Lifelong multilineage contribution by embryonic-born blood progenitors

SUMMARY

Hematopoietic stem cells (HSCs) arise in the embryo from the arterial endothelium through a process known as the endothelial-to-hematopoietic transition (EHT)^1–4. This process generates hundreds of blood progenitors, of which a fraction goes on to become definitive HSCs. It is generally thought that most of adult blood is derived from those HSCs, but to what extent other progenitors contribute to adult hematopoiesis is not known. Here, we use in situ barcoding and classical fate mapping to assess the developmental and clonal origins of adult blood. Our analysis uncovers an early wave of progenitor specification, independent of traditional HSCs, that begins soon after EHT. These embryonic multipotent progenitors (eMPPs) predominantly drive hematopoiesis in the young adult, have a decreasing yet life-long contribution over time, and are the predominant source of lymphoid output. Putative eMPPs are specified within intra-arterial hematopoietic clusters and represent one fate of the earliest hematopoietic progenitors. Altogether, our results here reveal functional heterogeneity during the definitive wave that leads to distinct sources of adult blood.

INTRODUCTION

Blood production in the developing mouse embryo occurs through multiple waves in several anatomic locations^5,6. The definitive wave, during which adult-repopulating hematopoietic stem cells (HSCs) arise, begins around E10.5 within the major arteries of the embryo, including the aorta-gonad-mesonephros region (AGM) and the vitelline and umbilical arteries^2,4,7,8. At these sites, through a process known as the endothelial-to-hematopoietic transition (EHT), hundreds of hematopoietic progenitors are generated and accumulate in cell clusters^1–4. Some of these cells represent pre-HSCs, of which a fraction matures into definitive HSCs (dHSCs)^9–12. It is generally thought that these dHSCs, defined by their ability to reconstitute the hematopoietic system of an irradiated recipient, establish and drive blood production during the late fetal stages and in the adult¹³.

Recent studies have shown that a far greater number of cluster cells than estimated through transplantation participate in adult hematopoiesis^14–16, and that cluster cells are functionally and transcriptionally heterogeneous^17–19. However, the exact in vivo fates of these cells are not fully understood, largely because previous studies have utilized embryonic disruption and transplantation to characterize cell populations. Here, we utilize cellular barcoding and fate mapping to characterize the clonal contribution of embryonic blood precursors to the murine adult blood system.

RESULTS

Clonal analysis of in utero barcoded mice

To understand the clonal structure of the developing blood system, we first applied our previously described Sleeping Beauty (SB)-based system for induced cellular tagging²⁰ (Figure ED1B). In this model, cells can be genetically and uniquely labeled by their individual transposon (Tn) integration sites. A single retro-orbital (RO) injection of doxycycline (Dox) in pregnant dams labeled approximately 5–35% of the hematopoietic compartment (Figures ED1F and ED1G). By examining SB induction at the mRNA and protein levels, and measuring rates of transposition post-Dox, we estimated a window of cellular barcoding of 12–36 hours following Dox administration.

To capture key processes within blood development, we induced pregnant dams at five time points—iE6.5, iE7.5, iE8.5, iE9.5, and iE13.5—broadly spanning the stages of mesodermal specification, development of hemogenic endothelium, the early definitive wave, the mid-late definitive wave, and the fetal liver stage (Figure ED1A)⁶. We sorted LT-HSCs, multipotent progenitors (hereafter referred to as MPPs, subsets 3 and 4)²¹, and mature lineages (granulocytes (Gr), B cells (B), monocytes (Mo), and megakaryocyte progenitors (MkPs)) from embryonically induced mice and performed clonal analysis of these populations at approximately three months of age (Figures 1A and Supplementary Figure 2). Semi-quantitative transposon tag analysis was done using a modified Tn5-based protocol (Figure ED2). Our data revealed that at the earliest induction time point (iE6.5), most of adult blood is derived from an initial pool of precursors that produces HSCs, MPPs and multilineage progeny (Figure 1B). A similar pattern was observed from induction at E7.5. Surprisingly, at the later induction time points, and particularly beginning at iE9.5, we noted robust adult hematopoietic contribution not only from clones found in HSCs, but also by clones present in MPPs for which we could not detect a corresponding tag in the HSC population (hereafter referred to as embryonic MPPs (eMPPs)) (Figure 1B). Analysis of the fraction of total reads in the mature lineages that belonged to HSC or eMPP clones revealed that 93.8% of mature cells belonged to an HSC clone when induction was performed at iE6.5, with the fraction gradually decreasing as the induction time point advanced (78.2% at iE7.5 and 59.3% at iE8.5) (Figure 1C and ED1J). Conversely, the fraction of reads in mature lineages belonging to an eMPP clone increased to approximately 37.0% in iE9.5-induced mice (30.0% of MkP, 37.8% of Gr, 38.9% of Mo, and 41.3% of B reads). Interestingly, in E13.5-induced mice, eMPP clones predominate, accounting for 57% of total mature lineage reads versus 8.25% for HSC clones.

Figure 1: Developmental clonal analysis identifies a population of embryonic MPPs.

A: Experimental design of embryonic induction and blood sampling in adult mice. B: Depiction of transposon tags across both primitive and committed bone marrow populations, with each row representing a unique transposon insertion site (tag/clone/barcode), and each column representing the population sampled. The color of the tag reflects its frequency (in terms of log read count fraction) within that population. The total number of tags is depicted in parentheses. Each set of plots represents one mouse out of 3 or 5 examined for the five induction time points. iEx = induced at gestational day x. C: Quantification of the fraction of reads in each mature cell population (MkP, Gr, Mo, B) that can be traced to either an HSC or eMPP at 3 months of age represent individual mice and mean + S.D. (n=3–5 mice). D: HSC and eMPP contribution as a fraction of HSC + eMPP total output to mature lineages at 3 months vs 1 year of age (iE13.5 induction). Values represent the mean + S.E.M. (n=3 mice). E: Lineage bias across the various induction time points represented as the log of myeloid (Gr, Mo, and MkP) output / B output between HSCs and eMPPs. Output was first normalized to the total HSPC (HSC + eMPP) output per lineage. Values represent the mean + S.E.M. (n=3–5 mice).

We considered whether this observation of eMPP clones was a result of technical under-sampling caused by loss of HSCs either during the initial bone marrow harvesting, sorting, and DNA recovery, or through PCR amplification and sequencing. To address this, we developed a statistical model (Figure ED3 and Supplementary Note) to estimate the potential for dropout at each step of our barcode retrieval workflow (Figure ED3D). Even assuming a low capture efficiency of 40%, the model reports a 95% confidence interval of 20–36% for the in vivo fraction of eMPPs at the iE8.5 and iE9.5 time points (Figures ED3I and ED3L).

We further validated our observation of eMPPs using another in situ barcoding model, our recently described RNA-based CARLIN barcoding mouse model (Figure ED4A)²². In three iE10.5 mice evaluated at one year of age, we again observed a sizable and active eMPP population. eMPPs contributed to 43.8% (fraction of total reads) of myeloid cells (vs 43.1% by HSCs), 31.3% of MkPs (vs 56.6% by HSCs), and 53.9% of B cells (vs 35.3% by HSCs) (Figures ED4C–D). Taken together, our results support the existence of a population of eMPPs, without traceable ancestry to self-renewing, adult HSCs, beginning at E10 and extending into the fetal liver stage, that robustly contribute to multilineage hematopoiesis for at least 3 months post birth. Furthermore, considering the relatively low initial contribution of iE13.5-labeled HSCs, our data would suggest that most HSC divisions that occur within this niche are predominantly symmetric towards expansion, even following initial seeding and homing into the bone marrow.

Clonal behavior throughout adulthood

We further investigated the lineage output profiles of HSCs and eMPPs to uncover potential biases and changes with time. Comparing 3-month-old and 1 year old mice induced at E13.5, we observed an increase in overall HSC clonal contribution, particularly within the granulocyte (16.3% to 36.8% of HSPC output at 3 and 12 months, respectively) and megakaryocyte compartments (9.54% to 53.1%) (Figure 1D). However, relative HSC clonal contribution to the B cell pool remained low (10.5% to 8.34%). On the other hand, while eMPP clones displayed reduced myeloid output at 1 year of age (83.7% to 63.2% of HSPC output at 3 and 12 months, respectively), their relative contribution to the B cell compartment was unchanged (89.5% to 91.7%) (Figure 1E). This eMPP lymphoid bias was also present at the 3-month time point (Figure 1E) and within the CARLIN cohort analyzed at 1 year (Figure ED4E). These differences highlight potential, intrinsic biases in lineage production between eMPPs and HSCs, and suggest that the former is the predominant source of lifelong lymphoid production.

Fetal liver Flt3+ eMPPs contribution

To verify our findings of long-lived eMPPs and overcome sampling issues associated with our barcoding experiments (Supplementary Note), we utilized a classical driver-based lineage tracing system. We chose Flt3 (FMS-like Tyrosine Kinase 3) as an MPP driver considering that its expression labels MPPs with no long-term transplantable activity²³ and that a Flt3-Cre allele was previously used to constitutively label ST-HSCs and MPPs, but not LT-HSCs²⁴. We generated a BAC transgenic line carrying a knock-in of an EGFP-F2A-CreERT2 cassette in the Flt3 locus (Figure ED5A). We found that a single dose of tamoxifen (Tam) in adult Flt3^EGFP-CreERT2; Rosa26^LSL-tdTomato mice labels ~90% of MPPs, ~75% of ST-HSCs, and ~6% of LT-HSCs (Figure ED5C and Supplementary Figure 3), indicating selective MPP/ST-HSC labeling.

We initially focused our analysis on the fetal liver stage, at which time point our transposon analysis had indicated predominant blood production driven by eMPPs (Figure 2A). E14.5 Tam induction resulted in robust labeling of MPPs (88.48 ± 4.13%) and ST-HSCs (67.99 ± 5.91%) 24 hours later, and reduced labeling of LT-HSCs (13.84 ± 6.57%) (Figures 2B and 2C). Transplantation experiments at the early fetal liver stage confirmed limited long-term reconstitution activity of the fetal liver Tomato+ fraction (Figure ED5E). When embryonically induced mice were analyzed longitudinally, we observed that ~80% of lymphomyeloid blood cells were Tomato+ in the first three months of life, followed by a gradual depletion of label that stabilized at around 11 months of age (average of 35.7% in Gr, 45.5% in B cells, and 63.9% in T cells) and, remarkably, persisted for up to 21 months of life (Figure 2E). Bone marrow analysis across various time points demonstrated that the Tomato+ chimerism in mature lineages reflected labeling levels of MPPs and ST-HSCs, and not that of LT-HSCs (Figure 2F).

Figure 2: Fetal liver Flt3+ eMPPs are long-lived and contribute to multilineage adult hematopoiesis.

A: Experimental design of embryonic lineage tracing with the Flt3^EGFP-CreERT2 Rosa26^LSL-tdTomato mouse model. B: Representative E15.5 fetal liver flow analysis of E14.5-induced Flt3^EGFP-CreERT2 Rosa26^LSL-tdTomato embryos (n=7). C: Aggregate analysis of B. Values represent individual embryos and mean + S.D. (n=7). D: Peripheral blood analysis of E14.5-induced Flt3^EGFP-CreERT2 Rosa26^LSL-tdTomato mice sampled at various times post birth. Gray lines represent individual mice (n=13). Blue line represents average of all mice at in different populations. E: Representative flow analysis of bone marrow from a 21-month-old Flt3^EGFP-CreERT2 Rosa26 ^LSL-tdTomato mouse induced at E14.5 (n=8). F: Quantitative analysis of Tomato chimerism in fetal liver or bone marrow populations over time. Total mice analyzed: E15.5 fetal liver (n=7), 4–6 months (n=4), 9–12 months (n=5), and 18–22 months (n=8). Significance assessed through one-way ANOVA with Holm-Sidak-adjusted multiple comparisons for LT-HSC, MPP3/4, and Gr (p = 0.0052 at 4–6 months; p = 2.7⁻⁵ at 9–12 months; p = 0.0018 at 18–22 months; * < 0.05; ** < 0.005; *** < 0.0005). Box plots drawn from 25^th to 75^th percentiles, centered about the median, with whiskers representing the min and max points. G: Experimental design of embryonic lineage tracing using the EYFP reporter. H: E15.5 fetal liver flow analysis of iE14.5 Flt3^EGFP-CreERT2 Rosa26^{LSL -EYFP} embryos. Values represent individual embryos and mean + S.D. (n=5). I: Peripheral blood analysis of E14.5-induced Flt3^EGFP-CreERT2 Rosa26^LSL-EYFP mice as in D. Granulocyte population is shown (n=7). J: Representative flow analysis of bone marrow of E14.5-induced Flt3^EGFP-CreERT2 Rosa26^{loxP-stop-loxP-EYFP} mice at 13 months of age (n=5).

Given that a small fraction of LT-HSCs were Tomato labeled, we repeated our experiments using a different Cre reporter with a significantly lower efficiency of deletion²⁵. Fetal liver analysis following E14.5 Tam induction in Flt3^EGFP-CreERT2 Rosa26^LSL-EYFP mice demonstrated that while MPPs were labeled at only ~12%, LT-HSC labeling was 0.58% (Figures 2G, 2H, and ED6B). Upon chasing these mice into adulthood, we again observed long-term and stable EYFP labeling of both lymphoid and myeloid peripheral blood lineages for up to 13 months (Figures 2I and 2J), confirming that a fraction of fetal liver Flt3+ MPPs are indeed contributing long-term. Importantly, in some of the mice with substantial peripheral contribution, LT-HSC EYFP labeling was close to 0%, providing further support that these labeled peripheral blood cells are derived from Flt3+ expressing eMPPs, and not a subset of LT-HSCs (Figure 2J). Altogether, our data argue for the existence of long-lived eMPPs at the fetal liver stage that significantly contribute to blood maintenance throughout adulthood, at least within the native context.

To address whether eMPP versus HSC contribution would change in the context of hematological stress, we subjected iE14.5-induced mice to a double LPS challenge (Figure ED7A). Longitudinal follow-up revealed a significant drop in Tomato positivity among B and T cells (Figure ED7B). Within the bone marrow, the Tomato chimerism within the MkP population was significantly lower in the control versus the LPS cohort (Figure ED7D). However, we did not observe a significant reduction in the other mature lineage compartments, implying that while non-MPP populations (likely LT-HSCs) can be recruited to produce lymphoid cells post-LPS, the eMPP-derived compartment can renew and participate in regeneration.

Definitive wave Flt3+ eMPP contribution

We next determined whether long-lived eMPPs might also be specified during the definitive wave (Figure 3A). Induction of Flt3^EGFP-CreERT2; Rosa26^LSL-tdTomato embryos at E10.5 resulted in a similar, though less efficient, pattern of HSPC labeling as above, with ~40% of MPPs and 3.5% of LT-HSCs labeled (Figures 3B and ED6D). Remarkably, longitudinal follow-up of these mice revealed Tomato+ contribution (approximately 20%) to both myeloid and lymphoid lineages for at least 1 year of age (Figure 3C). As in E14.5-induced mice, Flt3-derived peripheral blood contribution peaked early after birth, decreased over the first three months, and stabilized at about 3–6 months. Bone marrow analysis of traced mice indicated that the fraction of Gr labeling correlated tightly with that of the MPP/ST-HSC population over time (Figures 3D and ED6E). These observations suggest that at least a fraction of Flt3-expressing progenitors that emerge during the definitive wave are long-lived in situ and participate in hematopoiesis for at least a year. Interestingly, Tam labeling at E9.5 resulting in lymphoid-only contribution in the adult (Figures ED6F–H), suggesting that this population precedes the specification of multipotent eMPPs.

Figure 3: Adult-contributing embryonic Flt3+ progenitors emerge as early as the AGM stage of blood development.

A: Experimental design of embryonic lineage tracing with the Flt3^EGFP-CreERT2 Rosa2^{LSL-TtdTomato} model. B: E15.5 fetal liver flow analysis of E10.5-induced Flt3^EGFP-CreERT2 Rosa26^LSL-tdTomato embryos examining labeling within stem cell and progenitor compartments Values represent individual embryos and mean + S.D. (n=8). C: Peripheral blood analysis of E10.5-induced Flt3^EGFP-CreERT2 Rosa26^LSL-tdTomato mice multiple months after birth. Each gray line represents an individual mouse (n=7). The blue line represents the average of all mice at each time point. Populations depicted are Gr, B cells, and T cells. D: Representative flow analysis of the MPP3/4, ST-HSC, and LT-HSC compartments of an E10.5-induced Flt3^EGFP-CreERT2 Rosa26^{loxP-stop-loxP-tdTomato} mice at 9 months of age (n=5).

Emergence of putative eMPPs

To gain more insight into the dynamics of eMPP emergence, we analyzed embryos at various developmental stages. Flow analysis of the yolk sac or AGM at E9.5 did not reveal Tomato+ cells (Figures 4A and ED8A), indicating that eMPPs do not arise concomitantly with previously described yolk sac-derived EMPs. Only at E10.5, in both the yolk sac and AGM, could small numbers of Tomato+ cells be observed; these numbers were significantly higher by E11.5 (Figure ED8A). These observations are consistent with prior work using a Flt3-Cre;reporter model that demonstrated emergence of Flt3+ cells beginning at E10.5²⁶. Most Tomato+ cells in the E10.5 and E11.5 AGM were cKit+, CD31+, CD41-, and CD45+ (Figures 4B, ED8C, and ED8D). Heterogeneity with respect to Il7r expression was noted, likely distinguishing true eMPPs versus lymphoid-only progenitors (Figure 4C). To evaluate where Tomato+ cells are localized in the embryo, we performed whole mount imaging of the embryo at E10.5. We observed individual Tomato+ cells within cKit+, CD31+ hematopoietic clusters in the aorta and umbilical and vitelline arteries, and that Tomato+ cells are largely CD45+ (Figure ED8B). Together, our data suggest eMPPs likely arise around E10.5 and are localized to the same hematopoietic clusters as those that give rise to dHSCs.

Figure 4: Flt3+ cells emerge as early as E10.5 and are found within intra-arterial hematopoietic clusters.

A: Diagram of experimental design to detect Flt3+ cells within the embryo. B: Expression of hematopoietic and endothelial markers (CD45, CD31, CD41, and cKit) within the Tomato+ subset in E10.5 and E11.5 AGM. Tomato+ cells are in black and superimposed upon all Ter119- cells. E10.5 plots represent a concatenation of 5 embryos independently analyzed. E11.5 plots represent a concatenation of 4 embryos independently analyzed. C: Whole-mount confocal merged images of E10.5 AGM and umbilical artery in embryos induced at E9.5 (n=4). Tomato+ (red) cells are detected within arterial cKit+ (white) CD31+ (teal) hematopoietic clusters. Arrows point to Tomato+ cells within these clusters. The rightmost images represent a magnification of the delineated area. Scale bars, clockwise from top left, in μm: 20, 5 10, 20.

Molecular profiling of prospective eMPPs

To clarify the relationship between eMPPs and other progenitors, including pre-HSCs, within the AGM landscape, we performed single cell RNA-sequencing (Figures 5A and ED9A). Unsupervised clustering analysis of sorted E11.5 populations identified eight clusters, representing a spectrum of primitive and committed hematopoietic cells (Figures 5B–C and ED9B–C). One cluster in particular, “Prog”, was enriched in primitive, pre-HSC genes (including Cdh5, Mecom, Procr, and Cd27), which we thus took represent the emerging pre-HSC/progenitor population (Figure ED9D). When examining the distribution of Flt3 expression within our map, we observed enrichment within the “Prog” cluster and a lymphoid-like (“Ly”) cluster marked by Il7r (Figures 5D and ED9D). Given their observed multipotency, we reasoned that eMPPs would be represented by Flt3+ cells within the “Prog” cluster.

Figure 5: Single-cell transcriptomic analysis of prospective eMPPs.

A: Experimental populations analyzed by scRNA-seq isolated from three pooled E11.5 Bl6 embryos. Sorting parameters are described in detail in the corresponding extended data figure. B: UMAP representation of filtered single cells and their corresponding population of origin. C: Unsupervised clustering of the single cell dataset. Clusters were named according to their differentially upregulated markers. “Prog”: uncommitted progenitors. “Ery”: erythroid-like cells, “My”: myeloid-like cells, “MegP”: megakaryocytic-like cells. “Endo”: endothelial-like cells. “Ly”: lymphoid-like cells. D: Normalized expression of Flt3 within the UMAP space and per cluster using a violin plot (y axis: cluster; x axis: normalized expression). E: Close-up of the “Prog” cluster. Normalized expression of Mecom and Flt3 within this cluster using the above UMAP representation. F: Pseudotime analysis (using Monocle3) of “Prog” cells visualized in the above UMAP space, with Flt3+ cells (black) and Cdh5+ Mecom+ co-expressing cells (pink) highlighted. Significance of pseudotime difference between the two populations assessed using a two-tailed, unpaired t-test with Welch correction. G: Adult LT-HSC, ST-HSC, and MPP4 gene module scores, derived from the top 50 differentially expressed genes within single-cell adult dataset from ref²⁹, for the Cdh5+ Mecom+ and Flt3+ populations. H: Heatmap of the top genes positively correlating with Flt3 expression that are also expressed in greater than 25% of “Prog” cells, ranked by decreasing Spearman’s rho.

On closer evaluation of this cluster, we observed a continuum of cell states wherein Cdh5+, Procr+, and Mecom+ cells, representing the majority of sorted pre-HSCs, were localized towards the tip of the cluster, whereas Flt3+ Mecom- cells were distributed more centrally (Figure 5E). Pseudotime analysis suggested a pre-HSC to Flt3+ trajectory (Figures 5E and 5F), indicating that Flt3+ eMPPs represent a potential fate output of this population. To test this, we made use of a previously described Mds1-GFP reporter mouse, in which GFP was targeted into the first transcriptional start site of Mecom²⁷, with the goal of using the stability of EGFP protein as a short-term lineage tracer. Within the E11.5 AGM, approximately 82% of cKit+ VEC+ ESAM+ pre-HSCs were GFP+. Among Flt3+ cKit+ ll7r- eMPPs, 61% of cells were GFP+ (Figure ED10A). We also transcriptionally compared our Flt3+ eMPPs to type 1 (T1) pre-HSCs, T2-preHSCs, and yolk sac progenitors from an E11 dataset¹⁹. We found greater similarity of Flt3+ “Prog” cells to T1/T2 pre-HSCs than to yolk sac progenitors (Figure ED10B). We also explored a recently published single-cell dataset of pre-circulation embryo and yolk sac tissues cultured ex vivo²⁸. We noted that Flt3 and other genes upregulated both within our Flt3 eMPPs and “Prog” cluster were enriched within a subset of cells that derived predominantly from the embryo proper (Figures ED10C and ED10D). Collectively, our data suggest that eMPP-like cells are preferentially generated in post-emergence arterial clusters and represent one type of progeny of pre-HSCs.

Using Flt3 expression within the progenitor cluster as a proxy for prospective eMPPs, we next sought to identify genes that may characterize these cells. To this end, we identified genes that positively correlated with Flt3 (Figure 5H). Among these genes were H2afy, Hlf, Sox4, and Hoxa9 (Figure 5H). Interestingly, several of these genes have been shown to mark adult ST-HSCs and MPP4s^29,30. Indeed, adult ST-HSC and MPP4 module scores were higher among Flt3+ than Cdh5+ Mecom+ co-expressing cells, suggesting that these MPP-like programs are active soon after pre-HSC emergence (Figures 5G and ED9E).

DISCUSSION

Our results here identify an early, embryonic wave of MPP specification that is responsible for a significant fraction of multilineage adult blood output. Our findings argue that definitive wave pre-HSCs can generate two different populations that make up the adult hematopoietic system – eMPPs and definitive HSCs. eMPP contribution is predominant early after birth, decreases with age, and is responsible for most lifelong lymphoid output. Adult HSC contribution, conversely, is more prominent with age, limited in its lymphoid production, and biased towards Mk generation. Whether these biases are epigenetically encoded based on their cellular ancestry or are a consequence of an aging niche remain to be further evaluated. Our observed eMPP contribution might also be partly explained by previously described transient HSCs. We posit here that the dynamics of eMPP- versus HSC-derived contribution might provide an explanation for the recognized loss of lymphoid production in the aged individual³¹ and might also underlie age-specific susceptibilities to leukemic transformation³². It is likely that other functional differences in mature blood cells derived from these two populations remain to be discovered.

While our lab and others have previously shown that adult MPPs are long-lived in the native adult setting, it was presumed that these MPPs descended from definitive HSCs at some point late in development or soon after birth. Instead, our data here indicate that lifelong eMPP specification initiates as early as E10.5 within intra-arterial hematopoietic clusters and possibly in other locations; this eMPP pool may also have contributions from previously described transient HSCs²⁶. Our data showing that this non-transplantable population can self-renew and produce multi-lineage progeny at the single cell level suggest that these eMPPs have the cardinal features of a tissue stem cell. Thus, transplantation agnostic approaches are advantageous to identify and categorize novel populations and hierarchies during hematopoietic development.

METHODS

Mouse lines, maintenance, and embryo generation

We utilized Flt3-EGFP-CreER BAC, M2/SB/Tn transgenic²⁰, R26loxP-stop-loxP-tdTomato³³, R26loxP-stop-loxP-EYFP²⁵, cCarlin²², iCas9²², Mds1-GFP²⁷, and NOD.CB17-Prkdcscid/J (NOD SCID) mice in this study. For the Flt3-EGFP-CreER BAC mice, which was a newly generated transgenic line, the ATG of the first exon of Flt3 was targeted with an EGFP-F2A-CreERT2-polyA cassette. The targeted BAC was then used to generate transgenic founders. F1 mice were backcrossed into a C57/BL6 background for at least 4 generations and genotyped using Cre primers²⁷. For timed mating experiments, female mice were placed with male mice nightly and evaluated for the presence of a vaginal plug the following morning; detection of a plug marked embryonic day 0.5 (E0.5). All mouse procedures and protocols were approved by the Animal Care and Use Committee of Boston Children’s Hospital and followed all relevant guidelines and regulations.

Tamoxifen and doxycycline inductions

For induction of Flt3-EGFP-CreER reporter mice, 1 mg tamoxifen (Sigma-Aldrich, catalog T5648), resuspended in corn oil, was injected intraperitoneally in adult mice or at various gestation ages in pregnant dams. Embryonically induced embryos were analyzed either later in development or followed into adulthood. For doxycycline (Dox; Sigma-Aldrich, catalog D9891) experiments using the M2/SB/Tn mouse, 25ug Dox/gr of mouse was administrated to pregnant dams through a retroorbital route. Dox powder was first resuspended in sterile phosphate-buffered saline (PBS). All mouse procedures and protocols were approved by the AAALAC-accredited animal care and use committee of Boston Children’s Hospital.

Isolation and evaluation of bone marrow blood populations

Bone marrow cells were isolated via bone crushing using a mortar and pestle in PBS supplemented with 2% fetal bovine serum (FBS) and 1x penicillin/streptomycin. Upon removal of red blood cells using an RBC lysis buffer, the cell suspension was filtered through a 70 μm strainer to obtain a single cell suspension and remove other debris. For experiments requiring lineage depletion, antibody staining using CD19-, Ter119-, Gr-1-, CD3-, and Mac1-biotin conjugated antibodies was first performed followed by incubation with anti-biotin beads (Miltenyi Biotec, catalog 130-090-485) and subsequent magnetic column depletion. Cells were subsequently stained and sorted according to the following antibody panels. For the transposon model, LT-HSCs were defined as Lin- cKit+ Sca-1+ CD150+ CD48-; STHSCs were defined a Lin- cKit+ Sca-1+ CD150- CD48-; and MPP3/4s were defined as Lin- cKit+ Sca-1+ CD150- CD48+. Megakaryocyte progenitors (MkPs) were defined as Lin- cKit+ Sca-1- CD150+ CD41+. Within the lineage positive compartment, granulocytes were sorted as B220- 7/4+ Ly6G+, monocytes as B220- 7/4+ Ly6G-, and B cells (broad definition) as B220+ 7/4-. DAPI was used for exclusion of dead cells. For the Flt3-CreER model, the stem and progenitor cell gating schemes were the same as the transposon model described above. Cell sorting was performed using a BD FACSAria II sorter. For FACS schemes, see Supplementary Figures 2 and 3. For antibodies, see Supplementary Table 3.

Isolation and evaluation of peripheral blood populations

For Tomato expression analysis within the Flt3-CreER model, peripheral blood was collected monthly (unless otherwise stated in the main text). Approximately 2–3 capillaries of blood were obtained from the retro-orbital sinus and density-separated to remove red blood cells. The resultant non-RBC fractions were collected, centrifuged, and resuspended in 2% FBS in PBS. Cells were stained as follows, and Tomato positivity was measured in each compartment of interest. B cells: CD19+ CD4/CD8-; T cells: CD19- CD4/CD8+; granulocytes: CD19- CD4/CD8- Ly6G+. DAPI was used for exclusion of dead cells. A BD LSRII Flow Cytometer was used for flow cytometry analysis. Sorting schemes noted in Supplementary Figure 3.

Flt3 transplantation assays

Adult bone marrow transplants, whole bone marrow

NOD.CB17-Prkdcscid/J (NOD SCID) mice were used as bone marrow transplant recipients. A Flt3^EGFP-CreERT2; Rosa26^LSL-Tomato mouse was induced at two months of age and sacrificed 48 hours later. Whole bone marrow was isolated and 500,000 cells were used for transplantation. Prior to transplantation, SCID mice received a sublethal 2.5 Gy dose. Cells were transplanted retro-orbitally. Peripheral blood analysis of transplanted mice was performed for 4 months following transplantation. Percent Tomato chimerism of the CD45.2 fraction was recorded.

Fetal liver transplants

8–12-week-old B6.SJL-PtprcaPepcb/BoyJ (CD45.1) mice were used as fetal liver transplant recipients. Flt3^EGFP-CreERT2; Rosa26^LSL-Tomato embryos were induced at E12.5 and sacrificed one day later at E13.5. Fetal livers were isolated, digested into a single-cell suspension, and pooled. cKit+ cells were enriched using cKit+ beads (Miltenyi Biotec, catalog 130-091-224) and combined with CD45.1 helper bone marrow cells at a ratio of 10,000 cKit+ cells + 100,000 helper cells. CD45.1 recipients received a split dose of 11 Gy lethal irradiation, with a three hour interval time between the two doses. Five mice were transplanted via retro-orbital injection. Peripheral blood analysis of transplanted mice was performed for 10 months following transplantation. Percent Tomato chimerism of the CD45.2 fraction was recorded.

Lipopolysaccharide challenge of adult, in utero labelled mice

Lipopolysaccharide (LPS, Sigma-Aldrich, catalog L2880) was prepared at a concentration 0.2 mg/mL in sterile PBS, and subsequently filtered using a 0.22 μm filter. Experimental mice were injected intra-peritoneally with weight-dosed LPS (1 mg/kg). Control mice received 200 μL sterile PBS, also delivered through an IP route. Peripheral blood analysis for evaluation of Tomato chimerism in different blood compartments was performed as above.

Aorta-gonad-mesonephros (AGM), yolk sac (YS) and fetal liver (FL) isolation and evaluation of blood populations

The AGM, YS and FL were isolated in cold PBS supplemented with 10% FBS. E9.5, E10.5 and E11.5 embryos were used for YS and AGM isolations; E12.5 and E14.5 embryos were used for FL isolation. YS and AGM isolation was performed as previously described³⁴. Briefly, the uterine horns containing the embryos were first isolated followed by separation of each conceptus. The placental connection was carefully severed, after which the YS was separated from the embryo, with care to maintain the umbilical and vitelline arteries. The embryo with attached arteries was used for AGM isolation. Subsequently, the head was removed by a sagittal cut, followed by removal of the dorsal somites, limbs, and intestinal and cardiac tissue. For FL isolation, individual embryos were removed from the uterus and separated from their YS and placenta. The head and lower abdominal regions were then removed followed by exposure of the abdominal organs by removal of the surrounding tissue. Finally, the fetal liver lobes were dissected and separated from surrounding structures. Embryonic tissues were dissociated in 0.125% collagenase type I at 37°C for 30 minutes, after which cells were further dissociated by gentle pipette and then passed through a 40 μm strainer. Single cells were washed with PBS and then incubated with anti-mouse CD16/32 at 4°C for 10 minutes for Fc blocking prior to antibody staining. For evaluation of fetal liver stem and progenitor cells, the gating schemes were the same as the adult schemes. Importantly, however, Mac1 was excluded from the lineage biotin panel. AGM and YS cells were stained with a broad panel of antibodies: specific panels are either described in the text or within the pertinent methods subsections. For FACS schemes, see Supplementary Figure 4.

Embryonic isolation for qPCR and western blotting

For measurement of Sleeping Beauty (SB) transposase expression, embryos were harvested at various time points after doxycycline (Dox) induction at E9.5. AGMs were dissected and processed as above, and VE-Cadherin+ Ter119- cells were sorted. Cells were lysed in Trizol (Life Technologies, catalog 15596018) and RNA was purified according to the manufacturer’s protocol. RNA was converted to cDNA using iScript (Bio-Rad, catalog 1708891BUN) according to the supplied protocol. qPCR analysis was performed using Fast SYBR Green (Thermo Fisher Scientific, catalog 4385617), with the following primers listed in Supplementary Table 1.

For SB protein detection, whole embryos were harvested at various time points following Dox induction at E9.5 (yolk sac and placenta were removed). Whole embryos were digested in ice-cold RIPA buffer containing cOmplete Protease Inhibitor Cocktail (Sigma-Aldrich, catalog 11873580001), aided with a homogenizer. Samples were centrifuged and precipitated debris was removed. Protein concentrations were measured with the Pierce BCA protein assay kit, and equal amounts were resolved on NuPAGE Novex 4–12% Bis-Tris protein gels (Invitrogen, NP0336BOX). Antibodies used were Sleeping Beauty transposase (R&D Systems, catalog AF2789-SP) and Histone H3 (Cell Signaling, 9715S).

Whole-mount AGM and YS immunostaining

YS and AGM whole-mount immunostaining was performed as previously described^4,35. Briefly, upon YS dissection, the embryos were fixed in 2% paraformaldehyde (PFA). Somites were counted immediately after fixation to estimate the embryonic stage. Fixed embryos were dehydrated in methanol and dissected to expose the AGM, followed by rehydration. Serum blocking was performed to ensure staining specificity followed by overnight primary antibody incubation. c-Kit purified rat (1:500), CD31 purified rabbit (1:200), mCherry purified goat (1:500), CD45 biotin rat (1:100) and CD41 biotin rat (1:100) primary antibodies were used for staining. Secondary antibody (1:2000) staining was performed overnight using donkey anti-rabbit 488, anti-goat 555 and anti-rat 647 AlexaFluor antibodies. Finally, the embryos were cleared using BABB (benzyl alcohol/benzyl benzoate, prepared according to ref³⁵), mounted in FastWell slides (Grace Bio-Labs, catalog 664113), and imaged using confocal microscopy (Zeiss 710). Images were acquired using Zeiss ZEN microscopy software.

Digital droplet PCR

To estimate the kinetics of transposition in the M2/SB/Tn model post-induction, the copy number of transposed DNA within an induced embryo was measured through ddPCR. Genomic DNA was isolated from embryos using a QIAamp DNA Mini Kit (Qiagen, catalog 51304). The probe sets, purchased through Integrated DNA Technologies, can be found in Supplementary Table 1.

The following PCR program was run: 5 min denaturation (94°C), followed by 35 cycles of denaturation (30 seconds at 94°C), annealing (30 seconds at 58°C), and elongation (1 minute at 72°C). The PCR plate was then read by a QX200 droplet reader.

Transposon insertion site analysis through TRACE

Our previous techniques for identification of transposon integration sites relied upon ligation-mediated PCR and T7-mediated linear amplification (TARIS)^20,29. Here, we sought to improve our library construction protocol so as to reduce both the time to library generation (from 1 week to 2 days) and the minimum input DNA. This method, which we call TRACE (TRansposase-Assisted Capture of transposable Elements), utilizes custom Tn5 transposomes to quickly and robustly tagment variable amounts of genomic DNA^36,37. Following tagmentation—wherein adaptors are now ligated to gDNA fragments—SB-Tn insertion sites can be amplified through a series of nested, suppression PCRs.

To generate custom transposomes, Tn5 transposase was purified according to a Picelli et al³⁶. Tn5 transposase was loaded with custom annealed adaptors (Supplementary Table 2). The length of these adaptors allowed for effective suppression of unwanted PCR products; shorter adaptors lacked this suppression feature and did not enrich for SB-Tn insertion sites (data not shown). Finally, we validated Tn5 activity through fragmentation of high-molecular weight genomic DNA and gel-based visualization (Figures ED2A and ED2B).

For transposition, varying amounts of Tn5 transposome were added to DNA in a 10 μL reaction volume; a 5X Tris-MgCl₂ buffer was used, which consisted of 50 mM Tris-HCl, pH 8.4–9.0, and 25 mM MgCl₂. These three components were sufficient for tagmentation, which was carried out at 55°C for 10 minutes. Following tagmentation, a first round of Tn-specific PCR was performed on the reaction volume (without purification or Tn5 inactivation). The products from the first round of PCR were purified with Agencourt AMPure beads (Beckman Coulter, catalog A63881), and half was loaded into a subsequent, nested PCR, as previously described. PCRs (20 cycles each) were performed using KAPA HiFi HotStart Ready Mix (Roche Applied Science, catalog 07958935001). The primers utilized for each PCR are noted below. Finally, a 12-cycle indexing PCR (MP1 and ID primers) was performed using Phusion High-Fidelity polymerase (New England Biolabs, catalog M0531S). Sequencing was performed with Illumina Miseq (PE 150 × 2), and SB-Tn tag identification and alignment was performed as previously described²⁰.

Guided by cumulative frequency plots of both mature (granulocytes) and primitive populations (HSCs) at each time point (Figure ED2D), we employed a read frequency threshold of 0.05% of post-mapped reads. We noted that at a read frequency threshold of 0.05%, approximately >=90% total reads/cell population could be captured, while eliminating spurious barcodes that may have arisen through PCR or other steps. From our spike-in analysis using HEK293T clones with single, unique transposon integration sites²⁰, this read cutoff correlated with a detection limit of approximately 2.5–5 cells per 1000 cells (Figure ED2C). We applied this threshold across all populations and time points, and the resulting data were used to quantify HSC and eMPP contributions to mature lineages, which were represented as a fraction of total lineage reads. We observed similar HSC and eMPP contributions at the various time points when employing a lower cutoff of 0.01%. For the transposon barcode heatmaps and pie charts (Figure 1), the top 0.1% of tags (read fraction) are depicted for viewability. For two mice, one at iE9.5 and one at iE13.5, MkPs were not sampled due to failed library preparation. All other populations were assessed for each mouse at each time point. Adaptor and primer sequences can be found in Supplementary Table 2. All primers were ordered from IDT DNA technologies, at 100 nano-mole scale and HPLC-purified.

DNA extraction from sorted cells

Sorted cells were first spun at 4°C for 7 minutes at 5000 rpm. Supernatant was carefully removed and resuspended with 10 μL of lysis buffer, composed of an NP40-based extraction buffer (10 mM Tris-HCl pH7.4, 10 mM NaCl, 3 mM MgCl2, 0.1% Igepal) and Qiagen protease (Qiagen, catalog 19155). Qiagen protease is added to a final concentration of 0.5mg/mL of the lysis buffer. Approximately 10–20 μL of lysis buffer is added to the cell pellet and gently pipetted several times to dislodge the pellet. The cells (in lysis buffer) are then transferred to a PCR tube, after which the following lysis program is run: 55°C for 8 hours, followed by 15 min protease inactivation at 75°C. The DNA is then immediately used for tagmentation, as described above. For stem and progenitor populations sorted from iE6.5, iE7.5, iE8.5, and iE9.5 time points, bone marrow granulocytes from Bl6 cells were spiked in to achieve a relatively equivalent total DNA amount (as measured through Qubit) of approximately 100 ng. Conversely, no spike-in was performed for the mature lineages due to their relative abundance. Importantly, for iE13.5 HSPC populations, due to low transposition efficiency at this time point, cells were first whole-genome amplified using NxGen phi29 DNA polymerase (Lucigen, catalog 30221–2). Amplification was performed according to the supplied protocol. 100 ng of amplified product, following DNA cleanup, was used for TRACE.

Transposon experimental dropout estimation

To estimate an experimental dropout rate for HSCs, we generated independent Tn libraries from whole-genome amplified DNA from iE9.5 adult LT-HSCs (DsRed+ fraction). We utilized the E9.5 time point as it was the time point with the most barcode diversity. We reasoned that a range of experimental dropout rates for HSCs could be estimated by sampling at various input amounts from the LT-HSC whole-genome amplified DNA pool. DsRed+ LT-HSCs were isolated as above from iE9.5 adult mice; DNA was isolated and whole genome amplified using NxGen phi29 DNA polymerase (Lucigen, catalog 30221–2). Libraries were then generated with TRACE using a range of DNA inputs (100, 75, 50, 20, 10, and 5 ng). The identified Tn tags were compared against a reference set representing the highest loading amount of 100 ng. In order to generate a reference set, two libraries were independently prepared using 100 ng of input DNA: tags that were present in both (high-confidence tags) were included in the final set. Dropout rates were calculated as 1 - the fraction of retrieved high confidence tags (measure in read number and tag count).

Statistical dropout model for quantifying eMPP clones

See accompanying Supplementary Note, titled “Statistical dropout model for quantifying eMPP clones.”

Single-cell cell isolation and sequencing

Three Bl6 E11.5 embryos, pooled, were used for the single-cell experiment. Three populations were sorted: AGM pre-HSCs and hemogenic endothelial cells (Lin- cKit+ CD45^neg/mid CD31+ VE-Cad+); AGM progenitors (Lin- cKit+ CD45+); and fetal liver progenitors (cKit^+/high CD45+ CD41- and cKit^+/high CD45+ CD41^high were separately sorted and mixed before encapsulation). Lineage markers used here were Ter119, Gr1, CD19, and CD3. Single-cell encapsulation was performed using 10X Chromium single cell 3’ technology. Whole-transcriptome libraries were prepared using the 10X v3 protocol and sequenced on Illumina NextSeq500 paired-end 150 cycles (Read 1: 28 cycles; Index Read 1 (i7): 8 cycles; Read 2: 91 cycles).

Single-cell RNA sequencing analysis

Sequencing data from above were pre-processed using 10x Genomics CellRanger 3.0.4 using the standardized workflow, using the mm10 reference genome for alignment. The generated count matrices were analyzed using Seurat v4.0.0³⁸ and merged into a single Seurat object. Cells with a UMI count < 1000 and/or mitochondrial gene percentage > 10% were filtered out. A total of 225 pre-HSCs/endothelial cells, 520 AGM progenitors, and 545 fetal liver progenitors passed filter and were used for downstream analysis. Dimensionality reduction and clustering were then performed (resolution of 0.4), and cells were visualized using UMAP plots. Differential gene expression was performed using the FindMarkers function: a table of differentially expressed genes per cluster can be found in the supplementary files.

Score analysis (LSK)

For adult “HSC” and “MPP3/4” scoring of embryonic cells, we utilized the AddModuleScore() function of Seurat using the top 50 differentially expressed genes found within LT-HSCs, ST-HSCs, and MPP3/4s (single cell dataset from ref²⁹). Module scores were plotted using a mincutoff = 0.

Monocle analysis

Our Seurat-loaded data was converted into Monocle’s cds format using the as.cell_data_set() function. Cells were clustered and a pseudotime trajectory was calculated. The cells within the “Prog” cluster were then selected from the cds dataset and then plotted using the plot_cells() function. Monocle3³⁹ was used for analysis.

Random forest classification

To help determine the developmental stages of our cells, we analyzed the dataset generated by Baron et al.¹⁹ profiling hematopoietic progenitor cells from the E11 AGM. We used cell type annotations provided by the original publication, and then used CellRouter⁴⁰ to perform quality control and identify differentially upregulated genes among selected cell types reported in Figure ED10B. The final dataset contained 174 cells. Then, we trained binary random forest classifiers based on the top 100 differentially upregulated genes as a training dataset. Next, we used the resulting random forest classifiers to assign a classification score for each cell within our “Prog” subset.

Ex vivo embryo proper and yolk sac analysis

We made use of existing data from Ganuza et al. (ref²⁸), in which single-cell RNA sequencing analysis was performed on ex vivo-cultured embryo proper and yolk sac tissues. The same processing parameters as above were used.

Gene correlation analysis

To identify genes correlated with Flt3 within our sub-population of interest, we made use of the correlatePairs() R package: https://bioconductor.org/packages/release/workflows/vignettes/simpleSingleCell/inst/doc/misc.html. We restricted our gene space to only those genes expressed in at least 25% of cells. Positively correlating genes with an FDR < 0.1 are shown in Figure 5, ranked by Spearman’s rho (decreasing).

CARLIN cell isolation, sequencing, and alignment

Isolation of cell populations was performed in the same manner as described above for the M2/SB/Tn model. For CARLIN amplification, RNA was first purified using the Arcturus PicoPure RNA isolation kit (Thermo Fisher Scientific, catalog KIT0204). Following the protocol described in ref²², the RNA was reverse transcribed using a common CARLIN primer and Superscript III Reverse Transcriptase (Thermo Fisher Scientific, catalog 18080–044), and then amplified using a nested PCR approach KAPA HiFi polymerase mix (Roche Applied Science, catalog 07958935001). The primers and cycle specifications used are derived from ref²². Finally, an indexing PCR was performed using the purified PCR product, and then sequenced on an Illumina MiSeq using paired-end 500 cycles v2 kits (Read 1: 250 cycles; Index Read: 6 cycles; Read 2: 250 cycles) with 10% PhiX sequencing control v3 (Illumina). Paired-end reads were merged using PEAR, as described in ref²², and then analyzed using the CARLIN pipeline (https://gitlab.com/hormozlab/carlin; analyze_CARLIN()). Outputted alleles were merged across samples from each mouse, resulting in a table of alleles and detected UMIs per population. A threshold of 5 UMIs was used as a filtering cutoff (alleles with < 5 UMIs were not counted for that specific cell population). These data were used to quantify HSC and MPP contribution to mature lineages, in the same manner as in the transposon model.

Deposited Data

Single-cell RNA sequencing

GSE180357: Single-cell transcriptomic interrogation of hematopoietic progenitors within the E11.5 AGM of the mouse

Extended Data

Extended Data Figure 1: Characterization of transposon barcoding within the embryo.

A: Diagram of several key processes involved in murine blood development. B: Model of how the M2/SB/Tn system can be used to infer developmental lineage relationships. C: Relative Sleeping Beauty (SB) transposase mRNA levels at various time points following IV Dox at E9.5, demonstrating strong induction and return to baseline SB mRNA levels around 24–36 hours. RNA was harvested from VE-Cadherin+ cells sorted from the AGM. Values represent individual embryos and mean (n=4 embryos). D: Protein levels of SB following induction (whole embryo lysates used for protein extraction; one representative embryo shown of n = 3). For gel source data, see Supplementary Figure 1. E: The fraction of transposed cells was estimated by digital droplet PCR, using a probe that spans the site of excision of the donor locus. Overall transposition plateaus after 36 hours, with an effective transposition window of 12–36 hours. Values were normalized to sorted DsRed+ granulocytes. Values represent individual embryos/mice and mean + S.D. (n=3–8 embryos). F: Percentage of DsRed positive cells within the E12.5, E14.5, and adult bone marrow LSK compartment following intravenous (IV) Dox administration at E9.5. Labeling is consistent across LSK compartments and mature lineages, with a slight reduction in monocyte and B cell labeling. Values represent individual embryos/mice and mean (n=4–6 mice). G: Transposition efficiency (as measured by DsRed percentage within the granulocyte compartment) at each induction time point. Values represent individual mice and mean + S.D. (n=3–5 mice). H: In utero induction labels various cell types, including endothelial cells and putative HSPCs emerging within intra-arterial clusters (white arrows). A representative hematopoietic cluster within the umbilical artery of an iE7.5-induced embryo imaged at E10.5 is shown (whole-mount preparation, imaged with confocal microscopy; n = 2). Scale lengths of 50 μm (left) and 10 μm (right). I: Representation of clone sizes of HSC- (gray) or eMPP-derived (blue) Gr and B cells at 3 months and 1 year of age (from iE7.5, 9.5, and 13.5). Each set of plots (Gr and B) is drawn from one representative mouse. J: Comparison of overall contribution to mature lineages by HSCs and eMPPs. Output was normalized to the total HSPC output per lineage, and then averaged to generate a single value per mouse and per compartment. p-values derived from unpaired two-tailed t-tests with Welch correction (p = 0.0015 for iE6.5, p = 0.023 for iE7.5, p = 0.042 for iE8.5, and p = 0.020 for iE13.5; * < 0.05; ** < 0.005). Values represent individual mice and mean + S.D. (n=3–5 mice).

Extended Data Figure 2: Development and validation of a new insertion site capture method (TRACE).

A: Experimental flow chart representing the design and development of TRACE (transposase-assisted capture of transposable elements). Briefly, custom transposomes were generated by combining annealed adaptors with purified Tn5 transposase. Through a two-step, nested, suppression PCR following tagmentation with the custom transposomes, sequencing libraries can be efficiently generated. B: Tn5 activity assay demonstrating an additive increase in fragmentation. Suppression PCR enables selective amplification of Sleeping Beauty Tn integrations (representative image of one reaction shown, n = 4). For gel source data, see Supplementary Figure 1. C: Experimental design of detection limit of TRACE, wherein DNA from three HEK293T clones are added at varying dilutions (approximating 5, 10, 25, 50, 100, 1000, 2500 cells/clone in a final pool of ~10000 cells) to a polyclonal DNA sample of integration sites. TRACE enables a semi-linear readout of clone size and has a reliable frequency detection limit of at least 25 cells in 10,000. Values represent individual replicates with a line drawn through their means. D: Cumulative frequency plots (left two graphs) of granulocyte and HSC barcode read count fractions across the various time points. Each line represents one representative mouse. Only the top 200 barcodes are depicted. The fraction of total reads captured using various read frequency thresholds (right two graphs) is shown for each population, using the same mice as depicted in the cumulative read frequency plots. A threshold of 0.0005 was selected for subsequent transposon analysis as >90% of reads were captured for each population at each time point. Values represent individual mice. E: Concordance of tags and reads between independently generated libraries from split samples of MPPs and granulocytes or whole-genome amplified DNA from HSCs, at E9.5 or E13.5. Tags that fell below the read threshold were assigned a frequency of 1×10⁻⁴. The percent of shared reads is depicted on each plot.

Extended Data Figure 3: Statistical dropout model for quantifying eMPP clones.

A: Schematic for assessing experimental rate of dropout. DNA from DsRed+ adult HSCs from an iE9.5 mouse were extracted and whole genome amplified. Limiting dilutions of DNA were used as input for downstream library generation by TRACE. Tags retrieved with each DNA amount were compared against a reference set of high confidence tags (defined as shared tags recovered from two independently generated libraries with 100 ng starting DNA). B: Retrieval rate (represented as both individual tags and total read counts) of high-confidence tags at limiting DNA input amounts. C: Distribution of HSC barcode reads across all MPP-containing clones from iE9.5. The mode at zero represents putative eMPPs. D: Generative model of barcode recovery and sequencing. E: Convergence of the log likelihood estimate for increasing quantity of $X_i^{\text{pre}-\text{PCR}}$ samples. Each dot represents an independent sample set for the given sample size. Results reflect the lognormal model applied to iE9.5 data with $r_{\text{min}}=0.7,p_e=0.35,\mu =-0.1,\sigma =3.6.$ F: Log likelihood of each MCMC chain, normalized to the max value. Each dot represents an independent chain. The lower values at the beginning represent the burn in period. G: Autocorrelation functions of the $p_e$ parameter in each MCMC chain. H: Comparison of the posterior moments of $p_e$ between two pools of MCMC restarts (1–25 vs. 25–50). Each dot represents a different combination of conditional parameters $r_{\text{min}},f_{\text{capture}}$ and model family (lognormal vs geometric). I: Posterior distribution of $p_e$ across a range of conditional parameter values, model families and time points. Dotted lines represent the 2.5^th and 97.5^th percentiles. J: Posterior predictive checks for four models, distinguished by two different values of $r_{\text{min}}$ and two different parametric families of in vivo clone size distribution. For each model, the modal parameter estimates for iE9.5 were used to generate simulated read-count distributions (top). Quantile-quantile plots (bottom) compare the simulated read-count distribution to the empirical distribution from iE9.5. K: Difference of Bayesian evidence for two different values of $r_{\text{min}}$ and two different parametric families of in vivo clone size distribution. Each bar represents a different combination of the remaining parameters (i.e. $f_{\text{capture}}$ and either $r_{min}$ or the in vivo clone size distribution). L: Difference of Bayesian evidence between models with and without eMPPs respectively. Each bar represents a different conditional value of $f_{\text{capture}}$. All models shown have $r_{\text{min}}=0.7$ and a lognormal distribution for clone sizes in vivo.

Extended Data Figure 4: Validation of eMPP clones using an alternative barcoding model, CARLIN.

A: Schematic of experiment, in which iCas9;cCarlin mice induced at E10.5 and sacrificed at 1 year of age for bone marrow analysis. Populations (LT-HSC, MPP, MkP, Gr, Mo, B) were isolated by FACS from bone marrow, processed, and analyzed for edited Carlin alleles. B: Editing frequencies within granulocytes (n = 3), with an average of approximately 7%. Mean + S.E.M. of the three samples are depicted. C: Contribution of HSCs and eMPPs to mature lineages, represented as the number of UMIs shared/total UMIs in that population. Values represent individual mice; bar plots depict mean + S.D. (n=3). D: Representative plot of one mouse demonstrating the Carlin alleles across the sampled populations (showing only those alleles found in either HSCs, eMPPs, or both). Each row represents a unique allele, and each column represents the population sampled. The color of the allele reflects its frequency (in terms of log UMI count fraction) within that population (n=3 adult mice). E: Lineage bias represented as the log of myeloid (Gr, Mo, and MkP) output / B output between HSCs and eMPPs. Output was first normalized to the total HSPC output per lineage. p-value derived from two-sided paired t-test (p = 0.0035). Values represent individual mice and mean + S.E.M.

Extended Data Figure 5: Generation and characterization of the Flt3^EGFP-CreERT2 mouse model.

A: Schematic of the Flt3 genetic locus targeting approach using a BAC targeting vector. The ATG start site of the first exon was targeted with an EGFP-CreERT2 resulting in a Flt3^EGFP-CreERT2 transgenic mouse. B: Bone marrow FACS analysis of adult Flt3^EGFP-CreERT2 for Flt3 and EGFP expression demonstrates that while EGFP correlates with endogenous Flt3 protein expression, it also tracks with a small population in the CD34+ and CD34- BM fraction that expresses Flt3 mRNA but undetectable protein levels. C: Representative bone marrow flow analysis of adult LSK populations in a Flt3^EGFP-CreERT2 Rosa26^{loxP-stop-loxP-tdTomato} mouse induced at 2 months of age and analyzed 48 hours later. Percent Tomato labeling was assessed within the MPP3/4, STHSCs and LT-HSCs compartments (top panels, without Tamoxifen; bottom panels, with Tamoxifen) (n=6). D: Scheme of experiment designed to test reconstitution potency of Tomato+ bone marrow cells within the adult. Forty-eight hours after induction of a 2-month-old Flt3^EGFP-CreERT2 Rosa26^{loxP-stop-loxP-tdTomato} mouse, whole bone marrow was isolated and 500,000 cells were transplanted into sub-lethally irradiated NOD-SCID mice. Percent Tomato chimerism of the CD45.2 fraction for granulocytes, B cells, and T cells was assessed monthly for 4 months post-transplantation. Each line represents an individual recipient mouse (n=4). E: Scheme of experiment designed to test long-term reconstitution ability of E13.5 cKit+ cells fetal liver cells from Flt3^EGFP-CreERT2 Rosa26^{loxP-stop-loxP-tdTomato} embryos induced at E12.5. 10,000 cKit+ fetal liver cells were transplanted into lethally irradiated CD45.1 recipients, along with 100,000 helper CD45.1 bone marrow cells. Percent Tomato chimerism of the CD45.2 fraction for granulocytes, B cells, and T cells was assessed monthly for 10 months post-transplantation. Each blue line represents an individual recipient mouse that was transplanted with 10,000 cKit+ cells (n=4).

Extended Data Figure 6: Additional characterization of iE14.5, iE10.5, and iE9.5 Flt3-CreER mice.

A: Experimental design of embryonic lineage tracing using the EYFP reporter, with induction at E14.5. B: Representative E15.5 fetal liver flow analysis of iE14.5 Flt3^EGFP-CreERT2 Rosa26^{LSL -EYFP} embryos (n=7). C: Experimental design of embryonic lineage tracing using the Tomato reporter, with induction at E10.5. D: Representative E15.5 fetal liver flow analysis of E10.5-induced Flt3^EGFP-CreERT2 Rosa26^LSL-tdTomato embryos examining labeling within stem cell and progenitor compartments (n=8). E: Quantitative analysis of Tomato chimerism in bone marrow populations over time, following induction at E10.5. Multiple embryos or mice were grouped together for each time point. Total mice analyzed: E15.5 fetal liver (n=8), 9–12 months (n=5), and >12 months (n=4). Significance assessed through one-way ANOVA with Holm-Sidak-adjusted multiple comparisons for LT-HSC, MPP3/4, and Gr (p = 0.0489 at 9–12 months; * < 0.05). Box plots drawn from 25^th to 75^th percentiles, centered about the median, with whiskers representing the min and max points. F: Experimental design of embryonic lineage tracing with the Flt3^EGFP-CreERT2 Rosa26^{loxP-stop-loxP-tdTomato} mouse model at iE9.5. G: Quantitative analysis of the percent Tomato labeling in MPP3/4, STHSC and LT-HSCs fractions of the E15.5 fetal liver. Values represent individual embryos and mean + S.D. (n=7). H: Quantitative analysis of percentage Tomato label in bone marrow (LSK) and peripheral blood populations at 1 month and 6–7 months of age in iE9.5-induced mice, demonstrating lymphoid-biased Tomato labeling and minimal MPP labeling over time. Values represent individual embryos and mean + S.D. (n=5 for each time point).

Extended Data Figure 7: Characterization of Tomato+ progenitors following an LPS challenge.

A: Experimental design of LPS challenge. Mice were induced in utero at iE14.5 and traced as adults. Peripheral blood was collected regularly, assessing for Tomato chimerism within granulocytes, B cells, and T cells. Two LPS challenges (IP administered) were given—one at 11 weeks of age and the other at 23 weeks of age. The control group received PBS delivered IP. B: Fraction of Tomato+ labeling relative to pre-LPS/PBS levels (at 10 weeks of age) for granulocytes, B cells, and T cells. Asterisks represent time points at which the difference between the PBS and LPS conditions were significant (p < 0.05; assessed using two-tailed, unpaired t-tests with Welch correction). Modified box plots are drawn from min to max, with a line at the mean. C: Frequencies of MPP3/4s, ST-HSCs, and LT-HSCs (as a fraction of the total LSK compartment) in the PBS and LPS conditions. Significance assessed using two-tailed, unpaired t-tests with Welch correction. Values represent individual mice and mean + S.D. D: Fraction of Tomato+ labeling relative to MPP3/4 (normalized per mouse). Values represent individual mice and mean + S.D. Significance assessed using two-tailed, unpaired t-tests with Welch correction. For all experiments, n = 3 (PBS arm) and n = 5 (LPS arm). The same cohort of mice were followed throughout.

Extended Data Figure 8: Flt3+ labeling within the E10.5 yolk sac and fetal liver.

A: Flow analysis of the AGM at E9.5, E10.5, and E11.5 and yolk sac (YS) at E9.5 and E10.5 (representative embryo; inductions performed one day earlier; n=3–5). Flow plots demonstrate GFP and Tomato expression of all live cells. Tomato+ cells are present within the AGM and YS beginning at E10.5. B: Whole-mount confocal merged images of the E10.5 fetal liver bud, AGM, and umbilical artery in an embryo induced at E9.5 (representative image; n=3). Top image: cKit and Tomato staining. Tomato+ (red) cells are detected within the fetal liver bud, in addition to other cKit+ (white) cells. Scale bar: 50 μm. Middle and bottom images: CD31, CD45, and Tomato staining. Arrows highlight Tomato+ CD45+ cells, while arrowheads highlight Tomato+ cells with diminished CD45+ labeling. Scale bars represent 50 μm (middle) and 20 μm (bottom). Sorting schemes can be found in Supplementary Figure 4. C: Expression of hematopoietic and endothelial markers (CD45, CD31, CD41, and cKit) within the Tomato+ compartment within the E10.5 yolk sac (YS). Tomato+ cells are in black and superimposed upon all Ter119- cells. E10.5 plots represent a concatenation of 5 embryos independently analyzed. D: Fraction of Il7r+ cells within the Tomato+ cKit+ CD45+ Lin- compartment at the E10.5 AGM (n=4, concatenated) and E11.5 AGM (n=3, concatenated).

Extended Data Figure 9: Transcriptional characterization of Flt3+ cells.

A: Sorting scheme for single-cell RNA sequencing experiment. Three populations were independently sorted and isolated from three, pooled E11.5 AGMs or fetal livers (AGM pre-HSCs and hemogenic endothelial cells; AGM progenitors; and fetal liver progenitors). B: The top five differentially expressed genes per cluster. C: Source composition of each cluster. D: Normalized expression pattern of key differentiated and pre-HSC genes, visualized in the UMAP space. E: Adult LT-HSC, ST-HSC, and MPP4 gene module scores per cell, derived from the top 50 differentially expressed genes within single-cell adult dataset from ref²⁹. Scale reflects normalized scores with a cutoff of 0.

Extended Data Figure 10: Characterization of putative eMPP cellular origins.

A: Gating scheme for MDS-GFP analysis: pre-HSCs were defined as Lin- CD45+ cKit+ ESAM+ VE-Cadherin+, and progenitors were defined as Lin- CD45+ cKit+ Il7r- Flt3+. GFP positivity was measured in these two compartments and is quantified in the bar plot; mean + S.D. depicted (n=4 embryos). A representative histogram from one of the four embryos is shown. B: Transcriptomic similarity of “Prog” cells to defined E11 populations from Baron et al.¹⁹ using a random forest classifier. (Non-HE/HE: Non-hemogenic endothelium/hemogenic endothelium; YS HSPCs: yolk sac hematopoietic stem and progenitor cells; T1 pre-HSCs; T2 pre-HSCs). Flt3 and Cd48 expression depicted below. C: UMAP plot of ex vivo derived cultures from pre-circulation yolk sac and embryo proper tissues (from Ganuza et al.²⁸). D: Normalized expression patterns of several putative eMPP genes identified from our single-cell analysis, including Sox4, Hlf, and Hoxa9. A dot plot of key genes shows preferential enrichment within cells derived from the embryo proper. E: Analysis of CD41+ cKit+ cells in the E9.5 yolk sac of MDS-GFP embryos reveals no GFP positivity (representative embryo shown; n=4).

Supplementary Material

Supplementary Figure 3

Supplementary Figure 3: Adult bone marrow and peripheral blood flow analysis scheme

Supplementary Figure 2

Supplementary Figure 2: Adult bone marrow sorting scheme (Transposon model)

Supplementary Figure 1

Supplementary Figure 1: Uncropped Western blots and DNA gels

Supplementary Figure 4

Supplementary Figure 4: Fetal liver, AGM, and yolk sac flow analysis schemes

Supplementary Table 1

Supplementary Table 1: qPCR and ddPCR primers

Supplementary Table 3

Supplementary Table 3: FACS and whole-mount staining antibodies

Supplementary Table 2

Supplementary Table 2: Tn5 adaptors and Sleeping Beauty transposon amplification primers

Supplementary Note

Supplementary Note: Statistical dropout model for quantifying eMPP clones

DATA AVAILABILITY

The Gene Expression Omnibus accession number is GSE180357, which contains the single-cell RNA sequencing data. Source data are provided with this paper.

CODE AVAILABILITY

Custom code for transposon alignment and analysis has been previously published²⁰.