Slow cycling and durable Flt3⁺ progenitors contribute to hematopoiesis under native conditions

Blood production in the bone marrow (BM) is classically viewed as a constant hematopoietic flux originating from a small pool of self-renewing hematopoietic stem cells (HSCs). In this model, quiescent HSCs regularly differentiate and give rise to a complex set of transient multipotent progenitors (MPPs) that serve as compartments of amplification through division and undergo progressive lineage specification (Haas et al., 2018; Laurenti and Göttgens, 2018). Recent studies, exploiting sophisticated genetic marking for in vivo cell-fate tracking, have challenged this paradigm and have suggested that HSCs have a limited contribution to steady-state blood production (Busch et al., 2015; Rodriguez-Fraticelli et al., 2018; Sun et al., 2014). Consistent with this idea, persistent reduction of HSC numbers does not impair steady-state blood cell generation for at least 1 yr (Schoedel et al., 2016; Sheikh et al., 2016). The interpretation of these results and the extent of the HSC contribution to native hematopoiesis remain a matter of debate (Chapple et al., 2018; Pucella et al., 2020; Rodriguez-Fraticelli and Camargo, 2021; Sawai et al., 2016; Säwen et al., 2018). Beyond this controversy, these studies highlight the underestimated contribution of the MPP compartment to the daily hematopoietic output in unperturbed conditions (Säwen et al., 2018; Sun et al., 2014). This new understanding raises questions about the functional properties of the MPP compartment in steady-state conditions and its ability to independently sustain prolonged hematopoietic cell production.

The early hematopoietic hierarchy is confined in the Lin⁻ Kit⁺ Sca1⁺ (LSK) fraction of the BM. In this fraction, the MPP compartment is historically defined based on its qualitative and quantitative output, particularly its ability to generate all hematopoietic lineages in in vitro/in vivo assays and its failure to sustain long-term hematopoietic production in regenerative conditions in vivo. The molecular and functional heterogeneity of the MPP compartment has been described with different subsets that present variable levels of self-renewal capacity and lineage potential (Cabezas-Wallscheid et al., 2014). At the interphase between HSCs (LSK Flt3⁻ CD48⁻ CD150⁺) and progenitors, a quiescent MPP5 subset (LSK Flt3⁻ CD48⁻ CD150⁻) shares some molecular programs with HSCs and possesses finite self-renewal ability (Pietras et al., 2015; Sommerkamp et al., 2021). Downstream in the hierarchy, proliferating MPP2–4 subsets (MPP2: LSK Flt3⁻ CD48⁺ CD150⁺; MPP3: LSK Flt3⁻ CD48⁺ CD150⁻; and MPP4: LSK Flt3⁺ CD48⁺ CD150⁻) display molecular priming and differentiation biases toward the megakaryocytic, myeloid, and lymphoid lineages, respectively (Cabezas-Wallscheid et al., 2014; Pietras et al., 2015). However, the developmental relationship between quiescent HSC/MPP5 and proliferative lineage-biased MPP2–4 remains to be described.

Here, we revisited the early hematopoietic hierarchy present in the LSK fraction. Using cellular indexing of transcriptomes and epitopes by sequencing (CITE-seq) analyses, we identified a unique CD62L⁻ MPP4 population that clusters within the MPP compartments but displays a cell cycle gene signature similar to the one found in quiescent HSCs. Through a panel of functional assays, we directly validated the existence of the unexpected cell cycle diversity in MPP4 during unperturbed hematopoiesis. Our results challenge the accepted dichotomy between quiescent HSCs and dividing progenitors. While differences in lineage potential have been historically a major criterion to classify MPP populations, we established a new dimension of MPP diversity based on their proliferative activity at steady-state and defined three new phenotypically identifiable MPP4 populations associated with different levels of proliferation and differentiation. These results particularly suggest a new path of differentiation directly linking MPP5 to quiescent/slow-cycling CD62L⁻ MPP4 cells upstream to the other lineage-biased MPPs. Based on these results, we proposed that quiescent/slow-cycling Flt3⁺ MPPs act as a key transitional compartment that buffers the hematopoietic cell production and therefore regulates hematopoietic homeostasis at steady-state.

We performed highly multiplexed CITE-seq analyses with 65 barcoded antibodies to characterize the hematopoietic hierarchy present in the BM LSK fraction. We simultaneously generated 10× Genomics 3′ single-cell RNA sequencing (scRNA-seq) and antibody-derived tag (ADT) sequencing libraries from over 12,000 LSK cells. To identify discrete subsets of hematopoietic stem and progenitor cells (HSPCs), we performed a joint unsupervised analysis of mRNAs and ADTs for all cells in the software ICGS2. Here, the software was parameterized to consider all ADTs and differentially expressed genes as putative guides to delineate cell states (Salomonis, 2019; Fig. 1 A; and Fig. S1, A and B). Genes associated with the 11 identified clusters (hereafter denoted C1–C11) were compared through gene set enrichment with prior annotated HSPC subsets, including high-output and low-output compartments from lineage barcoding in vivo (Cabezas-Wallscheid et al., 2017; Giladi et al., 2018; Rodriguez-Fraticelli et al., 2020; Sommerkamp et al., 2021). These results (GO-Elite) confirmed overlap with published signatures encompassing the early hematopoietic hierarchy from long-term HSCs (LT-HSCs) to multipotent progenitors (Fig. 1 B and Table S1). Thus, this analysis segregated low-output compartments, presumptively composed of quiescent HSCs (C1 and C3) from high-output compartments which correspond to dividing progenitors (C11, C8, C9, C10, C5). Cluster hierarchy was reinforced by expression patterns of genes encoding for self-renewal and differentiation markers (Fig. S1 C). Using the putative C1 HSC cluster as starting point, SlingShot and CellRank predicted multiple trajectories of differentiation, including those toward the myeloid, lymphoid, and erythroid-megakaryocytic outputs. These analyses infer additional trajectories that pointed to putative intermediary clusters (Fig. S1 D). We particularly focused on the C6 cluster as the specificity of this compartment was highlighted by the striking absence of expression for cell cycle genes and cell cycle–associated DNA repair genes (Fig. 1 C). This property was shared with the HSC-like C1 cluster. However, the C6 subset did not show an HSC-related gene signature and instead was defined by expression of Flt3 (Fig. 1 D). Accordingly, association of the single-cell clusters with prior defined MPP subsets confirmed that the C6 cluster correlates with the intermediary MPP4 (LSK Flt3⁺) compartment (Fig. S1, E and F; Cabezas-Wallscheid et al., 2014). This profile was characterized by the downregulation of Esam and Dach1 (Amann-Zalcenstein et al., 2020; Beaudin et al., 2014) and the upregulation of Ccr7 and Ccr9 (Zlotoff et al., 2010; Fig. 1 E). Consistent with a connection to the MPP4 compartment, the C6 cluster also showed sporadic expression of lymphoid genes such as Il7ra and Rag2 (Amann-Zalcenstein et al., 2020; Igarashi et al., 2002; Fig. 1 F). More globally, clustering of Pearson-correlation coefficients for transcription factor (TF) gene pairs revealed a specific regulatory state within the C6 cluster, which is demarcated by signal-induced TFs and is clearly distinct from adjacent C7 and C11 clusters (Fig. S1 G and Table S2). Gene expression results were confirmed by the CITE-seq analysis, which particularly suggests that the C6 cluster can be enriched by the cell surface expression of CD135 (Flt3) coupled with the absence of the CD62L marker (Fig. 1 G). Altogether, this unbiased analysis provides a complex picture of the heterogenous HSPC compartment. Intriguingly, it points to a singular quiescent/slow-cycling LSK population (C6 cluster) that does not belong to the HSC compartment, but rather correlates to a cell subset of the downstream Flt3⁺ MPPs (MPP4).

To confirm the cell surface phenotype associated with the C6 cluster, we plotted transformed CITE-seq ADT counts to define the MPP4 compartment (LSK Flt3⁺ CD150⁻). Based on our initial analysis, we further segregated this compartment based on the expression of CD62L-ADT, a marker previously associated with lymphoid priming in MPP4 (Cho and Spangrude, 2011; Perry et al., 2006). Overlay of individual cell cycle–associated genes such as Mki67, Foxm1, and Aurkb suggests that the CD62L⁻ MPP4 population is enriched for cells with quiescent gene signature, and therefore encompasses the C6 cluster (Fig. 2 A). To validate this molecular characterization and compare this subset to previously described HSPC compartments, we revisited the early hematopoietic hierarchy present in the BM LSK fraction using flow cytometry. We used a phenotypic scheme based on the cell surface expression of Flt3 and CD62L along with CD48 and CD150 (Cabezas-Wallscheid et al., 2014; Cho and Spangrude, 2011; Kiel et al., 2005; Pietras et al., 2015; Solomon et al., 2020; Sommerkamp et al., 2021; Fig. 2 B and Fig. S2 A). Bulk gene expression analysis indicated that CD62L⁻ and CD62L⁺ MPP4 cells express similar levels of the Flt3 gene (Fig. 2 C, top panel). In contrast, CD62L⁻ and CD62L⁺ MPP4 populations displayed distinct CD62L gene (Sell) expression, therefore ruling out a posttranscriptional regulation of this marker in MPP4. As predicted by CITE-seq, gene expression analysis showed that CD62L expression in MPP4 segregates two distinct populations based on cell cycle and DNA repair gene profiles (Fig. 2 C, bottom panel, and Fig. S2 B). To confirm this result at the single-cell level, we used a reporter mouse model for the Mki67 gene, a well-established marker of proliferating cells (Basak et al., 2014). This reporter system showed the expected gradation of Mki67 expression along the hematopoietic hierarchy (Fig. 2 D, left panel, and Fig. S2 C). Consistent with their quiescent properties, the HSC compartment, with its constitutive LT-HSC (CD34⁻ HSC) and MPP1 (CD34⁺ HSC) subpopulations, and the downstream MPP5 compartment showed limited Mki67 expression. Conversely, committed progenitor gates (i.e., common myeloid progenitor, CMP; granulocyte-macrophage progenitor, GMP; megakaryocyte-erythroid progenitors, MEP; Akashi et al., 2000) showed homogeneous high Mki67 expression, consistent with their status of highly dividing cells. Intermediate MPP2 and MPP3 cells were all Mki67 positive with a mix of low and high expression (Fig. S2 C). In contrast, we found that the MPP4 compartment displayed a specific pattern of Mki67 expression, which combines the quiescent characteristics of the HSCs and the active profile of the other MPP compartments (Fig. S2 C). Interestingly, CD62L markers segregated these two distinct profiles within this population (Fig. 2 D, middle/right panels). CD62L⁺ MPP4 cells were all Mki67⁺, similar to the other MPP2/3 populations. CD62L⁻ MPP4 cells were mainly Mki67⁻, therefore mimicking the HSC quiescent profile. We confirmed these results by measuring the ability of these populations to incorporate the nucleotide analog 5-ethynyl-2′-deoxyuridine (EdU) after 1-h pulse in vitro (Fig. 2 E). As expected, few HSC/MPP5 cells were found cycling, while downstream MPP2–4 compartments showed high-levels of EdU incorporation (Pietras et al., 2015). Consistent with the Mki67 profile, CD62L expression segregated low and high cycling cells in the MPP4 compartment, with the CD62L⁻ MPP4 subpopulation being characterized by low EdU incorporation. Of note, CD62L expression was not confined to the MPP4 fraction (Fig. S2 D). However, CD62L expression did not correlate with cell cycle in other populations (MPP3 and Lin⁻ Kit⁺ Sca1⁻ myeloid progenitors), therefore indicating that CD62L does not simply associate with cell activation in all hematopoietic cells (Fig. S2, E–G). Altogether, the results confirm a novel level of diversity within the phenotypically defined MPP4 population. Particularly, they uncover a phenotypically isolable CD62L⁻ MPP4 compartment that is enriched for progenitors with low proliferative activity in native conditions.

We next established the hematopoietic potential of the MPP4 subpopulations defined by the CD62L marker. Since MPP4 cells are defined by Flt3 expression, we first confirmed that CD62L⁻ and CD62L⁺ MPP4 compartments were both able to mount a molecular response to in vitro Flt3L exposure. As expected, we observed ERK phosphorylation in both CD62L⁻ and CD62L⁺ MPP4 populations but not in Flt3⁻ progenitors (Fig. 3 A). We then assayed the potential of these populations in vitro. In clonogenic methylcellulose assays, both populations showed similar plating efficiency, clearly distinct from the HSCs (Fig. S3 A). In contrast, we observed an ∼24-h delay in the timing of the first division when these cells were analyzed in liquid culture with self-renewing cytokines (stem cell factor [SCF], thrombopoietin [TPO]) and Flt3L (Fig. 3 B). In this condition, CD62L⁺ MPP4 cells underwent precocious division compared with CD62L⁻ MPP4 cells, which showed the same kinetics of division as HSCs. We next sought to determine the potential of these populations in regenerative conditions in vivo. We first established that the two populations display no significant differential homing properties (Fig. S3 B). We then performed competitive transplantation assays into lethally irradiated recipients. In non-limiting conditions (10,000 transplanted cells), both CD62L⁻ and CD62L⁺ MPP4 showed transient reconstitution activity, with production of myeloid cells that progressively fades away 1 mo after transplantation (Fig. 3 C). These results confirmed the short-term potential of reconstitution of these populations and ruled out any contamination of the sorting gates with upstream quiescent HSC/MPP5 populations. However, saturating conditions were not optimal to study the differential lineage potentials of these populations. Transplantation assays with a limiting number of donor cells (500 transplanted cells) revealed that CD62L⁻ MPP4 cells showed extended reconstitution potential compared with their CD62L⁺ MPP4 counterparts, which undergo early lymphoid differentiation (Cho and Spangrude, 2011; Perry et al., 2006; Fig. 3 D). Importantly, similar results were obtained when transplanting Ki67⁻ and Ki67⁺ MPP4, further confirming an overlap between the CD62L^– phenotype and the quiescent/slow-cycling properties (Fig. S3, C and D). Altogether, these results show that differences in cell cycle status in the MPP4 compartment define specific hematopoietic potential and kinetics that can be revealed in regenerative conditions.

To analyze the activity of the CD62L^−/+ MPP4 cells during native hematopoiesis, we repurposed the Tet-On H2B-GFP mouse model, previously used to study HSC quiescence (Foudi et al., 2009; Wilson et al., 2008). This system allows the HSPC compartment to be GFP-labeled following a 2-wk doxycycline pulse period (Fig. 4 A, left panel). Loss of fluorescence in HSPCs following doxycycline withdrawal is then monitored to infer cell divisional kinetics and history. This system was previously used to track the most quiescent HSC population based on their long-term ability (>20 wk) to retain GFP labeling. For progenitors, we anticipated a more complex scenario in which GFP levels at any given time will depend on (i) the division history within each compartment and (ii) the division history of the upstream compartment at the time of its differentiation into the compartment (Fig. 4 A, right panel). Because of our interest in intermediary progenitors, we performed the analysis on the entire hematopoietic hierarchy after 6–10 wk of chase. As expected, HSCs and MPP5 showed the highest level of GFP retention at this time point (Fig. 4, B and C). Conversely, highly cycling lineage-committed progenitors gates (common lymphoid progenitor [CLP], CMP, GMP, and MEP) showed reduced and fast diminishing GFP labeling. For the MPP compartment, we observed a similar GFP labeling for the MPP2, MPP3, and CD62L⁺ MPP4 cells consistent with their common cycling properties. Interestingly, labeling retention in CD62L⁻ MPP4 cells was intermediate between HSC/MPP5 compartments and the other MPP compartments. This increased labeling retention is compatible with the quiescent/slow-cycling property of the CD62L⁻ MPP4 cells. It may also reflect, in part, a link between CD62L⁻ MPP4 population and upstream compartments (HSC/MPP5), requiring less division history for a transition between compartments. In line with this possibility, we found heterogeneous expression of CD48 within the CD62L⁻ MPP4 cells that allow the definition of three MPP4 subsets: CD48⁻ CD62L⁻ MPP4, CD48⁺ CD62L⁻ MPP4, and CD48⁺ CD62L⁺ MPP4 (Fig. 4 D). Further analyses showed that CD48⁻ CD62L⁻ MPP4 cells expressed higher Sca1 expression and higher 6–10-wk labeling retention than the other MPP4 subsets, consistent with a more immature phenotype (Fig. 4 E). The progressive acquisition of Ki67 and the differential expression of key TFs such as c-Myc (Huang et al., 2008), Gfi1 (Hock et al., 2004; Lee et al., 2018), PU.1, and Irf8 (Hoppe et al., 2016; Koga et al., 2007; Kueh et al., 2013) further highlighted a hierarchy within these subsets (Fig. S4, A and B). This hierarchy was functionally confirmed in a classical competitive transplantation assay, in which CD48⁻ CD62L⁻ MPP4 cells show better reconstitution ability than CD48⁺ CD62L^+/− MPP4 cells (Fig. 4 F). Of note, we observed that the CD48⁻ CD62L⁻ MPP4 subset partially overlaps with a recently described MPP4 compartment that expresses the endothelial cell adhesion molecule ESAM (Klein et al., 2022; Fig. S4 C). However, ESAM expression in MPP4 did not strongly correlate with low Ki67 expression (Fig. S4 D). Altogether these results suggest a novel early path of differentiation active during undisturbed native hematopoiesis. This path involves the sequential acquisition of Flt3 and CD48 expression to generate quiescent CD48⁻ CD62L⁻ MPP4 (denoted MPP4a) and CD48⁺ CD62L⁻ MPP4 (denoted MPP4b) cells, which then give rise to a cycling CD48⁺ CD62L⁺ MPP4 (denoted MPP4c) cells.

To functionally characterize the MPP4 subsets, we analyzed their activity in physio-pathological conditions linked to hematopoietic stress. First, to establish the ability of the MPP4 subsets to respond to inflammatory stimulation, we mimic bacterial infection using the in vivo injection of a single dose of lipopolysaccharide (LPS; Fig. 5 A). As expected, LPS treatment led to a rapid but transient surge of neutrophils in the blood, associated with a reduction of BM cellularity and fewer BM myeloid cells (Fig. 5, B and C). In the HSPC compartment, LPS induced different kinetics of expansion of the MPP populations (MPP2, MPP3, and MPP4) and the myeloid progenitors (Fig. 5 D). While no alterations of CD62L expression were detectable in MPP3, we observed rapid changes in the CD62L⁻ MPP4 compartments (MPP4a and MPP4b) at day 1 after LPS treatment, suggesting a direct activation and differentiation response to inflammation (Fig. 5 E). This response was followed by a complete return to steady-state, 8 d after stimulation. Consistent with this idea, study of Mki67 expression showed that the CD48⁻ CD62L⁻ MPP4 (MPP4a) subset quickly regains its cell cycle characteristics following activation (Fig. 5 F). Altogether these results highlight the plasticity of the slow-cycling MPP4 compartments, particularly their ability to react to acute inflammatory stimulation but also to rapidly recover their steady-state properties when the stimulation subsides.

As aging has also been associated with phenotypic and functional alterations of the MPP compartments, we analyzed the MPP4 compartment in old mice (18 mo of age; Young et al., 2016). Correlating with a decrease in lymphopoiesis, we observed the expected reduction in the size of the MPP4 population (Fig. 5 G). Interestingly, the CD48⁻ CD62L⁻ MPP4 (MPP4a) population was not changed. However, aging affected the CD48⁺ CD62L⁻ MPP4 (MPP4b) and CD48⁺ CD62L⁺ MPP4 (MPP4c) and therefore led to the relative enrichment in the more quiescent CD48⁻ CD62L⁻ MPP4 (MPP4a) subset (Fig. 5, G and H). These results confirm the intrinsic heterogeneity of the MPP4 compartment that can be revealed by the CD62L marker. They also suggest that the reduction of lymphoid differentiation associated with aging originates from a defect of differentiation downstream to the CD48⁻ CD62L⁻ MPP4 (MPP4a) subset.

Together, our results suggested a hierarchy among the MPP4 subsets in their relationship with HSC/MPP5 compartments. To track the fate of these compartments in vivo, we transplanted purified cells in recipient mice treated with a non-myeloablative (NMA) conditioning regimen (Fig. 6 A; Chang et al., 2022). Conditioning using injection of anti-CD117 (ACK) antibody combined with low-grade irradiation did not trigger a major inflammatory response, and as a consequence, we observed minimal hematopoietic alterations by flow cytometry (Fig. S5 A). This allowed for the monitoring of not only the peripheric mature progeny but also the BM progenitor dynamics driven by the transplanted cells over time. In the blood, we observed the expected delayed cell production driven by HSC/MPP5 compared with the MPP4 populations (Fig. 6 B). As previously described in fully conditioned recipients (Fig. 4 F), we found a hierarchy of hematopoietic potential within the CD48/CD62L-defined MPP4 compartments based on the level of blood chimerism and the kinetics of lymphoid cell production. In the BM, transplanted HSCs remained in an undifferentiated state within their LSK Flt3⁻CD48⁻CD150⁺ defining gate for at least 2 wk after transplantation (Fig. 6 C and Fig. S5 B). HSCs then displayed a rapid propensity to generate MPP4 (Flt3⁺) cells compared with the production of MPP2/MPP3 cells. Consistent with a recent report (Sommerkamp et al., 2021), MPP5 cells were highly responsive to transplantation as this population did not retain its defining phenotype but either acquired CD150 expression or induced the rapid production of MPP2–4 cells. Interestingly, transplanted MPP4 subsets were able to maintain their defining LSK Flt3⁺ phenotype after transplantation before undergoing exhaustion. CD48⁻ CD62L⁻ MPP4 cells promoted the highest level of BM reconstitution among the MPP4 subsets and persisted over the 2 mo of analysis. During this period, the CD48⁻ CD62L⁻ MPP4 subset displayed a significant propensity to generate phenotypically defined MPP2/MPP3 cells and to produce platelets (Fig. S5 C). Unfortunately, transplantations under NMA condition failed to directly assess the in vivo transition between the CD62L^+/− MPP4 subsets, as these compartments were homogenously CD62L⁺ in most recipients (data not shown). Despite this limitation, these results were consistent with the specific positioning of CD62L⁻ MPP4 subsets at the transition between HSCs and other lineage-biased MPP2/MPP3 populations. NMA transplantation experiments also revealed the certain level plasticity within the MPP compartment, exemplified by the ability of the MPP3 to generate phenotypically defined MPP2 and MPP4 cells (Fig. S5 D).

Altogether, these results add a new dimension to the MPP diversity, based on their proliferative activity at steady-state. Notably, they uncover novel subsets of quiescent MPP4 distinct from the other proliferative MPP compartments. Fate tracking in recipients treated with an NMA conditioning regimen suggests new hematopoietic developmental paths, which put the slow-cycling/quiescent Flt3⁺ cells at the center of the balanced blood cell production that occurs in undisturbed conditions (Fig. 7).

This work uses CITE-seq analysis and a panel of functional assays to refine the composition of the MPP4 compartment based on cell cycle activity. We show that part of the MPP4 compartment is composed of quiescent/slow-cycling cells that can be phenotypically enriched based on the cell surface expression of the CD62L marker. CD62L⁻ MPP4 population displayed a cell cycle and DNA repair gene signature similar to HSCs but did not express the other classical genes associated with HSC quiescence (e.g., Fdg5, Hoxb5, Pdzk1ip1, Procr, and Rarb). The specific cell cycle status of the CD62L⁻ MPP4 subset was independently confirmed by measurement of Ki67 expression, Edu incorporation, and kinetics of division in vitro. Moreover, we found that the reduced cell cycle activity correlates with an increased hematopoietic output of this population at steady-state and in regenerative conditions. Altogether our results uncover a novel level of MPP heterogeneity, define a Flt3⁺ population with a prolonged hematopoietic activity, and suggest new paths of differentiation within the early hematopoietic progenitors.

These results raise the question of the mechanisms that regulate the cell division in the MPP4 compartment. CD62L⁻ MPP4 cells show the same Flt3 expression as their cycling CD62L⁺ MPP4 counterparts, both at mRNA and cell surface protein levels. This suggests that the quiescent/slow-cycling MPP4 cells are fully competent to respond to Flt3 proliferative signaling. In line with these observations, CD62L⁻ MPP4 cells showed normal signal transduction responses after in vitro Flt3L stimulation. We can speculate about the cell-intrinsic and/or environmental mechanisms that impair the activation of the quiescent/slow-cycling MPP4 in vivo. Reduced CD62L⁻ MPP4 cycling was not associated with the expression of known regulators of HSC quiescence such as Cdkn1c and Tcf15 (Matsumoto et al., 2011; Rodriguez-Fraticelli et al., 2020). In contrast, the CITE-seq analysis suggests other modes of cell cycle control based on antiproliferative molecules such as Btg2 (Yuniati et al., 2019) or JunB (Santaguida et al., 2009) or anti-inflammatory regulators such as Wfdc17 (Karlstetter et al., 2010). Alternatively, we can postulate that the quiescent/slow-cycling Flt3⁺ population is confined to a specific microenvironment and not exposed to the Flt3 ligand in vivo. This hypothesis is consistent with the recent recognition that distinct BM microenvironments regulate specific aspects of hematopoiesis. Such spatially organized niches were initially described to regulate the HSC activity (Pinho and Frenette, 2019). More recently, specialized environments have been described as controlling myeloid differentiation by local cytokine expression, such as CSF1 (Zhang et al., 2021). Although this possibility is currently difficult to assess due to the absence of a reliable method to specifically localize the MPP4 populations in the BM, it is tempting to speculate that similar specialized niches can modulate cell cycle activity in Flt3⁺ MPPs. We notice that CD62L (also known as L-selectin) expression partially correlates with cell cycle activation in the MPP4 compartment. In this context, CD62L mimics the impact of CD62E (E-selectin) on HSC activation (Winkler et al., 2012). This suggests that the adhesion molecule CD62L may contribute to the localization of MPP4 subsets to Flt3L-expressing niche and their subsequent activation.

Our work suggests a novel developmental hierarchy in the MPP compartment by defining three MPP4 subsets (MPP4a: CD62L⁻ CD48⁻; MPP4b: CD62L⁻ CD48⁺; and MPP4c: CD62L⁺ CD48⁺). The finding that Flt3 expression can precede the acquisition of CD48 indicates that Flt3 acquisition is an early event in the differentiation process. Consistent with this idea, label retention experiments suggest that the transition between the CD62L⁻ MPP4 and the upstream HSC/MPP5 compartments requires a limited number of divisions. Even though lineage-prediction algorithms are compatible with this observation, we acknowledge that the absence of direct lineage tracing of these populations constitutes a limitation of our study, as it did not formally preclude the existence of additional complex and dynamic differentiation trajectories in the LSK compartment. This point is highlighted by the apparent plasticity in the MPP3 compartment that can be revealed in NMA transplantation. However, the existence of the CD48⁻ CD62L⁻ MPP4 subset at the transition between HSC/MPP5 and MPPs is consistent with prior single-cell gene expression analyses, which establish the existence of cell clusters with shared MPP5 and MPP4 gene signatures (Rodriguez-Fraticelli et al., 2018; Sommerkamp et al., 2021). This model also matches the partial overlap observed between the CD48⁻ CD62L⁻ MPP4 subset and the recently described multipotent ESAM⁺ MPP4 (Klein et al., 2022). Accordingly, analyses using transplantation in non-myeloablative conditions show that CD62L⁻ CD48⁻ MPP4 cells can contribute to a lasting production of other lineage-biased MPP populations at steady-state. Altogether, these results reinforce previous studies that found that all hematopoietic cells can be derived from a Flt3⁺ progenitor (Boyer et al., 2011; Buza-Vidas et al., 2011). Finally, these results establish the quiescent/slow-cycling CD48⁻CD62L⁻ MPP4 at the center of the blood cell production that occurs under native conditions.

The stem/progenitor dynamics that contributes to blood cell production at steady-state remains controversial. Conflicting reports have debated the HSC contribution to native hematopoiesis, as different levels of HSC contributions were described based on different in vivo cell fate tracking approaches. Interestingly, experimental data and recent mathematical models built on these studies suggest the unexpected persistence of MPP populations over time, a feature which was interpreted as the result of a significant degree of MPP self-renewal activity (Säwen et al., 2018; Takahashi et al., 2021). We envision that the pseudo-quiescence/slow-cycling property of CD62L⁻ MPP4 associated with a low differentiation rate also contributes to MPP persistence. In this context, the early quiescent/slow-cycling CD48⁻CD62L⁻ MPP4 subset, which can persist for over 2 mo in NMA transplantation assays, could be key to sustaining steady-state blood cell production in the absence of HSC input (Schoedel et al., 2016; Sheikh et al., 2016). Based on its properties, this population could be related to the recently uncovered embryonic MPP, capable of lifelong hematopoietic contribution (Patel et al., 2022). This reservoir property could also be essential for the buffering of the hematopoietic stress response, as our results show that this population reacts to acute inflammatory signals but quickly returns to equilibrium upon resorption of these signals. As such, our findings contribute to the current reevaluation of the mechanics of the hematopoietic stress response (Munz et al., 2023). Finally, we can speculate that this quiescent/slow-cycling population at the heart of the MPP compartment could be a prime target for the early stages of leukemogenesis. As described in HSCs, the pseudo-quiescence property is associated with low expression of the DNA repair machinery, which may render these cells vulnerable to mutagenesis (Mohrin et al., 2010). This vulnerability could be compounded by the size of this population. In addition, while mutations or epigenetic alterations are washed out from proliferative cells that undergo differentiation, these persisting cells could accumulate alterations until reaching a leukemogenic state. This reservoir of mutant progenitor populations would be particularly relevant in the context of late-stage activating mutations of the FLT3 gene that occur in ∼30% of all acute myeloid leukemia. Future studies will address the relevance of the quiescent/slow-cycling intermediary population to the generation of the “cell of origin” associated with hematological malignancies.

Mice were bred and housed in an Association for Assessment and Accreditation of Laboratory Animal Care–accredited animal facility of the Cincinnati Children’s Hospital Medical Center (CCHMC). C57BL/6J (#000664), Pepcb/BoyJ (#002014), UBI-GFP (#004353), CAG-tdTomato-EGFP (#026932), Ki67-RFP (#029802), c-Myc-EGFP (#019075), Irf8-GFP (#027084), and R26-M2rtTA::TetOP-H2B-GFP (#016836) mice were purchased from The Jackson Laboratory. Gfi1:H2B-tdTomato reporter mice were from Dr. Georges Lacaud (University of Manchester, Manchester, UK; Thambyrajah et al., 2016) and PU.1-YFP reporter mice from Dr. Claus Nerlov (MRC Weatherall Institute, Oxford, UK; Kirstetter et al., 2006). All animal experiments were approved by the CCHMC Institutional Animal Care and Use Committee.

For homing assay, 10⁴ CD62L⁻ or CD62L⁺ MPP4 cells were alternatively isolated by flow cytometry from UBI-GFP or CAG-tdTomato mice and intravenously co-transplanted in lethally irradiated CD45.1⁺ C57BL/6 recipients (137Cs irradiator, 11 Gy delivered in two split doses). BM isolated from leg bones was analyzed for fluorescent cells 36 h after transplantation.

For competitive BM transplantation assay, CD62L^+/− MPP4 or Ki67-RFP^+/− MPP4 from 8-wk-old CD45.2⁺ C57BL/6 mouse BM was mixed with 3 × 10⁵ unfractionated BM cells from CD45.1⁺ C57BL/6 mice and injected retro-orbitally into a lethally irradiated CD45.1⁺ C57BL/6 recipient. Donor chimerism was monitored by flow cytometry in peripheral blood (PB).

For BM transplantation assay in NMA conditions (Chang et al., 2022), CD45.1⁺ C57BL/6 mice were successively treated intraperitoneally with 500 mg diphenhydramine HCl (D3630; Sigma-Aldrich) and intravenously with 500 mg ACK2 anti-CD117 antibody (#BE0293; Bio X Cell) 6 d before transplantation. Mice were then irradiated 3 d before transplantation with 200 cGy (137Cs irradiator). 1–1.5 × 10⁴ FACS-purified HSCs, MPP5, and MPP4 subsets isolated from UBI-GFP mice were injected retro-orbitally. Donor (GFP⁺ cells) chimerism was monitored, at different time points, by flow cytometry in PB and in autoMACS ckit⁺-enriched BM cells.

BM and PB from 8-wk-old wild type C57BL/6 or Ki67-RFP were analyzed by flow cytometry at time points of 1, 3, and 8 d following an intraperitoneal LPS injection (35 µg/kg, L2630; Millipore Sigma).

Transgene expression was induced for 2 wk in 6-wk-old H2B-GFP mice with 2 mg/ml doxycycline (D9891; Sigma-Aldrich) in sugared drinking water (10 mg/ml).

BM preparation and cell surface staining were performed as described previously (Solomon et al., 2020). In brief, BM cells were isolated by crushing long bones and hips, treated with red blood cell lysis buffer (150 mM NH₄CL and 10 mM KHCO₃), and washed with Hank’s Buffered Salt Solution (#14175-095; Gibco) supplemented with 2% heat-inactivated fetal bovine serum (hereafter referred to as staining media, SM). 8 × 10⁶ unfractionated BM cells were stained for flow cytometry analysis. For sorting analysis, pools of unfractionated BM cells were purified by Ficoll separation (Histopaque-11191, #11191; Sigma-Aldrich) and then enriched for ckit+ cells using magnetic beads/autoMACS separation (Miltenyi Biotec). BM cells were stained with unconjugated rat lineage-specific antibodies (Ter119, Mac1, Gr-1, B220, CD5, CD3, CD4, and CD8) or biotinylated lineage-specific antibodies (Gr-1, Mac-1, B220, Ter119, and CD3e), followed by staining with goat anti-rat PE-Cy5 or streptavidin-Efluor660 secondary antibodies, respectively. Cells were then stained using c-kit-APC-eFluor780, Sca1-PB, CD48-BV711 or CD48-PerCP-Cy5.5, CD150-PE or CD150-BV650, Flt3-Biotin or Flt3-PE, CD62L-BV510, FcgR-BV510 or FcgR-PerCP-eFluor710, CD34-FITC, EPCR-PerCP-eFluor710, CD41-BV605, CD105-APC, and CD127-BV785 antibodies in a 1:3 ratio of Brilliant Staining Buffer (#563784; BD biosciences):SM. Secondary staining was performed streptavidin-BV711/ streptavidin-PE-Cy7. For dead cell exclusion, cells were washed with phosphate-buffered solution (PBS, #21-031-CV; Corning) and stained with Zombie NIR fixable viability kit (BioLegend) for 5 min at room temperature. Reagents are presented in Table S3. Data were collected on a five-laser (16 ultraviolet [355 nm] channels, 16 violet [405 nm] channels, 14 blue [488 nm] channels, 10 yellow-green [561 nm] channels, 8 red [635 nm] channels) Aurora spectral flow cytometer (Cytek Biosciences). Data analysis was performed using FlowJo (BD biosciences) software.

Following cell surface staining, BM cells were cultured in StemSpan Media (Stemcell Technologies) supplemented with cytokines (SCF, TPO 50 ng/ml; PeproTech) and EdU (Kit #C10632; Thermo Fisher Scientific) for 60 min at 37°C. Cell fixation, permeabilization, and Click-iT reaction were performed according to provider’s instruction.

Following cell surface and viability staining, BM cells were cultured in StemSpan Media (Stemcell Technologies) supplemented with Flt3 ligand (50 ng/ml; PeproTech) for 60 min at 37°C. Cells were fixed/permeabilized using Cytofix/Cytoperm kit (#554722; BD Biosciences), stained with anti-pERK1/2 antibody (#612592; BD Biosciences), and analyzed by flow cytometry.

For division-tracking experiments, HSCs, CD62L⁻, or CD62L⁺ MPP4 were sorted directly into 60-well Terasaki plates (one cell per well) and cultured in StemSpan Media (Stemcell Technologies) complemented by cytokines (SCF, 20 ng/ml; TPO, 20 ng/ml; Flt3 ligand, 20 ng/ml). Individual wells were visually monitored at the indicated time points to assess the kinetics of the first division. For single-cell methylcellulose assay, HSCs, CD62L⁻, or CD62L⁺ MPP4 were sorted directly into 96-well culture plates (one cell per well) and cultured in methylcellulose (M3434; StemCell Technologies). Colonies were scored in individual wells on days 7–9 on an inverted microscope.

qRT-PCR analyses were performed using total RNA isolated from ∼5 × 10⁴ cells. RNA was treated with DNase I and reverse-transcribed using SuperScript III kit and random hexamers (Invitrogen). cDNA equivalent of 200 cells per reaction was analyzed on a 7,900 HT Fast Real-Time PCR System (Applied Biosystems) using gene-specific primers and SYBR green reagents (Applied Biosystems). Each measurement was performed in triplicate with a β-actin (Actb) housekeeping gene used for normalization. Sequences for qRT-PCR primers can be found in Table S4.

20,000 LSK cells were sorted from a pool of four 2-mo-old C57Bl/6J mice. The CITE-seq protocol described in Muench et al. (2020) was applied to LSK cells using a collection of 65 TotalSeq antibodies (Table S5). For staining optimization, individual antibodies were titrated on ckit⁺ bone marrow cells by first conjugating the TotalSeq-A antibodies to an oligo-dT conjugated to Alexa Fluor 647 at a 1:1.5 M ratio, before staining the cells at different dilutions. The prepared samples were washed and run on a BD LSRFortessa flow cytometer to determine optimal concentrations. For FACS sorting, total BM cells were collected from three C57/Bl6 male mice and enriched for ckit⁺ cells (Miltenyi). Cells were stained with fluorochrome-conjugated antibodies in staining buffer (1% FBS, PBS) for 30 min at 4°C and washed three times in staining buffer at 4°C. Cells were FACS sorted and resuspended in 50 μl of TotalSeq-A antibody cocktail for 30 min at 4°C and washed one time. Cells were resuspended at 1,000 cells/μl, and 16,000 cells were loaded in 1 port of 10× Genomics Chromium Controller Instrument. The 10× Genomics single-cell 3′ v3 assay protocol was performed according to manufacturer’s recommendation until the cDNA amplification step. The ADTs were processed according to the CITE-seq protocol detailed by BioLegend (https://www.protocols.io/view/totalseq-a-antibodies-and-cell-hashing-with-10x-si-261geo6mol47/v1; BioLegend Inc). The Gene Expression (GEX) library was generated following regular protocol of 10× Genomics. The final GEX library was sequenced at 50,000 reads per cell with paired-end 150 on Illumina NovaSeq platform (Novogene). The final ADT library was sequenced at 10,000 reads per cell with paired-end 50 on Illumina NovaSeq platform (Novogene).

The LSK CITE-seq 10× Genomics library was sequenced at a depth of 326 million and aligned to mouse genome (mm10), with unique molecular index quantified (UMI) gene counts obtained using the Cell Ranger workflow (version 3.1.0), considering both mRNAs and ADTs. Cell Ranger identified 12,391 filtered cell barcodes with >7,500 UMI per cell on average (Table S6). Ambient RNAs were excluded in the mRNAs using SoupX and an ambient correction threshold of 10% (Young and Behjati, 2020). Using the default filtered cellular barcodes, the associated sparse-filtered SoupX corrected matrix files were converted to log₂ scaled normalized gene counts (counts per 10,000 [CPTT] UMIs) using AltAnalyze’s Chromium Processing model and combined with ADT log₂ scaled reads into a single flat matrix. This combined matrix was input to AltAnalyze and ICGS2 (Iterative clustering and guide-gene selection version 2). ICGS2 was run using the default options (cosine clustering; Venkatasubramanian et al., 2020), and exclusion with >500 genes expressed (CPTT>1) per cell was retained for further analysis (>90% of cell barcodes for all captures; see summary metrics in Table S7) This analysis autodetects ADTs with cell-barcode IDs in the name and forces the use of these ADTs as guide-genes in the iterative clustering analysis steps to prioritize markers for downstream sparse-non-negative matrix factorization analysis. GO-Elite (Zambon et al., 2012) was run using a custom database of HSCP-discriminating gene sets described in prior studies (Cabezas-Wallscheid et al., 2017; Giladi et al., 2018; Rodriguez-Fraticelli et al., 2020; Table S8). Cell cycle gene sets were obtained from the ToppFun database and used as input to the intracorrelation function in AltAnalyze’s hierarchical clustering module (Pearson rho > 0.3) to segregate gene expression patterns by cell cycle phase. Cell cycle phase was inferred from gene set enrichment against the default PathwayCommons database in GO-Elite (Rodchenkov et al., 2020). We used Slingshot and CellRank to computationally infer the pseudotime-based and RNA velocity-based lineage trajectories, respectively (Lange et al., 2022; Street et al., 2018). For Slingshot (v.1.8.0), we provided the uniform manifold approximation and projection (UMAP) coordinates (shown in Fig. 1 A) for all the cell populations (clusters C1–C11). Additionally, cluster C1 was provided as the starting point for lineage trajectory inference. For CellRank, the lineage trajectories were inferred by following the RNA velocity kernel protocol. Specifically, prior to applying CellRank, we used velocyto to compute the spliced and unspliced counts. For velocyto, we used “run” command and provided the Cell Ranger–generated bam file (possorted_genome_bam.bam) and the latest CellRanger’s mouse genome annotation file (mm10: GENCODE vM23/Ensembl 98). For CellRank analysis, first, the loom file, output from velocyto, containing the spliced and unspliced counts, was merged with the anndata object containing gene counts obtained from CellRanger (filtered_feature_bc_matrix). Only the cells that have a cluster assignment by ICGS2 (shown in Fig. 1 A) were considered for further analysis. Genes that did not have enough spliced and unspliced counts were removed (“min_shared_counts” set to 20 in filter_and_normalize function). The top 3,000 highly variable genes were calculated and the gene counts matrix was normalized and log-transformed (pp.filter_and_normalize from scvelo package and pp.log1p from scanpy package). To estimate the velocity, first, the principal components and moments were calculated using the author-recommended parameters provided on CellRank tutorial. RNA velocity was calculated using the stochastic model in the scvelo package. Cell-to-cell transition matrices from the RNA velocity (VelocityKernel) and gene expression similarity (ConnectivityKernel) kernels were computed. Finally, a combined kernel of the aforementioned kernels was used to calculate and visualize cell-to-cell transition. Random walks were simulated starting in the putative C1 HSC cluster (all other parameters set to default values) and show the visualization of the same in Fig. S1 D. To infer gene regulatory associations, we assembled a list of annotated murine transcription factors from multiple sources (Lambert et al., 2018; Zhou et al., 2017; ToppFun) dynamically expressed in MPP4 predicted cell populations (C6, C7, C11). Dynamically expressed genes were those determined using the software MarkerFinder with a Pearson correlation coefficient >0.2. A pairwise correlation matrix was generated in the software AltAnalyze (TFCorrelationPlot.py) using all dynamically expressed genes and transcription factors and then clustered using the software HOPACH. For comparison with prior defined MPP subsets, bulk RNA-seq profiles from Cabezas-Wallscheid et al. (2014) were obtained from supplemental materials and supplied to the software MarkerFinder to identify the top 200 marker genes for each HSC or MPP population. These genes were used for gene set enrichment in the software GO-Elite of the top 200 marker genes for each ICGS2 scRNA-seq cluster. CIBERSORTx was used to impute cell type fractions from the bulk RNA-seq data (Newman et al., 2019). Using the interactive web portal, we first built the signature matrix from our scRNA-seq data by providing normalized counts of all genes and cells and the cell type groups of all cells as inputs to the program. Other CIBERSORTx parameters for this analysis were set at the default value. Using the CIBERSORTx signature matrix, cell types were imputed in the bulk RNA-seq data by providing the normalized counts of all genes and samples, referred to as the mixture matrix. Since the signature matrix and the mixture matrices are on different platforms and technologies, as per the recommendation of CIBERSORTx authors, batch correction was applied in the “S-mode.” In the imputation of cell type fractions, permutations for significance analysis were set to 100 as per the authors’ recommendations. The output matrix of cell type fractions in each sample from bulk RNA-seq is visualized using the “pheatmap” package in R. For covisualization of ADTs and cell cycle genes, ADT count matrix was transformed by centered-log-ratio (CLR) using the built-in function in R package, Compositions. Select cell cycle genes (in normalized counts) were merged with the CLR-transformed ADT matrix. The matrix was saved as a comma-separated values file and analyzed in FlowJo (v10.8.1).

All results are expressed as means with error bars reflecting standard deviation. n represents the number of repeats or animals per experiment. Statistical analyses were performed using GraphPad Prism (v9; GraphPad). Differences between two groups were assessed using unpaired two-tailed Student’s t tests. Data involving more than two groups were assessed by one- or two-way analysis of variance (ANOVA) with Tukey’s or Sidak’s post-hoc test.

Fig. S1 describes CITE-seq analysis of the LSK cells at steady-state. Fig. S2 shows that heterogeneity in the cell cycle is a specific feature of the MPP4 population. Fig. S3 documents the functional properties of slow- and fasting-cycling MPP4 subsets. Fig. S4 describes MPP4 heterogeneity based on CD62L and CD48 expression. Fig. S5 shows hematopoietic reconstitution in recipient mice treated with NMA conditioning regimen. Table S1 shows GO-Elite enrichments with prior annotated HSPC subsets. Table S2 shows TF-to-gene correlation analysis in clusters C6, C7, and C11. Table S3 lists the antibodies used for flow cytometry analysis. Table S4 lists the sequences of primers used for RT-PCR analysis. Table S5 lists the antibodies used for CITE-seq analysis. Table S6 shows the RNA metrics of the scRNA-seq. Table S7 shows a summary metrics used for data clustering by AltAnalyze and ICGS2. Table S8 indicates a custom database of HSPC-discriminating gene sets used for pathway analysis by GO-Elite.

The data are available from the corresponding author upon reasonable request. CITE-seq datasets have been deposited into the Gene Expression Omnibus (GSE243197).

The authors would like to thank Drs. Daniel Lucas and Linde Miles for valuable feedback on the manuscript. We acknowledge the assistance of Kenneth Quayle for the setting of flow cytometry sorting experiments.

This work was partially supported by the National Institutes of Health (NIH) grants R01HL141418 and R01DK133145 to D. Reynaud, R01HL122661 to H.L. Grimes, and R01DK121062 to H.L. Grimes and M.-D. Filippi. We acknowledge the assistance of the CCHMC Comprehensive Mouse and Cancer Core and the CCHMC Research Flow Cytometry core (supported by an NIH S10OD025045 grant). This work benefited from funding from the Cooperative Center for Excellence in Hematology (Cincinnati Center of Excellence in Molecular Hematology: award number U54 DK126108).

Author contributions: M. Solomon and B. Song performed and analyzed all the experiments with the help of V. Govindarajah, S. Good, A. Arasu, and E.B. Hinton; K. Thakkar contributed to the bioinformatic analyses. J. Bartram contributed to the setting of the label retention experiments. M.-D. Filippi and J.A. Cancelas provided critical insights. D. Reynaud, H.L. Grimes, and N. Salomonis designed, interpreted, analyzed, and supervised the studies and wrote the manuscript.