Multipotent progenitors with distinct origins, clonal lineage fates, transcriptomes, and surface markers yield two hematopoietic trees

Significance

The hematopoietic system replenishes blood and immune cells life-long; it has been thought to be fueled only by hematopoietic stem cells (HSC). However, recent experiments showed two distinct sources of hematopoiesis, one proceeding via HSC and HSC-derived multipotent progenitors (hMPP), the other via embryonically specified, HSC-independent MPPs (eMPP). We developed a cell barcoding approach that allowed us to trace the output of hundreds of hMPP and eMPP clones in normal, unperturbed mice. We find that eMPP are a substantial, but slowly declining, source of hematopoiesis that is enriched for lymphoid cell fates. By contrast, the HSC-hMPP route supplies hematopoiesis long-term, and has myeloid bias. We further identify molecular markers of eMPP that allow their prospective isolation and functional characterization.

Abstract

Multipotent progenitors (MPP) are the quantitative source of native hematopoiesis that have been thought to be replenished slowly by hematopoietic stem cells (HSC). However, recent fate mapping studies have revealed two developmentally distinct populations of MPP, HSC-derived MPP (hMPP), and HSC-independent, embryonic MPP (eMPP). These data raise fundamental questions on the distinctions and functions of these progenitors. Here, we mapped the clonal dynamics of the two independent MPP systems, using in situ barcoding, and barcode linkage (hMPP), or disconnect (eMPP), with HSC. The cumulative output of eMPP to hematopoiesis was 35%, and their output was enriched for lymphoid fates. Conversely, hMPP output was enriched for myeloid-restricted fates. Distinguishing HSC from eMPP outputs revealed that only ~15% of adult HSC clones underwent multilineage differentiation (lymphoid, myeloid, and erythroid). To prospectively identify eMPP, we developed PolySMART for joint profiling of PolyloxExpress RNA barcodes, surface markers, and transcriptomes, and we found that the plasma cell marker CD138 enriches for eMPP. CD138⁺ MPP are primed for self-renewal and toward lymphoid fate, and become largely but not completely replaced by CD138⁻ MPP over time, which may contribute to the loss of lymphoid output with age. Taken together, adult hematopoiesis consists of two distinct lineage trees. The source of the “eMPP tree” substantially contributes to hematopoiesis before it declines, while the HSC-hMPP tree supplies hematopoiesis life-long. Our molecular determinants distinguishing the two MPP systems may open avenues to further explore these unexpected layers of hematopoiesis.

The hematopoietic differentiation hierarchy is composed of hematopoietic stem cells (HSC) at the top, followed by multipotent progenitors (MPP), lineage-biased or -restricted progenitors, and mature blood and immune cells. While low and rare output from HSC into MPP sustains hematopoiesis in the long term, experimental evidence from in vivo lineage tracing identified the MPP stage as the driver of day-to-day hematopoiesis (1, 2) and of regeneration upon challenges, including sepsis (3). Consistent with this important function, MPP exhibit higher proliferation and differentiation rates compared to HSC. In posttransplant hematopoiesis, MPP are active only in the short-term (in the order of 6 wk). In situ, MPP are more long-lived while exhibiting limited self-renewal, and in this manner generate rapid and constant output of new lineage-restricted progenitors (1, 2).

Importantly, the view of MPP as exclusive progeny of HSC has recently been revised by the discovery of a second lineage of MPP (4–8). These MPP are of embryonic origin, arise independently from HSC, and appear to persist in adult bone marrow (4, 5). Population fate mapping of embryonic hematopoiesis revealed that this subset of MPP, referred to as embryonic MPP (eMPP), is capable of producing myeloid and lymphoid cells both prenatally and postnatally, and arises from a Flt3⁺ precursor through a pathway independent of HSC beginning around embryonic day 10.5 (E10.5) (4). A caveat in these experiments is that Flt3⁺ precursors also give rise to adult HSC to some extent, requiring subtraction of HSC output in the data to estimate eMPP output; hence, new methods to directly identify eMPP are desirable.

In our extensive data based on in situ barcoding of embryonic, fetal, and adult hematopoiesis experiments (3, 9, 10), we noticed barcode disconnect between HSC and a subset of MPP, and barcode linkage between the remainder of MPP and HSC. These data support the existence of two developmentally distinct MPP subsets and, importantly, suggest that ontogeny and output of these subsets can be studied at single-cell resolution via barcode linkage (hMPP), or disconnect (eMPP), with HSC. Here, we develop this approach to trace and quantify the clonal dynamics emerging from eMPP and hMPP. Moreover, we combined the analysis of PolyloxExpress RNA barcodes, transcriptomes, and surface markers to identify surface markers that prospectively enrich for eMPP, allowing the study of eMPP without the need for an origin-based definition via barcode disconnect with HSC.

Our analyses provide insights into the hematopoietic mosaic composed of distinct HSC-hMPP and eMPP pathways. Remarkably, we find that only in the order of 15% of adult HSC were multilineage (lymphoid plus myeloid-erythroid), while the major fate of adult HSC clones was myeloid-erythroid-restricted. We next defined HSC-derived versus HSC-independent MPP lineages by testing for barcode overlaps with HSC. This separated hMPP, which shared barcodes with HSC, from eMPP, which had private barcodes not shared with HSC. We measured the clonal fates, clone sizes, and overall contributions emerging in adult mice from eMPP and hMPP. To identify cellular and molecular hallmarks of the two MPP lineages, we took a strictly fate-linked single-cell multiomics approach (PolyloxExpress RNA barcode, surface marker expression and transcriptome, PolySMART), which led to combined information on in situ fates, transcriptome, and cell surface markers. By screening several hundred cell surface molecules, we identified CD138 as a marker enriching eMPP and characterized the transcriptional signatures of each of the two MPP populations separated by CD138. EdU incorporation assay suggests that CD138⁺ MPP are more proliferative, indicating its capacity for self-renewal and long-term survival. Finally, we followed the dynamics of CD138⁺ MPP in normal mice, and found a marked decline of this population with age. More than 40% of CD138⁺ MPP possess lymphoid-restricted fates in situ, and the decline of this MPP subset coincides with the reduction in lymphocyte production with age. In summary, we provide a comprehensive view of in situ hematopoiesis incorporating two developmentally separate MPP lineages, their distinctive fates, overall contributions to lymphoid and myeloid-erythroid cells, and the molecular and cellular determinants of each MPP lineage.

Results

Embryonic Versus Adult Barcoding Unveils HSC-Independent Hematopoiesis.

The advancement of single cell barcode tracing experiments has led to the identification of major HSC clone fate patterns, notably lymphoid-myeloid-erythroid (termed here multilineage), myeloid-erythroid-restricted, megakaryocyte-restricted, and differentiation inactive (10–12). To explore whether these HSC fate classes are evident after both embryonic or adult barcoding, we induced Tie2^MCMRosa^{PolyloxExpress} mice at embryonic day 9.5 (E9.5), or at eight-weeks of adult life using tamoxifen (Fig. 1A). At 5 to 14 mo of age, we retrieved barcodes from long term (LT)-HSC, MPP, myeloid progenitor (MyP), precursor T cell (preT), progenitor B cell (proB), megakaryocyte progenitor (MkP), erythrocyte progenitor (EryP) as well as mature T and B lymphocyte, megakaryocyte, granulocyte, and monocyte (the complete phenotypes, including ST-HSC, and all gatings are listed in SI Appendix, Materials and Methods and Fig. S1). After barcode induction, in the order of 84% (average from 5 mice barcoded at E9.5) or 53% (average from 6 mice barcoded in adult age) of all cells in the adult hematopoietic system carried recombined barcodes which enabled quantitative analysis of the origins of all populations as described (for all experiments, numbers of barcodes, and their corresponding read counts are shown in SI Appendix, Tables S1 and S2) (9, 10). Regarding adult labeling, bone marrow HSC contribute only rarely and weakly to hematopoiesis (1, 2), and hence enrichment for barcode-bearing peripheral cells is essential for measuring HSC output. We took advantage of the fact that the PolyloxExpress locus harbors a tdTomato recombination reporter. Cre-mediated excision in the PolyloxExpress allele generates shorter 3’ untranslated regions (UTRs) which correlate with higher tdTomato expression (3). tdTomato expression can therefore be used to distinguish and separate by flow cytometry barcoded from nonbarcoded cells (3). Barcode sequencing of tdTomato low and high sorted mature T and B cells, granulocytes, and monocytes confirmed that more than 90% of tdTomato low cells lacked recombined barcodes (SI Appendix, Fig. S2A) whereas tdTomato high cells were highly enriched for recombined barcodes (SI Appendix, Fig. S2B). Prior to further analysis, we filtered the retrieved barcodes using appropriate generation probability (Pgen) (9, 13) ensuring likelihood greater than 91% that barcodes were initially induced in single cells (SI Appendix, Materials and Methods and Tables S1 and S2). Because HSC divide, we considered the measured fates as HSC clone derived. In adult induced Tie2^MCMRosa^{PolyloxExpress} mice analyzed either 1 mo or 6 to 9 mo postinduction (SI Appendix, Fig. S2C), we observed an approximate two-fold decline in total numbers of barcodes (SI Appendix, Fig. S2D) and clonal barcodes (SI Appendix, Fig. S2E), suggesting clonal drift (or selection), accompanied by clonal loss, during aging in mice. Of note, overall clonal diversity was comparable between E9.5 induced and adult induced mice analyzed at similar age (6 to 9 mo old) (SI Appendix, Fig. S2F). In E9.5 induced mice, we found an average of 615 total barcodes with 418 rare barcodes. In adult induced mice, there were on average 646 total barcodes with 404 rare barcodes (SI Appendix, Fig. S2F). Hence, although the average labeling efficiency was higher in E9.5 induced mice compared to adult induced mice, there was no significant difference in clonal diversity (SI Appendix, Fig. S2F).

Fig. 1.

Clonal analysis reveals HSC fates and distinct HSC-dependent and -independent multipotent progenitors. (A) Experimental outline for barcode induction in embryonic (n = 5) and adult (n = 6) Tie2^MCMRosa^{PolyloxExpress} mice, followed by barcode analysis in adult life. Barcodes were induced at the indicated timepoints using tamoxifen (Tam). Barcodes were retrieved and analyzed for the indicated cell populations in the depicted organs. (B) Heatmaps indicating barcode propagation. After barcode induction on E9.5 (Top panel) or adult age (Bottom panel) four major fate categories were identified based on the distribution of HSC barcodes in the cell types indicated for both panels on the Bottom of the lower panel. A very rare subset of HSC clones shared barcodes only with mature lymphocytes (unclassified). Color scale (shown Top Right) represents frequencies of barcode sequencing reads. One mouse for each experiment (E9.5 and adult induction) is shown. LyP, lymphoid progenitor, corresponds to pooled preT and proB cells. (C) Percent of HSC clones with the indicated fates for mice analyzed in (A). Data from E9.5 and adult barcoding are shown side by side. (D) Experimental outline for barcoding experiments. Barcodes were induced at E14.5 using Tam, and adult mice were analyzed. (E) Percent of HSC clones with the indicated fates for mice analyzed in (A) (adult induction) and (D). (F) Experimental outline for barcoding experiments. Barcodes were induced at E14.5 using Tam, and mice were analyzed at E18.5 (n = 5), or in adult life (n = 7). Cells from E18.5 mice were harvested from fetal liver, and from the bone marrow, spleen, and thymus of adult mice. (G) Fraction of recombined barcodes in HSC and MPP in E14.5 labeled mice shown in (F). n = 5 for E18.5 analysis, and n = 7 for adult analysis. (H) Heatmaps of all MPP barcodes, classified as hMPP or eMPP based on barcodes sharing or nonsharing with HSC. Color scale (shown Top Right) represents frequencies of barcode sequencing reads. Two mice from each analysis timepoint are shown, for experiment see (F). (I) Percentage of eMPP or hMPP clones per all MPP clones for mice induced at E14.5 and analyzed at E18.5 (blue) or adult (red). (J) Effects of in silico HSC subsampling on eMPP classification. eMPP classification is shown as a function of HSC sampling in silico. The true fraction of eMPP clones is estimated from the merged data from seven mice (F) as 51.1% (dashed line). The average experimental HSC sample size is marked with the gray vertical line and shaded area. Data in (C) (E) (I) are shown from minimum to maximum values, with boxes representing the interquartile range and central lines indicating the median. Data in (G) is presented as mean ± SEM and in (J) as mean ± SD. P values for (C) (E) (I) were calculated using two-tailed, unpaired t test. *P ≤ 0.05, **P ≤ 0.01 ***P ≤ 0.001, ****P ≤ 0.0001; n.s, nonsignificant.

For both embryonic (n = 5) and adult (n = 6) barcode induction timepoints, we clearly identified the four major HSC fate categories of differentiation inactive, megakaryocyte-restricted, myeloid-erythroid-restricted, and multilineage HSC (displayed in this order from top to bottom in Fig. 1B for one mouse per induction timepoint, with additional mice shown in SI Appendix, Fig. S3). Calculating the proportions of HSC fates for the two induction timepoints, we found that the fraction of inactive, megakaryocyte-restricted, and myeloid-erythroid-restricted HSC clones were comparable based on the fractions of clones (Fig. 1C). In contrast to previous reports employing barcodes (11), we found on average no more than 10% of HSC clones were megakaryocyte-restricted (Fig. 1C and SI Appendix, Fig. S4A), which is in the range observed in transplantation experiments (14). We hence confirmed the existence of megakaryocyte-specific HSC clones in situ. However, this does not appear to be a prevailing HSC fate type.

Interestingly, we noticed a marked difference in the frequencies of multilineage HSC clones comparing E9.5 versus adult labeled Tie2^MCMRosa^{PolyloxExpress} mice (Fig. 1C and SI Appendix, Fig. S4A). While 42% of HSC clones had multilineage output after embryonic labeling, only 14% of HSC clones were multilineage in adult labeled mice (Fig. 1C and SI Appendix, Fig. S4A). Furthermore, by barcoding mice at E14.5 or adult age using Rosa^{CreERT2/PolyloxExpress} mice, i.e., unbiased barcoding (Fig. 1D and SI Appendix, Fig. S4B), and analysis after 6 to 9 mo, we also observed the low frequency (about 12%) of multilineage HSC clones (Fig. 1E and SI Appendix, Fig. S4C). A possible explanation could be that on E9.5, barcodes were induced in Tie2^MCMRosa^{PolyloxExpress} mice in common progenitors that generated adult HSC prior to E14.5, and a second population of non-HSC, which may correspond to the reported eMPP (4, 5, 8). This scenario would lead to the observed multilineage fates (Fig. 1C, induction on E9.5) to which, however, two lineages (HSC and eMPP) would have contributed.

To resolve these lineage uncertainties, we induced barcodes globally in Rosa^{CreERT2/PolyloxExpress} mice on E14.5 and analyzed adult mice at 6 to 9 mo of age, along with fetal livers on E18.5 (Fig. 1F and SI Appendix, Table S1). Total barcode analysis suggests that our barcode labeling covered the vast majority of HSC and MPP populations (Fig. 1G). Next, we visualized all MPP barcodes and defined eMPP and hMPP based on barcodes sharing or nonsharing between HSC and MPP (Fig. 1H). Barcode analysis showed that the vast majority of MPP derived from fetal liver were eMPP (Fig. 1I, blue), with the caveat that, owing to the short time between E14.5 and E18.5, a fraction of MPP could have been generated prior to E14.5 from HSC. A large proportion of eMPP clones were still evident in 6 to 9 mo old adult mice, although barcode sharing between HSC and MPP increased by this time (Fig. 1I, red). Examples of the numbers of shared and nonshared barcodes in HSC and MPP are shown for one E18.5 fetus and one adult mouse in SI Appendix, Fig. S5 A and B. Furthermore, to validate that our definition of eMPP and hMPP clones based on barcode sharing achieves a reliable separation of clones derived from HSC (hMPP) vs clones independently induced in MPP (eMPP), we performed barcode network analysis in E14.5-labeled Rosa^{CreERT2/PolyloxExpress} mice. Barcode network analysis addresses the evolutionary relationship between barcodes, reflecting the history of the barcode recombination process. In proliferating and differentiating populations, ongoing barcode recombination may produce related, rather than identical, barcodes, potentially masking parent–progeny relationships between cells. However, we were able to exclude this possibility in our data, since HSC and eMPP share barcodes only as expected from random connectivity (SI Appendix, Fig. S5 C–E). Thus, eMPP barcodes are not only unshared but also evolutionarily unrelated to HSC barcodes, confirming the HSC independent origin of eMPP. Collectively, our data support that at least a fraction of MPP is generated independent of HSC by E14.5, and that such embryonic MPP persist into adulthood. In addition, this MPP population is gradually replaced by an influx of MPP derived from HSC (hMPP). The total MPP population is thus a mosaic from two developmentally distinct subsets.

Sufficient HSC sampling is essential for the definition of eMPP and hMPP. To assess the influence of sampling quantitatively, we subsampled in silico HSC and MPP populations from the two-dimensional clone size distribution from merged experimental barcodes (SI Appendix, Fig. S5F) from seven Rosa^{CreERT2/PolyloxExpress} mice (Fig. 1F). Given the measured clone size distribution, we sampled computationally from an estimated MPP population of 66,000 cells and an HSC population of 30,000 cells. We drew a constant number of ca. 20,000 MPP (in the order of the average experimental sampling), and varied HSC sample sizes between 10% to 100% of the total HSC number (Fig. 1J). The corresponding scatter plots for the lowest (43.7%), average (78.4%), and highest (100%) HSC sampling rates are shown in SI Appendix, Fig. S5G. In brief, for all analyzed mice, our HSC sampling was well above the threshold below which false positive eMPP calling occurs. Therefore, very few, if any, eMPP were false positives.

Phenotypic and Functional Distinctions Between eMPP and hMPP.

The coexistence of eMPP and hMPP in adult bone marrow prompts inquiry into their distinctions, including phenotypic differences, their lineage outputs, and their relative contributions to native hematopoiesis. The adult MPP compartment contains MPP3 (Lin^-Kit⁺Sca1⁺CD150^-CD48⁺Flt3^-) and MPP4 (Lin^-Kit⁺Sca1⁺CD150^-CD48⁺Flt3⁺) subpopulations, which in transplantation assays are myeloid and lymphoid biased, respectively (15). To correlate eMPP and hMPP with MPP3 and MPP4, we quantified the proportion of eMPP or hMPP clones that overlapped phenotypically with MPP3 and MPP4, respectively. We found that the majority of eMPP clones were detected within the MPP4 subset, although this was not exclusive (Fig. 2A). By contrast, hMPP clones were equally distributed in all phenotypic MPP subsets (SI Appendix, Fig. S5H).

Fig. 2.

Phenotypic and functional comparison of eMPP and hMPP. (A) Percent of eMPP clones in phenotypically defined MPP subsets from adult mice (6 to 9 mo). (B) Outline of the experiments for analyzing emergence of HSC-label to downstream populations at 6 and 14 mo after adult label induction in Tie2^MCMRosa^{PolyloxExpress} mice. (C and D) tdTomato^high (HSC label) acquisition in downstream compartments after 6 (C) or 14 (D) months post label induction. (E) Lineage output from eMPP or hMPP clones into progenitors. LyP, MyP, and LyP + MyP fates are shown. (F–H) Clone sizes, based on relative frequencies of Pacbio sequencing reads, of eMPP (F) or hMPP (G) with distinct fates, or eMPP and hMPP across all fates (H). (I) Overall contribution, based on Pacbio sequencing reads, of eMPP and hMPP to lymphoid and myeloid progenitors, respectively. P values for (A) (E) (H) were calculated using two-tailed, unpaired t test. Data in (A, C, D, F, G–H) are shown from minimum to maximum values, with boxes representing the interquartile range and central lines indicating the median. Data in (E) (I) are presented as mean ± SEM. P values for (C, D, F, and G) were calculated using the one-way ANOVA with the Tukey multiple-comparison test. * P ≤ 0.05, ** P ≤ 0.01 ***P ≤ 0.001, **** P ≤ 0.0001; n.s, nonsignificant.

Because MPP4 but not MPP3 were enriched in eMPP, we studied label propagation from HSC to ST-HSC and to MPP3 and MPP4 in adult induced Tie2^MCMRosa^{PolyloxExpress} mice (Fig. 2B). Both after 6 (Fig. 2C) and 14 (Fig. 2D) months, ST-HSC and MPP3 were label equilibrated to each other, while much less label was found in MPP4. This indicates that ST-HSC and MPP3 are more of HSC-origin compared to MPP4. This is consistent with the predominance of eMPP in MPP4 (Fig. 2A). In 14-mo chase mice, MPP4 labeling “catches up” faster than MPP3, potentially suggesting that many MPP4 are of non-HSC origin, which leads to lower label compared to ST-HSC. At the later time point, label in MPP4 increases because the HSC-derived contribution (which has HSC label) becomes larger at the expense of the fading embryonic contribution (which lacks HSC label). Of note, label in ST-HSC and MPP3 was very similar to label in myeloid-erythroid progeny (Fig. 2 C and D). In contrast, the lower MPP4 label was characteristic of lymphocyte progeny label (Fig. 2 C and D). We interpret these data to indicate a differentiation path from LT-HSC to ST-HSC, and MPP3. These label observations also suggest that ST-HSC–MPP3 is the dominant route to myeloid-erythroid lineages, while MPP4 (mostly eMPP) are feeding the lymphoid lineages.

To determine the fates emerging from eMPP and hMPP in vivo, we analyzed adult Rosa^{CreERT2/PolyloxExpress} mice barcoded on E14.5 (Fig. 1F). Barcodes were retrieved from HSC, MPP, MyP, lymphoid progenitors, and mature lineages (Fig. 1F). We first focused on downstream lymphoid and myeloid progenitors, which provide the most current view on cell production from MPP. Frequencies of clones with lymphoid-restricted fates were significantly higher in eMPP (~32%) compared to hMPP (~6%) (Fig. 2E). Frequencies of clones with myeloid-restricted fates were higher in hMPP (~72%) compared to eMPP (~52%) (Fig. 2E). No significant difference was observed for clones with multilineage fate (lymphoid plus myeloid) comparing the two populations (Fig. 2E).

Next, we compared the clone sizes of MPP populations using the relative frequencies (single clone reads over all clone reads) of PacBio sequencing reads as a proxy. Clone sizes across distinct fate-defined eMPP were similar (Fig. 2F). In contrast, multilineage hMPP clones were significantly larger than unilineage hMPP clones (Fig. 2G). When MPP clones were combined regardless of their fates, hMPP clones were on average four times larger than eMPP clones (Fig. 2H).

Finally, we addressed the overall contributions to hematopoiesis of the two MPP lineages. In 6 to 9 mo old Rosa^{CreERT2/PolyloxExpress} mice barcoded on E14.5 (Fig. 1F), based on barcode reads, in the order of 30% (myeloid progenitors) and 40% (lymphoid progenitors) originated from eMPP, the remainder were hMPP derived (Fig. 2I). When, additionally, mature cells of each lineage were included to estimate the cumulative contributions over 6 to 9 mo of eMPP and hMPP, similarly in the order of 40% of lymphoid and myeloid cells arising from eMPP were found (SI Appendix, Fig. S5I). In conclusion, eMPP and hMPP have different clone sizes, and distinct fates. While eMPP still produce new progenitors even after 6 to 9 mo, the overall contribution of hMPP (roughly 60% of native hematopoiesis) prevails.

PolySMART Linking Cellular Origins, Surface Markers, and Gene Expression.

Until now, the definition and analysis of eMPP is dependent on barcoding. To link fates with molecular determinants, barcoding needs to be combined with other readouts. However, in the process of barcode retrieval cells are disrupted, which impedes their further characterization. In an effort to find markers associated with MPP origins, we screened for expression of 253 surface markers using LEGENDScreen^TM antibodies (experimental outline and populations shown in Fig. 3A). Focusing on markers showing bimodal expression on MPP4 (which are enriched in eMPP), we selected 58 candidates for CITEseq experiments (Fig. 3B). To simultaneously profile marker expression and PolyloxExpress barcodes in the same cell, we integrated PolyloxExpress and CITEseq workflows (10, 16). This enabled joint analyses of PolyloxExpress RNA barcode, Surface Marker expression, and Transcriptome at single-cell resolution. PolySMART works by staining cells bearing in vivo generated barcodes with a pool of oligonucleotide-conjugated antibodies, followed by 10x single-cell RNAseq (scRNAseq) to capture RNA barcodes, antibody-derived tags (ADT), and transcriptomes (Fig. 3C).

Fig. 3.

Screening for surface markers on hematopoietic stem and progenitor cells (HSPC) and the development of PolySMART. (A) Workflow for screening for expression of 253 candidate protein markers on HSPC. Bone marrow cells, prestained with HSPC markers, were aliquoted and individually stained with 253 PE-conjugated antibodies from LEGENDScreen™ (SI Appendix, Materials and Methods). (B) Expression profiles of 58 markers selected in (A). Two columns of each population represent result from two independent screening experiments. Dot size represents the percentage of cells expressing the given marker. Color scale represents the MFI recorded by flow cytometry. (C) Workflow for PolySMART. Both mRNA and ADT molecules contain a polyA tail which is captured by polyT sequence on the indexed beads. The molecules are reverse transcribed (RT) into cDNA, and globally amplified using a common adapter sequence (SI Appendix, Materials and Methods). (D) Experiment to validate the function of PolySMART. LSK and LK, stained with oligo-conjugated antibodies tags (Kit, Sca1, CD150, CD48, CD135, CD16/32, CD34), were isolated and utilized as input for the workflow depicted in (C). (E) Proteo-transcriptional landscape derived from experiment in (D) (considering both single cell transcriptome and oligo-conjugated antibody staining) of HSPC, with LSK and LK color coded. (F) Proteo-transcriptional landscape highlights populations defined by ADT-based gating (SI Appendix, Fig. S6A). (G) Projection of indicated HSPC markers onto the landscape, with the color scale representing the normalized ADT abundance.

We induced barcodes on E9.5 in Tie2^MCMRosa^{PolyloxExpress} embryos, and sorted LSK (Lin^-Kit⁺Sca1⁺) and LK (Lin^-Kit⁺Sca1^-) progenitors from adult bone marrow after 4 mo (Fig. 3D). Cells were stained with oligonucleotide-conjugated antibodies against established HSPC markers CD34, CD150, CD48, CD135, and CD16/32. At single cell resolution, LSK and LK populations were then analyzed by PolySMART. We integrated ADT and gene expression datasets for pooled LSK and LK. We applied an unsupervised approach to generate a proteo-transcriptional landscape for joint visualization of LSK and LK (the two sorted populations are highlighted by different colors in Fig. 3E). We then used ADT reads as a proxy of marker expression for gating, and subdivided LSK/LK into known HSPC subsets based on established marker combinations. ADT read-based gating plots are shown in SI Appendix, Fig. S6A. HSPC subsets (SI Appendix, Fig. S6A) were projected onto the entire LSK/LK proteo-transcriptional landscape (Fig. 3F). We observed the expected transition from LT-HSC to downstream progenitors which bifurcated at the CMP stage into GMP and MEP (Fig. 3F). For further validation, we also projected ADT markers CD150, CD48, CD135, and CD16/32 onto this landscape (Fig. 3G). Individual surface markers were highly expressed in their phenotypically associated regions, validating the specificity of our ADT data (Fig. 3G). The transcriptome of each ADT-defined subset faithfully recapitulated published gene expression data based on flow cytometry-purified population sorting (3, 10, 11) (SI Appendix, Fig. S6B), demonstrating that ADT can reidentify known immunophenotypically defined populations. Finally, considering PolyloxExpress barcodes, we found between 17 and 38% of cells which contained joint information on barcodes, markers, and transcriptomes (SI Appendix, Fig. S6C). We depicted cells with PolyloxExpress barcodes on the proteo-transcriptional landscape (SI Appendix, Fig. S6D), and observed a distribution similar to cells without PolyExpress barcodes (SI Appendix, Fig. S6E). In summary, with the development of PolySMART, the stage is set for simultaneous identification of eMPP versus hMPP via barcodes, and the search for cell surface markers and transcriptomic signatures distinguishing the two MPP lineages.

Surface Marker Expression Associated With eMPP and hMPP.

We analyzed, using PolySMART, two 6-mo-old Rosa^{CreERT2/PolyloxExpress} mice labeled on E14.5 (Fig. 4A). To correlate barcodes with surface phenotypes, we stained lineage-depleted bone marrow cells with oligonucleotide-conjugated antibodies selected from the surface marker screening experiments (Fig. 3B), and sorted total MPP for the PolySMART workflow (Fig. 4A). In parallel, we purified HSC and subjected them in bulk to DNA barcode analysis (Fig. 4A). eMPP and hMPP were defined based on barcode sharing or nonsharing with HSC (SI Appendix, Fig. S7A). These eMPP and hMPP were analyzed for differential binding to the 58 antibodies which revealed on each end of the scale markers enriched for either eMPP or hMPP (Fig. 4B). These data suggest that markers exist that may help to prospectively isolate the developmentally distinct MPP lineages.

Fig. 4.

Identification of markers differentially expressed by eMPP and hMPP. (A) Experimental outline for PolySMART experiments (n = 2). Total MPP, stained with oligo-conjugated antibodies, were sorted and analyzed using PolySMART. HSC were sorted in parallel for DNA barcode retrieval. (B) Differentially expressed markers associated with MPP not sharing (Left, red bars), or MPP sharing (Right, blue bars) barcodes with HSC. (C) UMAP plot showing the dimensionality reduction results for MPP based on ADT data from (A). MPP3/4, defined by expression of CD135 (Flt3) using ADT, are color marked. (D) UMAP plot showing the dimensionality reduction results of MPP based on ADT data. Unsupervised clustering revealed three clusters (termed 0, 1, and 2) highlighted by color marks. (E) Projection of eMPP and hMPP marker signatures on the total MPP UMAP. eMPP (Left panel), and hMPP (Right panel) signature distributions are highlighted. The level of the signature is indicated by the color scales (Upper Right). (F) Relative eMPP and hMPP marker signature expression in clusters shown in (D). Y axis shows Z score. (G) Expression of individual markers in HSPC. MFI values for each population of the indicated markers are shown. (H) Experiment for validating the correlation between CD138 expression and eMPP (n = 4). (I) Percent of eMPP clones observed in CD138⁺ and CD138⁻ MPP4 subsets. (J) Experimental outline for Flt3^CreERT2Rosa^{PolyloxExpress}-driven MPP fate mapping. (K) Labeling frequencies of the indicated progenitors in the E15.5 fetal livers from E9.5 induced Flt3^CreERT2Rosa^{PolyloxExpress} embryos. (L) Fraction of CD138⁺ and CD138⁻ cells in the Flt3-fate mapped MPP4 (tdTomato^high). Data in (F, I, K, and L) are shown from minimum to maximum values, with boxes representing the interquartile range and central lines indicating the median. P values for (F, I, and L) were calculated using two-tailed, unpaired t test. *P ≤ 0.05, **P ≤ 0.01 ***P ≤ 0.001, ****P ≤ 0.0001; n.s, nonsignificant.

We next visualized the total MPP landscape in the dimensionality-reduced UMAP-space based on the 58 markers, in which we highlighted for orientation the known MPP3 and MPP4 populations based on expression of the CD135 (Flt3) ADT (Fig. 4C). We next searched for MPP subpopulations among total MPP using an unsupervised clustering approach in the original surface-marker space. This uncovered three clusters, termed 0, 1, and 2 (Fig. 4D). To further elucidate the regions where eMPP or hMPP are located, we took the top markers identified in Fig. 4B, which were derived from the direct eMPP-to-hMPP comparison, to calculate eMPP and hMPP signature scores for all MPPs. We visualized these signature scores in the UMAP projection (Fig. 4E). The eMPP signature was largely enriched in cluster 0, whereas the hMPP signature was predominantly located in clusters 1 and 2 (Fig. 4F). In summary, cells with eMPP and hMPP signatures occupy spatially distinct areas within the MPP compartment.

Among the differential markers enriched in eMPP in the comparison shown in Fig. 4B, we identified several lymphocyte lineage-associated markers, including the B cell marker CD5, the plasma cell marker CD138 and the B and T cell attenuator CD272, none of which have thus far been described to dissect MPP subpopulations. MPP that transcriptionally express CD5 and CD138 (SI Appendix, Fig. S7B) support the surface marker identification. The potential association of markers with eMPP requires further analysis by cell sorting and downstream analysis.

CD138, the Plasma Cell Protein, Enriches eMPP.

To test for surface expression of markers identified by barcode-based (Fig. 4B) analysis, we stained by flow cytometry MPP3 and MPP4 from normal mice with antibodies against ESAM, CD138, CD5, and CD272 (Fig. 4G). We also included LT-HSC, and ST-HSC to look for MPP specificity (Fig. 4G). While CD5 and CD272 had a tendency for uniformly higher expression in MPP4 compared to other populations (see MFI in the histogram panels in Fig. 4G), CD138 showed bimodal expression in MPP4 (Fig. 4G). ESAM, as predicted from Fig. 4B, was reduced in MPP4. None of the markers were expressed in MPP3 despite the fact that some eMPP could be included in MPP3 (Fig. 2A).

Because eMPP were enriched in MPP4 (Fig. 2A), and MPP4 contained a “CD138 subset” (Fig. 4G), we asked whether CD138 expression enriches for MPP that do not share barcodes with HSC. We sorted from E14.5-labeled adult Rosa^{CreERT2/PolyloxExpress} mice (n = 4) CD138⁺ and CD138⁻ MPP4, along with HSC and mature cells (Fig. 4H). Staining and postsort analysis of CD138⁺ and CD138⁻ MPP4 subsets are shown in SI Appendix, Fig. S7C. Confirming our marker analysis, we found a significantly higher fraction of eMPP clones (non-HSC barcode sharing) in the CD138⁺ compared to the CD138⁻ subset (Fig. 4I). This demonstrates a correlation between CD138 expression and the HSC-independent origin of MPP.

Given that eMPP have been characterized previously to arise from a Flt3-expressing embryonic progenitor (4), we next asked whether Flt3 fate mapping and CD138 expression “intersected.” We induced Flt3^CreERT2Rosa^{PolyloxExpress} mice at E9.5 with tamoxifen and analyzed fetal liver on E15.5 or bone marrow of 10 to 14 wk old adult mice (Fig. 4J). We chose this analysis time point because the eMPP and hMPP lineages are separated by E14.5 (Fig. 1I). Progenitors derived from Flt3 labeling were identified by tdTomato^high expression (SI Appendix, Fig. S7D). While HSC and ST-HSC exhibited very low labeling (LT 2.9%; ST 6.4%) at E15.5, a large proportion of MPP4 (55.0%) and a subset of MPP3 (25.1%) were derived from Flt3⁺ embryonic progenitors (Fig. 4K). The labeling also persists into adulthood, although the labeled fraction is smaller (SI Appendix, Fig. S7E). Therefore, our data support again the HSC independent eMPP origin. The higher abundance of fate-mapped MPP4 compared to MPP3 is consistent with the enrichment of eMPP that we described based on barcode analysis in Fig. 2A. Importantly, the vast majority of Flt3-labeled MPP4 (80.5%) were CD138⁺ (Fig. 4L and SI Appendix, Fig. S7 F and G), confirming CD138 as a marker that enriches for eMPP.

CD138 Expression Marks Transcriptionally Distinct MPP Subsets With Key Functional Implications.

To explore potential transcriptional differences between CD138⁺ and CD138⁻ MPP4, we utilized our PolySMART data (Fig. 4A). We first defined the two subsets using ADT-CD138 gating (SI Appendix, Fig. S8A). We then analyzed the transcriptional data that were linked to each subset at single cell resolution. Differential gene expression (DEG) was visualized as fold change vs P value in a volcano plot (Fig. 5A). Expression of 347 genes was elevated in the CD138⁺ subset, and 392 genes in the CD138⁻ subset (SI Appendix, Table S3). Based on fold changes, among the top 20 genes in the CD138⁺ subset we found genes related to lymphocyte development (Dntt, Il7r), genome stability, and DNA repair (Blm, Hmgb2, Rad51b) (Fig. 5A). Moreover, this subset expressed genes involved in cell cycle regulation (Cdca8, Top2a) and cellular proliferation (Ki-67, Haus3) (Fig. 5A). Emb, which was highly and differentially expressed in the CD138⁺ subset, encodes a transmembrane glycoprotein involved in cell growth and development, potentially mediating adhesion, however, its role in hematopoiesis has not been clarified. The CD138⁻ subset, in contrast, was characterized by expression of a more diverse array of genes, including genes related to metabolism (Impa1, Car2, Csgalnact1, Gsto1, and Ogt), immune response (Irf2, Ifitm3), and development (Mycn, JunD) (Fig. 5A). Of note, Hemogen (Hemgn), which was highly and differentially expressed in the CD138⁻ subset, is a hematopoietic-specific nuclear protein driving expansion of myeloid progenitors (17).

Fig. 5.

Transcriptional characterization of MPP4 subsets defined by CD138. (A) Volcano plot showing the Log2 fold change of differentially expressed genes (DEG), and their P values comparing CD138⁺ MPP4 and CD138⁻ MPP4. (B) Top ranked GO terms based on DEG in (A) in CD138⁺ (Right) and CD138⁻ (Left) MPP4 subsets. The value in each bar represents −log10(P value). (C) Transcriptional landscape of phenotypically defined stem/progenitor populations (color coded) reconstructed using the diffusion map. Each dot is a single cell. (D) Projection of CD138⁺ and CD138⁻ MPP4 onto the transcriptome landscape. (E) Comparison of the DC_1 value from (D) (X axis) versus relative cell counts (Y axis) of CD138⁺ and CD138⁻ MPP4 subsets. (F) Lineage output from CD138⁺ or CD138⁻ MPP4 into progenitors (LyP, MyP, and LyP + MyP). Data from experiment shown in Fig. 4H. (G and H) Cell cycle G2M (G) or S (H) score comparison between CD138⁺ and CD138⁻ MPP4. (I) EdU incorporation in vivo of CD138⁺ and CD138⁻ MPP4 subsets. (J) Percent EdU⁺ cells within each MPP4 subset. (K and L) Absolute cell numbers of CD138⁺ MPP4 (K), and relative frequencies of CD138⁺ MPP4 per total MPP4 population (L) as a function of mouse age. P value in (B) was calculated using the hypergeometric test. Data in (F) are presented as mean ± SEM. Data in (G, H, J, K, and L) are shown from minimum to maximum values, with boxes representing the interquartile range and central lines indicating the median. P values in (F, J, K, and L) were calculated using two-tailed, unpaired t test. P values in (E, G, and H) were calculated using the 2-sided Wilcoxon–Mann–Whitney rank-sum test. *P ≤ 0.05, **P ≤ 0.01 ***P ≤ 0.001, ****P ≤ 0.0001; n.s, nonsignificant.

To gain further insights into cellular functions operating in each subset, we performed Gene Ontology (GO) enrichment analysis using the DEGs. Enriched GO terms of the CD138⁺ subset encompassed several biological processes, including negative regulation of apoptosis, DNA repair, and cell cycle (Fig. 5B and SI Appendix, Fig. S8B). In contrast, GO terms enriched in the CD138⁻ subset most of all pointed at metabolic pathways and their regulation (Fig. 5B). Hence, the CD138⁺ and CD138⁻ MPP4 subsets are transcriptionally different, and their GO term analysis suggests distinct cellular functions and fate biases.

Toward a snapshot of cell positions on lineage trajectories, we examined the single-cell transcriptomes of CD138⁺ and CD138⁻ MPP4 by projecting both subsets onto our previously published transcriptome landscape (10). This transcriptome landscape of hematopoietic stem and progenitor cell (HSPC) was constructed using 3000 highly variable genes identified among HSC, MPP, CLP, and CMP. The landscape shows an HSC/MPP center and the bifurcation into the myeloid lineages (DC_2) and the lymphoid lineages (DC_1) (Fig. 5C). As expected, both CD138⁺ and CD138⁻ MPP4 subsets were primarily found in the MPP compartment (Fig. 5D). Using DC_1 as a metric for the extent of lymphoid differentiation, we found significantly more CD138⁺ MPP4 with higher DC_1 value (Fig. 5E), which, at the transcriptome level, suggest that CD138⁺ MPP4 are more predisposed to lymphoid development. In direct support of this notion, fate analysis (from the experiment in Fig. 4H) showed that CD138⁺ subset was more lymphoid-biased than the CD138⁻ subset, and, in fact, harbored 40% lymphoid-restricted fates (Fig. 5F). Indeed, the fate bias emerging from the CD138⁺ subset (Fig. 5F) recapitulated the fates of eMPP defined by barcode linkage (Fig. 2E).

DEG and GO terms (Fig. 5 A and B) indicated differences in proliferation comparing the CD138 subsets. To explore this further, we compared the transcriptomes for cell cycle scores (SI Appendix, Table S4) for cells in G2M (Fig. 5G), and S (Fig. 5H), summarized as non-G1 (SI Appendix, Fig. S8C). These data indicate that CD138⁺ cells are more proliferative compared to CD138⁻ cells. We next compared the in vivo proliferation of the two MPP subsets using the EdU incorporation assay (Fig. 5I). Indeed, a significantly higher proportion of CD138⁺ MPP4 was EdU⁺ (Fig. 5J and SI Appendix, Fig. S8D). In brief, based on transcriptome analysis and in vivo EdU incorporation, CD138⁺ MPP4 cells are more proliferative than CD138⁻ MPP4.

To compare two MPP4 subsets by transplantation, we adoptively transferred equal numbers (700 to 1,000 cells per subset) of CD138⁺ MPP4 (CD45.2) together with CD138⁻ MPP4 (CD45.2) along with one million total bone marrow cells (CD45.1) for survival into lethally irradiated CD45.2 recipient mice (SI Appendix, Fig. S8E). CD138⁺ MPP4 and CD138⁻ MPP4 donor cells were distinguished by panYFP or panRFP expression (SI Appendix, Fig. S8E). Periphery blood was collected weekly from the transplanted recipients for four consecutive weeks, and bone marrow was analyzed at week five (SI Appendix, Fig. S8E). The overall contribution after transplantation of both MPP subsets to nucleated cells was similar (SI Appendix, Fig. S8F). Except for day 21, the B cell contribution was higher from CD138⁺ MPP4 compared to CD138⁻ MPP4 (SI Appendix, Fig. S8G). In contrast, early myeloid output was higher from CD138⁻ MPP4 compared to CD138⁺ MPP4 (SI Appendix, Fig. S8H). While these data were consistent with the lymphoid and myeloid bias of CD138⁺ MPP4 and CD138⁻ MPP4, respectively, this distinction became much more obvious with the endpoint analysis of proB cells and myeloid progenitors (CMP, GMP, and MEP) at 5 wk (SI Appendix, Fig. S8 I and J). Finally, and in line with the known posttransplant behavior of MPP, both subsets had vanished from the bone marrow by week 5 (SI Appendix, Fig. S8K).

Loss of lymphoid cell production (myeloid bias) occurs during aging. Label propagation studies have shown that myeloid bias is, at least partially, attributed to reduced lymphoid production rather than increased myeloid output (2). However, the cellular basis of decreased lymphoid production remains unclear (2, 18). Given the lack of input from HSC into the eMPP compartment in adult mice, we reasoned that the CD138⁺ subset, which serves as a source of lymphoid production, may decline with age. Therefore, we quantified absolute numbers of CD138⁺ MPP4 (Fig. 5K) and CD138⁻ MPP4 (SI Appendix, Fig. S9A), as well as the relative proportions of each MPP4 subset within the MPP4 compartment (Fig. 5L and SI Appendix, Fig. S9B). While there was a decline of CD138⁺ MPP4 (Fig. 5 K and L and SI Appendix, Fig. S9C), we observed a plateau of CD138⁺ MPP4 persisting up to at least 18 mo of age, pointing at the persistence of CD138⁺ MPP4 long into the adult mouse life. To examine the lineage potential of MPP4 subsets during aging, we analyzed the fate of CD138⁺ and CD138⁻ MPP4 24 mo after E14.5 labeling using progenitor barcodes (SI Appendix, Fig. S9D). The fate pattern indeed resembled the fates observed after 6 to 9 mo of age (Fig. 5F), however, the lymphoid bias was somewhat less pronounced (SI Appendix, Fig. S9 E and F). Given the overall lymphoid contribution of eMPP in the order of 40% (Fig. 2I), the loss of CD138⁺ MPP4 may contribute to reduced lymphoid output, and thus the myeloid bias during aging.

Discussion

Via infrequent differentiation, HSC give rise to MPP, which in recent years have been recognized as long-lived drivers of day-to-day hematopoiesis (1, 2). Thus, rather than short-lived, transit amplifying cells, MPP have a more central role in hematopoiesis than previously appreciated, including their rapid response to demand for challenges, notably sepsis (3, 19). Interestingly, recent lineage tracing studies uncovered the existence of a new type of MPP that is derived from embryonic hemogenic progenitors, that is independent of HSC, and that persists postnatally in adult bone marrow (4, 5, 7, 8). Hence, the MPP compartment in adult mice is a mosaic of two ontogenetically distinct, coexisting subsets, one HSC-derived and one HSC-independent. These findings are not a minor modification but may necessitate a revision in our understanding of structure and function of the hematopoietic tree.

Here, we used in situ barcoding, linked with single-cell surface marker and transcriptome analyses, to trace and quantify the output of the separate eMPP and hMPP systems fueling adult hematopoiesis. Our findings show distinct fates of HSC clones in native hematopoiesis, belonging to four major fate categories: lymphoid-myeloid-erythroid (multilineage), myeloid-erythroid-restricted, megakaryocyte-restricted, and differentiation inactive, which is in agreement with previous work by us and others (10, 11, 20). In contrast to a previous report, however, which concluded that megakaryocyte-restricted HSC represent the dominant fate (11), we find this fate to be the rarest (~10%). While the basis for this discrepancy remains to be determined, it is conceivable that limited sampling may have missed nonmegakaryocyte lineage output, overestimating megakaryocyte-restricted HSC.

Remarkably, frequencies of multilineage HSC clones dropped from ~42% after labeling on E9.5 to ~14% after adult labeling. This discrepancy was not due to Tie2-driven barcoding in embryonic or adult HSC. We also observed a similarly low fraction of multilineage HSC clones (12%) when HSC were holistically labeled using Rosa^{CreERT2/PolyloxExpress} on E14.5 fetal mice (Fig. 1 D and E). Hence, in contrast to multipotency as a hallmark of HSC in bone marrow, our data now show that the majority of adult HSC have only myeloid-erythroid but not lymphoid fates.

The fact that HSC barcoding at E9.5 led to more multilineage barcodes compared to HSC barcode induction in the adult mice (Fig. 1C) suggested that at E9.5 we also barcoded another, non-HSC, source of adult hematopoiesis, conceivably the reported eMPP. In support of this interpretation, ubiquitous barcode induction in Rosa^CreERT2 mice at E14.5 directly demonstrated that the vast majority of MPP did not share barcodes with HSC at E18.5. This implies that such MPP were mostly generated in the embryo prior to E14.5, and independently from HSC. The barcode dissociation between HSC and MPP at E18.5 also indicates that HSC minimally contributed to MPP, at least until this fetal stage. Minimal HSC contribution to progenitors at late gestation has also been reported by others (5, 7). Barcode sharing increased between HSC and MPP over time in adult mice, suggesting gradual but not complete eMPP replacement by an influx of HSC-derived MPP. In conclusion, for extended periods of time (at least 18 mo), adult bone marrow harbors two ontogenetically distinct MPP populations which each generate their downstream hematopoietic hierarchy. Functional implications of this “dual MPP structure”, as opposed to an exclusively HSC-driven model, remain to be revealed.

What are the relative contributions of each MPP type to hematopoiesis over time? A recent fate mapping study concluded that eMPP are the major source of lifelong lymphoid cells production (4). We find that eMPP contain more lymphoid-restricted fates compared to hMPP, and that the overall contribution to both lymphoid or myeloid progenitors, which reflects recent eMPP activity, is in the order of 40% in 6 to 9 mo old mice. Nevertheless, eMPP could strongly contribute to hematopoiesis in early life before they are gradually replaced by the slow but steady generation of hMPP from HSC. Hence, the relative contributions to hematopoiesis from eMPP and hMPP are not fixed, but underlie dynamic changes. In brief, the HSC-hMPP route remains the major source for myeloid and lymphoid cells in adult hematopoiesis.

Currently, HSC-independent eMPP can only be inferred from fate mapping or barcoding data, and the underlying experiments depend on specific mouse mutants. Toward understanding the cellular and molecular properties of the two MPP systems, and to make studies in normal mice feasible, it would be desirable to obtain markers distinguishing the ontogenetically different MPP types. We developed a single-cell multiomics lineage tracing technique, PolySMART, which allows joint profiling of cell fate, surface marker expression, and transcriptome. These experiments were first guided by the strict criterion of barcode linkage or nonlinkage with HSC to define hMPP and eMPP, which led to a set of differentially expressed cell surface markers for each MPP subset. These included lymphoid-related markers (CD138, CD5, CD272) that correlated with the HSC-independent MPP origin. Interestingly, the plasma cell marker CD138 showed bimodal expression in the MPP4 compartment and enriched for eMPP. While our barcoding data suggest that other embryo-derived MPP may also exist outside the MPP4 compartment (e.g MPP3), low expression of potential markers in these compartments prohibited further analysis. The eMPP enriched CD138⁺ MPP4 is lymphoid-biased by barcode analysis which is supported by their position in a stem and progenitor transcriptional landscape. This subset expressed genes related to cell proliferation and antiapoptosis and was more proliferative in vivo. CD138⁺ MPP4 also expressed higher levels of Dntt and Il7r, suggesting that this subset may be developmentally closer to lymphoid progenitors (e.g., CLP) than CD138⁻ MPP4. Consistent with eMPP receiving no input from HSC, expression of these genes may indicate their degree of self-renewal and long-term survival. In contrast, the CD138⁻ subset is more metabolically active and expresses genes related to myeloid differentiation. Hence, ontogenetically distinct MPP subsets are indeed phenotypically and transcriptionally different. While there are limitations of CD138 in distinguishing eMPP from hMPP, and better or complementary markers may further improve resolution, sorting MPP from normal mice for and against CD138 yielded two populations differing in transcriptomes (GO terms, Fig. 5B), and proliferation (Fig. 5 G–J). Therefore, CD138 expression, while not an absolute marker for eMPP, represents in our view a useful MPP marker associated with given functional characteristics. Whether hMPP and eMPP also exist in the human bone marrow, and CD138 also marks an HSC-independent human MPP remains to be determined. Along these lines, a yolk sac-derived, HSC-independent and lymphoid-biased progenitor has recently been reported (21).

Future work will be required to potentially translate these findings into, for instance, differential immune responsiveness arising from the two MPP systems. Enriching markers falling short of purifying populations to homogeneity have been instrumental in stem and progenitor research (22, 23). The identification of CD138 expression in adult bone marrow on MPP of embryonic origin may hence stimulate future research on the functions of two phenotypically, functionally, transcriptionally distinct MPP subsets. Given their longevity in situ (at least for up to 18 mo), and their lymphoid plus myeloid fates, eMPP may also be considered as “HSC2,” next to the classical HSC in bone marrow. However, such nomenclature would need consensus in the field.

Materials and Methods

Transgenic mice were treated with tamoxifen at defined embryonic, fetal, or adult stages to induce barcode recombination. Hematopoietic progenitor and mature cells were isolated from bone marrow and lymphoid organs, stained with fluorescent or oligonucleotide-conjugated antibodies, purified by fluorescence activated cell sorting, and analyzed by barcode PCR on cell lysates (for DNA barcodes), or by scRNAseq with target enrichment PCR (for RNA barcodes, transcriptome, and ADT). Data analysis included retrieval of PolyloxExpress barcodes, computational filtering of clonal (i.e., rare) barcodes based on generation probability (Pgen), single-cell transcriptomic and proteomic analysis, and in silico subsampling of HSC and MPP to evaluate experimental sampling effects. Full experimental details, including protocols, key reagents, and resources, are provided in SI Appendix, Materials and Methods.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

Single-cell transcriptome and ADT sequencing data for hematopoietic stem and progenitor cells from PolyloxExpress mice, PolyloxExpress barcode sequencing data (single-cell and bulk analysis) have been deposited in Gene Expression Omnibus (GEO) under the accession number GSE272588 (24). All other data are included in the manuscript and/or SI Appendix.