifications that enable cis-regulatory elements to exert gene regulatory activities in development.

Interrogating the functional consequences of chromatin architecture rewiring

Numerous studies have looked into factors underlying chromatin structures and how they influence transcription, including the binding of architectural proteins such as cohesin and CTCF [17,33,74], chromatin remodelers such as Polycomb complexes [75–77], and transcription factors [49,75,78–80]. Whether the architectural protein CTCF has a role in gene regulation has attracted intense research interest, especially in developmental contexts where insulation of active and repressed genes is crucial to the maintenance and specification of cell identity. In hPSCs, CTCF-CTCF loop perturbation through deletion of CTCF binding sites allowed super enhancers originally contained within the loop to ectopically activate genes outside the loop, and in other cases caused derepression of genes originally inside the loop [17,74]. In the context of neuron differentiation, approximately 40% of CTCF binding peaks are pruned during mouse PSCs’ commitment to neural precursors, many overlapping with lost pluripotency gene-enhancer interactions and gained interactions anchored by other proteins [81]. Interestingly the further differentiation of human neural precursors into neurons was accompanied by an increase in CTCF expression [82], suggesting dynamic regulation of CTCF expression and possibly genomic occupancy during differentiation. Additional works in mouse PSC differentiation showed that CTCF binding sites are required for proper demarcation of repressive and active chromatin and organized enhancer-enhancer interactions that ensure cell-type-specific gene expression [83,84]. Together these results support a role of CTCF loops for safeguarding cell identity.

Several questions remain regarding the interplay between chromatin interactions and transcription regulation. First, at the level of enhancer-promoter loops, it remains to be determined whether they are sufficient for transcriptional activation. Heterodimerizing dCas9-based constructs have been engineered to generate artificial loops connecting a silenced gene with an active enhancer element to induce gene expression (Table 1; Fig. 2). However, the level of induction was relatively modest and the number of genes examined was limited, so the exact roles of looping in transcriptional activation requires further investigation [85–87]. Second, regarding longer-range chromatin interactions, while those anchored by architectural proteins (e.g., CTCF) are well recognized, it is unclear how other mechanisms, such as TF binding and DNA methylation, contribute to loop organization. CRISPRs have been applied to perturb the binding of lineage-specific TFs at chromatin loop anchors in mouse PSCs [49] and during human skeletal muscle transdifferentiation [79]. These works showed reduction in promoter-enhancer contact and modest decrease in expression of genes within interrogated loops [49,79]. However, the direct effects of perturbing TF binding on chromatin modifications and CTCF binding were not determined, thus it is unclear whether the observed gene expression changes were due to alterations in local chromatin state or reduced enhancer-promoter contacts. Another possible mechanism of chromatin looping is DNA methylation, as extended genomic regions (≥ 7.3 kb) with low levels of methylation, termed DNA methylation grand canyons, were recently reported to form loops spanning dozens of Mb and sometimes between chromosomes in hematopoietic stem and progenitor cells [88]. Notably, these regions are also enriched for Polycomb-deposited H3K27me3, and both Polycomb complexes and H3K27me3 have been implicated in long-range interactions [76,77,89,90]. Therefore, the relative contribution and interdependency of DNA methylation, Polycomb complexes, and H3K27me3 in regulating chromatin architecture remains to be determined. Addressing these questions will help the field bridge local chromatin modifications and 3D structures for a more comprehensive understanding of how these mechanisms jointly regulate gene expression and cell identity.

Epigenome rewiring in the context of human disease

The flip side of the well-orchestrated epigenomic rewiring in development is its dysregulation in disease. Epigenomic mapping in healthy and diseased tissue could guide the assignment of noncoding risk variants to target genes, and discover potential ectopic contacts between regulatory elements and disease-associated genes (Fig. 3). An example of the first application is the identification of enhancers that underlie the craniofacial disorder Pierre Robin Sequence (PRS) [91]. Mapping chromatin marks associated with active enhancers and extreme long-range interactions with the SOX9 promoter, a gene known to regulate craniofacial development, led to the discovery of enhancer clusters in hPSC-derived cranial neural crest cells (hCNCCs) located in a large noncoding region associated with PRS. Utilizing a mouse model harboring enhancer deletion complemented by in vitro hCNCC differentiation, the authors demonstrated that these enhancers regulated craniofacial development through modulating SOX9 dosage. Beyond focused investigation on specific disease-associated loci, a growing number of studies now utilize Hi-C-type interaction data obtained from various disease-relevant cells including primary human islets [92], hPSC-derived cardiomyocytes [93] and neurons [94,95] to globally identify target genes of risk variants from genome-wide association studies (GWAS). In addition to connecting disease-associated sequence variants to target genes, epigenomic mapping could also uncover structural variants, where abnormal contact between regulatory elements and target genes results in the latter’s misexpression. This type of aberrant rewiring is demonstrated in the TAD-spanning WNT6/IHH/EPHA4/PAX3 locus, where the TAD boundaries were disrupted due to deletion, inversion, or duplication in patients with limb malformations [96]. These structural mutations allowed genes flanking the central TAD to gain ectopic interaction with a cluster of limb enhancers normally insulated by TAD boundaries, resulting in pathogenic misexpression. These discoveries were made possible by accessible patient tissues combined with mouse models that were engineered to recapitulate structural variants found in patients and reproduced corresponding phenotypes (Table 1). When patient samples are not easily obtained, hPSC-differentiated cells can offer a powerful resource for molecular characterization as well as epigenome engineering to directly study the regulatory elements or patient genomic variants, as shown in the study of SOX9 enhancers in hCNCCs. We expect to see increasing application of epigenome mapping and perturbation methods in hPSC disease models in the future, which will greatly accelerate the discovery of disease-relevant genomic variants and contribute to the mechanistic understanding of pathogenesis.

Figure 3. Application of epigenomic profiling in disease.

a. Mapping chromatin interactions helps to assign target genes to risk-associated noncoding regions. Dashed curve indicates risk-associated sequence might impact chromatin contact in diseased condition.

b. Structural variant that disrupts the insulation between TADs can result in ectopic enhancer-promoter contact in pathological conditions.

Concluding Remarks

Rapid advances in sequencing technologies are revealing epigenomic features with increasingly higher resolution. Studies of hPSC differentiation are likely to further benefit from the integration of emerging single-cell (sc) epigenomic profiling approaches, including single-cell ATAC-seq [97–100], methylome sequencing [101,102], ChIP-seq [103], and Hi-C [104,105] (Table 1). As differentiating cells may progress asynchronously along multiple developmental trajectories, single-cell methods are well suited to deconstruct the heterogeneity during development and unravel how quantifiable epigenomic variance is correlated with variance in transcriptome and cellular function (Table 1). In terminally differentiated cells and embryos, the application of scATAC-seq coupled with scRNA-seq uncovered co-variance between the expression levels of lineage-specific TFs and accessibility of their bound regions, thereby linking transcriptional heterogeneity to epigenomic heterogeneity [98,99]. Further integrating functional heterogeneity, comparison of β cell scATAC-seq data and exocytosis activity measured by an earlier Patch-seq study found correlation between promoter accessibility of cell state-specific genes and cell function [100,106]. These studies support a vision where cell function, epigenome, and transcriptome can be simultaneously measured at the single-cell level to deepen our understanding of the molecular mechanisms governing cell identity in human development.

While the majority of studies covered in this review focus on the interplay between epigenomic rewiring and transcriptional regulation, how the epigenome itself is regulated remains to be fully understood. The challenges in studying the molecular mechanisms governing epigenomic rewiring include confounding secondary effects from genetic knockout of important epigenetic regulators and the difficulty in finding an accurate and high-throughput readout for discovering previously unknown regulators. The first limitation can be circumvented by applying targeted protein degradation tools, as shown in recent studies investigating the acute effect of depleting CTCF and cohesin release factor WAPL on genome organization by degron-mediated degradation [107,108], an approach that could be extended to the study of other epigenetic regulators. One solution to the second limitation can be engineering locus-specific reporters for epigenetic changes, as demonstrated by the use of a knockin fluorescent DNA methylation reporter at the PAX6 promoter in a genome-wide CRISPR-Cas9 screen [10]. The study identified a previously uncharacterized protein QSER1 that protects bivalent promoters, many of which are located in DMVs, from hypermethylation-mediated silencing [10]. These findings suggest that non-enzymatic proteins could provide an additional layer of control, beyond the actions of enzymatic regulators, to ensure proper epigenomic regulation in a genomic and cellular context dependent manner. In the future, it would be exciting to see the development of reporters for epigenomic features beyond biochemical modifications, such as enhancer-promoter loops or TADs. Taken together, we foresee future research on the regulation of the epigenome will benefit from the use of an expanded repertoire of experimental approaches, including acute genetic and epigenomic perturbation, large-scale CRISPR interrogation, and the use of live-cell reporters to track alterations in local and global epigenome features. With consortium efforts such as the 4D Nucleome Program from the National Institute of Health (https://www.4dnucleome.org/) devoted to further understanding the roles of the 3D genome organization and the effect of perturbing regulatory elements in human health and disease, we expect that the development of epigenomic mapping and editing technologies will accelerate the discovery of pathogenic mechanisms using hPSC-based disease models and open opportunities for developing treatment.

Outstanding Questions Box.

How are different levels of the epigenome rewired during human development? What is the relationship between the dynamics of chromatin modification and chromatin architecture during cell fate transition, and how do they functionally contribute to transcriptional regulation? How do enzymatic and non-enzymatic chromatin factors work in concert to regulate the epigenome?

How should biochemically annotated regulatory elements be functionally validated in a high throughput manner? How can validation data be used to optimize computational prediction of functional regulatory elements in diverse cell types?

What epigenomic aberrations, in particular of the 3D genome, can be found in diseases? Can they be reversed through genetic or epigenetic editing? How do chromatin interactions bridge risk variants with their target genes? How do we use epigenomic data to better understand human diseases?

Highlights:

1. Rewiring of the epigenome contributes to human development.

2. Stepwise hPSC-differentiation both facilitates and benefits from the understanding of epigenome rewiring.

3. Epigenomic rewiring during hPSC differentiation occurs at both local and structural levels.

4. The functional consequences of epigenomic rewiring on gene regulation can be interrogated with CRISPR technologies.

5. Epigenomic information can predict disease causal risk variants and reveal aberrant epigenomic rewiring in disease.