Abstract
Lineage commitment and differentiation is driven by the concerted action of master transcriptional regulators at their target chromatin sites. Multiple efforts have characterized the key transcription factors (TFs) that determine the various hematopoietic lineages. However, the temporal interactions between individual TFs and their chromatin targets during differentiation and how these interactions dictate lineage commitment remains poorly understood. We performed dense, daily, temporal profiling of chromatin accessibility (DNase I-seq) and gene expression changes (total RNA-seq) along ex vivo human erythropoiesis to comprehensively define developmentally regulated DNase I hypersensitive sites (DHSs) and transcripts. We link both distal DHSs to their target gene promoters and individual TFs to their target DHSs, revealing that the regulatory landscape is organized in distinct sequential regulatory modules that regulate lineage restriction and maturation. Finally, direct comparison of transcriptional dynamics (bulk and single-cell) and lineage potential between erythropoiesis and megakaryopoiesis illuminates the fine-scale temporal dynamics of these regulatory modules during lineage-resolution between these two fates. Collectively, these data provide novel insights into the global regulatory landscape during hematopoiesis.
Introduction
The temporal activation of stage-specific regulatory DNA instructs lineage specific gene expression programs that underpin cellular fate and potential. The establishment and maintenance of regulatory DNA is mediated by the combinatorial engagement of sequencespecific transcription factors (TFs) that bind in the place of a canonical nucleosome. Over the course of cellular differentiation programmed shifts in the global transcription factor milieu drive extensive re-organization of chromatin1,2, where silencing of regulatory DNA associated with alternate lineage and the de novo activation of lineage-restricted elements result in the narrowing of the epigenetic and functional landscape3. However, it is unclear how and when regulatory DNA is dynamically activated and silenced during cell state transitions to establish lineage restricted gene expression programs and how these epigenetic changes relate to developmental potential.
Hematopoiesis is a prototypical system to study how genetically and epigenetically encoded programs are established during cellular differentiation4–6. Conventionally, hematopoiesis is depicted as a discrete hierarchical process where a multipotent hematopoietic stem and progenitor cell (HSPC) traverses a sequence of bifurcating decisions, mediated by the expression of lineage-specific TFs, with each decision resulting in an increasingly restricted fate potential. Historically, the characterization of the gene regulatory programs involved in the transition from HSPCs to terminal fates has relied on the identification of differential transcriptional programs from isolated discrete populations using defined cell surface markers7–11. While this approach has led to the identification of master regulatory transcription factors10,12 that define many of the major hematopoietic cell lineages and has enabled a systematic mapping of their steady-state regulatory landscapes9,11,13, interrogation of discretely defined populations cannot elucidate the dynamic regulatory events that mark cell-state transitions.
Recently, single-cell chromatin and transcriptional profiling assays have attempted to resolve the spatio-temporal cis- and trans- dynamics in different stages of hematopoiesis14–17. Typically, these studies have relied on the analysis of either bulk or immunophenotypically isolated populations of steady-state peripheral blood or bone marrow derived cells, whereby hierarchical relationships and developmental trajectories between cell states are predicted computationally. While these experimental approaches have aided in defining major subpopulations of hematopoietic cells and their respective epigenetic and transcriptional landscapes, definition of developmental trajectories within individual lineages from population snapshots is challenging due to the limited sensitivity and the resulting technical and analytical artifacts associated with single-cell genomic assays18,19. Furthermore, because developmental trajectories are predicted in silico, direct association of functional changes (i.e., lineage potential) to intermediate cellular states is not possible20.
In order to investigate the dynamics of regulatory and functional events during differentiation, we use human erythropoiesis as a proxy for hematopoietic development. The transition from HSPCs to terminally differentiated enucleated red blood cells involves a series of morphologically, functionally, and phenotypically distinguishable states. While multiple efforts relying on the isolation of these states have characterized key transcriptional regulators21,22 and chromatin elements implicated in erythropoiesis9,11, our knowledge on the temporal interplay between individual cis- and trans- elements and how these establish stage-specific transcriptional programs and lineage commitment during erythropoiesis, and hematopoiesis in general, remains only rudimentary. Furthermore, because erythrocytes share their developmental origins of with other myeloid lineages (granulocytic/monocytic and megakaryocytic), erythropoiesis represents an ideal system to study how lineage choice is genetically and epigenetically encoded.
Here, we capitalize on the ex vivo human differentiation scheme where dense unbiased sampling of the populations allows us to capture the dynamics of chromatin accessibility and gene expression during differentiation with a completely defined developmental trajectory. DNase I-seq and gene expression profiling (bulk and single cell) time-course during erythropoiesis coupled with lineage potential assays and morphological characterization, enabled the assignment of distal elements (alone or in combination) to target genes and individual TFs to their target DHSs which collectively comprise discrete regulatory modules associated with lineage potential. Comparing the activity patterns of the TF regulatory modules in the erythroid lineage to the closely related megakaryocytic lineage, provides insights into how these modules instruct lineage commitment. Collectively, our findings provide key insights into the organization of the functional epigenetic landscape during differentiation and its relation to lineage-potential.
Dense mapping of the temporal dynamics of cis- and trans-elements during erythropoiesis
Human erythropoiesis was induced ex vivo for 12 days using an established differentiation protocol23 that faithfully recapitulates the major features of in vivo erythropoiesis. Starting from human adult CD34+-enriched HSPCs in mobilized peripheral blood derived from 3 healthy donors we cultured the cells in defined media for 12 days (Figure 1a and Methods). Characteristic features of developing erythroblast cells were confirmed by immunophenotyping using canonical cell-surface markers of early (CD117, C-Kit) and late (CD235a, Glycophorin A) erythropoiesis as well as morphologically by hematoxylin-eosin staining of cell smears (Supplementary Figure 1).
To densely map both chromatin accessibility and transcriptional dynamics during the transition from HSPCs to committed erythroblasts, we subsampled a single continuous culture each day (12 days) and performed DNase I analysis and total RNA-seq (Figure 1a,b). Biological replicates from CD34+ HSPCs from 3 donors were highly reproducible for both chromatin accessibility and gene expression profiles where the majority of the observed variability was accounted for by developmental trajectory (i.e., sampling days) (Figure 1b,c and Supplementary Figure 2a), as biological replicates were highly correlated (Supplementary Figure 2b,c). For many individual DHSs and genes, we observed quantitative changes in chromatin accessibility and expression over the course of differentiation highlighted by quantitative trajectories of opening or closing (Figure 1d,e). Notably, accessibility changes were mostly confined to compact regions of the genome (~200bp average DHS width). In many cases, we observed both opening and closing events within close proximity (Figure 1d), indicating focal regulation24 of chromatin structure in contrast to previous reports that chromatin changes during differentiation occur over large domains1,25.
To systematically identify developmentally responsive cis-elements, we leveraged the observed continuity of DHS signal over adjacent days (Figure 1d) and modelled DNase I cleavage density against differentiation time-points (Methods). We determined significance by comparing our full model to a reduced model (intercept-only; not accounting for developmental time) and performing a likelihood ratio test (Methods). Of the total 79,085 DHSs accessible in 2 or more samples/replicates, we conservatively identified 11,805 (14.9%) significantly changing DHSs (adjusted p < 10-5 and fold-change >2), nearly evenly grouped between activated and silenced (45% and 55%, respectively) (Supplementary Table 1). A similar analytical approach applied to the RNA expression data identified 5,769 developmentally regulated genes (adjusted p < 10-5 and fold-change > 2), of which 62% up-regulated and 38% down-regulated over the course of differentiation (Supplementary Table 2). Collectively, these data define a high-resolution and quantitative map of chromatin and gene expression dynamics during erythroid differentiation.
Stage-specific compartmentalization of the cis- and trans-landscape
PCA indicated that days 5-6 were associated with a critical developmental inflection point during ex vivo differentiation (Figure 1b,c). We next sought to characterize the relationship between temporal chromatin and gene expression dynamics with regards to the observed immunophenotypic and morphological changes present in the population of differentiating cells. We performed unsupervised clustering (K-means; k=5) on dynamically changing DHSs and developmentally responsive transcripts (Figure 2a, b). This analysis revealed a stark partitioning of activated and silenced genes and DHSs into non-overlapping sets that closely paralleled canonical developmental features of erythropoiesis. Particularly, DHSs rapidly silenced within the first days of differentiation (clusters E1 and E2) were found to preferentially harbor binding sequences utilized by the known HSPC regulators such as (HOXA926, RUNX27 and ERG28) (Supplementary Figure 3). Similarly, immediately downregulated transcripts upon induction of differentiation (cluster G1) include these transcription factors as well as structural genes characteristic of CD34+ HSPCs (Figure 2b). Consistent with PCA (Figure 1b,c), a rapid and marked turnover of chromatin and gene expression landscape is observed between days 5-7 where an early erythroid signature appears in both activated DHSs and gene expression, marked by the upregulation of GATA1, KLF1, PPARA and TFRC (cluster G4). Markers of mature erythropoiesis emerge later in the differentiation (after day 8; cluster G5) with the upregulation of hemoglobins, glycophorin A (GYPA) and ALAS2 (Figure 2b).
In addition to canonical activation and downregulation patterns observed, we found a subset of genes exhibiting reproducible transient upregulation (clusters G2 and G3) occurring prior to establishment of the erythroid signature (Figure 2b). Transiently upregulated genes are expressed in myeloid lineages and correspond to several myeloid markers (e.g. MPO, KIT), including the myeloid-specific transcription factor CEBPA. Compatible with gene expression, late closing DHSs in cluster E2 and E3 were enriched in CEBPA recognition sequences (Supplementary Figure 3), denoting a transient emergence of myeloid-related regulatory elements. Consistent with this, the majority (~80%) of DHSs in cluster E2 and E3 were found overlapping with DHSs active in other myeloid cell types (macrophages and monocytes) (Supplementary Figure 4).
Taken together these data indicate a sequence of developmentally related changes in both the cis- and trans-environment where an HSPC related chromatin and gene expression signature is succeeded by a transient activation of a broader myeloid program, prior to acquiring an erythroid-specific signature. Expectedly, activated DHSs (clusters E4-E5) were found to preferentially harbor red blood cell-related GWAS variants (1.36-fold enrichment over all detected DHSs), highlighting their functional role in regulating erythropoiesis (Supplementary Figure 5).
Connecting individual DHSs to genes
The overall dynamics of chromatin accessibility for individual DHSs closely mirrored that of the expression of nearby genes. To formulate this, we performed an enrichment test to investigate the DHS landscape around a gene promoter. Interestingly, we found that developmentally regulated genes are enriched for DHSs with a similar developmental profile (Figure 2c). For example, early closing genes (cluster G1) are significantly enriched for cluster E1 DHSs. Noteworthy, transient genes of cluster G3 are harboring DHSs belonging to both late closing DHS cluster E3 and early activated erythroid DHS cluster E4, suggesting that the transient nature of these genes is a result of the combinatorial activity of a closing and an opening chromatin landscape.
Because of fine-resolution afforded by our dense sampling approach, we sought to quantify the extent of genome-wide coactivation patterns that could potentially comprise physical regulatory links between DHSs and their target genes by correlating the temporal expression patterns of a gene to nearby (±500 kb from TSS) developmentally regulated DHSs. This analysis identified 25,624 connections at an absolute Pearson correlation coefficient (r) > 0.7 with the vast majority of gene-DHS links occurring within 50 kilobases of the transcription start site (Figure 2d). Overall, we connected 74.5% of the developmentally regulated genes with ≥1 DHS and 46% of changing DHSs were linked to ≥1 developmentally regulated gene. While on average 51 DHSs reside within 1 Mb of a given gene, only 5 DHSs (±4.4 SD) were found to be linked with a changing gene.
We therefore sought to functionally validate these associations by genetic perturbation of these gene-DHS links. Specifically, we disrupted two correlated DHSs (HS1 and HS2) with CDH1 expression, situated upstream (5kb and 12kb, respectively) of the promoter of CDH1 gene which is upregulated during differentiation (Figure 2e). Homozygous deletion of these DHSs as well as the promoter in the human derived erythroid progenitor cell line HUDEP-2 where these DHSs are also active, resulted in complete ablation of the CDH1 expression as determined by flow-cytometry (Figure 2f). These results suggest that both elements as predicted by the correlation analysis as regulators of CDH1, indeed drive the expression of the gene and their deletion confers effects similar to the deletion of the gene promoter.
Overall, these findings suggest that the majority of changes in transcription during development are regulated by a limited number of cis-regulatory inputs, situated within close proximity to the genes they regulate.
Distinct and sequential regulatory modules encode developmental stages
Clustering of dynamically changing DHSs revealed that chromatin active at different stages of hematopoiesis display differential enrichment for recognition sequences for transcription factors. indicating stage-specific transcriptional regulation of cis-elements. This however, does not resolve the temporal interactions between individual DHSs and individual trans- regulators and how this relationship shapes the developmental response of a DHS. Given the observed global correlated changes between the transcription factor expression levels and the accessibility of the DHSs containing their cognate recognition sequences (Figure 3a) we sought to quantify the contribution of individual TFs to the dynamic changes in DNase I density at individual regulatory cis-elements. We capitalized on our dense sampling approach and applied a regression strategy where the activity of an individual regulatory element (i.e. DNase I cleavage density) is modelled as a function of the gene expression profiles of developmentally regulated TFs with a compatible recognition sequence harbored within each DHS (Figure 3b and Methods). We controlled for weak and ambiguous association of TFs recognizing degenerate motifs using elastic-net regularization (Methods). We applied this approach to all of the 11,805 dynamically changing DHSs, identifying 11,734 (>99% of changing DHSs) with at least one explanatory TF regulator (Methods) where the regression coefficients broadly correspond the strength of association of a TF with an individual DHS (Supplementary Figure 7). Overall, 7 TFs on average (out of the total 214 tested), were positively associated with each DHS, suggesting that a small subset of TFs regulate the developmental activity of individual ciselements.
We next asked to what extent the activity of DHSs with similar temporal accessibility patterns are regulated by a coherent set of TF regulators. We identified 52 TFs positively associated with at least 200 DHSs and performed unsupervised hierarchical clustering based on their regression coefficients computed for each DHS (Figure 3c and Methods). This analysis resolved the temporal associations between transcription factors and their target DHS into a sequence of five discrete and largely non-overlapping regulatory modules, reflective of developmental stages of erythropoiesis (Figure 3d). Module 1 consists of known HSPC transcriptional regulators (e.g. ERG28, MEIS129, MYCN30) which are positively associated with early closing DHSs in clusters E1 and E2. In modules 2 and 3, transcription factors associated with commitment of hematopoietic progenitors to the different myeloid lineages (e.g. CEBPA31, MYB32, FLI133, RUNX134) interact with DHSs in clusters E2 and E3. Modules 4 and 5 define the erythroid-specific regulatory landscape as known erythroid regulators (e.g. GATA1, KLF121, RXRA35 and FOXO336) positively interact with activated DHSs in clusters E4 and E5.
Plotting the fraction of DHSs in each cluster positively associated with each TF (Figure 3d) highlights the major drivers of chromatin accessibility in each developmental stage. Particularly, ERG appears as a major regulator of the HPSC stage as it is positively associated with ~25% of DHSs in clusters E1 and E2. Although ERG has been long implicated in HSPC regulation, it is only recently its role as a critical regulator of HSPC survival has been appreciated37. Interestingly, KLF12 also appears to share a significant proportion of the early chromatin landscape, although its role in HSPC regulation is not known. Overexpression of the critical HSC regulator Evi-1 in mice, resulted in maintenance of the quiescent phenotype of murine HSCs along with the more than 12-fold increase in Klf12 expression38. In another experiment, sustained expression of Hlf in mice also resulted in enrichment of Klf12 in more primitive hematopoietic compartments39, implicating KLF12 in the HSPC regulation. Apart from the canonical erythroid transcription factors, we identified MXI1 among the top regulators of the erythroid chromatin landscape. Knockdown of Mxi1 in mice, blocks chromatin condensation and impairs enucleation of mouse erythroblasts, highlighting the role of MXI1 in erythroid maturation40. Additionally, we find CTCF to be positively associated with a large portion of the erythroid-specific chromatin (~25% of DHSs in clusters E4 and E5), an observation in line with the role of CTCF in the establishment of erythroid-specific functional chromatin loops41,42.
Taken together, these findings illustrate the dynamic interaction of the cis- and the trans- regulatory landscape during erythropoiesis and their organization into well-defined and discrete regulatory modules of associated DHSs with their cognate transcription factors, reflecting distinct stages of erythroid development.
A sequence of abrupt lineage restriction events marks erythropoiesis
The organization of chromatin and transcription factors into defined regulatory modules corresponding to distinct stages of erythropoiesis indicate a functional relationship between lineage potential and module activity. To gain insight into whether these modules underpin lineage decision events we determined the lineage potential of the erythroid cultures by daily sampling a population of cells and assaying their multipotent and unipotent capacity for different myeloid lineages (Figure 4a and Supplementary Figure 8a). Total number of colonies declined with the progress of differentiation resulting in an abrupt depletion of total progenitors on day 6 of differentiation (Supplementary Figure 8b). After 4 days of exposure to erythroid media, the most primitive and multipotent colonies (CFU-GEMM; granulocytic, erythroid, monocytic, megakaryocytic) were no longer detected (Figure 4b). Day 6 marked a second event of restriction of the fate potential as all colonies derived from progenitors with unilineage capacity were no longer detected in the cultures such that frequency of erythroid progenitors (BFU-E) rapidly declined from day 5 to day 6 (Figure 4c). Similarly, granulocytic/monocytic progenitors (CFU-GM) were depleted by day 6 of erythroid differentiation (Supplementary Figure 8c). Notably, none of the changes in clonogenic capacity were associated with any changes in the growth rate of the parental erythroid cultures (which remained constant throughout the differentiation) (Supplementary Figure 8d).
In addition to the above lineages, we specifically tested for the ability to differentiate into megakaryocytes during erythroid development by transferring cells from the primary erythroid culture, on a daily basis, to megakaryopoiesis-inducing suspension cultures and tested for their ability to give rise to CD41+ megakaryocytic populations (Supplementary Figure 9a and Methods). Consistent with the overall lineage restriction observed during colony-forming assays, erythroid cultures completely lose megakaryocytic potential on day 6 of the differentiation (Figure 4c and Supplementary Figure 9b).
The rapid changes observed in clonogenic capacity correspond to the transitions in the activity of regulatory modules (Figure 4d). Early depletion of primitive multipotent CFU-GEMM progenitors is concomitant with the transition from the HSPC-related modules (modules 1 and 2), while the decline of unipotent progenitors of all detectable myeloid lineages (granulocytic/monocytic, erythroid, megakaryocytic) coincides with the transition from a program with a broader myeloid signature to erythroid specific cis- and trans- landscape. Furthermore, because these rapid lineage restriction events are not associated with other abrupt changes in morphology or cell growth (Supplementary Figure 8d), these data suggest that the mechanism responsible for the exit from the progenitor stage is decoupled from maturation progress.
Exit from the HSPC-related transcriptional program is shared between erythropoiesis and megakaryopoiesis
The silencing of the HPSC regulatory modules prior to lineage commitment suggested that exit from the progenitor state is necessary for erythroid commitment to proceed. We therefore asked whether this represents a canonical feature of hematopoietic development to any lineage. To investigate this, we focused on megakaryocytic differentiation, a process that shares both close common developmental origins43 and key TF regulators with erythropoiesis44.
We induced ex vivo megakaryocytic differentiation and performed dense sampling of gene expression during development (Figure 5a and Methods). Developmentally regulated genes during megakaryopoiesis exhibit largely bipartite profiles similar to those observed during erythropoiesis (Supplementary Figure 10). To determine whether the transcriptional changes associated with exit from HSPC state during erythropoiesis are shared with megakaryopoiesis we examined the expression profiles of erythroid developmentally regulated genes during megakaryocytic differentiation. We observed highly correlated global expression profiles for early silenced transcripts (erythroid clusters G1 and G2) between the two lineages (median Spearman’s ρ=0.76 and 0.62, respectively) (Figure 5b), with the exception of key regulators and canonical markers of megakaryopoiesis (MEIS1, FLI1, PBX1, ITGA2B, etc.) (Figure 5c). In contrast, correlation for erythroid clusters G3-G5 was low (median Spearman’s ρ≤0.13).
Similar to erythropoiesis, the early suppression of HSPC-related gene signature resulted in abrupt restriction of lineage potential during megakaryopoiesis. We found that cells in megakaryocytic cultures abruptly lose erythroid potential on day 4 of differentiation, coinciding with the activity of the HSPC transcriptional program (Figure 5c right and Supplementary Figure 11). However, in contrast to erythropoiesis, megakaryocytic differentiation does not exhibit a transient activation of a myeloid signature. This finding is compatible with the recently revised hematopoietic tree according to which megakaryocytes directly emerge from the primitive HSPC compartments bypassing the common myeloid progenitor45,46.
Conclusively, these results indicate the existence of a shared mechanism between erythropoiesis and megakaryopoiesis driven by a common set of TFs which mediates the exit from HSPC state and signals the entry into lineage-restricted states.
Single-cell transcriptomics uncover discrete cell states corresponding to transcriptional programs
So far, observations based on population-level analysis suggest the existence of transcriptional programs responsible for changes in lineage potential. However, as lineage decision events occur in individual progenitor cells47, we analyzed transcriptional changes from ~50,000 single cells sampled from frequent intervals along both the ex vivo erythroid and megakaryocytic differentiation (Figure 6a). Overall, gene expression dynamics from aggregated single-cells were highly concordant to RNA-seq performed on bulk populations (Supplementary Figure 12).
Given the heterogeneity between cells we sought to explore the transcriptional dynamics of individual cells during the transition from HSPC state to lineage commitment. To this end we collectively analyzed all cells sampled from both the megakaryocytic and erythroid lineages and performed principal component analysis (PCA) using 166 lineage regulating TFs detected in the dataset and 11 marker genes of mature erythropoiesis and megakaryopoiesis (Figure 6a and Supplementary Table 5). PCA readily resolved the two primary axes of differentiation where PC2 reflected HSPC to terminally committed lineages, and PC1 distinguished the erythroid and megakaryocytic lineages (Figure 6a). The top positive and negative loadings from each component correctly identified known key TFs that regulate distinct aspects of differentiation such as known HSPC regulators (HOXA9 and SOX4), TF regulators shared between both lineages (GATA1 and GFIB), and lineage specific regulators (KLF1, NFIA, MEIS1, PBX1) (Figure 6b). Projection of the sampling time on the PCA plot, highlights day 4 as the bifurcation point where erythro-megakaryocytic lineages resolve (Figure 6a).
Analysis of the bulk RNA-seq and chromatin accessibility profiles revealed discrete regulatory modules associated with distinct functional states. We sought to explore whether these modules are reflected in discrete populations of individual cells with distinct transcriptional signature. To this end we performed Louvain clustering48,49 on all cells using the top 10,000 highly variable and highly expressed genes (Figure 6c and Methods). This resulted in 9 clusters with largely distinct gene expression signature, closely associated with sampling days where cells are partitioned into primitive clusters with strong HSPC signature (clusters 0 and 5), early progenitors with myeloid signature (clusters 6, 8) and committed/mature cells for each lineage (clusters 7, 1,2, 3 and 4) (Supplementary Figure 13). Compatible with earlier results, clusters that exhibit myeloid gene signature (clusters 6 and 8) are almost exclusively comprised of early erythroid cells (Supplementary Figure 13), while megakaryocytes are divided between early cells with HSPC signature and committed cells expressing markers of megakaryopoiesis (Figure 6c).
Bulk analysis and sampling time points might obscure cellular heterogeneity and true developmental time. Furthermore, as ex vivo erythropoiesis and megakaryopoiesis are two independent systems induced by different cytokine stimuli, the differential lineage commitment kinetics observed between the two lineages might be a result of the culture conditions and therefore not directly comparable. In order to address this and directly compare transcriptional developmental dynamics between erythroid and megakaryocytic differentiation, we sought to align the gene expression patterns of these two lineages along a common developmental coordinate system (independent of sampling time). Given the resolution of the developmental trajectories between erythropoiesis and megakaryopoiesis provided by PCA, we used the first two principal components to infer a metric of developmental pseudotime (Supplementary Figure 14 and Methods). Plotting gene expression of individual cells against pseudotime, readily identifies a sequence of non-overlapping, developmentally punctuated transcriptional states punctuated for both lineages (Figure 6e). Downregulation of HSPC signature for both lineages occurs over the same developmental interval, confirming that the exit from the HSPC state occurs for both lineages where the two lineages rapidly diverge after that. Erythroid development exhibits transient activation of myeloid-related transcripts prior to terminal commitment, while megakaryopoiesis rapidly acquires maturation signature. This observation becomes particularly evident when we place the Louvain clusters over developmental pseudotime. Transition from megakaryocytic progenitor cluster 0 to early committed cluster 2 occurs over a very short developmental interval. In contrast, committed erythroid cluster 7 appears much later (Supplementary Figure 15). Interestingly, annotating the sampling time over developmental pseudotime we observe that transcriptional inflection points coincide with the sampling time points where rapid lineage restriction was observed on population level. Day 4 of differentiation (Figure 6e, purple dashed line) demarcates the exit from HSPC for both lineages, while day 6 (Figure 6e, red dashed line) marks the switch from myeloid state to erythroid commitment.
Taken together, these findings suggest that exit from HSPC is an intrinsic mechanism, independent of the cytokine environment, that occurs during both early erythroid and megakaryocytic differentiation and it is driven by a common set of TFs. Additionally, rapid transitions in transcriptional states are a result of changes in expression of lineage regulating TFs, happening in individual cells, over short intervals of developmental time.
Discussion
Here, we systematically link individual transcription factors and their target cis-elements along ex vivo human erythropoiesis, resolving how these elements organize temporally, encoding lineage commitment and differentiation during hematopoiesis. Multiple efforts have extensively studied the individual (cis- and trans-) regulatory components involved in erythropoiesis11 as well as other diverse hematopoietic lineages9,50–53. The bulk of these efforts base their findings either on immunophenotypically defined hematopoietic populations, or single-cell dissection of steady state heterogeneous sources, where developmental relationships between cells within a heterogeneous steady-state population can only be inferred14,16,54,55. In this work we differ by capitalizing on the continuity of the differentiating populations during ex vivo erythropoiesis to finely map chromatin accessibility and gene expression dynamics while we overcome the sampling biases inherent to the conventional immunophenotypic definitions of hematopoiesis46,56. Sequential sampling of a continuously differentiating culture allows the direct and repeated measurement of the dynamic epigenetic landscape along a defined lineage trajectory. In addition, a dense sampling approach enables the unbiased detection of transient events occurring over short intervals that would otherwise be missed by sparse sampling methodologies.
Integrative analysis of chromatin accessibility and gene expression during erythropoiesis revealed a sequence of discrete, non-overlapping regulatory modules comprising of interacting transcription factors and individual cis-regulatory elements, corresponding to distinct stages of erythroid development. Strikingly, the transition between the activity of these modules coincides with a sequence of experimentally validated rapid lineage restriction events. We found that the exit from the program associated with the HSPC state occurred independent of lineage outcome, as it was identified during ex vivo megakaryopoiesis. Furthermore, comparison of developmental transcriptomics of single-cells along erythropoiesis and megakaryopoiesis reveals that exit from HSPC occurs over the same developmental interval for both lineages, indicating an intrinsic mechanism independent of the cytokine environment.
We found the two lineages to exhibit differential commitment kinetics after exit from HSPC state. Erythroid differentiation maintains unipotent myeloid capacity until terminal erythroid commitment, as it transiently activates a broader regulatory program involving canonical myeloid transcription factors (FLI1, C/EBPs, IRF8, etc). There are several lines of evidence from single-cell assays in both mouse and human hematopoiesis where they converge on the existence of a population early on erythroid development with a myelo-erythroid signature14,17,57. Notably, transgenic mice lacking a set of the C/EBP family of myeloid regulators exhibit decreased erythroid output, in addition to the granulocytic/monocytic58. Although all three myeloid lineages (Ery, Mk, G/M) are considered to derive from a common myeloid progenitor (CMP), we found megakaryocytic commitment to occur rapidly upon HSPC exit, bypassing the transient expression (and presumably chromatin) program. This observation is compatible with the growing evidence suggesting that megakaryocytic lineage arises directly from the primitive hematopoietic compartments46,59–61.
The single-cell RNA-seq findings presented here, represent the most comprehensive analysis of single-cell gene expression along a closely monitored developmental system, so far. Knowledge of the actual time point where individual cells were sampled allowed us to align changes recorded from population level experiments to developmental changes in lineage regulating TF expression in single-cells. Strikingly, our single-cell based observations faithfully recapitulated the population-derived findings as the inferred developmental pseudotime aligns with the time-points where lineage restriction events were observed. This suggests that highly synchronized rapid shifts in gene expression levels of lineage regulators across individual cells, occurring over short intervals of developmental time, underpin the changes observed in bulk populations. This finding is in contrast to observations based on single-cell analyses of steadystate hematopoiesis whereby variability in the chromatin and transcriptional landscapes are interpreted as gradients of continuous regulatory states14,15,55,62.
We present novel insights into the developmental regulatory dynamics during hematopoiesis whereby lineage regulating TFs organize in a developmental manner with their individual chromatin elements into discrete regulatory modules associated with distinct developmental stages. The transitions between these modules are demarcated by rapid shifts in lineage potential during differentiation. Furthermore, this work provides a comprehensive approach to studying the cis- and trans-regulatory element dynamics, in the context of lineage commitment and differentiation utilizing a well-defined ex vivo differentiation system. Although these findings might not entirely recapitulate the regulatory events occurring during in vivo hematopoietic development, they provide a generalizable model of how the trans- environment interacts with chromatin along the developmental axis to instruct fate choice and lineage commitment. Additionally, the dense sampling and the systematic linkage between distal elements and target promoters provides sufficient resolution to identify stage-specific activity of regulatory elements. Such elements can prove particularly useful in transgene-based therapies where the efficacy of these methods relies on the precise modulation of gene expression in a developmental and lineage-specific manner.
Author contributions
G.G., M.I., and N.P. performed the experiments. G.G., A.N., T.S., and J.V. performed data analysis. G.G., N.P., and J.V. designed the experiments. J.S., M.Y., and J.V. supervised the study and provided consultation. G.G. and J.V. wrote the manuscript. J.V. conceived the study. All authors approved the manuscript.
Conflict of interest
The authors state no conflicts of interest.
Acknowledgments
The authors would like to thank Daniel Bates, Morgan Diegel, Douglas Dunn, Fidencio Neri, Ericka Otterman, Shinny Vong, and Alister Funnell from the Altius Institute for Biomedical Sciences for help in sequencing and genome editing. We also thank John Lazar and Thalia Papayannopoulou for critical review of the manuscript.
Footnotes
This version of the manuscript has been revised to update reference list and include a Conflict of interest section.