Integrating transcription-factor abundance with chromatin accessibility in human erythroid lineage commitment

Design

The main challenge with combining ATAC-seq with intracellular protein quantification lies in the fixation and permeabilization of cells required for direct measurement of intracellular proteins using affinity reagents, such as antibodies, and subsequent paraformaldehyde crosslinking reversal at high temperatures prior to library amplification. The previously applied 65°C crosslink reversal step results in dissociation and loss of shorter fragments that contribute to a large fraction of the final ATAC-seq library, resulting in reduced library complexity and lower TSS enrichment scores (Chen et al., 2018). To improve data quality, the InTAC-seq protocol uses a shorter fixation time with mild permeabilization and reversal of formaldehyde crosslinks with a catalyst to allow crosslink reversal to occur at 37°C (Karmakar et al., 2015; Figure 1A).

InTAC-seq libraries are comparable in quality to ATAC-seq libraries generated from fresh cells

Libraries generated from fixed GM12878 cells using InTAC-seq exhibited enrichment in Tn5 insertions at TSSs comparable to TSS enrichment in ATAC-seq libraries from unfixed live cells and approximately 2-fold greater than that in published bulk piATAC libraries (Chen et al., 2018; Figure 1C). The fragment length distribution for InTAC-seq libraries was also similar to the distribution in libraries from live cells due to preservation of sub-nucleosomal fragments at 37°C (Figure S1A). The retention of sub-nucleosomal fragments translated into a larger estimated number of unique ATAC-seq fragments from InTAC-seq and live ATAC libraries relative to piATAC libraries (Figure 1D). To compare InTAC-seq libraries with unfixed libraries, we plotted the correlation of reads in consensus peaks and found a strong correlation between InTAC-seq data generated from fixed cells and ATAC-seq data generated from live cells (Figures 1E and S1B). Genome Browser tracks further illustrated the concordance between InTAC-seq data and live-cell ATAC-seq data (Figure 1B). We further demonstrated that InTAC-seq can be performed on as low as 100 cells, enabling profiling of rare cell types (Figure S1C). Together, these results show that the InTAC-seq protocol produces data on par with live ATAC-seq, with comparable signal-to-background (as defined by TSS enrichment scores) and library complexity.

InTAC-seq is sufficiently sensitive to detect differences in chromatin accessibility associated with varying TF protein abundance

We next aimed to use InTAC-seq to detect chromatin accessibility differences in cells endogenously expressing different levels of chromatin-binding protein such as TFs. We compared the chromatin-accessibility profiles in K562 cells expressing the highest or lowest 15% of GATA-1 levels using InTAC-seq (Figures 1F and 1G). With ChromVAR (Schep et al., 2017), we found that GATA-binding motifs were among the most variable in accessibility across samples (Figure S1E). Specifically, we observed increased accessibility in cells with high levels of GATA-1 at GATA-1 motif sites (Figures S1G, S1H, and 1H). Differentially accessible peaks (false discovery rate [FDR] <0.1) between cells with high versus low GATA-1 were almost exclusively more accessible in GATA-1-high cells, and they were most significantly enriched for GATA motifs relative to all other peaks present across samples (Figures 1H and S1F), suggesting that accessibility differences are likely due to differences in GATA-1 abundance. However, we observed slight increases in accessibility at motif sites for other erythroid TFs in GATA-1-high cells, which could indicate that GATA-1-high cells within a K562 culture are further along in erythroid differentiation. It therefore remains possible that the increased accessibility at GATA-1 sites could be due to factors other than GATA-1.

We next asked how increases in accessibility at GATA-1 motif sites can vary based on the predicted binding affinity of the putative GATA-1 binding site. To address this, we grouped all consensus peaks into 20 bins of equal size based on the quality of their GATA-1 motif score and measured the average accessibility across these bins in cells with high versus low GATA-1 abundance. We found that cells with high GATA-1 levels exhibit a general increase in accessibility at GATA-1 motif sites above a threshold motif score that likely represents the minimum score for a true GATA-1 binding site (Figure 1I). However, we observed that the greatest change in accessibility between cells with high versus lower GATA-1 occurred at sites of moderate predicted affinity (i.e., bin 18), suggesting that the highest-affinity GATA-1 binding sites may be saturated even at lower GATA-1 levels (Figure 1J). These results showcase the ability of InTAC to measure subtle differences in chromatin accessibility among populations, allowing us to link natural variation in TF abundance to accessibility differences at specific genomic loci such as putative TF-binding motif sites.

InTAC-seq links high expression of GATA-1 protein to erythroid-committed bone marrow (BM) progenitors

While clinically significant for both cell-based therapies and hematopoietic dysplasia (Zivot et al., 2018), the regulatory landscape of erythropoietic cell homeostasis in the human bone marrow (BM) compartment is not well understood. BM progenitor populations are traditionally defined and isolated based on expression of cell-surface proteins, which viably preserves these cells for downstream functional assays (Akashi et al., 2000; Manz et al., 2002; Mori et al., 2015; Seita and Weissman, 2010). However, these surface molecules are not always functionally related to the cellular state that we associate them with, thereby resulting in oversimplification of the complex hematopoietic system (Paul et al., 2015). We therefore hypothesized that intracellular regulatory protein abundance would identify cellular states with higher fidelity and enable more accurate molecular characterization. To test this, we focused on human erythroid progenitor cells, which are regulated by GATA-1 (Gutiérrez et al., 2020; Stachura et al., 2006; Suzuki et al., 2003) but are generally defined by unrelated surface molecules, such as a CD45RA-negative population separated based on FLT3 (Doulatov et al., 2010) or IL3RA (CD123) (Manz et al., 2002), with the latter gating subsequently referred to as CD123^– megakaryocyte-erythroid progenitor cells (MEPs; fully defined as CD34⁺/CD38⁺/CD10^–/CD45RA^–/CD123^–). We applied InTAC-seq to interrogate the link between the abundance of the lineage-defining TF, GATA-1, and epigenetic changes related to GATA-1 abundance differences in RBC development.

First, we used InTAC-seq to profile the accessible chromatin landscape of GATA-1-high- (top 8%) and GATA-1-mid and low-expressing cells (referred to as GATA-1-mid/low cells in the lower 87%) within the landscape of the general BM progenitor compartment (i.e., CD34⁺CD38⁺) to determine if GATA-1-high cells were enriched for erythroid epigenetic signatures (Figure 2A). InTAC-seq libraries generated from these isolated subpopulations had TSS enrichment scores similar to ATAC-seq libraries from live GM12878 cells, indicating that InTAC-seq performs well on primary human samples (Figures S2A and S2B). We observed a marked increase in accessibility at regulatory regions surrounding the GATA-1 locus in GATA-1-high versus mid/low progenitors, consistent with GATA-1 expression levels, along with a decrease in accessibility at regulatory regions within the SPI1 locus, which encodes a TF known to antagonize GATA-1 activity and repress erythroid commitment (Arinobu et al., 2007; Nerlov et al., 2000; Rekhtman et al., 1999; Zhang et al., 2000) (Figure 2B). The binding motifs for these two TFs also exhibited strong differences in accessibility between GATA-1-high and GATA-1-mid/low progenitors (Figure 2C). We further observed a broader trend of increased accessibility surrounding the motifs for TFs that drive erythroid fates (e.g., GATA-1, Mecom [Shimizu et al., 2002]) and decreased accessibility at motifs for TFs that drive myeloid and lymphoid lineages (e.g., SPI1 [Hromas et al., 1993; Klemsz et al., 1990], EBF1 [Nechanitzky et al., 2013]) in GATA-1-high progenitors (Figure 2C). Analysis of differentially accessible peaks showed enrichment of these erythroid TF motifs in sites more accessible in GATA-1-high progenitors and enrichment of a variety of myeloid and lymphoid TFs in sites more accessible in GATA-1-mid/low-expressing progenitors (Figures 2D, 2E, and S2C). Deeper analysis into accessibility at GATA-1 motif sites of varying binding affinities across the genome showed that high GATA-1 abundance in progenitors is associated with marked increase in accessibility above a threshold motif score (∼12; Figure 2F). The greatest increases were observed at sites with the highest predicted GATA-1 affinity, suggesting the preferential binding of GATA-1 to the genome at these sites. Altogether, these data suggest that high GATA-1 expression in human hematopoietic progenitors is likely linked to an erythroid epigenetic program while repressing alternative lineage fates.

To further assess the position of these progenitors within the hematopoietic hierarchy, we projected our InTAC-seq samples onto a scATAC-seq uniform manifold approximation and projection (UMAP) of filtered BM mononuclear cells (BMMCs) from published datasets to identify their closest hematopoietic cell type using ArchR (Granja et al., 2019, 2020). Given our focus on erythropoiesis, we performed a more fine-grained analysis of the erythroid cluster and separated it into early, mid, and late erythroid progenitor populations based on differences in gene accessibility and chromVAR deviation scores of key TFs (Figures 2G, S2D, and S2F). Despite diffuse GATA1 expression across hematopoietic populations, we see highest enrichment of GATA1 gene expression at the erythroid arm of UMAP (Figure 2H). Likewise, when projected, we found that progenitors with high GATA-1 protein abundance, represented by the top 8% of GATA-1-expressing BM progenitor cells, were positioned within the mid erythroid progenitor branch while mid/low-GATA-1-abundance progenitors spanned scRNA-annotated granulocyte/monocyte progenitor (GMP), lymphomyeloid primed progenitor (LMPP), and common myeloid progenitor (CMP) clusters (Figure 2I). Strikingly, the position of GATA-1-high progenitors at the mid erythroid cluster (Figure 2I) combined with the overall enrichment of erythropoietic programs (Figures 2C–2E) suggests that the GATA-1 highest expressing cells are most similar to progenitors in the erythroid fate arm of the BM map.

Single-cell proteomic map of erythroid development complements InTAC-seq to reveal GATA-1 co-expression patterns

We next sought to better understand GATA-1 protein abundance dynamics in relation to other key TFs in erythroid progenitor development. We curated a 35-plex antibody panel comprised of key TFs implicated in hematopoiesis, such as GATA-1, GATA-2, BMI-1, and PBX-1 (Figure 3A). The panel also included surface markers for BM progenitor gating strategies and markers previously shown to be involved in erythropoiesis to better understand their dynamics relative to GATA-1 in erythroid commitment (Figure 3A). In order to construct a map of the BM progenitor space, we embedded our density-downsampled dataset from ∼1 million BM progenitors into a force-directed layout (Jacomy et al., 2014) and visualized the topology of cell-state distributions in high-dimensional space (Traag et al., 2019; Wolf et al., 2018) (Figures 3B and S3B). We then overlaid manually gated, previously defined progenitor population labels (Manz et al., 2002) (Figures 3B, left, and S3A, progenitor gating). Expectedly, reference populations do not project in the previously established hierarchical manner and instead present as a continuum of states from early progenitor to distinct endpoints characterized by an erythroid-primed branch (indicated by GATA-1), CMPs or GMPs (indicated by CD123), common lymphoid progenitors (CLPs; indicated by CD10), and myeloid cells (immature myeloid indicated by CD117 and mature myeloid by CD33) (Paul et al., 2015) (Figures 3B, right, 3C, and S3C, surface marker distributions).

The CD123^– MEP population (Manz et al., 2002) occupies a large portion of the map indicating heterogeneity of states within this population, which is also exemplified by numerous Leiden clusters in this space that show enrichment of distinct cell states (Figures 3B, bottom right, 3D, 3E, S3E, and S3F). Previous work has shown myeloid-primed states existing within CD123^– MEPs (Manz et al., 2002), and we also observe these cells in clusters 3 and 4, marked by high CD117 and CD33 (Psaila et al., 2016) (Figure S3F). The large number of BM cells assayed (about 1 million) also enabled us to detect a small frequency of lymphoid-primed CD123^– MEPs marked by mid to high CD45RA and high CD10 expression in cluster 5 (Figure S3F). In contrast, cells expressing the highest GATA-1 are primarily localized within the erythroid arm (Figure 3C, orange arrow). Together, these results confirm the lineage-priming heterogeneity present in the CD123^– MEP population and localize cells with high expression of GATA-1 to erythroid-primed progenitor space.

Next, we aimed to better understand changes in levels of and relationships between functional markers in erythroid progenitor development with respect to GATA-1. We constructed a differentiation trajectory from hematopoietic stem cells (HSCs) to high GATA-1 cells (i.e., putative erythroid progenitors) using computed diffusion pseudotime (DPT) (Haghverdi et al., 2016) (Figures 3F, red arrow, and 3G). Notably, the erythroid arm is adjacent to early progenitors (both in single-cell proteomic and transcriptomic space), corroborating previous claims of erythroid progenitors arising directly from HSCs without intermediate states (Grinenko et al., 2018) (Figures 3F and S3G). When binned across normalized DPT, the ordering of cells enabled us to quantify key TF and surface marker trends through erythroid commitment (Figure 3G). We observe that CD71 and CD84, both known markers of erythroid commitment (Psaila et al., 2016; Sanada et al., 2016; Zaiss et al., 2003), increase during erythroid differentiation (Figures 3G and S3D). Our trajectory also showed expected trends in erythroid lineage-priming TFs, with GATA-2 being expressed earlier than GATA-1 in early progenitors (clusters 2 and 7, respectively) (Suzuki et al., 2013) (Figures 3G and S3F). Interestingly, stemness maintenance TFs, PBX-1 and BMI-1, that have previously been implicated in mouse RBC development (Kim et al., 2015; Manavathi et al., 2012) showed expression in erythroid progenitor clusters, indicating a potential role in human erythroid development (Figures 3G, 3C, and S3D). PBX-1 is likely acting to promote self-renewal in lineage-primed progenitor pools as indicated by its expression in early myeloid and early erythroid progenitor clusters (clusters 3 and 7) (Figure S3F). PBX-1 and GATA-1 show the highest mutual information only in cluster 8 despite lower PBX-1 abundance, supporting their known co-regulation through human PBX-1-interacting protein (PBXIP1/HPIP) (Manavathi et al., 2012) (Figure 3H). Additionally, we observe co-expression of BMI-1 with GATA-1, which is supported by previous work implicating BMI-1 in regulating self-renewal and ribosome biogenesis in erythroid progenitors at late-stage erythroid commitment (Gao et al., 2015; Kim et al., 2015; Liu et al., 2021) (Figure S3D). BMI-1 abundance increases with GATA-1 expression only toward later-stage erythroid progenitor states (cluster 8), hinting at unipotency at this stage (Figure 3G). Our single-cell, high-dimensional proteomic map complements our InTAC-seq assay in revealing trends of relevant surface markers and TFs in GATA-1 high cells within the context of erythroid progenitor development in human BM.

GATA-1-high BM progenitors exhibit significant clonal enrichment for erythroid-committed cells

To functionally assess erythroid progenitor commitment in high-GATA-1-expressing cells, we sought to benchmark them against CD123^– MEPs (fully gated as CD34⁺CD38⁺CD10⁻CD123^–CD45RA^–) using a colony-forming assay for clonal hematopoietic lineage potential. In order to isolate viable cells with high GATA-1 abundance for the assay, we first identified surface-protein surrogates for GATA-1 using our BM cytometry by time-of-flight (CyTOF) data. We found that CD71 and CD84, both known to be enriched in erythroid cells (Psaila et al., 2016; Sanada et al., 2016; Zaiss et al., 2003), best represented high GATA-1 expression (Pearson correlation = 0.372 and 0.378, respectively; Figure 4A). The myeloid-enriched surface protein CD33 was also mildly anti-correlated with GATA-1 (Pearson correlation = −0.079). Additionally, when we selected the highest expressing GATA-1-high progenitors (Figure 4B) at a frequency similar to that used to sort for GATA-1-high progenitors in the InTAC-seq experiment (∼8%), we observed an enrichment for CD71 and CD84 and a depletion for CD33 in the high-GATA-1-expressing cells relative to mid/low-GATA-1 progenitors (Figures 4B and 4C). A data-driven, back-gating approach (Aghaeepour et al., 2018) was employed, and CD33, CD84, and CD71 were predicted to best gate for the high-GATA-1-expressing target population with an F score of 0.99 (Figures S4A and S4B, red cells). By selecting for CD71⁺CD84⁺CD33⁻ cells within the CD34⁺CD38⁺ compartment, we confirmed that these cells have higher expression of GATA-1 relative to cells in the CD123^– MEP gate (Figures S4C and S4D).

We then compared the hematopoietic colony-forming potential of the CD71⁺CD84⁺CD33⁻ (i.e., GATA-1-high) population with CD123^– MEPs using a methylcellulose assay. While it is known that CD123^– MEPs are heterogeneous (Mori et al., 2015; Sanada et al., 2016) and contain erythroid progenitors as well as other lineage-primed progenitors (Figures 4E and S4D), in comparison, our GATA-1-high surrogate cell population (CD71⁺CD84⁺CD33⁻) was significantly (p = 0.048) enriched for erythroid (BFU-E) colonies (Figures 4E and S4D). These data confirm that the highest GATA-1 expressing cells within human CD34⁺CD38⁺ progenitors have erythroid potential by functional colony-forming assay and are consistent with the epigenetic programs we observed in our InTAC-seq data.

Consistent with their mixed-lineage potential demonstrated by colony-forming unit (CFU) analysis, the CD123^– MEPs have a spread of expression for GATA-1 (Figure 5A). To examine the nature of high-GATA-1 expressing cells from within CD123^– MEPs and how they compare to the high GATA-1 CD34⁺CD38⁺ progenitors, we isolated them and carried out InTAC-seq (Figure 5A). Despite the low cell numbers obtained from these rare primary BM populations (down to ∼250 cells), the resulting data were of high quality (Figures S5A and S5B). As expected, projection of high GATA-1-expressing cells isolated from CD34⁺CD38⁺ progenitors and CD123^– MEP compartments onto the scATAC UMAP of BMMCs demonstrated similar embedding within the mid-erythroid population (Figure 5B). These data suggest that they are indeed similar cells, further confirming that high GATA-1 expression in CD34⁺CD38⁺ BM progenitors is sufficient to enrich for erythroid-committed progenitors without further gating.

High GATA-1 protein expression overlaps with epigenetic switch in BM progenitors to erythroid lineage

During erythropoiesis, BM progenitors undergo global changes in TF activity and chromatin accessibility in order to restrict non-erythroid lineage potential and drive the erythroid program. In order to model this process and understand the epigenetic transitions associated with GATA-1 acquisition, and erythroid commitment, we calculated a pseudotime ordering of cells from HSCs to the late erythroid cluster from BM scATAC-seq data (Granja et al., 2019) (Figure 5C). Using our InTAC-seq data from high GATA-1-expressing cells as a landmark, we identified the positions of their closest scATAC BM cells within the erythroid trajectory (Figure S5H). To assess the chromatin accessibility changes across erythroid differentiation, we then plotted chromVAR deviations for variable TF motifs along this derived erythroid trajectory. We observed a coordinated decrease in accessibility at motif sites for TFs that drive alternative lineages and an increase in accessibility at erythroid TF motif sites precisely where our high GATA-1-expressing progenitors are positioned within the trajectory (Figure 5D). Interestingly, there is a global increase in accessibility of the GATA TF family motifs during erythroid differentiation coinciding with high GATA-1 protein expression (Figure 5D). The preceding stage of erythroid differentiation still shows persistence of non-erythroid TF motif accessibility, suggesting that functional lineage restriction and erythroid fate determination is tied to GATA-1 protein abundance.

Using existing scRNA-seq data from the same BM samples, we next integrated measured TF protein abundance data with gene expression, gene accessibility (gene score), and motif accessibility along a differentiation trajectory. We observe that gene accessibility changes before and after our experimentally determined GATA-1-high trajectory point concur with expected trends for lymphoid/myeloid and erythroid genes, respectively (Figure S5J). Differential analysis of integrated scRNA-seq data at these 3 stages of erythroid differentiation reveal key hemoglobin subunit genes (e.g., HBD, HBA1, HBB) and heme metabolism (TMEM14B/C) after our determined GATA-1 high point (Figure 5E, green and red). We also observe differential upregulation of histidine decarboxylase (HDC), an enzyme in the histamine-synthesis pathways, in cells at the GATA-1 high point, with decreasing HDC expression as cells move into the final erythroid stage. HDC has been implicated in RBC development in mice (Otsuka et al., 2021), and its expression in adult BM coincides with our InTAC-seq-identified GATA-1 high point in the erythroid developmental trajectory. In contrast, preceding the GATA-1 high point, we see enrichment for genes involved in various non-erythroid lineages, including NRIP1 for hematopoietic stem cell quiescence (Forsberg et al., 2010; Huang et al., 2008) (Figure 5E). Our integrated plots further reveal the discord between the different regulatory layers for GATA1 and GATA2 genes and highlight the importance of directly measuring protein levels of key TFs and associated chromatin accessibility (Figure S5I).

Given that CD123^– MEPs have mixed GATA-1 expression (Figure 5A) and correspond to mixed commitment to the erythroid lineage (Figure 4E), we also examined how the high and mid GATA-1-expressing CD123^– MEPs differed. For example, we observed differences in accessibility at the gene locus for the erythropoiesis-enhancing TF MYB, a known target of GATA-1 binding and repression, between these two populations, consistent with their GATA-1 abundance levels (Figure S5C). Globally, we found that GATA-1 expression marked distinct cellular subsets with thousands of differentially accessible sites between GATA-1 high and mid GATA-1-expressing CD123^– MEP populations (Figures S5D and S5E). While megakaryocyte and erythroid TF motifs were enriched in GATA-1 high cells, motifs for TFs involved in myeloid and B cell development were enriched in other subsets (Figures S5F and S5G). These results suggest that mid GATA-1-expressing cells within the CD123^– MEP population retain some myeloid and/or lymphoid potential, likely resulting in their mixed lineage functional output in clonal assays. These observations are consistent with a previous study that described a subpopulation of cells within the CD123^– MEP compartment exhibiting both erythroid and myeloid potential based on functional assays, although lymphoid potential was not tested (Psaila et al., 2016).

Limitations of study

Our method combines fluorescence-activated cell sorting (FACS) with a modified ATAC-seq protocol that can robustly capture chromatin-accessibility profiles of low numbers of fixed cells. Thereby, we note assay limitations shared by chromatin accessibility profiling methods such as ATAC-seq (Buenrostro et al., 2013; Corces et al., 2017) and by antibody-based quantification methods such as flow cytometry (McKinnon, 2018). As we measure genome-wide chromatin accessibility, we infer activity of multiple TFs through changes in accessibility of DNA regions flanking putative TF motif sites. Our method cannot determine definitive TF occupancy on chromatin unlike chromatin immunoprecipitation sequencing (ChIP-seq) or CUT&Tag assays (Kaya-Okur et al., 2019; Park, 2009) that directly measure TF binding. As a result, this method cannot resolve or detect the previously characterized GATA switch phenomenon in erythroid development (Suzuki et al., 2013). Additionally, our method relies on the validated affinity and binding specificity of the FACS antibody to transcription or chromatin binding factor-of-interest under staining conditions. We also note that our method captures chromatin accessibility in bulk populations and, while being able to profile down to a few (∼100) cells, is not a single-cell method. Other methods, namely piATAC (Chen et al., 2018), have linked single-cell chromatin accessibility to protein abundance; however, the complexity and high costs of such methods are not practical for most labs to apply at scale. Thus, InTAC-seq fills that niche by providing similar information at significantly lower cost and higher data quality.

Though we are able to measure genome-wide chromatin profiles associated with specific TF levels such as GATA-1, we cannot assert direct interaction with chromatin for driving accessibility, including TF motif changes. In complex developmental systems such as the human BM, there are multiple TFs and gene regulatory and chromatin binding proteins affecting chromatin structure and accessibility for expression and subsequent cell-fate determination (Wang et al., 2021). We cannot rule out that the above-mentioned epigenetic profiles measured, while associated with GATA-1 protein abundance, could be a result of differentiation and/or other TF activity. Further experimentation with overexpression of TF, for example, would have to be carried out to parse out direct causal links. Instead, our study uses endogenous levels of GATA-1 as a “lineage reporter” in human BM to understand the epigenetic and functional state of high GATA-1-expressing BM progenitors. Finally, though we show significant erythroid commitment through differentiation in clonal assay, we do not contend that this cell population represents true MEPs or that they are the definitive erythroid progenitor population in BM. Further work is needed to assess our GATA-1-high population in comparison to other MEP and erythroid progenitor populations as defined in the literature (Doulatov et al., 2010; Edvardsson et al., 2006; Mori et al., 2015; Notta et al., 2016; Psaila et al., 2016; Sanada et al., 2016).

Key resources table

Resource availability

Experimental models and subject details

Method details

Quantification and statistical analysis

DESeq2 was used for identifying differentially accessible peaks as described in the Method Details section, with FDR cut-off of 0.05 (for all comparisons) and log2 fold change of 2 (for GATA-1 high vs mid/low progenitor comparison). For comparing differences in colony forming ability from different bone marrow populations, a student’s t-test was performed. The number of biological and technical replicates for each experiment are indicated in the figures and/or figure legends.