Enhancer and Transcription Factor Dynamics during Myeloid Differentiation Reveal an Early Differentiation Block in Cebpa null Progenitors

Enhancer and Transcription Factor Dynamics during Myeloid Differentiation Reveal an Early Differentiation Block in Cebpa null Progenitors


INTRODUCTION
Myeloid cells such as macrophages and neutrophilic granulocytes constitute crucial components of the innate immune system.According to the classical models for the hierarchical architecture of the hematopoietic system, these cells develop from hematopoietic stem cells (HSCs) through a series of intermediate progenitor cells including common myeloid progenitors (CMPs)-or their near equivalents, the pre-granulocytemonocyte progenitors (preGMs)-and granulocytic-monocytic progenitors (GMPs).Specifically, these progenitors have been found to expe-rience increasing degrees of lineage restriction as they progress along their path of differentiation (Orkin and Zon, 2008).The hierarchical organization of the hematopoietic system has been challenged by a number of studies that focus on single cells demonstrating extensive heterogeneity in progenitor populations previously thought to be homogeneous (Drissen et al., 2016;Paul et al., 2015).As an example, immunophenotypically defined CMPs harbor progenitors primed toward the megakaryocyte lineage and the monocytic lineage, as well as several granulocytic lineages (Paul et al., 2015).This suggests that lineage priming is an extensive phenomenon and that lineage fate decisions occur at earlier differentiation stages than was previously thought.
The dynamic patterns of gene expression during hematopoietic differentiation are governed by interactions between transcription factors (TFs) and proximal (promoter) and distal (enhancer) cis-regulatory elements (Boyle et al., 2008;Heintzman et al., 2009).Activities of distal enhancers are highly dynamic, as exemplified by global analyses of enhancer dynamics during hematopoietic differentiation (Lara-Astiaso et al., 2014;Luyten et al., 2014).More important, enhancer accessibility, as judged by nucleosome-deficient regions or H3K4me1/me2 marking, precedes gene expression.
PU.1 and CEBPA are key TFs during GM-lineage (myeloid) differentiation (Rosenbauer and Tenen, 2007).PU.1 is expressed throughout the hematopoietic hierarchy and is essential for the formation of CMPs/preGMs (Iwasaki et al., 2005;Scott et al., 1994).PU.1 binds to a vast number of enhancers and has been suggested to act as a so-called pioneering factor at a subset of these (Heinz et al., 2010).CEBPA is expressed during the latter stages of GM differentiation and is indispensable for the formation of GMPs (Zhang et al., 2004).Similar to PU.1, a role for CEBPA as a pioneering transcription factor has been suggested based in part on its ability to enhance the formation of induced pluripotent stem cells (iPSC) (Di Stefano et al., 2016;Ohlsson et al., 2016).However, the extent to which CEBPA plays a similar role during GM-lineage differentiation remains to be explored.
In line with their crucial importance during GM-lineage differentiation, the expression of both Cebpa and Spi1 (encoding CEBPA and PU.1, respectively) is tightly controlled.In humans, 14 different enhancers drive the expression of CEBPA in different tissues, with the À42 kb enhancer being responsible for the CEBPA expression in the GM-lineage (Avellino et al., 2016).Indeed, deletion of the corresponding and highly conserved À37 kb enhancer in the mouse efficiently ablates Cebpa expression in myeloid cells and leads to a phenotype that is essentially indistinguishable from that of knocking out Cebpa in the entire hematopoietic system (Avellino et al., 2016;Guo et al., 2016;Hasemann et al., 2014).Spi1 expression is to a large extent controlled by a À14 kb enhancer, which in mice reduces the levels of PU.1 by 80% upon deletion (Rosenbauer et al., 2004).It is interesting that both PU.1 and CEBPA bind and activate this enhancer, demonstrating both autoregulation and CEBPAdependent regulation of Spi1 expression (Yeamans et al., 2007), although the importance of the latter has not been addressed in a proper in vivo setting.
Despite the evidence for extensive functional interplay between PU.1 and CEBPA, little is known of how these TFs work together to control the chronology of enhancer establishment during GM differentiation in a proper in vivo context.Here, we use a genome-wide chromatin immunoprecipitation-sequencing (ChIP-seq) approach to delineate the exact chronology of enhancer establishment during GM differentiation and its relation with the activity of PU.1 and CEBPA.We show that PU.1 predominantly binds at enhancers established during the early stages of differentiation and that late-established enhancers are gradually less dependent upon PU.1 binding.We observed instead that predominant CEBPA binding at late-established enhancers pointed toward a pioneering function of CEBPA at these elements.The distinct kinetics in the binding profiles of both PU.1 and CEBPA correlate with the enrichment of their respective sequence-binding motifs at enhancers.Analysis of Cebpa À/À (knockout [KO]) cells revealed a marked reduction in PU.1 binding at many enhancers, uncovering a surprising CEBPA dependency for binding of PU.1.Finally, we used these analyses to demonstrate that the loss of CEBPA leads to a GM differentiation block preceding preGMs.
We used our recently developed peak-valley-peak (PVP) method, which relies on the local H3K4me1 distribution for the detection of active enhancers (Pundhir et al., 2016), to predict a total of 44,002 enhancers across the four populations (Figure S2A; Table S2), most of which (86%) were located distally (>500 bp) to transcription start sites (TSSs).Consistent with our previous observations, our PVP-predicted enhancers only display modest overlap (18%) with those derived from a classical read density-based approach (Lara-Astiaso et al., 2014;Pundhir et al., 2016).However, the PVP enhancers were associated with higher levels of accessibility (H3K4me1 and assay for transposase-accessible chromatin using sequencing [ATAC-seq]) and activity (H3K27ac), demonstrating that the H3K4me1 PVP pattern is efficient in capturing active enhancers (Figure S3A).Thus, our enhancer predictions provide an accurate map of the regulatory complexity and dynamics during GM differentiation, with enhancer regions showing distinct patterns of accessibility (Figures 1B [early], 1C [throughout], and 1D [late]).
Globally, we observed two major groups of enhancers that either maintain or lose H3K4me1 accessibility beyond the GMP stage (GRN-active and GRN-inactive) (Figure 2A).Within each group, we detected enhancer groups that gain and maintain H3K4me1 accessibility across distinct GM-lineage stages.These are exemplified by group 1 and group 5 enhancers, which are established in LSKs, whereas group 3 and group 7 enhancers are established at the GMP stage (Figure 2B).The accessibility profile of enhancers from each group correlated with a gradual gain in their activity as measured by H3K27ac levels and proximal gene expression, both of which followed similar profiles across the four stages (Figures 2C-2E and S3B).
Our enhancer classification was corroborated further by a distinct enrichment profile of previously reported myeloid enhancers in the GRN-active group and of stem cell enhancers in the GRN-inactive group (Lara-Astiaso et al., 2014) (Figure S3C).This observation suggests an active role for GRN-active enhancers in GM-lineage differentiation and a role for GRN-inactive enhancers during stem cell fate decisions.Indeed, we observed an enrichment of myeloid, lymphoid, and erythroid enhancers in the GRN-inactive group, consistent with the multipotent potential of LSK cells.However, we also note that even in this group, the frequency of myeloid-specific enhancers supersedes that of their lymphoid and erythroid counterparts, suggesting that priming toward the myeloid lineage is a key feature of the most immature blood cells.Finally, and consistent with their importance during myeloid differentiation, GRN-active enhancers are enriched for H3K27ac when compared to their GRN-inactive counterparts (Figures S2B-S2E).Taken together, we delineated enhancers into distinct groups that exert their regulatory functions across different stages of GM differentiation.
PU.1 and CEBPA Bind to a Large Fraction of GM Enhancers but Are Associated with Separate Groups of Enhancer Accessibility To determine the TFs that are potentially associated with enhancer dynamics during GM differentiation, we compared the enrichments of de novo motifs in GRN-active and GRN-inactive enhancers.This demonstrates enrichment of the motifs that are associated with the myeloid master regulator CEBPA in GRN-active enhancers and with PU.1 in both GRN-inactive and GRN-active enhancers, which is consistent with the pan-hematopoietic importance of PU.1 (Figure S4A).
To elucidate further the roles of these factors during GM differentiation, we performed ChIP-seq in LSKs, preGMs, and GMPs and identified a total of 52,338 and 22,582 regions bound by PU.1 and CEBPA, respectively (Figure S4).Whereas 49% of PU.1-bound regions were identified in all populations, the number of CEBPA-bound regions increased gradually along the GM differentiation path.These findings are consistent with the expression patterns for these factors because CEBPA is upregulated as cells progress from LSKs to GMPs, while PU.1 is highly expressed at all stages.Well-known binding motifs for PU.1 and CEBPA were enriched in PU.1-and CEBPA-bound regions, respectively, along with a prominent enrichment of a PU.1-binding motif in CEBPA-bound regions.We find 31% of CEBPA-bound regions to be co-bound by PU.1 (Figure S4H) in GMPs, suggesting an extensive degree of cooperation between PU.1 and CEBPA.
Next, we categorized our enhancer set into those bound by either PU.substantial portion (36%) of these enhancers was bound by PU.1 and/or CEBPA and showed significantly higher levels of activity (H3K27ac; highest for co-bound) than the levels displayed by enhancers devoid of the two TFs (Figures 3A-3C).We used these data to determine the stages at which PU.1 and CEBPA exerted their main functions during GM-lineage differentiation, focusing on the GRN-active enhancers.Here, we observed a striking correlation between the timing of enhancer establishment and PU.1-binding levels (i.e., enhancers that were established in LSKs displayed higher levels of PU.1 in GMPs) (Figure 3E, group 1).Conversely, late-established enhancers displayed comparably lower levels of PU.1 (groups 2 and 3).These observations Cell Reports 23, 2744-2757, May 29, 2018 2747 correlated directly with the fraction of enhancers bound by PU.1 (Figure 3D).In contrast, CEBPA binding did not correlate with early enhancer establishment but was instead found mainly on group 3 enhancers that became established in GMPs (Figure 3F).Again, a higher fraction of these enhancers was associated with CEBPA binding compared to the other groups (Figure 3D).PU.1binding motifs were consistently mainly enriched in early established enhancers, whereas CEBP motifs were preferably enriched in group 3 enhancers (Figure 3G).Collectively, we have shown that PU.1 and CEBPA exert their main effects on enhancers that are established in LSKs and GMPs, respectively, and that the combined action of these two factors controls the activities of a substantial proportion of GM-lineage enhancers.Thus, our data uncover the underlying molecular basis for the critical roles of PU.1 and CEBPA in the formation of CMPs/preGMs and GMPs, respectively.

Loss of CEBPA Has a Strong Impact on PU.1 Binding
The current model of the PU.1/CEBPA interplay during late stages of GM differentiation holds that PU.1 facilitates the binding of CEBPA to numerous genomic loci by acting as a pioneering factor (Heinz et al., 2010).This model is consistent with both the expression patterns of PU.1 (early and maintained) and CEBPA (induced late) during GM differentiation and the extensive overlap of their binding profiles.However, given the early differentiation block in Spi1 À/À myeloid progenitors, it has not been possible to test this model in a proper in vivo context, and it remains a formal possibility that CEBPA also controls PU.1 binding and levels.
Loss of CEBPA in murine bone marrow (BM) cells leads to a differentiation block upstream of GMPs (Hasemann et al., 2014;Zhang et al., 2004).To test the possibility that CEBPA may affect PU.1 function or levels, we isolated preGMs from Cebpa À/À mice (preGM-KO) and generated genome-wide binding profiles for PU.1, H3K4me1, and H3K27ac.The loss of CEBPA was associated with a significant loss of PU.1 peaks in preGM-KO (30,893 versus 44,813 in preGM-wild-type [WT]), suggesting that CEBPA controls the genome-wide association of PU.1 (Figure 4A).Moreover, we observed an overall reduction in PU.1 levels on the 44,002 GM-lineage enhancers when comparing preGM-KO to their WT counterpart (Figure 4B).Experiments in cell lines have shown that CEBPA binds and activates the À14 kb Spi1 enhancer, which is a main driver of PU.1 transcription; this raises the possibility that the loss of CEBPA reduces the levels of PU.1 (Yeamans et al., 2007).We find that CEBPA binds this enhancer in LSK cells and that the loss of CEBPA is associated with a decrease in activity (H3K27ac; Figure 4C).We find reduced Spi1 mRNA levels in preGM-KOs translating consistently into an approximate 2-fold reduction in PU.1 protein levels, as determined by quantitative mass spectrometry comparisons of WT and preGM-KOs (Figure 4D).Consistent with a previous report, loss of CEBPA leads to upregulation of the Cebpg mRNA and CEBPG protein (Figure 4D) (Alberich-Jorda èt al., 2012).These findings clearly demonstrate that in addition to the effect on CEBPA-bound enhancers, the loss of CEBPA also affects enhancers bound by PU.1 because of the overall reduction of PU.1 levels in preGM-KOs.
CEBPA Binds to GM-Lineage Enhancers in PU.1-Dependent and -Independent Manners Our analyses so far have demonstrated a complex interplay between PU.1 and CEBPA during GM differentiation.Because loss of GMPs represents the main feature of CEBPA ablation, we decided to focus on changes in enhancer accessibility during the preGM to GMP transition.
To account for the recently reported heterogeneity in GMPs (Olsson et al., 2016), we further fractionated this population into colony-forming unit-granulocytes (CFU-Gs) and colonyforming unit-macrophages (CFU-Ms) (primed toward GRN and MONO production, respectively) and subjected these to TFs and histone profiling (Figures S1C and S1D).Focusing on GRN differentiation, we next ranked our 44,002 GM-lineage enhancers (from Figure 2) on the basis of their changes in H3K4me1 marks during the preGM to GMP transition.For each of the resulting 13 subclusters, we assessed the levels of PU.1, CEBPA, H3K4me1, H3K27ac, and gene expression in LSKs, preGMs, preGM-KOs, GMPs, and CFU-Gs (Figures 5A-5E).We further investigated the degree of change in these features during the transition from LSK to preGM, from preGM to GMP, and between WT and Cebpa À/À preGM (Figures 5F-5H).Enhancers belonging to cluster A gradually lose H3K4me1 from LSKs to GMPs, correlating with a reduction in H3K27ac and gene expression.Only a minor fraction of these enhancers is bound by PU.1 and/or CEBPA and mainly constitute GRN-inactive enhancers enriched for stem cell and non-GM-lineage enhancers (Figures 5I, 5J, and S3C).Cluster B is constituted by enhancers that either maintain (Bi) or moderately increase (Bii) their H3K4me1 levels as cells differentiate from LSKs to GMPs.A quarter of these enhancers (26%; 9,839 out of 37,519) are bound by PU.1 and this cluster also is characterized by the highest fraction of PU.1/CEBPA co-bound enhancers and by a transient gain of PU.1 binding from LSK to preGM.Most subcluster Bi enhancers lose PU.1 binding when differentiating onward to GMPs, whereas the Bii subcluster either maintains or further acquires PU.1 binding.Cluster B enhancers gain H3K27ac, but only the Bii subcluster enhancers activate gene expression in GMPs.Subcluster Bi enhancers overlap with stem and non-myeloid GRN-inactive enhancers, likely containing enhancers that are active in a variety of lineages (Figure 5J).It is interesting that the transcription of cluster B enhancerdriven genes is induced mainly transiently, suggesting that they may be important for promoting preGM formation.Of note, the gain of H3K27ac on subcluster Bii enhancers correlates with CEBPA binding, suggesting that CEBPA activates enhancers already marked by PU.1.However, because these enhancers are bound by PU.1 already in LSKs, any potential pioneering function of PU.1 will have to occur at earlier developmental stages.Finally, cluster C contains enhancers that are established mainly in GMPs.These enhancers are characterized further by low levels of PU.1 in preGM and acquire CEBPA binding in GMPs, followed by an increase in PU.1 levels (Figures S5A  and S5B).This correlates strictly with the acquisition of H3K27ac and activation of gene expression.Indeed, cluster C enhancers are highly enriched for CEBPA, suggesting that CEBPA is able to assess closed chromatin in the absence of PU.1.The behavior of clusters A-C enhancers are further stratified based on their PU.1 and CEBPA binding status (Figure S6).
To gain further insight into the enhancers controlled by CEBPA at the preGM to GMP transition we took advantage of our gene expression data to define a group of CEBPA-dependent genes (up during the preGMP to GMP transition and down in preGM-KOs) and CEBPA-independent genes (down during the preGMP to GMP transition and up in preGM-KOs compared to preGMs; Figure S5C).CEBPA-dependent genes are markedly enriched for genes involved in GM-specific gene programs and are

CEBPA Loss Blocks GM Differentiation Upstream of the preGM Stage
Previous studies, including our own, have demonstrated that Cebpa À/À mice exhibit a differentiation block upstream of the GMP (Hasemann et al., 2014;Mancini et al., 2012;Zhang et al., 2004).The identity of the upstream progenitor depends on the immunophenotyping strategies used in these particular studies, and both CMPs and preGMs have been shown to accumulate in the absence of CEBPA.However, the immunophenotypes of CMPs and preGMs only differ from the immunophenotype of upstream LSK cells by the loss of Sca-1 expression, thus rendering the definition of these progenitors sensitive to the expression kinetics of this marker.Moreover, because the functional definition of Cebpa À/À progenitors is prohibited by the inability of these cells to sustain GM output, we decided to take a more global approach to characterizing preGM-KO cells.The changes across our ranked enhancers (clusters A-C) comparing KO to WT preGMs were almost the complete opposite of the changes occurring at the LSK to preGM transition (Figure 5H, compare to Figure 5F), thereby placing immunophenically defined preGM-KOs near the LSK compartment.
Gene set enrichment analysis (GSEA) demonstrated consistently that preGM-KOs were enriched for the expression of a pre-megakaryocyte erythrocyte progenitors (preMegEs) signature and depleted of the expression of a preGM signature relative to their WT counterparts (Figure 6A).This, along with the observed increased expression of a stem cell signature in preGM-KOs, places this population upstream of its WT counterpart (Figure 6A).Because our analysis of enhancer usage demonstrates extensive dynamics of PU.1 binding and H3K4me1 levels during differentiation along the GM-lineage, we next used principal-component analysis of these features as well as of global gene expression to further assign the differentiation block of Cebpa À/À GM progenitors.All three features placed preGM-KOs upstream of preGM-WTs and immediately downstream of LSKs (Figures 6B-6D).Overall, these findings suggest that loss of CEBPA results in the inability of LSKs to progress beyond the point of downregulation of Sca-1 expression, thereby moving the CEBPA restriction point to a position well before the preGM population (Figure 6E).
The Distinct Chromatin Landscapes of Granulocytic and Monocytic Progenitors Prime Their Differentiation toward Mature Populations CEBPA and PU.1 are known to drive differentiation toward the granulocytic and monocytic lineages, respectively, but how this differential behavior is coordinated and how it relates to the recently described functional heterogeneity within the GMP population remains unanswered.To address these questions, we first classified 37,221 enhancers as common or specific to GRN and MONO lineages based on the ratio of their H3K4me1 levels in granulocyte and monocyte populations, respectively (Figure S7A; Experimental Procedures).Next, we assessed the activity and TF occupancy at these enhancers in their corresponding upstream progenitors.In support of a priming model, we find that both GRN-and MONO-specific enhancers displayed higher levels of activity (H3K27ac) and proximal gene expression already in their respective progenitor populations (CFU-Gs and CFU-Ms) and that their activity gradually increased as cells differentiate into granulocytes or monocytes (Figure 7A).Furthermore, PU.1 binding was relatively higher at MONOcompared to GRN-specific enhancers already in preGMs, whereas a weak but opposite trend was observed for CEBPA (Figure 7A).These findings are corroborated further by the observed enrichment of CEBPA and PU.1 sequence motifs in GRN-and MONO-specific enhancers, respectively (Figure 7D).
GRN-specific enhancers are highly enriched for the previously defined CEBPA-dependent cluster C enhancers (Figure 7B).Likewise, these enhancers are activated (H3K27ac) concomitantly with the acquisition of CEBPA binding in CFU-Gs and are associated with an increase in target gene expression (Fig- ures 7A-7C).Although CEBPA binding to GRN-specific enhancers also increases in CFU-Ms, this increase is much less pronounced and is not accompanied by increases in H3K27ac and gene expression levels (Figure 7A).
For the MONO-specific enhancers, CEBPA binding remains low while PU.1 occupancy increases, albeit to the same levels in both CFU-Gs and CFU-Ms.The latter observation suggests that other TFs are instrumental for monocyte differentiation in line with the well-established role for interferon regulatory factor (IRF) family members such as IRF8 (Figures 7A, S7B, and S7C) (Becker et al., 2012).Indeed, several IRF motifs are selectively enriched at MONO-specific enhancers, and IRF8 is the most upregulated factor in CFU-Ms relative to CFU-Gs (Figures 7D  and 7E).Finally, a major fraction of enhancers (N = 22,332) was common to both granulocytes and monocytes.These enhancers displayed high levels of H3K27ac and gene expression in preGM, which remained constant as the cells differentiated toward CFU-Gs and CFU-Ms.While PU.1 occupancy at these enhancers decreased as preGMs differentiate into CFU-Gs or CFU-Ms (Figure 7A), CEBPA binding increased, albeit more profoundly in CFU-Gs (Figure 7A).The kinetics of PU.1 and CEBPA are similar to our previous observation for PU.1-driven enhancers (cluster B in Figure 5) that govern the differentiation from LSKs to preGMs.Indeed, we found a significant overlap between cluster B enhancers with this common group of GM enhancers (Figure 7B).
Finally, to discern the transcriptional program regulated by GRN-and MONO-specific enhancers, we defined 376 (CFU-G) and 457 (CFU-M) genes that are differentially expressed between the CFU-G and CFU-M populations (Figures 7E and  S7D).We observed an enrichment of functional categories that are crucial for the immune function of each of the two lineages (Figure S7E).We also detected a relative increase in the expression of cell-cycle genes in CFU-Ms compared to that of CFU-Gs (Figures 7E and S7E).We hypothesize that the selective repression of cell-cycle genes in CFU-Gs is associated with the increased expression of CEBPA in this compartment, which would be consistent with its recognized function as a negative regulator of proliferation (Johansen et al., 2001;Porse et al., 2001).
In conclusion, MONO-and GRN-specific enhancers are established as preGMs differentiate toward CFU-Ms and CFU-Gs, respectively, and are differentially occupied by PU.1 and CEBPA during GM-lineage differentiation (Figure 7F).

DISCUSSION
Defining how gene regulatory networks serve to shape biological properties is key to understanding how cells progress from immature stem and progenitor cells into their mature progeny.Differentiation processes are governed by the actions of TFs, some of which exhibit highly lineage-restricted expression patterns.Within the hematopoietic system, CEBPA is expressed almost exclusively in the GM branch of the hematopoietic hierarchy and is ultimately required for the development of mature myeloid cells, but the underlying mechanisms for this requirement have remained elusive.By integrated analysis of enhancer dynamics, TF binding, and proximal gene expression during GMlineage differentiation in the context of a normal and a Cebpa À/À hematopoietic system, we showed an extensive functional interplay between CEBPA and PU.1 in driving GM-lineage differentiation.Moreover, our analyses also demonstrate that the loss of CEBPA leads to a differentiation block at an earlier stage than previously reported.Thus, our work yields insights into how TFs functionally interact to drive differentiation along the GMlineage.
Enhancer Dynamics during Differentiation along the GM-Lineage Using our recently developed PVP approach for the analysis of histone modifications, we defined the enhancer landscapes during differentiation from multipotent progenitors (LSKs) to mature neutrophilic granulocytes.A large fraction of identified enhancers was accessible already at the LSK stage and became either more accessible (GRN-active) or was shut down (GRNinactive) during GM differentiation, demonstrating both a high degree of enhancer dynamics and a high level of lineage priming in immature populations (Lara-Astiaso et al., 2014;Paul et al., 2015).Indeed, enhancers that became closed in mature cells were assigned mainly to non-GM-lineage genes and genes associated with stem cell functions.
De novo motif analysis identified PU.1 and CEBPA as potential key TFs at GRN-active enhancers.Consistent with this notion, PU.1 and CEBPA bound pervasively to GM-lineage enhancers, and enhancers bound by either or both of these factors were enriched for enhancer activity.PU.1 preferred enhancers that were accessible at earlier stages, likely reflecting its role as a pan-hematopoietic master TF (Wilson et al., 2010).In contrast, CEBPA bound mainly to late-accessible enhancers, which is consistent with its expression pattern and its importance in inducing the expression of genes involved in granulocyte function and development.Finally, by additional subfractionation of GMPs into CFU-Ms and CFU-Gs, we were able to provide evidence for the selective importance of CEBPA in the latter population, which is consistent with the view of CEBPA as the key TF during granulocytic differentiation.Overall, these analyses reveal the underlying molecular basis for the importance of both PU.1 and CEBPA in setting up gene regulatory programs during GM-lineage differentiation.
Several Layers of Functional Interaction between PU.1 and CEBPA PU.1 has been reported to serve as a pioneering factor during GMlineage differentiation, facilitating H3K4me1-driven enhancer formation and subsequent recruitment of TFs, including CEBPs (Heinz et al., 2010).Although these data clearly demonstrate the ability of PU.1 to bind and remodel closed chromatin, the fact that they were generated by re-expression of a PU.1-ER fusion protein in a PU.1 null context does not allow us to assess the extent to which this occurs during normal GM-lineage differentiation.In the present work, we observed little evidence for PU.1 binding to closed chromatin, most likely reflecting the fact that a vast amount of hematopoietic enhancers are marked by PU.1 already in HSCs (Wilson et al., 2010).However, our data do not exclude a pioneering function for PU.1 at early stages such as during HSC specification or in other hematopoietic lineages.
Cluster C enhancers constitute a group of enhancers that remains inaccessible in the absence of CEBPA.In contrast to PU.1, CEBPA binding clearly correlates with the opening and activation of these enhancers, and re-expression of CEBPA in cultured Cebpa null progenitors was associated with enhancer opening concomitant with CEBPA binding.These findings strongly suggest that CEBPA acts as a pioneering factor on selected enhancers, perhaps via its switch/sucrose nonfermentable (SWI/SNF) domain, which has been shown to be important for enhancer binding in adipocyte differentiation (Madsen et al., 2014).
PU.1 and CEBPA exhibit extensive co-binding at GM-lineage enhancers.A large degree of CEBPA binding occurs at enhancers already bound by PU.1, but also we detect examples of enhancers in which PU.1 binding is dependent on CEBPA.These examples include co-bound enhancers in which PU.1 binding follows CEBPA occupancy (Figures 5,S5A,and S5B).This dependency may reflect the lower expression of PU.1 in Cebpa À/À cells, where we observe an overall reduction in PU.1 binding.The reduction in PU.1 expression could be traced back to the À14 kb Sfi1 enhancer, where CEBPA binding was detected in LSK cells.Collectively, our data suggest not only that CEBPA binds to chromatin in PU.1-dependent and -independent manners but also that CEBPA is important for maintaining appropriate levels of PU.1.Hence, these findings extend the current view on the PU.1/CEBPA interplay during GM-lineage differentiation toward a more active role for CEBPA.
Loss of CEBPA Leads to an Early Block in GM Differentiation Previous studies have placed the CEBPA-mediated differentiation block during GM differentiation immediately upstream of the GMP (Hasemann et al., 2014;Mancini et al., 2012;Zhang et al., 2004).This was inferred from the accumulation of the immediate upstream population (i.e., preGM or CMP), depending on the specific immunophenotyping strategy.However, preGMs and CMPs differ from the broad LSK population only through their lack of Sca-1 expression.Using gene expression data and H3K4me1 and PU.1 ChIP-seq data, we showed that Cebpa À/À preGMs mapped closely to LSKs, suggesting that the loss of Cebpa leads to a differentiation block upstream of the preGM at a stage immediately after downregulation of Sca-1 expression (Figure 6E).Previous studies have shown that loss or functional inactivation of CEBPA in various settings frequently is associated with the aberrant expression of erythroid and lymphoid gene expression signatures (Bereshchenko et al., 2009;Mancini et al., 2012;Schuster et al., 2013;Wouters et al., 2007).Consistently, preGM/CMP expressing no or functionally compromised CEBPA shows an increased potential toward non-GM-lineages.Moreover, in a leukemic setting, epigenetic silencing of CEBPA sustains the production of leukemic blasts with both T cell and GM characteristics, whereas BCR-ABL, normally associated with GM leukemias, promotes the formation of erythroleukemias when overexpressed in Cebpa À/À progenitors (Figueroa et al., 2009;Wagner et al., 2006;Wouters et al., 2007).Whereas these non-GM-lineage characteristics of Cebpa À/À progenitors have been interpreted to reflect the loss of lineage fidelity imposed by the absence of Cebpa, they may as well reflect a less differentiated phenotype of Cebpa À/À progenitors, which is consistent with the model supported by our data.Indeed, this interpretation is supported further by single-cell RNA sequencing (scRNA-seq) analysis of Lin À ,Sca-1 À ,c-kit + progenitors, where loss of Cebpa leads to the formation of a transcriptional cluster characterized by the co-expression of GM, erythroid, and lymphoid genes (Paul et al., 2015).We hypothesize that this cluster represents a subset of our Cebpa À/À preGMs.
Reports have highlighted the importance of CEBPA as a transcriptional repressor in LSK cells, promoting the repression of target genes such as Sox4 and Cebpg (Alberich-Jorda `et al., 2012;Zhang et al., 2013).Consequently, the loss of CEBPA leads to increased expression of these factors and the GM differentiation block was rescued by short hairpin RNA (shRNA)-mediated knock down of either of these factors.Whereas the underlying mechanism of SOX4 remains unclear, CEBPG, which lacks a transactivation domain, was shown to control overall CEBP activity in Cebpa À/À progenitors through its ability to repress both the levels and transactivation potential of CEBPB (Alberich-Jorda `et al., 2012).The latter presumably occurs through the leucine zipper-mediated heterodimerization with CEBPB, a factor that is able to functionally substitute for CEBPA, as evidenced by the lack of any overt hematopoietic phenotypes of mice expressing Cebpb from the Cebpa locus (Jones et al., 2002).
Figure 1.Experimental Outline (A) Schematic diagram of the five GM differentiation stages profiled in the present study.Also shown are the datasets acquired from the different cell populations (gray boxes).(B-D) Examples of differential enhancer behavior illustrated by enhancers that become accessible early (B), accessible throughout (C), or accessible at late differentiation stages (D).Shown are the normalized H3K4me1 and H3K27ac signal profiles (TPM) at these enhancers (gray boxes), as well as the gene expression profiles of the nearest gene associated with these enhancers (boxes).See also Figure S1.

Figure 2 .
Figure 2. Enhancer and Gene Expression Dynamics during GM-Lineage Differentiation (A) Classification of 44,002 enhancers into two major groups based on their activity in granulocyte (GRN-active and GRN-inactive), and further subclassified into four and three groups, respectively, based on their accessibility profile across the four GM-lineage differentiation stages (LSK, preGM, GMP, granulocytes).Filled and unfilled circles represent accessible and non-accessible enhancers as determined by their H3K4me1-derived PVP patterns.(B) Normalized H3K4me1 signal (TPM) at midpoint-centered enhancers (rows, ±1,000 bp) across the four GM-lineage differentiation stages (columns; both as average and heatmap).(C) The same as (B), but for H3K27ac.(D) Scaled expression (Z score) of enhancer proximal genes across the four GM stages.Also shown is the median scaled signal at each stage above the heatmap.(E) Schematic illustration of the early to late establishment of enhancers from the seven groups.See also Figures S2 and S3.

Figure 3 .
Figure 3. Differential Behavior of PU.1 and CEBPA at GM-Lineage Enhancers (A) Enhancers (N = 44,002; bubbles) classified into four categories on the basis of normalized PU.1 (y axis) and CEBPA (x axis) signals.(B) Density plot of the H3K27ac signal at four enhancer categories.ECDF, empirical cumulative distribution function (Kolmogorov-Smirnov test).(C) Differentially enriched de novo sequence motifs at four enhancer categories.The top enriched and depleted motifs are shown along with their bestknown motif match names.(D) Fraction of enhancers from GRN-active groups bound by PU.1, CEBPA, both PU.1 and CEBPA, or other TFs (in GMPs).(E) PU.1 binding signal (TPM) at enhancers belonging to the four GRN-active groups from LSKs to GMPs.p values (PU.1) represent the significance level at which the PU.1 signal (preGM) at enhancers from an earlier established group is higher than at enhancers from a later-established group (Wilcoxon rank-sum test).(F) The same as (E), but for CEBPA binding.p values (CEBPA) represent the significance level at which CEBPA signal (GMP) at enhancers from two adjacent groups are different.(G) Differential enrichment of de novo sequence motifs in the four GRN-active groups.The top enriched or depleted motifs are shown along with their best-known motif match name.See also Figure S4.

Figure 4 .
Figure 4. Loss of CEBPA Affects PU.1 Levels and Binding Properties (A) Overlap between PU.1 peaks from WT and Cebpa KO preGMs.(B) PU.1 density plots in WT and Cebpa KO preGMs (Kolmogorov-Smirnov test).(C) University of California, Santa Cruz (UCSC) genome browser screenshot of PU.1 enhancers (gray).The À14 kb enhancer is indicated.(D) Spi1 and Cebpg expression (top) and corresponding protein levels (bottom) levels in WT and Cebpa KO preGMs (gene expression measured in DESeq2).Error bars represent the SD between the three replicates.

Figure 5 .Figure 6 .
Figure 5. PU.1 and CEBPA Binding Profiles during GM-Lineage Differentiation The 44,002 GM enhancers were ranked (N = 13) based on the fold change in H3K4me1 signal between the GMP and preGM stages (GMP/preGM).(A) Mean PU.1 signal at GM enhancers from each cluster.(B-E) The same as (A), but for CEBPA (B), H3K4me1 (C), and H3K27ac (D) and for the expression of the nearest associated gene (E).(F) Median fold change in PU.1, CEBPA, H3K4me1, and H3K27ac levels at enhancers from each cluster during the LSK to preGM transition along with gene expression fold changes of the nearest associated gene.(G and H) The same as (F), but for the preGM to GMP transition (G) and between WT and Cebpa KO preGMs (H).(I) Fraction of enhancers bound by PU.1, CEBPA, both PU.1 and CEBPA, or other TFs.(J) Differential enrichment of seven enhancer groups (from Figure 2) at enhancer clusters A, B, and C. See also Figures S5 and S6.

Figure 7 .
Figure 7. Monocytic Enhancers Are from Granulocytic Enhancers in Terms of Their Transcription Factor Environment and Target Genes (A) Enhancers classified on the basis of their activity in granulocytic, monocytic, or both lineages.The dynamics H3K4me1, H3K27ac, proximal gene expression, PU.1, and CEBPA binding levels (quantile normalized) during differentiation of preGMs granulocytes or monocytes the CFU-G or CFU-M intermediates, respectively.Also shown is the fraction of enhancers from the three classes that are bound by PU.1, CEBPA, or both TFs in the GMP stage.(B) Percentage of group C and B enhancers from Figure 5 that overlaps with common and granulocytic enhancers (Fisher's exact test).(C) A representative example of a CEBPA-bound enhancer active specifically in the granulocytic lineage.(D) Differentially enriched sequence motifs in the three enhancer classes.The top enriched and depleted motifs are shown.(E) Volcano plot for 833 differentially expressed (DE) genes between CFU-G and CFU-M.Names of selected DE genes are shown, along with the fraction of DE cell-cycle genes between the two populations.(F)Model illustrating the differential activity profile of PU.1 and CEBPA-dependent enhancers during GM differentiation.The CFU-GM population is indicated for the sake of completion, although it was not assayed here.See also FigureS7.