DNMT3A regulates murine megakaryocyte-biased hematopoietic stem cell fate decisions

Hematopoietic stem cells (HSCs) continuously replenish the differentiated cells of peripheral blood (PB). Recent studies established that HSCs are a heterogeneous population with different biases toward regeneration of specific lineages^1-7 (reviewed in Hass et al⁸). However, little is known about how these stem cell subtypes are differentially regulated. One well-defined subtype is biased toward the generation of megakaryocytes (Mks)/platelets.⁹ These Mk-biased (Mk^bi) HSCs give rise to non-Mk^bi HSCs and all other cell types, but in transplantation, they preferentially reconstitute the platelet lineage.^9,10 They are conserved between mice and humans,¹¹ increase with age,¹² and contribute to increased thrombotic risk in aged individuals.¹³ 

Mk^bi HSCs bypass certain progenitor states in the pathway to Mk and platelet production.¹⁴ This accelerated differentiation serves as a stress-response pathway to increase platelet counts in response to thrombocytopenia, inflammation, or infection.^15-17 Genetic tracing studies also indicate that the direct Mk differentiation pathway responds to niche-damaging stressors, such as chemotherapy, because it produces niche-supporting Mks to directly support HSCs.¹⁸ Mk bias is thought to be epigenetically encoded in part through silencing of lymphoid upstream regulatory elements,¹⁴ but specific mechanisms of Mk^bi HSC lineage fate programming remain to be defined.

DNA methyltransferase 3A (DNMT3A) is an epigenetic regulator of HSC lineage fate decisions. Loss of DNMT3A confers a self-renewal advantage upon HSCs and results in myeloid-biased differentiation,^19,20 ultimately leading to differentiation block after serial transplantation.²¹ DNMT3A loss also leads to erosion at the edges of DNA methylation canyons, large regions of low methylation associated with stemness genes.²² The resulting hypomethylation impairs the downregulation of HSC-identity genes and hinders progress toward differentiation. Additionally, the self-renewal advantage of Dnmt3a-mutant HSCs can be modulated by external factors, including inflammation.^23,24 However, despite its position as a key gatekeeper of HSC differentiation, the effects of DNMT3A loss have largely been examined in either bulk progenitors or the total long-term HSC (LT-HSC) population rather than in subtypes of lineage-biased or lineage-restricted LT-HSCs.

DNMT3A is one of the most commonly mutated genes in clonal hematopoiesis (CH), in which variant HSCs acquire a fitness advantage that leads to their clonal expansion over time,^25-27 which is associated with numerous adverse health outcomes.^25,26,28-32 Recent studies demonstrated that the epigenetic landscape primes hematopoietic cells for clonal dominance upon mutation acquisition, highlighting the importance of the cell’s epigenetic identity before mutation.³³ However, it is unknown whether acquiring a DNMT3A mutation in a lineage-biased or lineage-restricted stem cell would have different effects on the outcomes of CH in the downstream progeny or whether loss of DNMT3A simply reprograms any LT-HSC, altering its properties or fate trajectory.

In this work, we explored the effect of DNMT3A loss on murine Mk^bi HSCs and found functional deficits specific to this subpopulation that also alter the function and molecular phenotypes of downstream progeny. Additionally, we implicate DNMT3A as a regulator of the Mk differentiation pathway.

Detailed versions of all methods are available in the supplemental Material.

Mice were housed in Association for Assessment and Accreditation of Laboratory Animal Care-certified facilities at Baylor College of Medicine and approved by the Institutional Animal Care and Use Committee. The von Willebrand factor–green fluorescent protein (VWF-GFP) mouse⁹ was crossed with Dnmt3a^+/– mice³⁴ to obtain VWF-GFP/3aHET mice, maintained on a CD45.2 C57Bl/6 background. Littermates from multiple pooled litters were used in all experiments. CD45.2⁺ HSCs were sorted and combined with CD45.1⁺ whole bone marrow (WBM). Transplant donors were aged 6 to 16 weeks, and the ages of mice in all other experiments are stated in the text or figure legends. Recipients received a split dose of 11-Gy gamma irradiation and were retro-orbitally injected. Recipients with <0.5% CD45.2 engraftment were excluded from analysis.

WBM was collected from hindlimbs and pelvis. PB mononuclear cells were isolated using a heparin-dextran gradient and red blood cell lysis. Antibodies and markers are listed in the supplement. Samples were analyzed and sorted on a BD LSR II or FACSAria.

Mouse bone marrow was cultured with thrombopoietin and recombinant hirudin for 3 days. Mks were purified using a bovine serum albumin gradient. Proplatelet formation was visualized using time-lapse microscopy for 24 hours on a Leica TCS SP8 and quantified.³⁵ 

Mouse femurs were fixed in 4% paraformaldehyde, decalcified (0.2-M EDTA), and cryoprotected with sucrose. Frozen sections of 14 μm were obtained on the Cryostar Nx70 and stained with a-CD41 ± a-laminin. Imaging was performed on the Keyence BZ-X800 or Zeiss LSM900. Mks were manually counted across the length of the femur. Mk size was analyzed in ImageJ or Photoshop.

Blood was collected into acid citrate dextrose, and washed platelets were isolated by differential centrifugation and quantified on a complete blood count machine (Balio OV-360). For RNA sequencing (RNAseq), platelets were purified with a-Ter119 and a-CD45. For transmission electron microscopy, platelets³⁶ were fixed and suspended in agarose, then treated with 1% tannic acid, followed by 1% osmium tetroxide (aqueous), dehydrated in an acetone series, and embedded in Embed-812 resin. Sections of 100 nm were imaged on a Tecnai G2 Spirit BioTWIN EM. Platelet aggregation assays were performed³⁷ after treatment with indicated thrombin concentrations and staining with a-Ter119, a-CD41, and a-CD62P at room temperature for 20 minutes. For proteomics, the platelet pellet was lysed and digested, and the tryptic peptides were subjected to liquid chromatography with tandem mass spectrometry analysis using a nano-LC 1200 system coupled to Orbitrap Exploris 480.

One to 10 ng of DNA was used for the NEBNext Enzymatic Methyl-Seq kit. More than 200 ng of RNA was used for Tru-Seq messenger RNA kit. The Takara Bio SMARTer Stranded Total RNA kit (v3, Pico Mammalian Input) was used for low input RNAseq. Data analysis is described in the supplemental Material.

Striatal hemorrhage was induced by stereotactic injection of 250 nL of 0.1 U/μL type IV collagenase. Twenty-four hours later, total hemoglobin within each hemisphere (hemorrhage and nonhemorrhage) was determined at 560 nm after combining brain homogenate with Drabkin reagent.³⁸ 

Statistics were performed using GraphPad Prism and R studio. Results that reached statistical significance are indicated in figures as follows: ∗P ≤ .05; ∗∗P ≤ .01; ∗∗∗P ≤ .001; ∗∗∗∗P ≤ .0001.

We began by characterizing the effect of heterozygous DNMT3A loss using competitive WBM transplantation. To track platelets, we crossed germ line Dnmt3a^+/– mice (referred to throughout as 3aHET) with the VWF-GFP reporter mouse line, in which GFP is expressed under the VWF promoter, resulting in GFP-labeled platelets, Mks, and Mk^bi HSCs.⁹ We mixed WBM from VWF-GFP 3aHET donors with that from CD45.1 wild-type (WT) donors at a 20:80 ratio and transplanted it into lethally irradiated CD45.1 recipients (Figure 1A). We monitored donor chimerism as the percentage of CD45.2⁺ leukocytes or GFP⁺ platelets. Over the course of 20 weeks, there were no significant differences in PB donor chimerism (supplemental Figure 1A), donor chimerism in the B cell, T cell, or myeloid compartments in the bone marrow (supplemental Figure 1B), LT-HSC donor chimerism (supplemental Figure 1C), or the percentage of CD45.2⁺GFP⁺ (Mk^bi) LT-HSCs (supplemental Figure 1D).

Because serial transplantation has been shown to exacerbate DNMT3A phenotypes,^19,21 we pooled WBM from primary recipients and retransplanted it into secondary recipients (Figure 1A). There was no significant difference in overall PB donor chimerism (Figure 1B) or donor chimerism within B cells, T cells, myeloid cells, or platelets (Figure 1C). However, within the CD45.2⁺ progenitors (Lineage^neg, Sca1⁺ cKit⁺ [LSK]; see the supplemental Methods for definitions), the 3aHET group demonstrated significant skewing toward the LT-HSC compartment at the expense of the more differentiated multipotent progenitor-3/4 (MPP3/4) population³⁹ (Figure 1D-E). These progenitor-level changes did not translate into differences in the composition of differentiated bone marrow (Figure 1F). However, within the test LT-HSC population, the 3aHET cells exhibited a lower percentage of GFP⁺ (Mk^bi) HSCs (Figure 1G). When we quantified Mks based on GFP and CD41 staining in cryosectioned femurs of transplant recipients (supplemental Figure 1E), we found no difference in the percentage of GFP⁺ Mks (supplemental Figure 1F), Mk area, (supplemental Figure 1G), or the total number of Mks (supplemental Figure 1H). Additionally, there was no difference in platelet or total white blood cell counts (supplemental Figure 1I). Together, these data show that DNMT3A loss promotes progenitor skewing toward LT-HSCs but may have differential effects on Mk^bi vs non-Mk^bi HSCs.

To clarify the differential effects of DNMT3A loss on different subpopulations of HSCs, we transplanted GFP⁺ (Mk^bi) or GFP^– (non-Mk^bi) LT-HSCs from either WT or 3aHET VWF-GFP donors (Figure 2A; supplemental Figure 2A). Unexpectedly, the 3aHET Mk^bi HSCs failed to engraft in over half of the recipient mice (Figure 2B). In contrast, WT Mk^bi HSCs, WT non-Mk^bi HSCs, and 3aHET non-Mk^bi HSCs exhibited no engraftment deficiencies. These results were consistent across male and female sorted donor pools, as well as across multiple transplants performed months apart.

In engrafted recipient mice, the contribution of 3aHET Mk^bi HSCs to overall PB donor chimerism was significantly delayed (supplemental Figure 2B). The 3aHET Mk^bi condition exhibited delayed ability to reconstitute the B cell lineage compared with WT Mk^bi HSCs but robustly generated the myeloid/platelet lineages (Figure 2C). 3aHET Mk^bi HSCs were also delayed in their T cell reconstitution capacity but not more so than WT Mk^bi HSCs. Although it is well established that Mk^bi HSCs exhibit platelet/myeloid skewing,^9,10,14 this phenomenon was more pronounced in 3aHET Mk^bi HSCs (Figure 2C). The 3aHET Mk^bi HSC condition also exhibited high absolute platelet counts by complete blood count at the 24-week end point (supplemental Figure 2C). At this time, there were no differences in LSK or LT-HSC donor chimerism in Mk^bi recipients, suggesting that the delayed lymphoid reconstitution in the 3aHET Mk^bi condition was not due to a failure of HSCs to seed the bone marrow but rather due to functional deficits and delayed differentiation (supplemental Figure 2D).

We also assessed the composition of the progenitor compartment. Skewing toward LT-HSCs at the expense of MPP3/4 within the CD45.2⁺ LSK compartment occurred in recipients of 3aHET Mk^bi HSCs (Figure 2D). These results are consistent with findings in the competitive WBM transplant setting, demonstrating the distinct reliance of Mk^bi HSCs on DNMT3A. Mk^bi HSCs of both genotypes were better at reconstituting Mk^bi HSCs in recipients than their non-Mk^bi counterparts (Figure 2E; supplemental Figure 2E).

Finally, we examined Mk production. Although donor chimerism within the Mk populations was similar (supplemental Figure 2F), the total number of Mks per area was higher in the 3aHET non-Mk^bi condition than the WT non-Mk^bi or WT Mk^bi conditions (supplemental Figure 2G). The fact that the 3aHET Mk^bi HSC condition produces more absolute platelets (supplemental Figure 2C) without having a larger number of Mks suggests that Mks derived from these cells are more productive. In total, these data indicate that Mk^bi 3aHET HSCs exhibit engraftment deficiencies, delayed B cell reconstitution, and platelet bias without gross Mk expansion.

To better understand the altered hematopoietic reconstitution properties of Mk^bi 3aHET HSCs, we analyzed previously generated single-cell RNAseq data from WT or 3aHET LSK cells 10 months after WBM transplantation.⁴⁰ We identified Mk^bi HSCs as cells expressing Vwf transcripts (supplemental Figure 3A-B). As expected within the LSK compartment, these were mostly confined to the population annotated as LT-HSCs (Figure 3A; supplemental Figure 3A,C). Differential gene expression analysis between WT and 3aHET Vwf⁺ LT-HSCs suggested higher expression of genes that suppress cell motility in 3aHET HSCs, a feature of quiescence⁴¹ (Figure 3B; supplemental Table 1). Supportive of these findings, the Vwf⁺ 3aHET cells more closely matched an Mk-biased (Figure 3C) and low-output HSC signature (Figure 3D).⁵ These findings are consistent with our previous work demonstrating that Dnmt3a-knockout HSCs have difficulty exiting G0 and express lower levels of Ki67 in response to stimulation⁴² but indicate that this phenotype may be more prominent in mutant Vwf⁺ HSCs.

However, we noticed that the number of differentially expressed genes unique to Vwf⁺ 3aHET vs WT and not shared with Vwf- 3aHET vs WT was relatively small (supplemental Figure 3D-I), although these small differences led to very different pathway enrichment annotations between the 2 comparisons (Figure 3B; supplemental Figure 3J). Area under the curve scores for Mk-biased, low-output, and high-output signatures (supplemental Figure 3K-M), as well as the expression of key HSC marker genes (supplemental Figure 3C), gave us confidence that our Vwf expression cutoff was biologically meaningful. However, we recognized that the population might slightly differ from the one we sorted in our transplants. Therefore, we performed bulk RNAseq on sorted WT and 3aHET Mk^bi and non-Mk^bi HSCs. High expression of Vwf and Slamf1 (CD150) and low expression of Flt3 in the Mk^bi HSC samples validated the quality of our sorting strategy (Figure 4A). We compared Mk^bi to non-Mk^bi HSCs within the WT and 3aHET genotypes and then looked for differentially expressed genes that were unique to each comparison (Figure 4B-C; supplemental Table 2). Pathways related to cell motility and chemokine response were upregulated uniquely in Mk^bi 3aHET HSCs (Figure 4B), similar to the single-cell data. The bulk RNAseq data also revealed an enrichment of platelet-derived growth factor signaling unique to Mk^bi 3aHET HSCs (Figure 4B). Very few genes were uniquely downregulated in Mk^bi 3aHET HSCs (Figure 4C). Overall, transcriptional HSC identity in the steady state was influenced much more significantly by intrinsic HSC bias than by Dnmt3a genotype.

We wondered whether the observed changes were unique to the transplant context or whether phenotypic differences might be present in germ line 3aHET bone marrow. We sectioned mouse femurs and found that the number of Mks per area of bone marrow was significantly increased in 3aHETs (Figure 5A). Furthermore, 3aHET Mk morphology appeared qualitatively different from WT Mks, with abnormal borders that were similar in appearance to those described in mice fed a high-fat diet,³⁵ suggesting a possible change in membrane morphology (Figure 5B). However, unlike in the transplant context, there were only very minor differences in LSK composition, with slightly less contribution from 3aHET MPP3/4 (Figure 5C). In young animals, there were no differences in the proportion of Mk^bi LT-HSCs (Figure 5D). Although in aged WT animals, we observed the expected^11-13 increase in Mk^bi HSCs, this increase was significantly muted in 3aHET animals (Figure 5D). We assessed a different stressor by administering 5-fluorouracil (5-FU) or phosphate-buffered saline to WT and 3aHET mice and examined their LT-HSC compartment 7 days after treatment (Figure 5E). Although the Mk^bi HSC percentage increased in both genotypes after 5-FU treatment compared with phosphate-buffered saline, there was no difference between the genotypes, suggesting that 3aHET Mk^bi HSCs are equally capable of surviving and responding to this short-term stress as WT Mk^bi HSCs.

We next investigated the function of 3aHET platelets. We did not observe any differences in ex vivo activation or aggregation (supplemental Figure 4A-C). They had no visible or quantifiable (supplemental Figure 4D-E) morphological abnormalities when spread on fibrinogen-coated coverslips. Moreover, by transmission electron microscopy, we did not observe significant morphological differences (supplemental Figure 4F). Finally, in Mks differentiated ex vivo, there was no difference in proplatelet area or the percent of Mks forming proplatelets (supplemental Figure 4G-H). However, proteomics analysis and RNAseq of platelets from 3aHET mice indicated numerous molecular changes. Notably, in proteomics data, several structural components, including numerous members of the α-tubulin family, were downregulated in 3aHET compared with WT (Figure 5F). In RNAseq data, metabolic genes were downregulated, as well as components of the coagulation and complement cascades (Figure 5G). Because patients with DNMT3A-mutant CH are at increased risk of hemorrhagic stroke,²⁸ we transplanted WT and 3aHET WBM into irradiated WT recipients (to avoid potential effects of Dnmt3a loss in the vasculature of germ line 3aHETs) and induced hemorrhagic strokes via collagenase injection. However, there were no differences in intracerebral hematoma volume between mice with WT or 3aHET bone marrow (supplemental Figure 4I). Thus, although 3aHET mice have more Mks, the structure and function of 3aHET platelets are grossly normal in mice and in ex vivo activation assays.

We sought to determine how de novo DNA methylation might be altered in the 3aHET context in Mk^bi and non-Mk^bi HSCs, as well as in downstream cells during Mk lineage specification. We sorted Mk^bi and non-Mk^bi HSCs and Mk-erythroid progenitors (MEPs) from WT and 3aHET mice and generated Mks via ex vivo culture for whole-genome enzymatic methyl sequencing (supplemental Tables 3 and 4).

Both Mk^bi and non-Mk^bi 3aHET HSCs exhibited >1000 differentially methylated regions (DMRs) compared with their WT counterparts, and most were areas of decreased methylation (hypoDMRs), as expected given the heterozygous loss of Dnmt3a (Figure 6A). However, these DMRs were mostly unique to each HSC type, with only 2 shared areas of increased methylation (hyperDMRs) and 38 shared hypoDMRs between Mk^bi and non-Mk^bi 3aHET HSCs compared with WT (Figure 6A). This finding suggests that Dnmt3a loss has almost completely different effects on the methylation landscape of the 2 types of LT-HSCs, emphasizing the importance of considering them separately rather than in bulk. Downstream, 3aHET MEPs mostly exhibited hypoDMRs compared with WT MEPs, whereas 3aHET Mks had regions of both significantly increased and significantly decreased methylation (Figure 6B). There is also limited overlap between genes that exhibited hypomethylation or hypermethylation in 3aHET MEPs and 3aHET Mks compared with their WT counterparts (supplemental Figure 5A), further supporting the cell type specificity of DNMT3A-mediated methylation.

We first focused on DMRs in Mk^bi and non-Mk^bi HSCs. Integrative clustering revealed groups of non-Mk^bi–specific hyperDMRs and hypoDMRs, Mk^bi-specific hyperDMRs and hypoDMRs, and common hypoDMRs (Figure 6C). We explored the functional annotations associated with these clusters (Figure 6D). We were most interested in DMRs specific to 3aHET Mk^bi HSCs, because these cells were functionally deficient in transplant. 3aHET Mk^bi-specific hyperDMRs involved adenosine triphosphate and glycoprotein metabolism, whereas 3aHET Mk^bi-specific hypoDMRs involved cell adhesion, secretion, and several signaling pathways (platelet-derived growth factor and Rap1) (Figure 6D), some of which were also identified in the matched bulk RNAseq data (Figure 4B).

We also wondered how methylation changes in 3aHET Mk^bi HSCs might compare with those in 3aHET Mks (Figure 6E). Common hypoDMRs between 3aHET Mk^bi HSCs and Mks involved oxidative stress response and cytoskeletal organization. Mk^bi-specific hypoDMRs involved negative regulation of cell motility, whereas Mk-specific hypoDMRs were nearly all related to various cell signaling pathways (Figure 6F).

We next compared 3aHET and WT MEPs. The functional annotations associated with 3aHET MEP hypoDMRs were striking and involved cell junction and cell projection organization, calcium signaling, and phosphorylation (supplemental Figure 5B). The relatively small number of hypermethylated DMRs showed less significance and apparent functional relevance (supplemental Figure 5C). Finally, we compared 3aHET and WT Mks. 3aHET Mk hypoDMRs involved secretion, cell migration, and proliferation (supplemental Figure 5D), whereas 3aHET Mk hyperDMRs included cell junction and cell projection organization (supplemental Figure 5E).

To better understand the differences common to both MEPs and Mks in the 3aHET context, we performed integrative clustering of DMRs (supplemental Figure 5F) and focused on cluster 5, which represents hypoDMRs that are shared between 3aHET MEPs and Mks (supplemental Figure 5F, red box). Cluster 5 genes are involved in platelet/Mk development and platelet production (supplemental Figure 5G). Some genes implicated in this pathway, such as Gata3, are hypomethylated specifically in 3aHET Mk^bi HSCs but not in 3aHET non-Mk^bi HSCs. Others, such as Jmjd1c, are hypomethylated specifically in 3aHET non-Mk^bi HSCs but not in 3aHET Mk^bi HSCs. In total, our methylation sequencing results depict a landscape of epigenetic dysregulation in 3aHET mice that is unique to each cell type and differs vastly even within the LT-HSC compartment.

Here, we demonstrate that heterozygous loss of Dnmt3a differentially affects LT-HSC subtypes. It has been previously reported that Mk^bi HSCs are more quiescent than their non-Mk^bi counterparts.^5,14 However, this has largely been reported to enhance their transplantability.⁵ With loss of just 1 allele of Dnmt3a, 3aHET Mk^bi HSCs appear to have delayed hematopoietic reconstitution. Furthermore, we found that 3aHET Vwf-expressing LT-HSCs corresponded to an Mk-biased and low-output signature⁵ more closely than their WT counterparts. This is further evidence that the Mk^bi state of LT-HSCs is epigenetically encoded,¹⁴ and DNMT3A is a key regulator of LT-HSC identity.

It is well known that complete ablation of DNMT3A enhances stem cell self-renewal at the expense of differentiation.^19,21 Even though in our study, the HSCs retain 1 intact Dnmt3a allele, we can observe the essence of this phenotype; however, it is mostly confined to the specific Mk^bi subset of HSCs rather than the bulk population. The LT-HSC skewing was most pronounced in Mk^bi 3aHET HSCs when purified cells were transplanted, although it occurred to a lesser degree in non-Mk^bi 3aHET HSCs. The delayed B cell production is unique to 3aHET Mk^bi HSCs. Our work, therefore, demonstrates that DNMT3A loss affects different subsets of the LT-HSC population to different degrees and may differentially regulate pluripotency exit based on the HSC’s intrinsic identity. The classic “myeloid bias” of HSCs lacking DNMT3A may be a hallmark of 3aHET Mk^bi HSCs rather than non-Mk^bi LT-HSCs. It is notable that when bulk WBM was transplanted, 3aHET Mk^bi HSCs were depleted, and this depletion compared with WT also occurred with aging but not with 5-FU treatment. 3aHET Mk^bi HSCs thus appear intolerant of long-term stressors but capable of persisting under short-term duress.

It is also striking that many recipients of 3aHET Mk^bi HSCs did not survive, despite receiving an amount of supporting WBM cells that should have supported hematopoiesis. We suggest that 3aHET Mk^bi HSCs may exhibit suppressive effects on the competitor WBM but fail to differentiate efficiently, hindering the regeneration of remaining marrow and blood components necessary for rescue from lethal irradiation. The potential for interaction between 3aHET Mk^bi HSCs and competitor bone marrow is an intriguing possibility that warrants further follow-up. Additionally, single HSC transplants or barcoding experiments might further elucidate whether the observed phenotypes result from altered composition or bias within the Vwf⁺ HSC pool, which is known to still exhibit somewhat heterogeneous potential,⁴³ or whether they are caused by uniform intrinsic functional deficits within the Mk^bi HSC compartment.

In addition to the effects on 3aHET Mk^bi HSC function, we report more subtle changes downstream in Mks and platelets, with more Mks but no substantial platelet phenotypes. It is possible that the in vivo functions of mutant platelets might be subtly altered; however, due to the extremely high platelet counts in mouse blood (1000 × 10⁹/L to 1500 × 10⁹/L in mice compared with 150 × 10⁹/L to 400 × 10⁹/L in humans)⁴⁴ and a number of physiologic peculiarities, mice can be difficult models for detecting subtle thrombosis and hemostasis phenotypes.⁴⁵ Additionally, metabolic defects might not be apparent in culture conditions with serum or might only be present in cells derived from aged animals.

One of our most striking findings is that there are many more DMRs than differentially expressed genes when comparing WT and 3aHET HSC subsets. 3aHET HSCs function relatively normally in the steady state, and thus, we are unsurprised by the relatively minor transcriptional changes. However, as our functional data demonstrate, the substantial methylation changes that are present at the HSC level likely impair its ability to mount appropriate responses to transplant and aging, resulting in altered differentiation trajectories. Additionally, Mk^bi and non-Mk^bi HSCs exhibit almost completely different DMRs in 3aHET compared with WT, emphasizing the differential epigenetic effects exerted by DNMT3A loss in these 2 cell types.

DNMT3A likely plays a specific role in the Mk/platelet lineage, because we show methylation changes at key loci in the normal transition from progenitors to Mks, which are altered in 3aHET mice. De novo DNA methylation has previously been implicated in Mk specification using methyl-binding domain sequencing,⁴⁶ which showed that differentiation of Mks compared with the erythroid lineage involved a massive reorganization of the methylome. However, the genome-wide Mk methylation landscape has not been previously examined, and no enzymes have been shown to be responsible for its reorganization. The regions that failed to become methylated in 3aHET MEPs and 3aHET Mks relative to WT were grouped into relevant functional categories, such as hemostasis, Mk development, and platelet production. It is interesting to consider whether Mks derived from direct and indirect differentiation pathways might exhibit different methylation patterns and whether they might more closely resemble the methylation patterns of one subset of HSCs over another.

In the germ line context, we were unable to distinguish between the output of Mk^bi HSCs and non-Mk^bi HSCs, because this requires lineage tracing.^13,18,47 Therefore, it is unclear which population contributes to the increased number of Mks per area of bone marrow in germ line 3aHET mice or whether other environmental factors, such as increased inflammatory signaling,^34,40 contribute to the output differences. However, in the transplant setting, it is clearly the non-Mk^bi 3aHET HSCs that produce elevated numbers of Mks per area of bone marrow rather than the Mk^bi 3aHET HSCs (supplemental Figure 2G). This finding may be consistent with the idea that the traditional differentiation pathway, from the HSC through a cascade of progenitors, is responsible for producing immune Mks that respond to inflammatory stress.¹⁸ 

Overall, our results suggest that DNMT3A loss might have pleiotropic effects depending on the type of stem cell that acquires the mutation. Notably, DNMT3A loss higher in the HSC hierarchy appears to cause the most substantial changes in self-renewal and differentiation potential. This finding may have implications for CH, suggesting a relevance for the type of HSC carrying the mutation. Indeed, the order of DNMT3A vs additional mutations can affect outcomes.^33,48 We speculate that our work might explain the frequent occurrence of DNMT3A mutations in essential thrombocythemia, a myeloproliferative neoplasm characterized by excessive platelet production,⁴⁸ in which Mk^bi HSCs are thought to be the source of the malignancy.^11,49 These data reinforce the concept that DNMT3A is important for promoting normal hematopoietic differentiation into all lineages in a rapid fashion and underscore the exquisite dosage sensitivity of HSCs to the presence of DNMT3A.

The authors thank Kwangwon Lee and Antrix Jain for their assistance with the sample preparation and data analysis. The authors are grateful to Sten Eirik Jacobsen for sharing the VWF-GFP reporter mouse line. Figure schematics were created with BioRender.com.

This project was supported by the Cytometry and Cell Sorting Core at Baylor College of Medicine (supported by Cancer Prevention and Research Institute of Texas grant CPRIT-RP180672, National Intitutes of Health [NIH]/National Cancer Institute [NCI] grant CA125123, and NIH grant RR024574) and the assistance of Joel M. Sederstrom. The Mass Spectrometry Proteomics Core is supported by NIH/NCI grant CA125123, Cancer Prevention and Research Institute of Texas grant RP210227, and NIH grant S10 OD026804. The Genomics and RNA Profiling Core is supported by NIH/National Institute of Diabetes and Digestive and Kidney Diseases (NIDDKD) grant NIDDK-DK56338 and NIH/NCI grant CA125123, NIH/National Institute of Environment and Health Sciences grant NIEHS-5P30ES030285-03, Cancer Prevention and Research Institute of Texas grant 1S10OD02346901.

The authors gratefully acknowledge the following funding sources: the Eunice Kennedy Shriver National Institute of Child Health and Human Development (F30HD111129; S.M.W.), the Robert and Janice McNair Foundation MD/PhD Scholars program (S.M.W.), and Baylor Research Advocates for Student Scientists (S.M.W.). The Goodell Lab is supported by NIH/NCI grants CA183252, CA265748, and CA237291, NIDDKD grant DK092883, and NIH/National Institute of Aging grant AG036695. The study was supported in part by Merit Review Award I01 BX002551 from the Department of Veterans Affairs Biomedical Laboratory Research & Development Service (R.E.R.). J.A.C. was supported by the NIH/National Heart, Lung, and Blood Institute (NHLBI) T32 Training program HL139425 and VA CDA-2 IK2BX006462. A.R.B. is supported by NIH grant P30 EY007551. The Machlus Lab is supported by NIDDKD grant R01DK139341; K.R.M., NHLBI grants R01HL151494 (K.R.M.) and K99HL175037 (M.N.B.), the American Society of Hematology (ASH) Scholar Award (V.C.), and American Heart Association grant 24IPA1274573 (K.R.M.). The Deneen Lab is supported by NIH/National Institute of Neurological Disorders and Strokes grant R35-NS132230 (B.D.) and American Heart Association grant AHA-23POST1019413 (M.R.W.). A.G.L. is supported via the Baylor College of Medicine PREP Scholars program NIH/National Institute of General Medical Sciences grant R25GM069234.

The contents presented herein do not represent the views of the US Department of Veterans Affairs or the US Government.

Contribution: S.M.W., K.R.M., and M.A.G. contributed to conceptualization; S.M.W., V.C., A.G.G., A.G.-M., R.R., A.G.L., J.A.C., M.R.W., J.I.H., S.P., E.K., C.-W.C., and J.R.P. conducted investigations; S.M.W., D.F., J.D.L.F., J.R., M.N.B., E.C., J.A.C., E.K., and J.R.P. performed formal analysis; S.M.W. and M.A.G. wrote the original draft of the manuscript; S.M.W., V.C., M.A.G., K.R.M., and J.R.P. reviewed and edited the manuscript; and R.E.R., A.R.B., M.A.G., K.R.M., B.D., and J.S. provided supervision.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

Correspondence: Margaret A. Goodell, Baylor College of Medicine, 1 Baylor Plaza, Alkek N1030, Houston, TX 77030; email: [email protected].