MLL3 and MLL4 sustain hematopoietic stem cell multipotency by opposing a B-cell default state

Key Points

• •MLL3 and MLL4 act redundantly in HSCs to sustain transcription factor and enhancer networks that support multipotency.
• •Simultaneous loss of MLL3 and MLL4 drives hematopoietic progenitors into a uniform B-cell–like default state.

Visual Abstract

Abstract

Hematopoietic stem cells (HSCs) and multipotent progenitors (MPPs) are sustained by networks of transcription factors and epigenetic regulators that prime lineage-specific programs yet maintain multipotency. Two epigenetic regulators, MLL3 and MLL4, play important but distinct roles in maintaining this balance. MLL3 promotes HSC differentiation, whereas MLL4 opposes differentiation. These activities are essential for both normal homeostasis and leukemia suppression, yet it is not clear how MLL3 and MLL4 regulate HSC and MPP gene expression to control HSC/MPP fate decisions. To resolve these mechanisms, we performed an extensive series of single-cell genomic studies after conditionally deleting Mll3, Mll4 or both genes together. Mll3 deletion had only limited effects on HSC/MPP enhancer networks at steady state, whereas Mll4 deletion led to precocious activation of myeloid enhancers. Surprisingly, compound Mll3/4 deletion eliminated all myeloid, erythroid, and megakaryocytic potential within the hematopoietic hierarchy and caused all progenitors to rapidly default to a B-cell–like identity. These changes were accompanied by widespread inactivation of HSC/MPP enhancers and superenhancers and ectopic activation of B-cell superenhancers. Disabling MLL3/4 histone methyltransferase activity did not recapitulate the pervasive changes in cell identity that were observed when MLL3 and MLL4 were fully inactivated, indicating that MLL3 and MLL4 activate HSC/MPP enhancers independently from their enzymatic activities. Our findings show that HSC/MPP multipotency requires sustained tension between MLL3/4-dependent enhancers that maintain myeloid, erythroid, and megakaryocyte potential and MLL3/4-independent enhancers that prime B-cell identity. MLL3 and MLL4 therefore serve as critical linchpins of multilineage hematopoiesis.

Chromatin regulators MLL3 and MLL4 act redundantly to preserve hematopoietic stem and progenitor cell multipotency while preventing inappropriate B-lineage differentiation. Wang and colleagues report that these functions are independent of their histone methyltransferase activity and the associated UTX demethylase, revealing a catalytic activity–independent mechanism of epigenetic control. Their findings identify MLL3 and MLL4 as key regulators of hematopoietic stem cell fate and provide a framework for understanding how chromatin regulatory networks maintain stem cell identity and contribute to hematologic disease.

Introduction

Throughout life, blood cells arise from hematopoietic stem cells (HSCs) and multipotent progenitors (MPPs) that either self-renew or differentiate into myeloid, lymphoid, megakaryocytic, or erythroid progenitors.^{1, 2, 3, 4, 5, 6} Maintaining this extensive degree of multipotency requires gene regulatory networks that prime several different lineage programs within a given stem cell while simultaneously preventing any one program from fully asserting itself up until the time of lineage commitment. Gene regulatory programs that prime and maintain the multipotent state have been scrutinized heavily, and many critical transcription factors have been identified.^{7, 8, 9} For example, the transcription factor MECOM recently has been shown to prevent HSC differentiation by directly repressing genes that encode myeloid transcription factors, such as CEBPA.^10,11 Other transcription factors, such as GATA2^12,13 and ERG,^14,15 are similarly necessary to maintain HSCs. Epigenetic regulators often interact with transcription factors to enforce cell fate decisions by covalently modifying histones or DNA, or by altering chromatin accessibility. These regulators play essential roles in maintaining normal HSC homeostasis and suppressing leukemic transformation, yet we have only a limited understanding of how they are deployed to coordinate HSC/MPP fate decisions.

MLL3 (also called KMT2C) and MLL4 (also called KMT2D and previously MLL2) are 2 highly homologous epigenetic regulators that have established roles in balancing HSC self-renewal and differentiation.^{16, 17, 18} They therefore offer potential insights into how individual epigenetic regulators coordinate gene expression to either reinforce the multipotent state or promote differentiation.^16,17,19 Each protein nucleates large chromatin-bound complexes that belong to the complex of proteins associated with SET1 (COMPASS) family (Figure 1A-B). COMPASS complexes bind enhancer elements to promote gene expression.²⁰ MLL3 and MLL4 contain SET domains that monomethylate histone H3 lysine 4 (H3K4me1) to prime enhancer elements for activation.^{21, 22, 23} MLL3/4 COMPASS complexes have also been shown to regulate gene expression independently of SET methyltransferase activity by nucleating transcriptional condensates and promoting RNA polymerase pause-release.^{24, 25, 26, 27} These diverse biochemical functions empower MLL3/4 complexes to drive cell identity changes across a range of tissues in all metazoan species^{21,23,28, 29, 30}; however, the proteins perform both redundant and nonredundant functions in vivo.^23,31,32

Figure 1.

MLL3 and MLL4 act redundantly to sustain myeloid colony formation and hematopoiesis. (A) MLL3 and MLL4 protein functional domains. Both proteins are characterized by clusters of plant homeodomains (PHD), a high mobility group (HMG) domain, phenylalanine/tyrosine-rich N-terminus/C-terminus (FYRN/C) domains, and a SET methyltransferase domain. MLL4 has a glutamine rich prion-like domain (PLD). (B) MLL3- and MLL4-COMPASS complexes. Purple indicates MLL3/4-specific cofactors. (C) Donor (CD45.2⁺) blood chimerism in secondary recipients of the indicated donor bone marrow genotypes (n = 15-20). (D) Donor HSC chimerism in secondary transplantation recipients of the indicated genotypes at 16 weeks after transplant (n = 15-20). (E) The percentage of donor HSCs with myeloid colony forming unit (CFU) potential (n = 3). (F) The percentage of myeloid colonies with genotypes that would be expected following complete deletion of floxed alleles (n = 3). (G-I) Donor HSC, MPP, and Lineage⁻Kit⁺Sca1⁻ numbers in recipients at 2 weeks after pIpC treatment (2 hindlimbs) (n = 16-21). (J) Representative bone marrow histology 30 weeks after pIpC treatment (n = 8-11). Scale bars, 100 μm. For all panels, the error bars reflect standard deviation. ∗P < .05; ∗∗P < .01; ∗∗∗P < .001; ∗∗∗∗P < .0001 by 1-way analysis of variance and Holm-Sidak post hoc test.

Mouse genetic studies have shown that MLL3 and MLL4 have distinct functions in the hematopoietic system that seem incongruent with their structural similarities. MLL3 restricts HSC self-renewal under conditions of inflammatory stress, proliferative stress, and serial transplantation by promoting differentiation into MPPs and myeloid progenitors.^17,19 In contrast, MLL4 sustains HSC self-renewal capacity by opposing myeloid differentiation.¹⁶ Furthermore, MLL3 suppresses myeloid leukemogenesis,^19,33 whereas MLL4 is required to sustain acute myeloid leukemia.¹⁶ It is not clear how HSCs and myeloid progenitors deploy MLL3 and MLL4 to distinct regulatory elements to enable them to selectively promote (MLL3) or suppress (MLL4) HSC differentiation. Furthermore, it remains unclear whether MLL3 and MLL4 also perform redundant, currently unidentified functions in HSCs beyond their established, antagonistic roles in myeloid commitment.

In this study, we conditionally deleted Mll3, Mll4, or both genes together to determine how each gene controls divergent HSC fate decisions. We found that the interactions between Mll3 and Mll4 are far more complex than a simple antagonism model would suggest, as the genes had unanticipated redundant roles in maintaining the multipotent state. In the absence of both Mll3 and Mll4, hematopoietic progenitors rapidly adopted a non–self-renewing, B-cell–like default identity, coupled with near-total loss of HSC and MPP superenhancer activity. MLL3 and MLL4 proteins did not require catalytically active SET domains or the histone demethylase activity of UTX to sustain the multipotent state. The data show that myeloid, erythroid, and lymphoid potentials are held in tension against one another within HSCs and other progenitors by distinct MLL3/4-dependent and MLL3/4-independent enhancer networks. Our data further illustrate how key functions of MLL3 and MLL4 can be decoupled from the enzymatic functions of the MLL3/4 complexes.

Methods

Mouse lines

All mouse lines have been described previously.^16,17,34,35 For all in vivo assays, cells were isolated, stained, and analyzed as described previously.^17,18 Mll3 and Mll4 alleles were deleted at the time points noted in the text and as described in greater detail in the supplemental Methods, available on the Blood website.

Flow cytometry

The cells were isolated, stained, analyzed, and transplanted as previously described.³⁶ Additional details can be found in the supplemental Methods.

CITE-seq and scATAC-seq library construction, sequencing, and analysis

Cellular indexing of transcripts and epitopes by sequencing (CITE-seq), 5′ single cell RNA-sequencing, single cell assay for transposon-accessible chromatin using sequencing (scATAC-seq), ChIPmentation, and CUT&RUN libraries were prepared as described previously^18,33,37 and analyzed using established computational pipelines.^{38, 39, 40, 41, 42, 43} Single-cell libraries were prepared using 10x Genomics kits according to the manufacturer instructions. Additional details can be found in the supplemental Methods.

Colony formation and B-cell potential assays

The myeloid colony forming potential was assessed by culturing single HSCs in MethoCult M3434 media (Stemcell Technologies). B-cell potential was measured by culturing cells, at the numbers indicated in the text, on OP9 stromal cells in media with FLT3-ligand and interleukin-7 (10 ng/mL each). Additional details can be found in the supplemental Methods.

All procedures were performed according to an institutional animal care and use committee–approved protocol at the Washington University School of Medicine.

Results

MLL3 and MLL4 act redundantly to enable myeloid potential and sustain hematopoiesis

The previously described, opposing phenotypes of Mll3- and Mll4-deficient HSCs raised the question of whether the encoded proteins interact within a common pathway, such that one gene is epistatic to the other, or whether they have additional redundant functions. To test these possibilities, we competitively transplanted 300 000 Mx1-Cre⁻ (control), Mll3^f/f; Mx1-Cre (Mll3^Δ/Δ), Mll4^f/f; Mx1-Cre (Mll4^Δ/Δ), or Mll3^f/f; Mll4^f/f; Mx1-Cre (Mll3/4^Δ/Δ) bone marrow cells (CD45.2⁺), along with 300 000 wildtype CD45.1⁺ competitor cells, into lethally irradiated CD45.1⁺ recipient mice (supplemental Figure 1A). We confirmed engraftment at 4 weeks after the transplant (supplemental Figure 1B), followed by administration of poly-inosine:poly-cytosine (pIpC) to delete the floxed alleles. This approach enabled us to delete Mll3/4 exclusively in hematopoietic cells without the confounding effects of systemic deletion. To assess the repopulating activity, we transplanted 3 million bone marrow cells per recipient from the conditionally deleted primary recipient mice into lethally irradiated secondary recipients. Mll3^Δ/Δ marrow engrafted the secondary recipient mice more efficiently than control marrow, whereas Mll4^Δ/Δ and Mll3/4^Δ/Δ marrow failed to engraft altogether (Figure 1C-D; supplemental Figure 1C). These phenotypes align with previous studies,^16,17,19 and the compound mutant engraftment pattern suggests that Mll4 may be epistatic to Mll3 rather than redundant. However, to confirm that the Mll3 and Mll4 alleles were truly deleted, we genotyped colonies from individual HSCs. Strikingly, none of the Mll3/4^Δ/Δ HSC-derived colonies had homozygous deletions of both Mll3 and Mll4 (Figure 1E-F). Thus, Mll3 and Mll4 act redundantly to enable myeloid colony formation from HSCs.

Based on these observations, we tested whether Mll3 and Mll4 are necessary to sustain hematopoiesis over time. We noncompetitively transplanted bone marrow from Mll3/4^Δ/Δ mice (before pIpC treatment), administered pIpC 6 weeks after transplantation, and analyzed the stem and progenitor cell frequencies 2 weeks later (supplemental Figure 1D-E). Mll3/4^Δ/Δ recipients had marked expansion of phenotypic HSCs, MPPs, and committed progenitor (Lineage⁻Kit⁺Sca1⁻) populations relative to control recipients (Figure 1G-I; supplemental Figure 1F-G). The degree of MPP and committed progenitor expansion was far greater in Mll3/4^Δ/Δ recipients than in single mutant recipients (Figure 1G-I). We followed a separate cohort for up to 30 weeks post-pIpC administration, at which point several mice in the Mll3/4^Δ/Δ recipient cohort became moribund. These mice had pancytopenia with absolute reductions in all bone marrow lineages and hypocellular morphology (Figure 1J; supplemental Figure 2). While Mll4^Δ/Δ recipients also had hypocellular marrow (supplemental Figure 2E-I), they did not have aberrant expansion of HSC, MPP, or committed progenitor populations, suggesting that cytopenias in these mice reflect progenitor attrition rather than a differentiation block. Altogether, the changes in the phenotypic progenitor numbers observed in Mll3/4^Δ/Δ recipients suggest redundant roles for MLL3 and MLL4 in HSC/MPP fate specification, in addition to their nonredundant, antagonistic functions.

MLL3 and MLL4 maintain multilineage hematopoiesis by opposing a B-cell–like default state

To better evaluate the molecular changes that ensue following compound Mll3/4 deletion, we performed CITE-seq on donor Lineage⁻Kit⁺Sca1⁺ (LSK) cells, as well as on Kit⁺ cells isolated from primary recipient bone marrow 2 weeks after pIpC treatment (supplemental Figure 1D; supplemental Table 1). Iterative Clustering with Guide-gene Selection identified 20 transcriptionally distinct cell clusters across all genotypes (Figure 2A-B; supplemental Figure 3A). In some cases, clusters from adjacent clades of the Iterative Clustering with Guide-gene Selection heat map were combined as superclusters (called HSC, MPP4, MLL4 ko-1, MLL4 ko-2, and DKO) to simplify description of the data (Figure 2A-B; supplemental Figure 3A). We identified populations within each sample group with previously described HSC/MPP immunophenotypes (Figure 2C; supplemental Figure 3B-C).⁵ Control and Mll3^Δ/Δ LSK cells had similar cluster distributions with modest expansion of the HSC supercluster in Mll3^Δ/Δ LSK (Figure 2D). Mll4^Δ/Δ and Mll3/4^Δ/Δ LSK cells had very different cluster distributions than controls (Figure 2B-C). Mll4^Δ/Δ LSK cells populated 2 superclusters (MLL4 ko-1 and 2) and ectopically expressed myeloid genes, including Elane, Mpo, and Il1rl1 (Figure 2E; supplemental Figure 3D). Mll3/4^Δ/Δ LSK cells populated a distinct supercluster (DKO) and ectopically expressed B-cell genes, including Il7r, Dntt, Pax5, and Ebf1 (Figure 2F; supplemental Figure 3E; supplemental Table 2). Thus, Mll3 limits the size of the HSC pool, Mll4 prevents precocious myeloid gene expression, and Mll3 and Mll4 act redundantly to prevent conversion of HSC/MPPs to a B-cell–like state.

Figure 2.

Compound Mll3/4 deletion converts hematopoietic progenitors to a B-cell–like state. (A-B) UMAP representing donor LSK single-cell transcriptomes, colored by cluster (A) or genotype (B). The 5 superclusters are outlined in black. (C) Distribution of HSC, MPP2, MPP3, and MPP4 immunophenotypes by genotype. (D) HSC, MPP2, MPP3, and MPP4 immunophenotype frequencies in each supercluster. (E-F) The expression of myeloid and lymphoid marker genes in single cells. (G-H) UMAPs representing donor Kit⁺ single-cell transcriptomes, colored by cluster (G) or genotype (H). The HSC/MPP, GMP, myeloid, lymphoid, erythroid, and megakaryocytic (Mk) annotations are indicated. (I) Myeloid identity quadratic programming (QP) scores for the HSC/MPP, GMP, and myeloid superclusters combined (left) and the HSC/MPP cluster alone (right). ∗∗∗∗P < .0001 for Mll4^Δ/Δ relative to control or Mll3^Δ/Δ by pairwise Wilcoxon test. (J) Myeloid, erythroid, and Mk marker gene expression by genotype. (K) B-cell GSVA enrichment scores by genotype. ∗∗∗∗P < .0001 in Mll3/4^Δ/Δ (higher) relative to all other genotypes; ∗P = .02 in Mll3^Δ/Δ (lower) relative to control. UMAP, uniform manifold approximation and projection.

We subsequently performed CITE-seq on donor Kit⁺ cells to capture the effects of Mll3 and Mll4 deletions on more differentiated progenitors. We annotated clusters that contained HSC/MPP, granulocyte-monocyte progenitor (GMP), myeloid, lymphoid, erythroid, and megakaryocytic cells based on gene and surface marker expression (Figure 2G-H; supplemental Figure 4A-B). Mll3 deletion had only modest effects on the cluster distributions (supplemental Figure 4C). Mll4 deletion increased myeloid cluster representation and reduced HSC/MPP and GMP cluster representation (supplemental Figure 4C). To more precisely test for enhanced myeloid bias in Mll4^Δ/Δ progenitors, we used quadratic programming to calculate myeloid identity scores for each cell within the HSC/MPP, GMP, and myeloid clusters. Scores ranging from 0 to 1 were assigned to each cell on the myeloid trajectory based on expression of genes that distinguish HSCs from committed myeloid cells.^33,40,44 Mll4 deletion significantly increased myeloid identity scores across all myeloid progenitor populations, including HSC/MPPs (Figure 2I; supplemental Figure 4D), indicating precocious myeloid differentiation. In contrast, Mll3/4^Δ/Δ cells clustered separately from all other cells (DKO cluster) and expressed B-cell genes, including Pax5 and Ebf1 (supplemental Figure 4E). The cells also expressed CD127, and most had a common lymphoid progenitor surface marker phenotype (supplemental Figure 4A-B). There was no evidence of myeloid, erythroid, or megakaryocyte gene expression in Mll3/4^Δ/Δ cells (Figure 2J). Instead, gene set variation analysis (GSVA) confirmed strong enrichment of a bone marrow B-cell signature within most Mll3/4^Δ/Δ cells (Figure 2K; supplemental Table 3). Similar changes in progenitor numbers, function, and gene expression were observed when Ubc-CreER was used to delete Mll3 and Mll4, indicating that they were not caused by pIpC (supplemental Figure 5A-I). Deleting Mll3 and Mll4 in committed myeloid progenitors with LysM-Cre did not impair terminal myelopoiesis or cause B-cell reprogramming (supplemental Figure 5J-M). Thus, once myeloid commitment is fully established, Mll3 and Mll4 become dispensable for later stages of myeloid differentiation. These data show that MLL3 and MLL4 act redundantly within HSC/MPPs to maintain myeloid, erythroid, and megakaryocytic potential, and loss of these proteins causes HSCs and MPPs, as well as lineage-negative myeloid, erythroid, and megakaryocytic progenitors, to convert to a B-cell–like default state.

Mll3/4-deficient progenitors do not resemble normal B-cell progenitors

We sought to characterize the B-cell–like, Mll3/4-deficient progenitor population in greater detail with the goal of understanding whether the population aligns with a stage of normal B lymphopoiesis or whether it reflects an emergent cell state that lacks a clear analog in normal hematopoiesis. We tested whether Mll3/4^Δ/Δ progenitors have B-cell potential by plating control, Mll3^Δ/Δ, Mll4^Δ/Δ, and Mll3/4^Δ/Δ LSK cells at limiting dilutions and culturing in conditions that support B-cell production. Mll3 deletion enhanced B-cell colony formation, Mll4 deletion impaired colony formation, and, despite B-cell priming, Mll3/4^Δ/Δ LSK cells lacked B-cell colony forming potential altogether (Figure 3A-B). Furthermore, Mll3/4^Δ/Δ mice had reduced donor pro–B, pre–B, and B220^hi recirculating B-cell numbers in the bone marrow (Figure 3C; supplemental Figure 6A), indicating B-cell maturation defects.

Figure 3.

Mll3/4 deficient HSC/MPPs lack mature B-cell potential and undergo immunoglobulin heavy chain, but not light chain, rearrangement. (A) Limiting dilution analysis of B-cell colony forming potential of donor LSK cells with the indicated genotypes (n = 3). (B) The percentage of wells with B-cell colonies when 10 donor cells per well were plated. ∗P < .05; ∗∗∗∗P < .0001 relative to control, as determined using extreme limiting dilution analysis. (C) Donor pro–B, pre–B, IgM⁺, and B220^hi recirculating B cells in the bone marrow of control or Mll3/4^Δ/Δ recipient mice (n = 7-16). (D) Distribution of the single-cell transcriptomes for the control and Mll3/4^Δ/Δ genotypes. Donor control B220^lo and Kit⁺ cells served as positive and negative controls for immunoglobulin rearrangement, respectively. (E) The distributions of cells with heavy chain rearrangements only (left, IgH) or heavy and light chain rearrangements (right, IgH + Igκ/λ). Distinct clusters of control B220^lo cells and Mll3/4^Δ/Δ Kit⁺ cells are circled. (F) The percentage of cells with IgH only or both IgH + Igκ/λ rearrangement by cell type. (G) IgM expression in the indicated cell populations and genotypes. NUP98 served as a loading control. (H) Volcano plot showing differentially enriched Hallmark gene sets between IgH-rearranged control B220⁺ and Mll3/4^Δ/Δ Kit⁺ cells. (I) The calculated cell cycle phases for IgH-rearranged cells. ∗∗∗∗P < .0001 as determined by the χ² test. (J) Bone marrow pre-B cells in mice of the indicated genotypes following deletion with Mb1-Cre (n = 6-10). (K) Serum IgG levels in the MBI-Cre conditional knockout mice (n = 3-4). The error bars reflect standard deviation. ∗P < .05; ∗∗P < .01; ∗∗∗P < .001; ∗∗∗∗P < .0001 as determined by 1-way analysis of variance, followed by the Holm-Sidak post hoc test. UMAP, uniform manifold approximation and projection.

We next used single-cell RNA-sequencing to test whether Mll3/4^Δ/Δ cells are capable of rearranging immunoglobulin loci. For these assays, we included control and Mll3/4^Δ/Δ Kit⁺ cells, as well as Cre-negative B220^lo bone marrow cells. Mll3/4^Δ/Δ cells clustered separately from both control Kit⁺ and B220^lo lymphoid progenitor cells (Figure 3D). They underwent immunoglobulin heavy chain (IgH) rearrangement without further light chain rearrangement, in contrast with control B220^lo cells (Figure 3E-F). Western blotting for IgM showed low levels of protein expression in the Mll3/4^Δ/Δ cells (Figure 3G), likely reflecting IgH but not mature IgM expression. Gene set enrichment analysis with pseudobulk populations revealed several differences between the control and Mll3/4^Δ/Δ IgH-rearranged cells, including evidence of reduced proliferation within the Mll3/4^Δ/Δ population (Figure 3H-I). Altogether, the data showed that the aberrant, B-cell–like default state of Mll3/4^Δ/Δ progenitors lacks a clear analog within the normal B-cell differentiation trajectory.

We subsequently tested whether deleting Mll3 and Mll4 in pro–B cells, using Mb1-Cre, would impede B-cell maturation. Mll4 deletion significantly reduced pro–B, pre–B, and B220^hi recirculating B-cell numbers in the bone marrow, as well as follicular, marginal zone, and transitional B-cell populations in the spleen (Figure 3J; supplemental Figure 6B-I). Mll4 deficiency also reduced serum IgG levels, consistent with the loss of functional mature B-cells (Figure 3K). The data illustrated that MLL3 and MLL4 have stage-specific roles during B-cell development. B-cell priming in HSC/MPPs occurs independently of MLL3/4, whereas B-cell maturation requires MLL3/4. Thus, MLL3/4-deficient progenitors default to a B-cell–like state, but B-cell production is still compromised in colony forming assays and in vivo.

MLL4 maintains the HSC/MPP enhancer landscape by preventing precocious activation of myeloid enhancers

To understand how MLL3 and MLL4 regulate enhancer activity to maintain HSC/MPP multipotency, restrict myeloid differentiation, and prevent ectopic B-cell priming, we performed single-cell ATAC-seq (scATAC-seq) on control, Mll3^Δ/Δ, Mll4^Δ/Δ, and Mll3/4^Δ/Δ LSK cells (supplemental Figure 1D). We clustered cells in ArchR³⁹ and used integrated CITE-seq data to annotate the clusters based on inferred transcript expression (Figure 4A-C). We identified HSC, MPP4, MLL4 ko-1, MLL4 ko-2, and DKO superclusters, analogous to the superclusters defined in Figure 2A and supplemental Figure 3A, based on integrated gene expression (supplemental Figure 7A-B). As anticipated, control and Mll3^Δ/Δ cells had very similar cluster distributions (Figure 4A; supplemental Figure 7B). In contrast, Mll4^Δ/Δ cells occupied the MLL4 ko-1 and MLL4 ko-2 superclusters, and Mll3/4^Δ/Δ cells occupied the DKO cluster almost exclusively (supplemental Figure 7B-C). These scATAC-seq profiles recapitulated the cluster distributions observed with CITE-seq.

Figure 4.

Mll4 deletion causes premature activation of GMP enhancers in HSC/MPPs. (A-B) scATAC-seq profiles of donor LSK cells, by genotype or ArchR clusters, with supercluster annotation based on integrated CITE-seq data. (C) Accessibility and integrated expression of representative HSC, MPP, myeloid, and B-lymphoid marker genes. (D) Distribution of scATAC-seq peaks relative to transcriptional start sites by supercluster. (E) Left: supercluster-specific aggregate scATAC-seq signals by genotype. Middle and right: H3K27ac and H3K4me1 signals at MLL4-ko-1 and MLL4-ko-2-specific elements for LSK or GMP. (F) ChromVAR deviation scores for FOS, PU.1, and CEBPA motifs. For all violin plots, ∗P < .05; ∗∗P < 10⁻¹⁰; ∗∗∗P < 10⁻¹⁰⁰ as determined by Wilcoxon signed rank test relative to control.

Cluster-specific, differentially accessible regions mapped primarily to distal enhancer elements. Approximately 80% to 90% of the elements were located in intragenic or intronic regions (Figure 4D; supplemental Table 4). HSC and MPP4 cluster-specific elements overlapped with the previously curated H3K4me1 and H3K27ac peaks from sorted HSCs and MPPs (supplemental Figure 7D).¹⁷ These elements therefore reflect active HSC/MPP enhancers. Mll3 deletions did not alter the H3K4me1 or H3K27ac levels at HSC- or MPP4-specific enhancers (supplemental Figure 7E), consistent with a previous study that showed that MLL3 regulates H3K4me1 and H3K27ac at only a small number of target enhancers that are activated by inflammatory and proliferative stress.¹⁷ Mll4^Δ/Δ LSK enhancers had H3K4me1 and H3K27ac patterns that more closely resembled GMPs than HSC/MPPs, even at steady state (Figure 4E). These enhancers were strongly enriched in FOS, PU.1, and CEBPA motifs (Figure 4F; supplemental Figure 7F-G; supplemental Table 5). Ectopically accessible regions mapped near myeloid genes, such as Mpo, Itgam/Cd11b, Csf1r, Elane, and Ctsg, consistent with precocious myeloid differentiation (Figure 4C; supplemental Figure 7H). Thus, MLL4 maintains the HSC/MPP enhancer landscape and prevents premature activation of GMP-associated enhancers.

MLL3 and MLL4 act redundantly to maintain HSC/MPP enhancers while suppressing the activity of B-cell regulatory elements

As observed in the CITE-seq studies, Mll3/4^Δ/Δ LSK cells clustered separately from all normal HSCs and MPPs based on their scATAC-seq profiles (DKO supercluster; Figure 4A-B). We confirmed that cells in the DKO supercluster had increased accessibility of B-cell–associated genes by GSVA (supplemental Figure 8A) and normal expression of other COMPASS proteins by western blot (supplemental Figure 8B). DKO-specific elements were enriched in EBF1 and PAX5 motifs (Figure 5A-B; supplemental Figure 8C-D), consistent with B-cell differentiation.⁴⁵ Analysis of previously described pro–B-cell ChIP-seq data confirmed binding of EBF1⁴⁶ and PAX5⁴⁷ to DKO-specific enhancers (Figure 5C). Motifs for critical effectors of HSC and myeloid identity (eg, NFIX, RUNX1 and GATA2) were negatively enriched in DKO-specific enhancers (Figure 5D; supplemental Figure 8C). We performed ChIPmentation on wild-type and Mll3/4^Δ/Δ LSK cells to evaluate H3K27ac (a mark of active enhancers and promoters) at HSC/MPP cis-regulatory elements. Mll3/4 deletion did not alter the total H3K27ac levels in LSK cells as determined by western blot (supplemental Figure 8E-F), but it dramatically reduced H3K27ac at a previously curated set of enhancers that are normally active in HSC/MPPs (Figure 5E).¹⁷ This reduction was offset by a gain of H3K27ac at DKO-specific enhancers (Figure 5F). In addition, the promoters and gene bodies of several B-cell regulators, including Ebf1 and Pax5, were hyperacetylated in Mll3/4^Δ/Δ LSK cells (Figure 5G-H). Thus, deletion of Mll3/4 causes dramatic realignment of HSC/MPP enhancer activities, such that enhancers that maintain stemness and myeloid identity are inactivated while enhancers that promote B-cell differentiation are ectopically activated.

Figure 5.

Loss of MLL3 and MLL4 inactivates HSC/MPP enhancer networks and ectopically activates B-cell enhancers. (A) DKO cluster-specific aggregate scATAC-seq signals by genotype. (B) ChromVAR deviation for EBF1 and PAX5 motif accessibility. (C) EBF1 and PAX5 binding in pro–B cells at cluster-specific enhancers. (D) ChromVAR deviation for RUNX1 and GATA2 showing reduced accessibility in Mll3/4^Δ/Δ LSK. ∗P < .05; ∗∗P < 10⁻¹⁰; ∗∗∗P < 10⁻¹⁰⁰; as determined by Wilcoxon tests. (E-F) H3K27ac signal at HSC/MPP and DKO-specific enhancers in control and Mll3/4^Δ/Δ LSK cells. (G) Volcano plot showing differential H3K27ac at DKO-specific enhancers in control and Mll3/4^Δ/Δ LSK cells. (H) H3K27ac tracks at Ebf1 and Pax5 loci. Red arrows indicate regions of significantly increased H3K27ac in Mll3/4^Δ/Δ LSK cells. (I) Rank Ordering of Super Enhancers (ROSE) analysis of HSC and MPP enhancers based on H3K27ac signals. (J) Aggregate H3K27ac levels at HSC/MPP superenhancers in control and Mll3/4^Δ/Δ LSK cells. (K) Tracks showing reduced accessibility and H3K27ac at Gata2 in Mll3/4^Δ/Δ LSK cells. Peaks to gene (P2G) indicates the association between peaks and the promoter. (L) ROSE analysis to identify MLL3/4-independent superenhancers in Mll3/4^Δ/Δ cells. (M) Aggregate H3K27ac levels at previously described pro–B-cell superenhancers in control and Mll3/4^Δ/Δ LSK cells. (N) Tracks showing increased accessibility and H3K27ac at Foxo1 in Mll3/4^Δ/Δ LSK cells. The red box indicates MLL3/4-independent superenhancers.

The transition from HSC/MPP to B-cell regulatory programs was even more pronounced when we evaluated superenhancers. Superenhancers encompass large genomic regions and are notable for very high H3K27ac levels. We identified HSC and MPP superenhancers by performing Rank Ordering of Super Enhancers⁴² and then evaluated the H3K27ac levels in Mll3/4^Δ/Δ LSK cells (Figure 5I; supplemental Table 6).¹⁷ Mll3/4 deletion dramatically reduced H3K27ac in these regions, including at superenhancers associated with critical HSC genes, such as Myct1, Gata2, Mecom, Cd34, and Hlf (Figure 5J-K; supplemental Figure 8G-H). These reductions in superenhancer activity were accompanied by the activation of pro–B-cell associated superenhancers in Mll3/4^Δ/Δ LSK cells,⁴² including at genes such as Ebf1, Foxo1, Rag1, Dntt, and Il7r (Figure 5L-N; supplemental Figure 8I-J). Thus, MLL3 and MLL4 are necessary to maintain normal HSC/MPP superenhancer activity and suppress B-cell superenhancer activity.

MLL3/4 directly bind enhancers near HSC/MPP transcription factor genes but not B-cell–specific genes

Single-cell transcript expression and chromatin profiling data predicted that MLL3 and MLL4 directly and redundantly regulate the expression of transcription factors that maintain HSC self-renewal and multipotency. To identify and characterize MLL3/4-bound regulatory elements, we performed CUT&RUN using antibodies directed at MLL3 or MLL4, as well as ChIPmentation to detect 3xFLAG-tagged MLL3 (Figure 6A; supplemental Figure 9A). CUT&RUN identified far more MLL3 bound regions than ChIPmentation, suggesting greater sensitivity (Figure 6B). Most MLL3/4-bound elements mapped to distal intragenic or intronic regions, and ∼75% of all overlapping peaks aligned with previously described HSC/MPP enhancers,¹⁷ GMP enhancers,¹⁸ or both (Figure 6C-D). MLL3/4-bound regions overlapped with UTX binding sites,⁴⁸ as expected for a COMPASS protein (Figure 6E), and were enriched in ETS, RUNX, and STAT5 binding motifs (supplemental Figure 9B). Deleting Mll3 in HSCs, MPPs, and GMPs had only modest effects on H3K27ac at MLL3/4-bound regions (Figure 6F; supplemental Figure 9C). In contrast, deleting Mll3 and Mll4 together reduced the H3K27ac levels nearly to background levels with a greater than fourfold reduction in the aggregate peak height (Figure 6F). These changes demonstrate redundancy between MLL3 and MLL4.

Figure 6.

MLL3 and MLL4 directly and redundantly regulate a subset of HSC/MPP transcription factors. (A) Top: overview of the CUT&RUN and ChIPmentation strategies as orthogonal methods to study MLL3/4 chromatin localization. Bottom: schematic of the 3xFLAG-MLL3 insertion and location of the inactivating exon 14 mutations in 3xFLAG-MLL3 and 3xFLAG-MLL3-null 32D cells, respectively. (B) Venn diagram showing the overlapping peaks among the MLL3 CUT&RUN, MLL4 CUT&RUN, and FLAG-MLL3 ChIPmentation data sets. (C) Distribution of the MLL3/4-bound peaks (the intersection in panel B) relative to gene bodies. (D) Association of MLL3/4-bound peaks with enhancers that are either primed (H3K4me1) or active (H3K4me1 and H3K27ac) in HSC/MPPs, GMPs, or both. (E) Tornado plots showing UTX binding in Lineage⁻ bone marrow at MLL3/4-bound enhancers. The FLAG-MLL3 signal is shown to the left. (F) Top: H3K4me1 at MLL3/4-bound enhancers in control and Mll3^Δ/Δ HSCs and MPPs. Bottom: H3K27ac at MLL3/4-bound enhancers in control and Mll3^Δ/Δ HSCs and control or Mll3/4^Δ/Δ LSK cells. (G) A schematic of the MLL3 and MLL4 target genes encoding key HSC/MPP transcription factors or epigenetic regulators. (H) Representative tracks showing MLL3/4 CUT&RUN, FLAG-MLL3 ChIPmentation, and H3K27ac at the Runx1 locus. Red boxes indicate MLL3/4-bound enhancers. (I) Tornado plots showing the MLL3 and MLL4 CUT&RUN signals at MLL3-specific peaks based on the MACS2 peaking calling algorithm. (J) Transcription factor binding motifs enriched in MLL3-specific peaks, based on HOMER analysis. (K) Tornado plots showing the MLL3 and MLL4 CUT&RUN signals at MLL4-specific peaks based on the MACS2 peaking calling algorithm. (L) Transcription factor binding motifs enriched in MLL4-specific peaks, based on HOMER analysis.

Many MLL3/4-bound elements mapped to near genes that either sustain HSC self-renewal or control multilineage differentiation, including Runx1, Cited2, Cux1, Zeb2, and Erg (Figure 6G-H; supplemental Figure 9D; supplemental Table 7). Some cis-regulatory elements, such as the Runx1 +24 kb and Erg +85 kb enhancers, have been characterized previously.^{49, 50, 51} MLL3/4-bound enhancers did not map near genes that prime B-cell identity, such as Ebf1 or Pax5, nor did they map near other B-cell identity genes (eg, Cd19, Cd79a, Dntt, etc). We observed minimal overlap between MLL3/4-bound enhancers and DKO cluster-specific regulatory elements (8 of the 1013 total elements). Altogether, the data imply that MLL3 and MLL4 directly and redundantly regulate the expression of transcription factors that sustain HSC/MPP multipotency. In their absence, MLL3/4-independent, B-cell–associated elements dominate the cis-regulatory landscape.

CUT&RUN also identified regions that were called distinctly as MLL3- or MLL4-bound peaks (Figure 6B). Most MLL3-enriched elements showed some degree of MLL4 binding and vice versa (Figure 6I,K). Called MLL3 peaks were enriched in IRF and Forkhead family motifs. Called MLL4 peaks were enriched in NRF and homeobox family motifs (Figure 6J,L). Specific transcription factors may therefore differentially recruit MLL3 and MLL4, but the differences seem to be quantitative rather than absolute. There are few truly MLL3- or MLL4-specific regulatory elements.

MLL3 and MLL4 sustain multilineage hematopoiesis independently of COMPASS catalytic activities

We tested whether MLL3 and MLL4 require histone methyltransferase activity, mediated by the SET domains in each protein, to sustain HSC/MPP multipotency. We crossed previously described Mll3^Y4792A and Mll4^Y5477A mutant alleles³⁴ with Mll3^f/f; Mll4^f/f; Mx1-Cre mice to generate Cre-negative (control), Mll3^Y4792A/f; Mx1-Cre (Mll3^SI), Mll4^Y5477A/f; Mx1-Cre (Mll4^SI), and Mll3^Y4792A/f; Mll4^Y5477A/f; Mx1-Cre (Mll3/4^SI) mice. The Mll3^Y4792A and Mll4^Y5477A mutations abolish H3K4me1 placement while preserving normal protein expression and binding interactions.²⁴ We transplanted bone marrow from control, Mll3^SI, Mll4^SI, and Mll3/4^SI mice and treated recipients with pIpC to delete the floxed alleles (Figure 7A-B). After pIpC treatment, the hematopoietic cells retained single copies of the SET-inactive alleles. SET-inactive LSK cells had reduced H3K4me1 levels that resembled complete MLL3/4 loss of function (Figure 7C). Phenotypic MPP and myeloid progenitor populations expanded modestly in Mll3/4^SI recipients, but the magnitude was lower than was observed in Mll3/4^Δ/Δ recipients (Figure 7D-F). The Mll3/4^SI HSCs had reduced but not absent myeloid colony forming potential, in contrast with Mll3/4^Δ/Δ LSK cells (Figure 7G). Thus, myeloid colony formation does not absolutely require MLL3/4 SET activity.

Figure 7.

MLL3 and MLL4 do not require histone methyltransferase activity to sustain HSC/MPP multipotency, prevent premature myeloid differentiation, or prevent conversion to a B-cell–like default state. (A) Schematics indicating the MLL3 and MLL4 SET inactivating (SI) mutations. (B) Overview of the noncompetitive transplantation strategy to assess dependence on MLL3/4 histone methyltransferase activity. (C) Western blots showing H3K4me1 in LSK cells of the indicated genotypes. The SET-inactivating alleles greatly diminished the total H3K4me1. (D-F) Donor HSC, MPP, and Lineage⁻Kit⁺Sca1⁻ numbers in recipient bone marrow 2 weeks after pIpC treatment (n = 5). The Mll3/4^Δ/Δ cell numbers from Figure 1 were included for comparison. The error bars reflect the standard deviation. ∗P < .05; ∗∗P < .01; ∗∗∗P < .001, as determined by 1-way analysis of variance and the Holm-Sidak post hoc test; ^##P < .01 for comparison between Mll3/4^Δ/Δ and Mll3/4^SI by the 2-tailed Student t test. (G) The percentage of donor HSCs with myeloid CFU potential (left) and percentage of myeloid colonies with genotypes that are expected following complete deletion of floxed alleles (right) (n = 3). (H) Distribution of single-cell transcriptomes of Mll3/4^SI Kit⁺ cells projected against Mll3/4^Δ/Δ Kit⁺ cells from Figure 2G by Symphony. Cells are colored by genotype (top) or reference cluster identity (bottom) as in Figure 2G. (I) The percentage of cells for each genotype that map to HSC/MPP, GMP, myeloid, and DKO superclusters by Symphony (n = 2). (J-K) Myeloid identity QP scores for HSC/MPP, GMP, and myeloid clusters combined (J) or the HSC/MPP cluster alone (K). (L) B-cell signature GSVA enrichment scores by genotypes. (M) The distribution of single-cell transcriptomes of control and Utx-deficient Kit⁺ cells projected against Mll3/4^Δ/Δ Kit⁺ cells from Figure 2G by Symphony. Cells are colored by genotype (top) or reference cluster identity (bottom) as in Figure 2G. (N) The percentage of cells for each Utx genotype that map to HSC/MPP, GMP, myeloid, and DKO superclusters by Symphony (n = 2). (O) B-cell signature GSVA enrichment scores by Utx genotypes. For panels J-L,O, the myeloid identity and B-cell enrichment scores did not significantly differ by genotype.

We performed CITE-seq on control, Mll3^SI, Mll4^SI, and Mll3/4^SI Kit⁺ cells, as in Figure 2G, to test whether MLL4 SET inactivation causes precocious myeloid differentiation and whether MLL3/4 SET inactivation drives cells toward a B-cell–like state, as was observed with complete loss-of-function. For these comparisons, we used Symphony⁴¹ to assign SET-inactive cells to the same clusters as were defined using the complete loss-of-function data set in Figure 2G, with the data from Figure 2G serving as the reference. We then performed quadratic programming and GSVA to assess myeloid differentiation and B-cell priming, respectively. In contrast with Mll4^Δ/Δ cells, Mll4^SI cells were not enriched in the myeloid supercluster and did not show evidence of precocious myeloid differentiation based on quadratic programming (Figure 7H-K). Likewise, in contrast with Mll3/4^Δ/Δ cells, Mll3/4^SI cells did not associate with the DKO supercluster or show evidence of ectopic B-cell gene expression based on GSVA analysis (Figure 7H-I,L).

We next tested whether UTX is required to sustain HSC/MPP multipotency. UTX is a histone demethylase that removes repressive H3K27me3 marks. It is the other enzymatic subunit of the MLL3/4 COMPASS.^48,52 We generated female Utx^f/f, Mx1-Cre (Utx^Δ/Δ) mice,³⁵ transplanted bone marrow cells, deleted Utx, and performed CITE-seq, as described earlier. Utx deletion modestly reduced MPP numbers without altering HSC or myeloid progenitor numbers, and myeloid colony forming unit potential was unaffected (supplemental Figure 10). It did not cause Kit⁺ cells to default to a B-cell–like state, in stark contrast with the Mll3/4 deletion (Figure 7M-O). Thus, MLL3 and MLL4 maintain HSC/MPP multipotency independently of COMPASS catalytic activities, including SET-mediated histone methyltransferase activities and UTX-mediated histone demethylase activity.

Discussion

This study illuminates critical, redundant roles for MLL3 and MLL4 in maintaining HSC/MPP multipotency. Simultaneous inactivation of MLL3 and MLL4 leads to complete loss of myeloid-, erythroid-, and megakaryocyte-biased progenitors, loss of self-renewal capacity, and rapid conversion of essentially all Kit⁺ progenitors into a non–self-renewing, B-cell–like population that is transcriptionally and functionally distinct from any normal B-cell progenitor. At a molecular level, MLL3 and MLL4 directly and redundantly regulate several transcription factors and other epigenetic regulators that are critical for HSC maintenance, and they thus sustain much of the HSC/MPP enhancer and superenhancer network. MLL3 and MLL4 therefore serve as critical linchpins for multilineage hematopoiesis.

Our results raise the question of why HSCs, MPPs, and more committed progenitors all default to a B-cell–like state uniformly when MLL3 and MLL4 are inactivated. One possibility is that MLL3/4 target genes, such as Runx1, Gata2, and Mecom, directly or indirectly antagonize B-cell master regulators, such as Pax5 and Ebf1. Consistent with this interpretation, Gata2 has been shown to suppress the expression of several B-cell–associated genes, including Ebf1 and Pax5, in mice.^53,54 Furthermore, deleting Gata2b in zebrafish leads to B-cell priming but impaired B-cell maturation.⁵⁵ When Mll3/4-dependent enhancer and superenhancer networks collapse after conditional Mll3/4 deletion, a feed forward circuit involving B-cell transcription factors (eg, PAX5 and EBF1) could rapidly assert B-cell identity. In this scenario, HSC/MPP multipotency reflects constant tension between MLL3/4-dependent stemness programs and MLL3/4-independent B-cell programs.

Our data also raise 2 additional questions. First, why do MLL3 and MLL4 have distinct, nonredundant functions in HSCs given that they have similar domain structures and bind largely overlapping sets of enhancers? Second, how are B-cell–associated enhancers activated independently of MLL3/4? Answers to these questions might be found in the unstructured regions of MLL3, MLL4, and B-cell–specific transcription factors rather than in the structured domains. MLL4 has been shown to partition chromatin, recruit the Mediator complex, and form phase-separated transcriptional condensates via a glutamine-rich, prion-like domain.²⁷ This domain is not conserved in MLL3, but other COMPASS cofactors, including UTX, NCOA6, and PTIP, all contain prion-like domains.^25,26,56 The absence of a prion-like domain in MLL3 might account for its distinct functions from MLL4, whereas its association with UTX, NCOA6, or PTIP may preserve some ability to activate redundant enhancers in the absence of MLL4. Interestingly, the B-cell master regulator EBF1 also contains a prion-like domain that enables it to form condensates.⁵⁷ Loss of MLL3 and MLL4 could impair condensate formation at MLL3/4-bound regulatory elements while simultaneously enabling EBF1-mediated condensate formation at B-cell regulatory elements. Further studies are needed to evaluate these interactions in vivo.

Finally, this study reveals the degree to which the MLL3 and MLL4 functions can be isolated from COMPASS catalytic activities, including SET-mediated histone methyltransferase activities and UTX-mediated histone demethylase activity. These observations align with previous work that showed that MLL3 and MLL4 have SET-independent functions during embryogenesis,^34,58 as well as with studies that showed that other chromatin modifying enzymes, including UTX, MLL1, and EZH2, can sustain normal hematopoiesis even when their catalytic domains are disabled.^{59, 60, 61} SET-independent mechanisms of gene regulation may include transcriptional condensate formation, as noted earlier. In addition, MLL3 and MLL4 interact with ASXL1/2 to recruit the histone deubiquitinase BAP1 to chromatin.^29,62 Furthermore, MLL3 and MLL4 can antagonize MLL1/MENIN interactions at promoter elements, and MLL3/4 mutations convey MENIN inhibitor resistance in acute myeloid leukemia.⁶³ Finally, MLL3 and MLL4 can promote RNA polymerase pause release independently of SET activity.²⁴ In principle, any of these mechanisms could account for the SET-independent roles of MLL3/4 in maintaining HSC multipotency. Further studies are needed isolate the individual mechanisms.

Conflict-of-interest disclosure: G.A.C. reports performing consultations for and receiving research funding from Incyte, Ajax Therapeutics, and ReNAgade Therapeutics Management and is a cofounder/shareholder of Pairidex, Inc. The remaining authors declare no competing financial interests.