Activation of a branched chain amino acid rheostat restores replication-dependent hematopoietic stem cell fitness

Summary

Adult hematopoietic stem cells (HSCs) sustain the lifelong production of all mature blood and immune cells. HSCs possess extensive regenerative potential, but their self-renewal is limited. A long-standing question has been why replicative history negatively impacts HSC functions. We found that accrued divisions alter HSC production; generating low-output bone-marrow landscapes that are highly variable in lineage contribution and transcriptionally divergent within individual lineages. Division-driven HSC functional alterations arise from redirecting branched chain amino acid (BCAA) usage from catabolic towards anabolic activity, causing faster HSC cell cycle kinetics. Adding a BCAA transamination product overcomes the BCAA catabolic checkpoint and slows down the cell cycle, durably rescuing balanced lineage output of HSCs with accrued divisions. Hence, our study suggests the paradigm whereby replicative history causes metabolic and transcriptional drift, generating divergent HSC output. Division-dependent HSC functional drift can be restored by metabolite replacement, which has long-term therapeutic implications for HSC regenerative medicine.

eTOC Blurb:

Bartram et al. show that HSCs with accrued divisions generate low-output progeny that is highly variable in lineage distribution and transcriptionally divergent within individual lineages. HSC functional decline arises from redirecting BCAA usage from catabolic towards anabolic activity. BCAA metabolite replacement restores division-dependent HSC fitness.

Graphical Abstract

Introduction

The lifelong production of all hematopoietic lineages is ensured by a pool of rare hematopoietic stem cells (HSCs) and their progeny. While residing in a quiescent state in the bone marrow, HSCs divide infrequently to produce another stem cell or progeny that commit to differentiation to generate all blood lineages.^1,2 Long thought to possess extensive self-renewal ability to sustain blood cell production throughout life, studies investigating the relationship between HSC divisional history and function using a histone 2B-green fluorescent protein (H2B-GFP) label-retention model have shown that HSC regenerative potential is inversely correlated with their division history.^3–6 Hence, although the quiescent state of HSCs is reversible and essential for HSC functions, it is insufficient to protect HSCs from functional loss upon replication. These studies provide experimental evidence for the generation-age hypothesis, which postulates that the lineage generation capacity of HSCs is intrinsically determined by their previous divisional history.⁷ How past cell division determines the regenerative capacity of HSCs is ill-defined.

Mitochondrial metabolism contributes to maintaining HSC functions through division. Quiescent HSCs with long-term repopulation potential possess low metabolic activity, low protein and transcriptomic synthesis, and rely on anerobic glycolysis, lysosome and catabolic activity to meet energy needs.^8–12 While all HSCs are quiescent, HSC repopulation potential is inversely correlated with the depth of quiescence.^4,5,13–18 The time to cell cycle entry progressively decreases in HSCs with intermediate and short-term repopulation potential, and this is accompanied by higher metabolic activity and a lineage-primed transcriptomic state.^19–22 In the H2B-GFP label-retention model, label-retaining cells (LRCs) have low metabolic activity, which can be inferred by reduced tetramethylrhodamine ethyl ester (TMRE) mitochondrial labeling, whereas non label-retaining cells (nLRCs) exhibit higher TMRE, are metabolically activated, and are primed for cell cycle and differentiation, at least at the transcriptional level.^9,23 During quiescence exit, HSC activation depends on major metabolic changes, including higher transcription and mitochondrial activity, and a switch to anabolic activity to promote their growth and differentiation.^12,24–31 Repeated HSC replication under acute regenerative conditions significantly alters mitochondrial functions that contribute to their functional exhaustion.³² Conceptually, it is thus postulated that division-driven HSC functional decline occurs through differentiating divisions in which Long-Term (LT)-HSCs generate Short-Term (ST)-HSCs that are more metabolically active and acquire differentiation characteristics. The metabolic and functional behaviors of HSCs with past division history under homeostatic conditions in vivo remain unresolved. Clear functional and transcriptomic differences driving HSC functional decline with accrued divisions are still lacking.

To examine HSC functional decline with division history, we integrated the H2B-GFP label-retention model with TMRE. First, the LRC compartment is heterogeneous with a wide range of GFP and TMRE levels. A subset of HSCs with past division history retains low TMRE. However, they exhibit unique functional behavior. Using a combination of long-term serial transplantation and single cell RNA-sequencing analyses, we show that these cells possess a low-output repopulation potential and give rise to highly variable lineage reconstitution patterns and incoherence in lineage cell state representation. Unexpectedly, the resulting individual lineages are also transcriptionally divergent, suggesting that that HSC replicative stress is propagated to the progeny. Mechanistically, division-associated HSC functional alterations are driven by reduced branched chain amino acid transaminase activity. Interestingly, it can be functionally reversed by the product of branched chain amino acid catabolism, alpha-ketoisocaproate. Hence, rather than exclusively differentiating divisions, our data suggest that accrued divisions also generate a pool of HSCs that retain stem-like features but undergo functional and molecular drift. The functional drift is driven by cell-intrinsic changes in nutrient usage. Importantly, metabolite replacement is sufficient to restore HSC functions, which has long-term clinical implications for HSC regenerative medicine.

Results

Division history drives HSC functional drift.

To understand mechanisms of HSC functional decline with division history, we used the doxycycline-inducible histone 2B (H2B)-GFP label-retaining cell tracking system. H2B-GFP mice were given doxycycline in their drinking water for two weeks to induce H2B-GFP expression, and the label-retaining cells were analyzed 5 months following doxycycline removal (Figure 1A). Under these conditions, the GFP-labeled fraction of the HSC compartment [LSK-SLAM], named label-retaining cells (LRCs), contain HSC activity. GFP-negative LSK-SLAM or non-label-retaining cells (nLRCs) are devoid of HSC activity.^4,5 We previously showed that the mitochondrial network is permanently remodeled in nLRCs compared with LRCs.³² At the single cell level, the transcriptome of nLRCs cluster closer to HSCs that have a disrupted mitochondrial network (Supplemental Figure S1A-B). Conversely, LRCs cluster with HSCs that have an intact mitochondrial network, prompting us to examine the relationship between division history and mitochondrial activity. We used TMRE to monitor differences in mitochondrial activity.^9,12 HSCs have a broad TMRE staining pattern. HSCs with high TMRE levels are in a more activated state compared to HSCs with low TMRE levels.^9,12 The TMRE staining was done without the dye efflux blocker, verapamil, because of cellular toxicity,³³ which would prevent subsequent functional analysis. Under these conditions, the TMRE levels do not reflect the exact HSC mitochondrial membrane potential, only distinct metabolic states.^34,35 As previously reported, LRCs can be separated into GFP high, low, and negative populations (Supplemental Figure S1C), representing incremental increases in divisional history.⁴ TMRE levels were higher in nLRCs compared with LRCs, as previously reported⁹ (Supplemental Figure S1C). When freshly isolated, cell cycle states of GFP^High and GFP^Low LRCs were not different (Supplemental Figure S1D). nLRCs had increased frequency of cells in S/G2/M phases of the cell cycle (Supplemental Figure S1D). Interestingly, TMRE levels separated the LRC compartment into 4 populations: GFP^High-TMRE^Low (G^HiT^Lo); GFP^High-TMRE^High (G^HiT^Hi); GFP^Low-TMRE^Low (G^LoT^Lo); GFP^Low-TMRE^High (G^LoT^Hi) (Figure 1B; Supplemental Figure S1E). TMRE staining is highly stable within one hour of staining, (Supplemental Figure S1F). EPCR and Sca-1 levels were high in G^HiT^Lo cells but gradually decline in G^HiT^Hi; G^LoT^Lo; G^LoT^Hi, respectively, whereas expression of the progenitor marker CD34 increased (Supplemental Figure S1G). T-Distributed Stochastic Neighbor Embedding (t-SNE) analyses (FlowJo) separated the LSK-SLAM cells into 5 distinct clusters when integrating division history and TMRE to surface markers (Supplemental Figure S1H). These data suggest the existence of distinct HSC subsets within LRCs that can be identified based on division history and TMRE status.

Figure 1. Division history and TMRE organize the LRC compartment into discrete cellular subsets.

(A) Schema of experiment using the Tet-on H2B-GFP mouse labeling model. (B) TMRE labeling of LSK-SLAM cells. Representative flow cytometry contour plot of H2B-GFP intensity and TMRE intensity within LSK-SLAM.

(C-D) Competitive serial transplantation. (C) PB donor-cell contribution over 16 weeks in primary recipients (mean ±SD, n=4–12) one-way ANOVA, multiple comparisons. (D) Proportion of PB lineages in primary recipients, (mean ±SD n=6–12). (E) PB donor-cell contribution over 16 weeks in secondary recipients (mean ±SD n=4–14) one-way ANOVA, multiple comparisons. (F) Longitudinal analyses of PB donor-derived lineages in primary and secondary recipients. Three representative recipients per group is shown, both secondary recipients per group are shown.

(G) Reconstitution patterns scored as long-term high-output (LT^Hi), long-term low-output (LT^Lo), intermediate-term (IT), short-term (ST); Fisher exact test of proportion of LT^Hi in GFP^LoTMRE^Lo versus GFP^HiTMRE^Lo. (H) Relative proportion of PB lineages at 16 weeks in each secondary recipient with overlay of total PB chimerism. (I) Reconstitution patterns of PB lineages scored as high-output and balanced (Ba^Hi), high-output and myeloid-bias (My^Hi), high-output and lymphoid-bias (LyHi), low-output and myeloid-bias (My^Lo). Fisher exact test of proportion of My^Lo in GFP^LoTMRE^Lo versus GFP^HiTMRE^Lo. (J) Bifurcation model of myeloid cell proportion versus total chimerism (n=24–26). (K) Schema of experiment of H2B labeling and chase in recipients of LRCs. (L) Proportion of donor-derived LSK-SLAM subsets in each recipient of donor LRCs, n=7,7,5,5 recipients per group).*p<0.05, **p<0.01, ***p<0.001, ****p<0.0001. Related to supplemental Figure S1-2 and Table S1.

In serial competitive transplant, recipients of G^HiT^Lo LRCs showed gradual increased donor-cell chimerism over time, with some recipients exhibiting over 70% chimerism by week 16 (Figure 1C). Recipients that received G^LoT^Lo LRCs had engraftment at around 40% chimerism (Figure 1C). Recipients of G^HiT^Hi or G^LoT^Hi LRCs showed the lowest level of chimerism at below 20% (Figure 1C). There were no significant differences in the mature blood lineages between the groups in the primary transplant (Figure 1D; Supplemental Figure S1I). In the bone marrow (BM), G^HiT^Lo LRCs gave rise to the highest donor-cell chimerism and reconstituted the HSC and progenitor compartments (Supplemental Figure S1J-L). At this time point, all cells are GFP^Neg, as expected due to proliferation of donor population, and TMRE is similar in the reconstituted LSK-SLAMs between the groups (Supplemental Figure S1M). Secondary transplanted mice from the original G^HiT^Lo LRC donors had the highest PB chimerism (Figure 1E). In sum, G^HiT^Lo LRCs possess the highest LT-HSC activity. Donor LRCs that have divided or have high TMRE exhibit lower reconstitution activity. Donor LRCs that have both divided and have high TMRE do not contain LT-HSC activity.

Analyzing donor-derived myeloid, B and T cells longitudinally,^36,37 we noted various reconstitution patterns according to the duration, amplitude and lineage composition of the reconstitution, reflecting the proliferation and differentiation output of 200 cells during serial competitive transplantation experiments. To quantify these differences, we adopted the following classification, adapted from classifications used for single cell transplant.^16,37,38 The donor populations were retrospectively defined as containing high-output long-term repopulation potential when giving rise to a durable donor cell-derived chimerism in each PB lineage at 10% or above throughout the primary or secondary transplants; low-output repopulation potential when the donor cell-derived chimerism of any of PB lineages was between 1% and 10% throughout the primary or secondary transplants; intermediate when the donor cell-derived chimerism in any PB lineage was above 1% in primary transplant but below 1% in secondary transplants; and, short-term when any PB lineage was below 1% by 24 weeks in primary transplant and remained below 1% or undetectable in secondary transplants. Accordingly, 80% of G^HiT^Lo LRCs contained high-output repopulation potential. Only G^HiT^Lo LRCs generated high output progeny with all lineages above 60% chimerism (Supplemental Figure S1N, and Supplemental Table S1). In contrast, G^LoT^Lo LRCs had 10% high-output repopulation potential, 40% low-output repopulation potential, and 30% intermediate. G^LoT^Hi LRCs had only intermediate and short-term repopulation potential (Figure 1F-G, and Supplemental Figure S1N, and supplemental Table S1).

The lineage compositions of the graft were also variable. The reconstitution pattern was termed balanced if the myeloid reconstitution was between 20–40%, myeloid-biased if the recipients had greater than 40% myeloid reconstitution, and lymphoid-biased if the lymphoid reconstitution was greater than 80%. Additionally, some mice exhibited a fourth reconstitution pattern where the myeloid reconstitution was greater than 50%; however, the donor-derived chimerism fell below 5–10%. We named this pattern low-output myeloid-biased. Donor-LRCs that possessed high-output repopulation potential mostly gave rise to a lineage balanced reconstitution. In contrast, intermediate or short-term LRCs could generate myeloid-bias or lymphoid-bias progeny. Remarkably, the low-output myeloid-bias reconstitution pattern was mainly found in mice that received G^LoT^Lo or G^LoT^Hi LRCs (Figure 1H-I and Supplemental Figure S1O and supplemental Table S1), as underscored with a correlation analysis of donor-cell chimerism and myeloid cell frequency (Figure 1J). Finally, some recipients of G^HiT^Lo, G^LoT^Lo or G^LoT^Hi LRCs exhibited T-cell dominance (Figure 1H). These data suggest that division history reduces the amplitude and duration of HSC lineage output, consistent with reduced HSC activity. Twelve months post-doxycycline removal, the LRC compartment contained G^HiT^Lo, G^HiT^Hi, G^LoT^Lo, G^LoT^Hi populations. G^HiT^Lo frequency was lower than 5M post-doxycycline removal, as expected (Supplemental Figure S2A-B). Only G^HiT^Lo and G^HiT^Hi engrafted, albeit with myeloid bias, perhaps due to aging (Supplemental Figure S2C-C’).

To gain insight into the relationship between LRC populations, we examined which LRC subset can produce G^HiT^Lo cells in transplant studies. Two months after transplantation, when the HSC pool has been reconstituted, recipient mice were treated with doxycycline to relabel cells, and GFP dilution was examined 5 months later (Figure 1K and Supplemental Figure S2D,E). Recipients of donor G^HiT^Lo cells had GFP high and low LSK-SLAMs, and TMRE high and low cells within each GFP fraction, suggesting that G^HiT^Lo LRCs can give rise to all phenotypically defined LRC subsets. In contrast, G^HiT^Hi and G^LoT^Lo LRCs generated G^LoT^Lo and G^LoT^Hi cells. G^LoT^Hi LRCs mostly generated non-LRCs (Figure 1L and Supplemental Figure S2F). Of note, one recipient of G^LoT^Lo LRCs had 75% of GFP^Neg LSK-SLAM. Donor-derived LSK-CD48+ cells from each LRCs were GFP^Neg (Supplemental Figure S2G). Noteworthy, some recipients of G^LoT^Lo LRCs exhibited 80–90% of G^LoT^Lo cells with little GFP^Neg LSK-SLAM (Figure 1L and Supplemental Figure S2F), suggesting that during the span of doxycycline chase, HSCs generated by G^LoT^Lo LRCs initially divided faster that those generated by G^HiT^Lo LRCs, but stopped dividing and accumulated as GFP low population. This is consistent with prior findings showing that during aging, LRCs divide few times, then go into dormancy.³ Hence, LRC subsets can be organized in a hierarchy with G^HiT^Lo LRCs at the apex.

HSC replicative history is associated with transcriptional drift within lineages.

To further examine the effect of divisions on hematopoietic cell production, we analyzed the transcriptome of BM c-Kit-enriched cells from mice reconstituted with G^HiT^Lo LRCs and G^LoT^Lo LRCs, using 10X single cell RNA-sequencing at 16- and 24-weeks following transplantation. Thirty-three cell states were identified in all samples, based on their transcriptional profiles using a dataset of broad cKit+ progenitor cells using cellHarmony software,^39,40 that represented myeloid, erythroid, megakaryocytic, and lymphoid lineages, and HSCs and multipotent progenitors (Figure 2A and Supplemental Figure S2H,I and Supplemental Table S2). The relative proportion of individual cell states was uniform across recipients of G^HiT^Lo LRCs and deviated little from the average of all samples (Figure 2B and Supplemental Table S2). In contrast, the proportions of the various cell states generated by G^LoT^Lo LRCs greatly diverge from the average of G^HiT^Lo BM (Figure 2B-C). The patterns were highly heterogenous across recipients of G^LoT^Lo LRCs with substantial overrepresentation or underrepresentation of various lineages (Figure 2B). Some recipients exhibited up to 9- or 16-fold increase in CD19+ cells or macrophages, respectively. Others had a 6-to-10-fold decrease in megakaryocyte progenitors (MKPs) or erythroid progenitors (ERPs), respectively. At week 24, myeloid precursors (ProNeu1–2 and PreNeu1–3)³⁹ were less represented in the BM derived from G^LoT^Lo LRCs than that derived from G^HiT^Lo LRCs; whereas immature neutrophils were enriched (Figure 2B). The divergence in cell state representation was recipient specific and appeared to be stochastic. Because single-cell capture of immature neutrophils can be problematic and was different between the 2 captures, we examined late-stage myeloid differentiation using spectral flow cytometry analysis.⁴¹ It confirmed increased relative abundance of myeloid precursors and immature neutrophils in BM generated by G^LoT^Lo (Figure 2D; Supplemental Figure S2J). Together, these findings suggest that HSCs with division history generate a highly heterogenous BM landscape with incoherent but random representation of different cell states, consistent with a loss in lineage reconstitution fidelity.

Figure 2. Division history alters the bone marrow transcriptomic landscape.

10X Single cell RNA sequencing on cKit+-enriched BM cells from GFP^HiTMRE^Lo and GFP^LoTMRE^Lo LRC recipients, at 16 weeks (capture 1, n=3) and 24 weeks (capture 2, n=2) post-primary transplantation. (A) UMAP plot showing all samples. (B) Normalized cell state proportion relative to the average of GFP^HiTMRE^Lo. Two-way ANOVA f=4.930, sample effect. (n=32 clusters per sample, n=5). (C) Same data as in (b) displayed as average cell state proportion relative to GFP^HiTMRE^Lo cells. (D) Flow cytometry analysis of cKit+-enriched bone-marrow cells from GFP^HiTMRE^Lo and GFP^LoTMRE^Lo LRC recipients at 24 weeks post transplantation (mean, n=3 biological replicates representative of 2 independent experiments, two-sided unpaired T-Test). (E) PCA of cell states from recipients of GFP^HiTMRE^Lo LRCs relative to GFP^LoTMRE^Lo LRCs from capture 1. (F) Heatmap of DEGs within each cell state from recipients of GFP^HiTMRE^Lo LRCs relative to GFP^LoTMRE^Lo LRCs from capture 1 and 2. (G) Pathway analysis of DEGs within each cell state cluster from recipients of GFP^HiTMRE^Lo LRCs relative to GFPLoTMRELo LRCs. Related to supplemental Figure S2 and Table S2-5.

Surprisingly, principal component analysis (PCA) showed that cell states generated by G^LoT^Lo LRCs clustered away from those generated by G^HiT^Lo LRCs, in almost every lineage (Figure 2E and Supplemental Figure S2K). Pseudobulk gene expression analyses identified numerous differentially expressed genes (DEGs) in each lineage generated by G^LoT^Lo LRCs compared with G^HiT^Lo LRCs (Figure 2F and Supplemental Figure S2L). DEGs categorized in protein translation, cell cycle, oxidoreduction, immune system, and response to stress, as well as in mitochondria, lysosomes (Figure 2G and Supplemental Figure S2M and Supplemental Table S3-5). Hence, the divergence in lineage output generated by G^LoT^Lo LRCs includes changes in the relative proportion of hematopoietic lineages, and in the transcriptome of individual cells within one given lineage along its differentiation path, suggesting that replicative stress is propagated to HSC progeny. The divergence could arise from high division history of donor G^LoT^Lo cells and high proliferation during transplantation.

HSC replicative history is associated with altered metabolic pathways.

To understand why HSCs with division history are less functional, we compared the transcriptome of each LRC subset to the corresponding unseparated LSK-SLAM population, using single cell RNA-sequencing (Figure 3A). We examined the transcriptome of cells during activation into cycle because HSC lineage fate decisions take place in actively cycling cells, and transcriptional differences related to replicative stress are manifested during cell cycle.³² Uniform manifold approximation and projection (UMAP) analysis of cells from all samples, performed in AltAnalyze^R using unbiased ICGS-2 clustering, uncovered nine clusters (C1–C9) (Figure 3B and Supplemental Table S6). Cluster annotation using ICGS-2 identified C7-C9 as being the least differentiated cells, expressing stem cell markers (Mllt3, Hlf, Procr, Meg3) and quiescent genes (Cdkn1a, Sesn2). C6 and C5 were intermediate with expression of stem cell markers (Mllt3, Hlf), lineage-primed genes (Meis1, Stmn1) and activated genes (Cdk6). C1-C3 were found in cell cycle (Mki67, Cdk6, Birc5, Dut) and C4 was megakaryocyte lineage-primed (Pf4, Itga2b) (Figure 3C-D and Supplemental Figure S3A-B). Consistently, differential gene expression between clusters indicated progressive increase in cell cycle, DNA replication and metabolism genes with concomitant decrease in stem cell marker genes and pathways known to be associated with stemness, including lysosomes and cholesterol metabolism^9,42–44 (Figure 3E). G^HiT^Lo LRCs mostly contained C7-C9, as well as C5 and C6. G^LoT^Lo LRCs also mostly contained C7-C9 but had a notable increase in C6 at the expense of C8. G^HiT^Hi contained all clusters. G^LoT^Hi LRCs separated from the other LRCs, mostly contained C1-C3, with increase in C4 (Figure 3F and Supplemental Figure S3C-E-E’). Notably, expression of cell cycle genes [ie, Cdk6] was higher in G^LoT^Lo compared to G^HiT^Lo LRCs, within C6–7-8 (supplemental Figure S3F). The data suggest heterogeneity within each LRC population. G^HiT^Lo and G^LoT^Lo are molecularly close, enriched in HSC states. Noticeably, G^LoT^Lo is slightly more enriched in cell cycle-primed cells. High TMRE populations are lineage- and cell cycle-primed.

Figure 3. Differences in division history and TMRE cause transcriptional drift in cultured activated LRCs.

(A-F) 10X scRNA-seq analysis of LRCs 15h after activation into cycle in culture. (A) Schema of experiment. (B). UMAP plot showing distinct cell clusters. (C) UMAP showing expression of indicated genes. (D) Expression card of genes and percent expressed in each cluster. (E) ZScore card of differential pathway analysis in each cluster compared to C7. (F) Proportion of C1-C9 in each LRC. (G-L) Bulk RNA-seq of LRCs 24h after activation into cycle in culture (n=3 replicates each). (G) Principal component analysis of the unsupervised clustering. (H) Unsupervised clustering analysis. (I) Supervised clustering of stemness, differentiation, and cell cycle genes. (J) Volcano plot of differentially expressed genes of GFP^LoTMRE^Lo versus GFP^HiTMRE^Lo LRCs. (K) Supervised clustering of metabolic genes in GFP^LoTMRE^Lo versus GFP^HiTMRE^Lo LRCs. (L) Network analyses of uniquely differentially expressed genes in indicated populations. Related to supplemental Figure S3 are supplemental table S6-7.

To gain more insights in pathway analysis, we performed bulk RNA-sequencing. PCA analyses indicated a clear separation between each group (Figure 3G). Unbiased hierarchical clustering in AltAnalyze^R revealed a specific transcriptome signature for each LRC population (Figure 3H-I and Supplemental Figure S3G and Supplemental Table S7). Compared to G^HiT^Lo LRCs, all other LRCs had more downregulated genes, suggesting that gene expression repression is coupled with stemness loss (Figure 3J and Supplemental Figure S3H). G^HiT^Lo LRCs had higher expression of genes related to ‘stem cell’, such as Runx1, Gfi1, and lysosomal genes (Figure 3I and Supplemental Figure S3I and Supplemental Table S7). Programs related to cell cycle and lineage-priming (Pf4, Cd48, Mpo, Epor, Gfi1b, Pbx1, Gata1) were overexpressed in TMRE^Hi LRCs, consistent with independent reports on the transcriptome of TMREHi HSCs⁹ (Figure 3H-I and Supplemental Figure S3I-J). These data suggest that lineage priming is linked to differential mitochondrial activity of HSCs rather than their division history, consistent with scRNA-sequencing analyses.

Network analyses of DEGs unique to G^HiT^Lo LRCs revealed enrichment for signaling and metabolic pathways, including one carbon folate pathway, mammalian target of rapamycin (mTOR) suppression, fatty acid metabolism, and amino-acid catabolism (Figure 3K-L and Supplemental Figure S3K). Notably, branched-chain amino acid (BCAA, valine, leucine, and isoleucine) degradation pathway, including genes related to leucine catabolism (Bcat2, Ivd and Mccc1, Bckdk), was higher in G^HiT^Lo LRCs, along with higher expression of mTOR suppressive genes (Sesn2 and Sesn3) (Figure 3K-L). In contrast, DEGs uniquely expressed in G^LoT^Lo LRCs were enriched in adhesion, immune response, and oxidoreductase activity genes, consistent with a myeloid-bias potential (Figure 3K-L).

We next performed metabolomic using liquid chromatography–mass spectrometry (LC–MS), optimized for low cell number HSC studies.^31,45 We obtained information on 150 metabolites for all LRCs but for G^HiT^Hi, due to low cell numbers. Consistently throughout the datasets, we detected different metabolite abundance related to amino acids, especially BCAAs, and nucleotides in G^LoT^Lo compared with G^HiT^Lo LRCs (Figure 4 and supplemental Table S8). Most notably, BCAA levels (leucine/isoleucine and valine) were higher in G^LoT^Lo compared with G^HiT^Lo LRCs whereas BCAA catabolism products [carnitine-conjugated ketoacids derived from BCAA metabolism, including acyl-c3 (from valine and isoleucine), acyl-c5:1 (from isoleucine)⁴⁶] tended to be reduced in G^LoT^Lo. The leucine catabolism product acyl-c5-OH was significantly lower in G^LoT^Lo LRCs. As a result, the ratio of BCAA metabolism products to BCAA was significantly lower in G^LoT^Lo LRCs, suggesting reduced BCAA metabolism. Other amino acids (histidine, threonine, glycine, and glutamine), ATP and nucleotides (guanine, inosine, and cytidine) were significantly more abundant in G^LoT^Lo LRCs. TCA cycle-related metabolites, like malate, were higher in G^LoT^Lo compared to G^HiT^Lo LRCs, whereas levels of succinate were similar. Finally, no consistent changes in glycolysis or free fatty acids were found between the groups. Metabolites of BCAA catabolism were also less abundant in G^LoT^Hi LRCs, along with increased fatty acid oxidation product acyl-c4. Amino acids, including glutamine and threonine, were also higher in G^LoT^Hi LRCs. The data suggest that increased BCAA metabolism is associated with stemness.

Figure 4. Division history rewires HSC metabolic activity, in cultured activated LRCs.

Untargeted metabolomics analysis of LRC subsets after 20h activation in culture. Metabolite abundance values were normalized to the average of GFP^HiTMRE^Lo within each experiment. (n=4 independent experiments, n=2 independent mass-spectrometry runs). One-way ANOVA with multiple comparison test. *p<0.05, **p<0.01, ***p<0.001. Related to supplemental Table S8.

This was intriguing as HSC activation is thought to be accompanied by increased anabolic processes.^8,30 BCAAs leucine, isoleucine, and valine can be used for anabolic or catabolic purposes.^47,48 Notably, leucine abundance is sensed by mTOR complex 1 leading to mTOR activation and subsequent anabolic processes, including protein, nucleotide, and lipid synthesis.⁴⁹ Alternatively, BCAAs can be catabolized into metabolite intermediates. The first step of BCAA catabolism is catalyzed by branched-chain aminotransferase (Bcat) in which BCAAs are converted into branched-chain α-keto acids. A series of enzymatic reactions then leads to the generation of acetyl-CoA or succinyl-CoA. Valine catabolism generates succinyl-CoA, leucine catabolism generates acetyl-CoA, and isoleucine catabolism generates both. Mammalian cells express two Bcat isoforms, Bcat1 and Bcat2, that are found in the cytosol or the mitochondria, respectively.⁴⁷ Bcat2 expression is ubiquitous whereas Bcat1 expression is limited to the brain tissue.⁴⁷ Interestingly, Bcat2 protein expression was significantly reduced in G^LoT^Lo LRCs compared to G^HiT^Lo LRCs (Figure 5A). Analysis of scRNA-seq datasets of activated LRCs suggested that bcat2 mRNA expression correlated with expression of Sca-1 (encoded by Ly6a) in LRCs (Supplemental Figure S3L). Analysis of prior scRNA-seq datasets comparing the transcriptome of LSK-SLAM before and after transplantation³² showed that Bcat2 mRNA expression increased in LSK-SLAM after activation into cycle, but less so in post-transplant HSCs, which have substantial proliferative history (Supplemental Figure S3M), suggesting replicative history reduces Bcat2 expression in HSCs. In Bcat-deficient models, BCAA oxidation can be restored by supplementing cells with the leucine transamination product 2-keto-isocaproate (KIC).^50–53 We tested the effect of KIC on G^LoT^Lo LRC metabolic activity. Due to limitations in cell number, it was not possible to examine the metabolome of G^LoT^Lo and G^LoT^Lo+KIC from the same group of cells. G^LoT^Lo and G^LoT^Lo +KIC conditions were obtained from different experiments and each compared to G^HiT^Lo LRCs; the data were normalized to G^HiT^Lo LRCs for comparative analyses. KIC-treated G^LoT^Lo LRCs exhibited levels of leucine/isoleucine, valine, and BCAA oxidation products similar to G^HiT^Lo LRCs (Figure 5B, supplemental Figure S3N and supplemental Table S8). Hence, KIC treatment boosts BCAA oxidation. Other metabolites were also restored in KIC-treated G^LoT^Lo LRCs to levels found in G^HiT^Lo LRCs, notably nucleotides (Supplemental Figure S6). Hence, BCAA oxidation is active in HSCs and decreases with division history.

Figure 5. Branched chain amino acid catabolism couples stemness and cell cycle speed.

(A) Bcat2 protein expression using IF (mean ±s.e.m., n=25 cells; two-sided unpaired t-test). (B) Normalized leucine/isoleucine, valine abundance, one-way ANOVA, multiple comparisons. (B’) Ratio of BCAA-derived acylcarnitines to BCAA levels, one-way ANOVA, multiple comparisons. (C) Schema of in vitro KIC supplementation. (D) Cumulative single cell division kinetics in vitro. (n=21, n= 44, n= 56 cells, respectively). (E) Proportion of EdU positive cells (n=84, n=85, n=86 cells, respectively), at 18 hours post-activation in vitro. Fisher exact test. (F) Cdk6 protein expression at 18 hours post activation in vitro, using IF (n=131, n=131, and n=95 cells, respectively), two-sided unpaired t-test. (G) Expression of pS6 at 12 hours post-activation in vitro, using IF (mean ±s.e.m., two-sided unpaired t-test). (H) Proportion of EdU positive cells in GFP^Hi TMRE^Lo HSCs treated with DMSO, KIC, or Rapamycin (n=49, 49, and 50 cells respectively) and GFP^Lo TMRE^Lo HSCs treated with DMSO, KIC, or Rapamycin (n= 50, 50, and 50 cells respectively) 18 hours post culture, Fisher exact test. (I) Multilineage differentiation potential of single cells in vitro, Fisher exact test. (J) Schema of experiments of Bcat inhibitor treatment in vitro. (K) Flow cytometry analysis of pS6 expression at 12 hours post-activation in vitro. Proportion of pS6-positive cells is shown (mean ± s.e.m, n=5, two-sided paired t-test). (L) Cumulative single cell division kinetics in vitro (n=114 and n=130, respectively). (M) EdU MFI at 18 hours post-activation in vitro (DMSO Control n=201, Bcat inhibitor n=199 cells), two-sided unpaired t-test (N) Multilineage differentiation potential of single cells in vitro. Fisher exact test. Related to supplemental Figure S3-4 and Table S8.

Branched-chain amino acid metabolism couples HSC cell cycle latency with stemness.

Given the cell cycle signature, we examined time-to-first division in single cells (Figure 5C).^54,55 The time to first-division of G^LoT^Lo LRCs was much faster than that of G^HiT^Lo LRCs (Figure 5D). The second and third divisions were also accelerated (Supplemental Figure S4A). G^LoT^Lo LRCs had increased EdU incorporation (Figure 5E) and expression of the G1-phase cyclin-dependent kinase Cdk6 (Figure 5F); consistent with HSC repopulation potential being inversely correlated with time-to-cell-cycle entry.^{5,13,14,19,56} This was corrected by KIC supplementation (Figure 5D-F). Supplementation with keto-isovalerate (KIV) or keto-methylvalerate (KMV), from valine and isoleucine catabolism, respectively, did not restore cell division kinetics of G^LoT^Lo LRCs (Supplemental Figure S4B), suggesting a specific role for leucine catabolism in cell cycle control in HSCs. Interestingly, mTOR activity, as assessed by pS6 immunostaining, was higher in G^LoT^Lo LRCs compared to G^HiT^Lo LRCs but not in KIC-treated G^LoT^Lo LRCs (Figure 5G). However, protein synthesis was similar in G^HiT^Lo and G^LoT^Lo LRCs (Supplemental Figure S4C). Increased EdU incorporation of G^LoT^Lo LRCs was dependent on mTOR activity, but Cdk6 was not (Figure 5H and Supplemental Figure S4D-E). Finally, G^HiT^Hi and G^LoT^Hi also divided much faster than G^HiT^Lo LRCs and exhibited higher EdU incorporation but no changes in Cdk6 or in protein synthesis (supplemental Figure S4F-I). In single cell differentiation assay in vitro, G^HiT^Lo LRCs produced more multipotent clones (containing monocytes, megakaryocytes, neutrophils, and erythroid cells) than G^LoT^Lo cells (Figure 5I and supplemental Figure S4J-L). KIC supplementation, but not KIV or KMV, improved multilineage differentiation of G^LoT^Lo cells in vitro (Supplemental Figure S4M). G^LoT^Hi cells produced fewer multipotent clones than G^HiT^Lo and G^HiT^Hi (supplemental Figure S4N). Together, stemness correlates with higher Bcat2 expression, higher BCAA catabolism, slower cell cycle and lower mTOR activity. Bcat2 expression is reduced in HSCs that have divided in vivo, which correlates with faster cell division, higher nucleotide incorporation and mTOR activity. Boosting BCAA oxidation with KIC is sufficient to reduce mTOR activity and slow the cell cycle. Hence, LRC functional heterogeneity is coupled to a proliferation hierarchy via BCAA catabolism.

We next examined the effects of inhibiting Bcat activity in LSK-SLAM functions using an in vitro pharmacological approach (Figure 5J). A reversible inhibitor approach enables us to inhibit Bcat in HSC during their active division only, without inhibiting Bcat in the progeny, which is not possible in traditional genetic approaches. Bcat inhibition increased pS6 of cultured LSK-SLAMs, without affecting cell survival, compared to vehicle-treated LSK-SLAMs (Figure 5K and Supplemental Figure S5A-B). Bcat inhibition accelerated LSK-SLAM time-to-first division at the single cell level (Figure 5L). It also reduced multilineage potential of LSK-SLAMs in single cell differentiation assay (Figure 5N). In bulk culture, Bcat inhibition increased LSK-SLAM proliferation, increased EdU incorporation, without changing Cdk6 expression (Supplemental Figure S5C-F). Despite increased proliferation, Bcat inhibition did not increase TMRE (Supplemental Figure S5G). In serial competitive transplantation, the cultured cells treated with Bcat inhibitor had similar engraftment in primary recipients than vehicle-treated cells (Supplemental Figure S5H). But they generated more myeloid cells than lymphoid cells in the BM of primary recipients, despite normal LSK-SLAM and MPP frequency (Supplemental Figure S5I-J). Most notably, secondary recipients of Bcat inhibitor-treated cells exhibited a greater variability in PB lineage reconstitution with abnormally high frequency of either myeloid, B, or T-cells, depending on the recipient, compared to primary recipients, and had slightly less total donor-derived chimerism (Supplemental Figure S5K-N), which is reflective of decreased HSC function. Hence, Bcat activity couples HSC cell cycle latency and low metabolic activity with LT-HSC activity and lineage reconstitution fidelity.

A metabolite rescues the regeneration potential of HSCs with past division history.

We next tested if KIC supplementation in vitro restored G^LoT^Lo LRC repopulation potential. Upon serial transplantation (Figure 6A), 50% of the recipients of G^HiT^Lo LRCs showed robust reconstitution with high donor cell contribution (> 30%) in all 3 peripheral blood lineages in secondary recipients, which was scored as reconstitution with high-output repopulation potential. Recipients of G^LoT^Lo LRCs showed great variability with a wide range of donor cell contribution within each PB lineage. Only 1 out of 14 secondary recipients of G^LoT^Lo LRCs had high-output repopulation potential in secondary recipients (Figure 6B and Supplemental Figure S5O-R). Remarkably, 40% recipients of KIC-treated G^LoT^Lo LRCs exhibited high output and balanced reconstitution (Figure 6B-D). Increased balanced repopulation and T-cell competitiveness was apparent (Figure 6D). Hence, KIC supplementation is sufficient to restore HSC activity of LRCs that have history of divisions.

Figure 6. Alpha-Ketoisocaproate restores repopulation potential of HSCs with high divisional history.

(A) Schema of in vitro treatment of LRCs with KIC. (B) Donor-cell contribution in each PB lineage in secondary recipients at 20 weeks post-transplant (n=8, 14 , 17; respectively). Number of mice showing reconstitution above 30% in all 3 lineages, fisher exact test p=0.0454. (C) Frequency of secondary recipients showing reconstitution above 30% in all 3 lineages at 20 weeks post-transplant. (D) Donor cell contribution in each peripheral blood lineage in secondary recipients at 20 weeks post-transplant. Two-way ANOVA F=9.007, p=0.0039, sample effect. (E) Schema of in vivo KIC treatment. (F) Donor-cell contribution in each PB lineage in secondary recipients at 20 weeks post-transplant (n=15 and 14). (G) Proportion of secondary recipients exhibiting donor-cell contribution to all 3 PB lineages above 1%. Fisher exact test. Related to supplemental Figure S5-6.

To determine if KIC can improve the regenerative capacity of HSCs in vivo under replicative stress, in a clinically relevant model, mice were challenged with one dose of the myeloablative agent 5-Fluorouracil (5FU) and treated with KIC (Figure 6E). KIC treatment did not alter the peripheral blood cell recovery from 5FU-induced myeloablation (Supplemental Figure S6A-B). The BM was collected on Day 11 for competitive serial transplantation. Under these conditions, the HSC pool from 5FU-treated mice is less potent than non-5FU treated HSCs so that only 40% of secondary recipients were engrafted with donor-cell contribution above 1% in all 3 PB lineages, consistent with a negative effect of replicative stress on HSC functions (Figure 6F-G and Supplemental Figure S6E-G). In contrast, KIC treatment enhanced the long-term repopulation potential of the HSC pool reconstituted after 5FU challenge, enabling engraftment in 70% of secondary recipients (Figure 6F-G and Supplemental Figure S6E-G). These data suggest that KIC treatment protects HSCs from replicative stress in vivo without negatively impacting short-term PB recovery following myelosuppression.

Discussion

Our findings uncover paradigms that address long-standing issues of HSC biology. It has been unclear whether HSCs can perpetuate themselves (self-renewal) for a continuous and stable generation of blood cells, or whether HSC functions progressively decline through differentiating divisions, known as the generation-age hypothesis.^3–7,57,58 In this study, we provide evidence that HSC self-renewal is limited. Division history creates infidelity in blood lineage regeneration that carry a memory of replicative history. HSC functional decline with division history is driven by a change in branched chain amino acid usage from catabolic to anabolic purposes. This is reversible by replacing metabolites that the cells no longer produce. Our findings uncover a paradigm in which HSC functional decline with division history occurs through transcriptional and metabolic drift to generate a divergent bone marrow landscape but is plastic and can be reversed through metabolite replacement.

Integrating the label retention model with TMRE reveals heterogeneity in the LRC compartment that is comprised of HSC populations that have distinct behaviors in their repopulation activity, cell cycle, lineage output, and transcriptomic and metabolic state. Cells with low TMRE levels but different division history are molecularly close, and thus unlikely to represent completely unrelated HSC subsets. G^HiT^Lo cells can generate all phenotypically defined LRC subsets, as well as reconstitution patterns generated by G^LoT^Lo LRCs, in transplant studies. G^HiT^Lo but not G^LoT^Lo LRCs generate high output in each peripheral blood lineage. LRCs with higher levels of TMRE but low division history are cell cycle primed and produce fewer progeny but retain balanced lineage reconstitution. LRCs with high levels of TMRE and high division history are transcriptionally distinct, are cell cycle and megakaryocyte-primed, exhibit limited reconstitution potential. Hence, both division history and TMRE levels drive LRC heterogeneity and HSC functional decline. The findings suggest that the HSC pool is self-organized in a proliferation and metabolic hierarchy with G^HiT^Lo at the apex. Division history is one important factor that determines the lineage generation capacity of HSCs to limit the lifespan of the hematopoietic system. We found that HSCs with past divisions can maintain low TMRE, are not lineage-primed, differing from differentiating HSCs,^{4,9,13–15,23,59} but have faster time-to-first-division. HSCs with past divisions differ from old or exhausted post-transplanted HSCs, which tend to have low and myeloid-biased output, but exhibit a slower cell cycle entry and lower TMRE. At the population level, HSCs with past divisions can have long-term but less robust repopulation potential. The resulting mature lineage output is highly variable in relative mature lineage composition, but more surprisingly, is highly variable in relative proportion of discrete cell states within a given lineage along its differentiation path. In addition, individual cells derived from HSCs with past division history exhibit transcriptional divergence. Hence, division history causes functional and molecular heterogeneity that is associated with random lineage infidelity along the differentiation path rather than mere commitment to differentiation. Since HSCs with past divisions contain fewer repopulation units, they may need to replicate more in transplant studies⁵⁸ which could contribute to output variability. Future single cell transplantation will determine whether this reduced repopulation potential is homogenous. This study provides insights into how HSCs with distinct division history can coexist in the same environment and adopt distinct functional behavior within an organized model while theoretically not exposed to selection pressure, thus driving HSC functional heterogeneity. These findings suggest that the lifespan of the HSC pool and BM lineage development fidelity can be intrinsically pre-determined within the HSC pool and actively controlled by prior division history.

Prospective isolation of these cells allowed us to identify causes that drive HSC functional decline with division, and points to differences in how metabolic information is processed during active cell cycle, notably a change in usage of BCAAs, including leucine. Bcat2 expression is low in quiescent HSCs but drastically increases in activated HSCs along with high BCAA oxidation. Interestingly, the mitochondrial-targeted 2C-type Ser/Thr protein phosphatase (PPM1K), which promotes BCAA degradation, is critically important to maintain HSC functions in transplant settings.⁶⁰ BCAAs are building blocks for protein synthesis; leucine also directly controls mTOR activation. Conversely, BCAAs can be used as a source of energy being catabolized into metabolite intermediates in the mitochondria. One consequence of reduced Bcat2 activity is increased leucine availability to activate mTOR.^47,48 Hence, leucine availability serves as rheostat of mTOR activation.^61,62 mTOR is essential for HSC cell cycle.^54,55 Our findings suggest that BCAA catabolism is important during HSC activation to maintain its function, and that division history shifts BCAA usage towards mTOR activation to accelerate cell cycle entry and limit HSC activity. Hence, the division-dependent functional drift is driven by a cell-autonomous metabolic checkpoint, upstream of nutrient sensing, that couples HSC cell cycle latency with HSC activity (Supplemental Figure S9H). It will be important to determine how BCAA catabolism intermediates couple the cell cycle to stemness and cell division memory. Acetyl-CoA, generated downstream of BCAA catabolism, can enter the TCA cycle and fuel energy production. Alternatively, acetyl-coA can be transferred to the cytoplasm through the citrate shuttle and be used for de novo lipid synthesis.⁵¹ Lipids are needed for membrane-bounded organelle functions, can be used for energy storage or for energy production through fatty acid oxidation,⁶³ all of which are critical for HSC functions.^31,45,64,65 Production of BCAA intermediates can thus modulate energy and biomass to impact the cell cycle. Organelle segregation during mitosis is important for division memory and this can be impacted by mitochondrial activity.^32,44,66 AcetylCoA is also used for DNA or RNA epigenetic modification, an essential step for maintaining cell state and cell division memory.^67,68 Interestingly, the cell-autonomous model of HSC functional decline is plastic. We show that HSCs are highly responsive to changes in nutrients that shape lineage development and impact the lifespan of the hematopoietic tissue, such that a single metabolite can overcome the metabolic checkpoint and restore and/or protect HSCs from mitotic-driven functional decline. Interestingly, high fat diet, valine depletion, calory restriction or ketogenic diet, alter stem cell fates.^69,70 HSCs respond to subtle changes in vitamin A or C.^23,71 Some lipid intermediates, such as gamma-linoleic fatty acid, can enhance HSC regenerative potential in vivo when autophagy is defective.⁴³ This has important clinical implications for the use of dietary interventions for therapeutic purposes. In conclusion, our findings uncover a paradigm of stem cell biology that addresses a long-standing question in our understanding of HSC generation-aging. Importantly, it proposes metabolite replacement to overcome generation-aging functional decline, which has long-term clinical implications for HSC regenerative medicine.

Limitation of the study

Analysis of mitochondrial membrane potential (MMP) in HSCs is complex. While TMRE serves as a measure of MMP, HSCs express ATP-binding cassette transporters that efflux dyes such as TMRE, which can lead to lower fluorescence.³⁴ Without inhibiting efflux pumps, MMP quantification is less accurate. Efflux pump inhibitors such as verapamil, are toxic.³³ Verapamil inhibits calcium signaling, which is critical for HSC cell cycle entry and mitochondrial function. This would interfere with HSC functional studies, hence it was not used in this study.³³ We refer to TMREhi and TMRElo LRCs, as TMRE is not intended to be used to measure MMP accurately. Metabolomics remains challenging for small cell numbers, preventing us comparing directly G^LoT^Lo LRCs with and without KIC or perform stable isotope tracing from the same biological sample. The data can be highly variable and may underestimate differences between populations.

Resource availability.

Lead Contact:

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Marie-Dominique Filippi (marie-dominique.filippi@cchmc.org).

Materials Availability:

This study did not generate new unique reagents.

Data and Code Availability:

RNA-seq data were deposited into the Gene Expression Omnibus database under accession number GSE256529 for bulk RNA-seq at: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE256529 and GSE256530 for scRNA-seq at https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE256530

Metabolomics: This study is available at the NIH Common Fund’s National Metabolomics Data Repository (NMDR) website, the Metabolomics Workbench, https://www.metabolomicsworkbench.org where it has been assigned Study ID ST004298. The data can be accessed directly via its Project DOI: http://dx.doi.org/10.21228/M8V26M

The paper did not generate any new custom code.

All data supporting the findings of this study are available within the paper and its Supplementary Information. Summary of transplantation data is provided in Supplementary Table S1. Summary of cell clusters is provided in Supplementary Table S2.

STAR Methods:

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Mice:

Young (6–12 weeks) C57Bl/6 and transgenic R26-M2rtTA;Col1a1-tetO-H2B-GFP (H2B-EGFP) mice of both genders were bred in house. All animals were bred in house in pathogen-free environment. B6.SJL- Ptprc^a (BoyJ) mice were used as recipients of donor cells. Naïve H2B-EGFP mice (6–10 weeks old) were treated with doxycycline (1g/500ml, Sigma) and sucrose (5g/500ml, Sigma) in drinking water for 2 weeks to label the cells; mice were used for experiments 5 months following doxycycline removal. All studies were conducted with a protocol approved by the Institutional Animal Care and Use Committee of Cincinnati Children’s Hospital Medical Center.

METHOD DETAILS

Flow cytometry staining:

Hematopoietic stem cells were analyzed and isolated as described previously. ^32,72 BM cells were obtained by flushing the bones in Hanks’ Buffered Saline Solution (HBSS) containing 10% heat-inactivated fetal bovine serum (Atlanta). For LSK-SLAM-H2B-EGFP TMRE high or low isolation, bone marrow cells were enriched for c-Kit+ cells using c-Kit magnetic microbeads. C-Kit+ cells were separated using AutoMACS pro cell separator machine (Miltenyi biotec). C-Kit+ enriched cells were immuno-stained for specific surface markers for LSK-SLAM. All antibodies were from BD bioscience unless specified. Briefly, cells were incubated with biotin-labeled anti-mouse lineage antibodies (Ter119 (Cat 51–09082J), B220 (Cat– 51–01122J), Gr-1 (Cat - 51–01212J,), CD11b (Cat- 51–01712J), CD3e (Cat- 51–01082J) followed by staining for streptavidin, Sca-1, c-Kit, CD48, CD150 (eBioscience). Different combinations of fluorochromes were used, such as lineage V500/APCCy7 (Cat-561419and for APCCy7 Cat- 554063), Sca-1 PECy7 (clone-D7, Cat-558162), c-Kit – APCCy7/PE (clone-2B8, Cat-47–1171-82, Invitrogen and for PE Cat– 553355, , BD biosciences), CD48 AF700/FITC/BV605 (clone- HM48–1, Cat- 103426, , Biolegend, for FITC Cat – 557484, BD bioscience, and for BV605 Cat- 740353, BD bioscience), CD150 APC/FITC/PE (Clone – 9D1, Cat −17–1501-81, Invitrogen, for FITC Cat – 11–1501-82, eBioscience and for PE Cat – 12–1501-82, eBioscience). Then cells were washed and incubated at 37⁰C for 30 minutes with 0.1μM TMRE to measure mitochondrial membrane potential (Cat-T669 Invitrogen). Cells were washed and sorted using BD FACSAria II or BD/FACSymphony S6 cell sorter (BD Bioscience) within 60 minutes of mitochondrial staining.

BM analysis of mice following transplantation: for combined HSC and MPP analysis, unfractionated BM cells stained with various combination of fluorochromes such as streptavidin lineage V500/ APCCy7, Sca-1 PECy7, c-Kit – APCCy7/PE, CD48 AF700/FITC/BV605, CD150 APC/FITC/PE and CD45.2 Percpcy5.5/AF700, CD34 efluor 450 (clone- RAM34, Cat – 48–0341-82, eBioscience) and CD16/32 AF700 (clone-93, Cat- 12–0161-82) and CD45.2 Percpcy5.5/AF700 (Clone –104, Cat- 560693, and for percpcy5.5 Cat – 552950, BD bioscience). For post-transplant analysis of GFP and TMRE, BoyJ recipient mice were treated with doxycycline 2 months post-transplant for a 2-week period as described above. 5 months following doxycycline removal, bone marrow was isolated from mice as described above. Unfractionated BM cells were stained with biotin-labelled lineage antibodies followed by streptavidin V500, cKit APCCy7, Sca1 PECy7, CD48 AF700, CD150 APC, and CD452 BV421 (Cat-562895, BD, see above for all other antibodies). Cells were then washed and stained with TMRE as described above and CD45.2+ LSK SLAM and progenitors were analyzed for GFP vs TMRE expression.

Peripheral blood analysis of mice following transplantation: Peripheral blood was collected monthly for up to 24 months following transplantation. Red blood cells were lysed using lysis buffer (Cat - 555899, BD bioscience). Cells were immuno-stained for CD45.1 PE/BV605 (clone- A20, Cat- 553776 and for BV605 Cat- 563010, both BD Bioscience), CD45.2

Percpcy5.5/AF700/FITC ((Clone –104, Cat – 552950, for AF700 Cat- 560693, and for FITC Cat- 553772, BD bioscience) to identify recipient versus donor population. Cells were immuno-stained for blood mature lineages Gr-1 AF700/ FITC (clone- RB6–8C5, Cat- 557979, BD Bioscience and for FITC, Cat- 108419, Biolegend), Mac1 AF700/FITC (clone- M1–70, Cat - 557960, and for FITC Cat – 553310, BD bioscience), B220 PECy7 (clone -CD45R/B220, Cat- 552772,BD bioscience) , CD3e APC/PE (clone – 145–2C11, Cat- 561826, and for PE Cat- 553063, both BD bioscience) and CD8 APC/PE (clone – 53–6.7, Cat - 553035, and for PE Cat – 553032, both BD bioscience). Similar protocol was used for mature cell analysis in BM.

Transplantation:

Cells from each H2B-EGFP-TMRE group (200) were transplanted into lethally irradiated BoyJ recipient mice (1175 rads irradiation dose, 700+475 split dose 3 hours apart) with BoyJ whole BM competitor cells (200,000). Transplanted mice were used for experiments for 6 months after transplantation. Peripheral blood was collected by retro-orbital bleeding monthly to analyze percent donor chimerism in the myeloid and lymphoid population by flow cytometry. For secondary transplants, 2×10⁶ whole BM cells from each primary recipient was transplanted into lethally irradiated secondary recipients (2 secondary recipients per individual donor).

Single cell RNA sequencing using C1 fluidigm platform to examine the transcriptome of label retention cells:

Naïve H2B-eGFP mice (6–10 weeks old) were treated with doxycycline for two weeks to label the cells. Lin−Sca-1⁺c-Kit⁺CD48−CD150⁺ H2B-eGFP⁺ cells (LSK-SLAM eGFP⁺) and LSK-SLAM eGFP- were isolated (BD FACS Aria II, BD Biosciences). Cells were incubated in stemspan medium (Stem cell technology) supplemented with murine stem cell factor (SCF) and murine thrombopoietin (TPO; 100ng/ml each; Peprotech) at 37⁰C in 5% CO₂ incubator. After 38–40 hours in culture, cells were re-isolated based on H2B-eGFP dilution for single cell RNA sequencing to analyze transcriptome of cells that have divided once. The cells were loaded on 5–10micron chip. Single cells were confirmed visually under the microscope; cDNA and libraries were prepared by the CCHMC Gene Expression Core. Deep-sequencing was done by the CCHMC DNA core. Bioinformatic analyses – unsupervised ICGS, pairwise comparative analyses and PCA were performed to compare transcriptome between the groups using AltAnalyze. ³⁹ Two tailed empirical Bates moderated t-test from AltAnalyze was used with p<0.05 and fold>2 and minimum number of changed genes was 3. Gene ontology analyses of PCA genes was performed using ToppGene Suite and EnrichR softwares.

Single cell RNA sequencing using the 10X genomic platform:

scRNA-seq on murine bone marrow hematopoietic stem cell progenitors isolated from GHiTLo and GLoTLo-transplanted mice was done using droplet-based single-cell RNA-Seq (scRNA-Seq) analysis. All captures were performed using the Chromium 3’ version 3.1 kit (10X Genomics). Emulsion, Gel Beads-in-emulsion (GEM) collection, clean-up, and cDNA amplification were performed according to 10X Genomics protocols. All libraries were constructed with single index and sequenced on multiple Illumina S4 flow cells with PE150+10+10 setting (Illumina). Raw RNA-Seq BCL files were demultiplexed into FASTQ files for alignment and quantification in the 10x Cell Ranger version 6.1.2 workflow. Transcriptome was mapped to mm10 reference (mm10–2.1.0). Ambient RNA contamination was corrected for (15% contamination threshold) using the software SoupX v1.3.0. Corrected matrices were filtered for cells with >500 gene expressed, following counts per ten thousand (CPTT) normalization and log2 scaling in the software AltAnalyze version 2.1.4. To annotate single-cell barcodes in this dataset, we applied label transfer from a curated bone marrow cell atlas. This atlas is comprised of distinct bone marrow stem cell and progenitor flow cytometry captures. For single-cell label transfer, marker genes and centroids from this reference were derived using the software MarkerFinder and cellHarmony modules of the software AltAnalyze version 2.1.4, respectively, using default settings. Prior to label transfer, genes associated with cell cycle were excluded, to minimize cell classification bias during in vivo stimulation. Specifically, MarkerFinder gene sets enriched in cell cycle pathway genes (WikiPathways:WP190), using a Fisher Exact test p < 0.05, FDR corrected, for the were excluded from the final reference centroids. Stringent cellHarmony filtering (alignment score > 0.5) was applied to exclude potential doublets and other artifacts. Proportion of each cluster was normalized to 100; average cluster proportion in G^HiT^Low cells was calculated, ratio of cluster proportion in G^LoT^Low cells to the G^HiT^Low average was then calculated and shown in figure 2B. Average of G^LoT^Low clusters relative to G^HiT^Low is shown in figure 2C.

To identify differentially expressed genes, we averaged the gene expression for all cells for individual replicate per cell-type (pseudobulk) prior to computing differential expression. To address potential batch effects, differential expression analyses were performed separately on each batch of replicates (two batches – n=3 for batch 1, n=2 for batch 2) using the cellHarmony workflow (empirical Bayes t-test p<0.1), requiring that the final differentially expressed genes match across batches (cell-type, direction of regulation). Results are viewed through the second batch results for consistency. Significantly differentially expressed genes were computed using the software AltAnalyze. All gene expression heatmaps were produced in AltAnalyze using HOPACH clustering.

scRNA-seq performed on isolated LRCs was performed using the Chromium 3’ version 4 kit (10X Genomics) and similar pipeline as described above. Clusters were identified using ICGS-2 in AltAnalyze. All gene expression heatmaps were produced in AltAnalyze using HOPACH clustering. R package ggplot2 was used to visualize the data.

Bulk RNA-seq:

H2B-GFP SLAM (500) from each group were cultured in stemspan medium plus SCF and TPO for 15 hours. Three biological replicates were processed. Cells were lysed and cDNA was generated using the Smart-seq v4 Ultra Low Input RNA Kit (Takara/Clontech). A barcoded DNA library was then made using the Nextera XT DNA Library Preparation Kit (Illumina). The quality of the DNA library was assessed using the Agilent High Sensitivity DNA kit (Agilent Technologies) and an Agilent 2100 Bioanalyzer (Agilent Technologies). Sequencing was then done by Novogene. The open-source software Alt-analyze was then used to perform supervised hierarchical clustering and principal component analysis of the samples, as previously described ^32,39. Gene ontology and pathway analysis were performed in Alt-Analyze. ^32,39 Differentially expressed genes were then analyzed using the ENRICHR database. ⁷³

Metabolomics:

1200 H2B-GFP SLAM from each group were cultured in StemSpan medium plus SCF and TPO for 20 hours. Cells were washed with PBS and cell pellets were snap frozen using liquid nitrogen. Sample extraction and UHPLC was performed by the Mass Spectrometry Metabolomics Shared Resource Facility at the University of Colorado Anschutz Medical Campus. Cell pellets were suspended in 100 ul of methanol/acetonitrile/water (5/3/2 v/v/v) pre-chilled to −20°C, vortexed for 30 minutes at 4°C, and spun down at 18,000 x g for 15 minutes, 4°C. The entire supernatant fraction was transferred into autosampler vials (Thermo, product 03–340-621) and dried completely under vacuum centrifugation (Labconco, cat #7810016). Dried extracts were resuspended in 12 ul of 0.1% formic acid and 5 ul of extract was injected per polarity mode. Extracts were analyzed by UHPLC-MS (Vanquish, Exploris 120 – Thermo Fisher, San Antonio, CA, USA) as detailed in previous methods papers. ⁷⁴ For analysis of each metabolite, abundance values for each sample were normalized to the average of the values for G^HiT^Low cells for each experiment and mass spectrometry test.

Ex vivo culture:

cKit-enriched or sorted LSK-SLAM were cultured in Stemspan containing SCF+TPO (100 ng/ml each, Peprotech) and IL-6+IL-11 (20ng/ml each, Peprotech) at 37⁰C in 5% CO₂ incubator, for up to 3 days. For in vitro experiments, cells were treated with either DMSO or PBS controls, a-KIC (1mM, Cat-21052, Cayman Chemical), BCAT inhibitor (10uM, Cat-9002002, Cayman Chemical), or Rapamycin (10nM, MedChemExpress #HY-10219). For post-culture transplantation studies. 500 LRCs, cultured with or without KIC were then mixed with 250,000 unfractionated BoyJ BM cells and transplanted into lethally irradiated BoyJ recipients. 1,000 LSK-SLAM, treated with Bcat Inhibitor or vehicle, were then mixed with 250,000 unfractionated BoyJ BM cells and transplanted into lethally irradiated BoyJ recipients. Peripheral blood was collected by retro-orbital bleeding at 4, 8, 12, 16, 20, and 24 weeks post-transplantation to analyze donor chimerism. We calculated the ratio of donor contribution to myeloid, B, and T cells in each recipient mouse, and classified the reconstitution as myeloid-dominant, balanced, or lymphoid-dominant according to the criteria previously assigned to similar patterns.³⁶ For the secondary transplant, 2 million unfractionated bone marrow cells were transplanted per recipient, two recipients per donor.

Immunofluorescence:

LSK-SLAM from the indicated group of cells were cultured in StemSpan with SCF and TPO for the indicated amount of time. For BCAT2 and pS6, cells were seeded on retronectin-coated chamber slides and fixed using 4% paraformaldehyde for 15 minutes. Cells were then permeabilized using 0.1% Triton X-100 and blocked with 2% BSA in PBS for 20 minutes at room temperature. Primary or conjugated antibody staining was performed overnight at a 1:100 dilution at 4C (BCAT2- abcam ab95976 and pS6- BD #560432). Secondary antibody staining was performed using goat anti-rabbit 546 at a 1:100 dilution (Life Technologies, Cat- A11010). Cells were mounted using ProLong Glass with NucBlue (Invitrogen,Cat-P36981). EdU incorporation was detected using click chemistry and immunofluorescence (Cat-K1076, APEx Bio). LSK-SLAM from each group were cultured as described above in the indicated conditions. After 15 hours, half the media was removed from each well and replaced with fresh StemSpan media containing 20uM EdU for a final concentration of 10uM EdU for 3 hours at 37C, 5% CO2. For the final hour of incubation, cells were seeded on glass bottom dishes coated with retronectin. Fixation, permeabilization, click reaction, and nuclear labelling were performed according to manufacturer’s instructions. Cells were imaged using a 60X objective on a Nikon C2 confocal system on a Nikon Ti Inverted microscope. Acquisition was performed using Nikon Elements. The proportion of EdU positive cells in each condition was quantified. For CDK6 analysis, cells were incubated with primary antibody at a 1:200 dilution overnight at 4C (Cat- 14052–1-AP, Proteintech). Cells were stained with goat anti-rabbit 546 (Life Technologies, Cat-A11010) 1:100 for 1h at room temperature and cells were imaged as described above. OP-Puro incorporation was detected using click chemistry and immunofluorescence (Cat- C10458, Fisher Scientific). LSK-SLAM from each group were cultured as described above in the indicated conditions for 10–12 hours. For the final hour of incubation, cells were seeded on glass bottom dishes coated with retronectin. Media was replaced with fresh StemSpan media containing 10uM OP-Puro and cells were incubated for 10 minutes at 37C. Fixation, permeabilization, click reaction, and nuclear labelling were performed according to manufacturer’s instructions. Cells were imaged using a 60X objective on a Nikon C2 confocal system on a Nikon Ti Inverted microscope. Acquisition was performed using Nikon Elements. Mean fluorescence intensity of the OPP signal was calculated using Imaris image analysis software.

In vitro division kinetic and multilineage differentiation at the single cell level:

Single LSK-SLAM cells from indicated group of cells were sorted in 60-well terasaki plates (Greiner-Bio-One). Single cells were visually confirmed under light microscope. Cells were cultured in serum free Stemspan medium (Stem Cell Technology) supplemented with murine SCF and murine TPO (100 ng/ml, each, Peprotech) during 72 hours at 37⁰C in 5% CO₂ incubator. Numbers of cells per well were scored every 12 hours, under the light microscope to determine the division kinetic. A first division was scored when two cells could be observed; a second division was scored if three or four cells were observed. Data was expressed as percent cumulative division at every interval. Resulting clones were then individually cultured in Iscove’s Modified Dulbeco’s Medium (IMDM) containing 10% fetal bovine serum (Omega) and a cocktail of cytokines allowing for myeloid differentiation (murine SCF, murine TPO, human G-CSF (20 ng/ml), murine IL-3 (50 ng/ml) and EPO (4 U ml⁻¹; Espogen)) for 14 days at 37⁰C in 5% CO₂ incubator. Clones were harvested and used for cytospin preparation. Cells of various lineages were identified based on their morphology after Diff-quick staining (Siemens). Clones were examined for the presence of neutrophils (n), erythroid cells (e), macrophages (m) and megakaryocytes (M; nemM) and scored as multipotent clones if contain nemM cells. In some experiments, cells were incubated with Bcat Inh (10uM), KIC (1mM), KIV (1mM, Sigma, Cat-198994), or KMV (1mM. Sigma, Cat- K7125) or vehicle control DMSO or PBS for the duration of the division kinetic; then clones were cultured in differentiation media for 2 weeks in the absence of treatments.

Cell Cycle analysis.

BM Cells from H2B-GFP 5 months after dox treatment were stained for LSK-SLAM cell surface markers (see above), fixed, permeabilized (Cat- 554714, BD) and then stained for Ki67 (1:100 dilution, eBioscience) overnight. Cells were then incubated with Hoechst 33342 (10 ug ml⁻¹ , Invitrogen) for 30 minutes at 4 C in the dark.

In vivo KIC treatment:

5 fluorouracil was injected intravenously, 5-FU (150mg/kg). Mice were injected with alpha-ketoisocaproate (KIC) (1mg/kg per mouse, Cayman) or PBS intraperitoneally every day for seven days following 5-FU challenge. All mice were sacrificed at the indicated time for analysis. BM was collected at day 11 and unfractionated BM cells (140,000) were transplanted into lethally irradiated BoyJ recipients with 200,000 BM BoyJ competitor cells. Secondary transplantation was performed as described above.

QUANTIFICATION AND STATISTICAL ANALYSIS

The results are presented as mean ± SD or as mean ± sem with individual values. Unless specifically indicated, data were analyzed using unpaired student’s T-test; or ANOVA, p values are indicated for each analysis. Analysis of repopulation activity was analyzed using contingency table Fisher exact test comparing frequency of mice presenting with high regenerative potential or low regenerative or myeloid-dominant, balanced or lymphoid-dominant reconstitution. For immunofluorescence, pS6, EdU, OPP, and CDK6 expression were analyzed either as percent of positive cells or MFI and paired student’s T-Test of inhibitor or supplement condition relative to vehicle control, or one-way ANOVA with multiple comparisons if there were 2+ conditions. Percent of multipotent and non-multipotent clones was analyzed using 2x contingency table Fisher exact test. Comparative analysis of proportion of donor-derived cells within each lineage of G^LoM^Lo and G^LoM^Lo+KIC in figure 6 was done using two-way ANOVA test.

Supplementary Material

2

Table S1: Summary-transplant-related to Figure 1

3

Table S2: Summary of cell clusters-related to Figure 2

4

Table S3: Pseudobulk-capture-1–2-related to Figure 2

5

Table S4: Differentials-scRNA-seq- related to Figure 2

6

Table S5: S2M- hierarchical-clustering- related to Figure 2

7

Table S6: Markers-scRNAseq-4groups- related to Figure 3

8

Table S7: Bulk-RNA-seq-matrix-clusering- related to Figure 3

9

Table S8: Metabolomics-related to Figure 4

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies: Flow cytometry
Ter119 – biotin labeled	BD Bioscience	51–09082J; RRID:AB_10053179
B220 - biotin labeled	BD Bioscience	51–01122J; RRID:AB_10053179
B220 – PECy7	BD Bioscience	552772; RRID:AB_394458
Gr-1 - biotin labeled	BD Bioscience	51–01212J; RRID:AB_10053179
Gr-1 – AF700	BD Bioscience	557979; RRID:AB_396971
Gr-1 - FITC	Biolegend	108419; RRID:AB_493480
Mac1 - biotin labeled	BD Bioscience	51–01712J; RRID:AB_10053179
Mac1 – AF700	BD Bioscience	557960; RRID:AB_396960
Mac1 - FITC	BD Bioscience	553310; RRID:AB_394774
CD3e - biotin labeled	BD Bioscience	51–01082J; RRID:AB_10053179
CD3e - APC	BD Bioscience	561826; RRID:AB_10896663
CD3e - PE	BD Bioscience	553063; RRID:AB_394597
CD8 - APC	BD Bioscience	553035; RRID:AB_398527
CD8 - PE	BD Bioscience	553032; RRID:AB_394570
Streptavidin – V500	BD Bioscience	561419; RRID:AB_10611863
Streptavidin – APCCy7	BD Bioscience	554063; RRID: AB_10054651
Sca-1 - PECy7	BD Bioscience	558162; RRID:AB_647253
c-Kit – APCef780	Invitrogen	47–1171-82; RRID:AB_1272177
c-Kit -PE	BD Bioscience	553355; RRID:AB_394806
CD48 – AF700	Biolegend	103426; RRID:AB_10612755
CD48 - FITC	BD Bioscience	557484; RRID:AB_396724
CD48 – BV605	BD Bioscience	740353; RRID:AB_2740086
CD150 - FITC	eBioscience	11–1501-82; RRID:AB_465209
CD150 - PE	eBioscience	12–1501-82; RRID:AB_465873
CD150 - APC	Invitrogen	17–1501-81; RRID:AB_469441)
CD45.2 – percpcy5.5	BD Bioscience	552950; RRID:AB_394528
CD45.2 – AF700	BD Bioscience	560693; RRID:AB_172749
CD45.2 – FITC	BD Bioscience	553772; RRID:AB_395041
CD45.2 – BV421	BD Bioscience	562895; RRID:AB_2737873
CD45.2 - PE	BD Bioscience	560695; RRID:AB_1727493
CD34 – ef450	eBioscience	48–0341-82; RRID:AB_2043837
CD16/32	Invitrogen	12–0161-82; RRID:AB_465568
CD45.1 - PE	BD Bioscience	553776; RRID:AB_395044
CD45.1 – BV605	BD Bioscience	563010; RRID:AB_2737948
CD45.1 – BV421	BD Biosciences	563983; RRID:AB_2738523
EPCR-PerCP-eFluor710	eBioscience	46–2012-80; RRID:AB_10718383
CD41-BV605	Biolegend	133921; RRID:AB_2563933
Ki-67 PE	Biolegend	652404; RRID:AB_2561525
Mitotracker dyes: Flow cytometry
TMRE	Invitrogen	T669
Antibodies: Microscopy / Imaging
anti-rabbit – AF488	Invitrogen	A21206; RRID:AB_2535792
anti-rabbit -- AF546	Life technologies	A11010; RRID:AB_2534077
anti-mouse -- AF555	Life technologies	A21422; RRID:AB_2535844
Anti-BCAT2	Abcam,Inc	Ab95976; RRID:AB_10677595
Anti-pS6	Cell Signaling Technology	4858S; RRID:AB_916156
Anti-pS6	BD Biosciences	560432; RRID:AB_2736905
Anti-CDK6	Fisher Scientific	14052–1-AP; RRID:AB_10642144
Chemicals
Doxycycline	Sigma- Aldrich	D9891–25G
Sucrose	Sigma- Aldrich	501212905
DMSO	Fisher Scientific	BP231–100
Paraformaldehyde	Electron Microscopy Sciences	15713
Triton X-100	Sigma-Aldrich	X10–100ml
Bovine serum albumin	Roche	3117332001
SlowFade gold antifade mountant with DAPI	Invitrogen	S36939
Prolong Glass Antifade Mountant with NucBlue	Invitrogen	P36981
Fetal bovine serum	Atlanta	S11550
Fetal bovine serum	Omega	FB-01
Diffquick staining	Siemens	B4132-A
Bcat Inhibitor	Cayman Chemical	9002002
KIC	Cayman Chemical	21052
KMV	Sigma-Aldrich	K7125
KIV	Sigma-Aldrich	198994
Rapamycin	MedChemExpress	HY-10219
OPP-AF647 Click-it Kit	Invitrogen	C10458
EdU Imaging Kit Cy5	APEx Bio	K1076
Media:
Stemspan Serum-Free Expansion Medium	Stem Cell Technologies	9650
Hank’s buffered saline solution	Fisher Scientific	21–020-CV
Iscove’s Modified Dulbeco’s medium	Fisher Scientific	10–016-CV
Cytokines:
Murine stem cell factor	Peprotech	250–03
Murine thrombopoietin	Peprotech	315–14
Murine IL-6	Peprotech	216–16
Murine IL-11	Peprotech	220–11
Human granulocyte colony stimulating factor		GF05–20ug
Murine interleukin-3	Peprotech	213–13
Erythropoietin	Espogen	NDC55513–126-10
Experimental Models: Strains
B6.SJL- Ptprca (BoyJ)	In house, CCHMC	RRID:IMSR_JAX:002014
C57Bl/6	In house, CCHMC	RRID:MGI:3028467
Transgenic R26-M2rtTA;Col1a1-tetO-H2B-GFP	Jackson laboratories	Stock number: 016836
Software and Algorithms
FlowJo	FlowJo™ Software	https://www.flowjo.com/
GraphPad Prism 5.0	GraphPad Software Inc.	http://www.graphpad.com/scientific-software/prism/
ImageJ	NIH	https://imagej.nih.gov/ij/
Imaris	Oxford Instruments	https://imaris.oxinst.com
AltAnalyze	Dorothea Emig, Nathan Salomonis et al	http://www.altanalyze.org
Toppgene	Chen et al	https://toppgene.cchmc.org
EnrichR	Chen et al; Kuleshov et al	https://amp.pharm.mssm.edu/Enrichr/
Metaboanalyst	Pang et al	https://www.metaboanalyst.ca/MetaboAnalyst/
RStudio	Posit, PBC	RStudio Desktop - Posit
Deposited Data
scRNA-seq	This Paper	GEO: GSE256530
Bulk RNA-seq	This Paper	GEO: GSE256529
Metabolomics	This Paper	http://dx.doi.org/10.21228/M8V26M

Highlights:

• Accrued divisions causes HSC functional drift with lineage output infidelity

• Accrued division causes transcriptional drift in HSCs and in resulting lineages

• Accrued division redirects BCAA usage from catabolic to anabolic activity in HSCs

• Division-dependent HSC functional drift can be restored by metabolite replacement