Host factor TIMP1 sustains long-lasting myeloid-biased hematopoiesis after severe infection

Immunity against pathogens relies on lymphoid and myeloid cells, which are maintained in a balanced ratio under a steady state through a stepwise developmental process from hematopoietic stem cells (HSCs) and multipotent progenitor (MPP) cells to the downstream lymphoid or myeloid lineage–committed progenitor cells (Pietras et al., 2015; Rodriguez-Fraticelli et al., 2018). The lymphoid/myeloid equilibrium, however, can be precipitously disrupted by systemic bacterial or viral infection. Upon sensing the pathogen-associated molecular patterns (PAMPs) and stimulated by proinflammatory cytokines, quiescent HSCs are activated and hyperproliferative HSCs preferentially differentiate toward the myeloid lineage cells to meet the increased demand for innate immune defense during acute infection (Trumpp et al., 2010; Baldridge et al., 2011; Takizawa et al., 2012; Chavakis et al., 2019; Hormaechea-Agulla et al., 2020; Caiado et al., 2021). Such an emergency myelopoiesis is considered transitory since PAMPs are eliminated after pathogen clearance and proinflammatory cytokines decrease to basal levels when the body returns to the normal and healthy state. However, in some cases, enhanced myelopoiesis persists in the absence of inciting infectious agents and the known inflammatory mediators after the resolution of infection (Kaufmann et al., 2018). While a persistently high output of myeloid cells following infection could improve innate immunity (Ciarlo et al., 2020; Moorlag et al., 2020), enhanced myelopoiesis is often associated with chronic inflammatory conditions, such as autoimmune diseases, cancer, and “inflammaging” (Caiado et al., 2021; Niu et al., 2013; Pang et al., 2011; Schultze et al., 2019; Sudo et al., 2000). Thus, to modulate innate immunity and control the harmful side effects of inflammation, it is important to understand whether infection-experienced animals manage the long-term enhanced myelopoiesis by altering the HSC intrinsic differentiation program or the bone marrow (BM) hematopoietic niche microenvironment that controls the fate decision during HSC differentiation.

Hematopoietic stem and progenitor cells (HSPCs) reside in the BM in a specialized niche that is comprised of multiple populations of cells, including adipocytes, endothelial cells, fibroblasts, osteoblasts, and other mesenchymal stromal cells (Baccin et al., 2020; Baryawno et al., 2019; Morrison and Scadden, 2014; Tikhonova et al., 2019). These cells provide structural integrity, cellular communication, and local instructive signals to maintain HSC quiescence and self-renewal, as well as to regulate HSPC differentiation. Infection induces a variety of damages to the BM hematopoietic niche (Batsivari et al., 2020). For instance, sepsis can lead to the ablation of IL-7–producing osteoblasts that support lymphopoiesis (Terashima et al., 2016) and causes endothelial malfunction (Maneta et al., 2023). The sepsis-caused damage to other cellular components of the BM niche has not been well determined. It is also unclear whether sepsis-injured BM niche can be fully restored and to what extent it affects the myeloid/lymphoid lineage balance after severe infection.

Here, we explored the long-term impact of severe infection on HSCs per se, as well as on the hematopoietic microenvironment in mice surviving polymicrobial peritonitis (sepsis). We observed a persistent increase in myelopoiesis in mice, even months after the initial infection had resolved. We demonstrate that the tissue inhibitor of metalloproteinases 1 (TIMP1) is constitutively upregulated in the infection-experienced mice, which facilitates HSPCs to adopt a myeloid-biased differentiation program. Such a regulation is dependent on the TIMP1-mediated inhibition of Notch signaling in HSPCs.

B6 mice subjected to cecal ligation and puncture (CL&P) developed polymicrobial peritonitis, bacteremia, septic shock, multiorgan dysfunction (Cuenca et al., 2010; Fink, 2014), and ∼50% of treated mice became moribund under our surgical procedure (Fig. S1 A). The hematopoietic system is severely damaged in septic mice, as lymphopenia and emergency myelopoiesis develop within days following CL&P surgery (Hotchkiss et al., 2013; van der Poll et al., 2021; Wichterman et al., 1980). Thus, to investigate whether the generation of lymphoid and myeloid cells in mice surviving severe infection could be restored to the preinfection balanced state, we decided to use mice surviving CL&P surgery as our model. We examined lymphoid and myeloid cells in the peripheral lymphoid organs of CL&P and sham-treated mice at different time points after surgery. We found that splenic B cell counts normalized 2 wk after CL&P surgery (Fig. S1 B), consistent with a previous report (Terashima et al., 2016). T cell numbers in CL&P mice recovered at a slower rate relative to B cells, with a complete normalization 10 wk after surgery (Fig. S1 C). In contrast to lymphoid cells, the counts of peripheral monocytes and granulocytes remained higher in CL&P mice than in controls 12 wk after surgery (Fig. S1 D), and there was an increase in circulating myeloid cells even 7 mo after CL&P surgery (Fig. S1 E). Given the relatively short lifespan of peripheral myeloid cells, it is likely that infection has altered the BM myeloid output.

To determine whether the increase in the peripheral myeloid cells was due to an enhanced BM myelopoiesis, we analyzed the BM compartment in mice surviving sepsis. The BM of CL&P mice had higher cellularity compared to sham mice, with a similar number of B cells but approximately a twofold higher count of monocytes (CD11b⁺Ly6C⁺) and granulocytes (Ly6G⁺) (Fig. 1 A). There was also a 1.8-fold increase in the number of HSPCs (LSK: Lineage marker [Lin]⁻ c-Kit⁺ Sca-1⁺) in the BM of CL&P mice relative to sham mice (Fig. 1 B). Within the LSK population, lymphoid-primed MPP4 cells and myeloid-biased MPP2 and MPP3 cells, as well as long-term (LT-) and short-term (ST-) HSCs can be identified using CD150, CD48, Flt3, and CD34 (Fig. S1 F). We found a comparable number of MPP4 cells in sham and CL&P mice, however, the number of MPP2 and MPP3 cells was increased 4.6- and 3.6-fold respectively in CL&P mice over sham mice (Fig. 1 B). Overall, the increased LSK cells in CL&P mice contained higher proportions of MPP2 and MPP3 subsets and decreased MPP4 and ST-HSCs, whereas the percentage of LT-HSCs remained unchanged in the CL&P BM compared with control mice (Fig. 1 C). Thus, the cellular expansion of MPP3 cells resulted in a drastically biased ratio of MPP3 to MPP4 cells in the BM of CL&P mice compared with sham mice (Fig. 1 D). Collectively, these data highlight a different impact of severe systemic infection on the production of lymphoid and myeloid cells, with lymphopoiesis only transiently impaired in the septic phase, whereas the myelopoiesis is continuously augmented during septic shock and after infection has resolved.

To understand the mechanisms underlying the myeloid-biased HSPC differentiation in mice surviving severe infection, we performed single-cell RNA sequencing (scRNA-seq) on LSK cells sorted from the BM of mice at week 12 after sham or CL&P surgery (Fig. S2, A and B). After quality control and exclusion of contaminating cells expressing lineage-restricted genes (Table S1), a dataset containing 3,021 and 3,843 LSK cells respectively of sham and CL&P mice was processed to generate graph-based clusters using the smart local moving (SLM) algorithm (Fig. S2 C). Based on the top differentially expressed genes (DEGs) representing signature genes for HSCs and MPPs, we conducted a reference-based classification using published datasets to annotate the seven unsupervised clusters as five distinct cell states (Fig. 2 A; Fig. S2, D and E; and Table S2). The cluster identity was further validated as UMAP plots displayed nearly identical gene expression profiles between the top features of annotated clusters and the published signature genes of HSC and MPP subsets (Fig. S2, E and F) (Klein et al., 2022). Cell-type proportion analysis of the scRNA-seq data revealed an expansion of MPP2 and MPP3 cells and a decrease in MPP4 cells in the CL&P sample (Fig. 2 B), consistent with the flow cytometry results (Fig. 1 C).

Next, we conducted trajectory analyses to infer cell-fate specification and differentiation of HSPCs after infection. We designated cells expressing signature genes of dormant LT-HSCs (e.g., Cdkn1c, Tcf15, Neo1, Pdzk1ip1, Fgfr1, Mmrn1, and Tinagl1) as the starting cell state (the root node 0) and used Monocle 3 to build a trajectory tree, which identified three branch points leading to five trajectory outcomes (leaf nodes 1–5) over pseudotime (Fig. 2 C). By comparing the pseudotime-highlighted trajectory plot to the cell state-highlighted one, we observed that HSCs were enriched in leaf node 1, which was closest to the root, whereas MPP2, MPP3, and MPP4 diverged from three branch points and were distributed over the remainder of the trajectories, and cycling cells were coalesced in leaf node 3 at the extremity of the trajectory tree (Fig. 2 C). Over pseudotime, cells progressively downregulated HSC markers (e.g., Kit and Fgfr1) along with an overall increase in the expression of genes associated with differentiation and the cell cycle (Fig. 2 D). To quantitatively assess cell state transition across the trajectory, we plotted cell states over pseudotime and calculated the frequencies of HSCs and MPPs of sham or CL&P samples at each leaf node and between branch points (Fig. 2 E). We found an equivalent frequency of sham and CL&P HSCs in the root node (node 0). However, the frequency of CL&P HSCs in leaf node 1 was reduced to 60% of that in the sham control (Fig. 2 E), similar to the flow cytometry results demonstrating that the percentage of LT-HSC is unchanged while ST-HSC is proportionally decreased (Fig. 1 C). These data suggest defective self-renewal or accelerated cell cycle progression and differentiation of HSCs in CL&P mice. As expected, there was a drastic reduction in CL&P MPP4 frequency throughout the entire trajectory. Compared with sham MPP3 cells, half of which emerged in an earlier pseudotime domain than MPP4 cells did, CL&P MPP3 differentiation occurred at later branch points and expanded at leaf nodes 2 and 4 (Fig. 2 E). The distinct topological positionings of MPP3 and MPP4 cells of sham and CL&P mice on the developmental trajectories suggest that increased myelopoiesis in mice surviving severe infection is associated with altered differentiation programs at multiple cell-fate decision points during development.

To understand the biological and molecular underpinnings of the altered HSPC differentiation in mice surviving severe infection, we first identified genes that were differentially expressed in sham and CL&P LSK cells (Fig. S3 A) and then applied Gene Ontology (GO) analysis for genes that were up- or downregulated in CL&P LSK cells. We found that genes underrepresented in CL&P LSK cells included MHC molecules that were expressed by lymphoid-primed MPP4 cells (Fig. S3, A and C), whereas genes upregulated in CL&P LSK cells were associated with cellular biosynthesis and mitochondrial activities (Fig. S3 B). Gene set enrichment analysis (GSEA) complemented the results of GO analysis and showed that CL&P LSK cells were enriched for mitochondria biogenesis, tricarboxylic acid cycle, regulation of cell cycle transition and myeloid cell function (e.g., innate immune response) (Fig. 3 A). In contrast, sham LSK cells were enriched for autolysosome formation, somatic stem cell maintenance, and lymphocyte-mediated immunity (Fig. 3 A). In line with previous literature that high mitochondrial activity propels HSCs to exit dormancy (Chen et al., 2008; Nakamura-Ishizu et al., 2020; Vannini et al., 2016), we saw notable increases in pathways regulating the cell cycle in CL&P LSK cells (Fig. 3 A and Fig. S3 B). Consistent with these data, cell cycle profiling of sham and CL&P LSK cells with Pcna (expressed in all cell cycle phases except G0) and Ccnb2 (expression level peaked in the G2/M phase) showed a decrease in the frequency of CL&P LSK cells in the G0 phases of the cell cycle compared with sham cells (Fig. S3, D and E). Flow cytometry analysis, which measured DNA content and the protein level of the cell proliferation marker Ki67, confirmed that there were indeed fewer quiescent CL&P LSK cells relative to sham cells (Fig. 3 B), with a notable reduction of LT-HSCs and MPP4 cells in the G0 phase in the CL&P BM (Fig. S3, F and G). This decrease in CL&P LSK cells in the G0 phase was correlated with an increased expression of the cell cycle activator Cdk6 (Fig. S3 H). Expression of genes involved in the licensing of DNA replication, such as Mcm3 and Mcm5, was particularly upregulated in CL&P LSK cells in the G0 phase. Conversely, Foxo3, known to regulate autophagy for maintaining HSC quiescence (Miyamoto et al., 2007), was expressed in fewer CL&P LSK cells and at a relatively lower level, and Sirt7 that suppresses mitochondrial metabolism and myelopoiesis (Mohrin et al., 2015) was also downregulated in CL&P LSK cells in the G0 phase of the cell cycle compared to sham cells (Fig. S3 H). These analyses thus reveal that CL&P LSK cells are enriched for biological processes, distinct from those associated with quiescent sham LSK cells, and of note, preferentially use metabolic pathways that are known to facilitate myeloid differentiation (Ito et al., 2019).

Next, we interrogated lineage specification and fate decisions of sham and CL&P LSK cells. Principal component (PC) analysis showed that sham and CL&P LSK cells shared an overall similar distribution pattern, with HSCs high in PC1 scores, MPP4 cells high with PC2 scores, and MPP3 cells predominantly high in PC1 scores and low in PC2 scores (Fig. 3 C). Notably, a cluster of PC1^low PC2^high PC3^high cells expanded in the CL&P sample relative to sham control, and this cluster of cells predominantly resided in the leaf node 2 domain of the pseudotime rank (Fig. 3 D). We curated core signatures that specify each leaf node (Fig. S3, I and J; and Table S3) and profiled sham and CL&P LSK cells expressing leaf node 2 signature genes because the leaf node 2 core signature was likely associated with the aberrantly differentiated CL&P LSK cells. Compared with the expression profile of sham cells, CL&P cells singularly expressed a group of genes specific for granulocyte–monocyte progenitors (GMPs) that were developmentally downstream of LSK cells (Fig. 3 E). In particular, Ms4a3, a marker for GMPs committed to the myeloid lineage (Ishibashi et al., 2018; Liu et al., 2019), was aberrantly expressed in CL&P LSK cells, while its expression was barely detectable in sham LSK cells as previously reported (Fig. 3, E and F). To evaluate the lineage specification of Ms4a3-expressing CL&P LSK cells, we examined the expression of a lymphoid commitment marker, Satb1, in these cells (Satoh et al., 2013). We found that while Satb1 expression was indeed restricted in sham MPP4 cells, it was expressed not only by CL&P MPP4 cells but also by the aberrant cluster of CL&P cells in the leaf node 2 domain (Fig. 3 F). As a matter of fact, a substantial amount of CL&P LSK cells co-expressed Satb1 and Ms4a3 (Fig. 3 G), as well as other myeloid and lymphoid-specific genes, such as Ctsg and Dntt (Fig. S3 K). These results reveal increased lineage plasticity of HSPCs in CL&P mice long after infection has resolved and suggest that a significant proportion of HSPCs that have begun lymphoid-lineage specification may eventually escape the lymphoid cell fate and develop into the myeloid lineage.

The above analyses show that HSPCs in mice surviving severe infection shifted from a quiescent to an active state, altered multiple biological processes, and adopted a myeloid-primed differential pathway. These functional changes could result from altered cell-autonomous developmental programs or be inflicted by infection-injured hematopoietic niche. To examine HSPC-intrinsic differentiation potential after infection, we transferred 2 × 10⁶ BM cells from CD45.2⁺ B6 mice 12 wk after sham or CL&P surgery into irradiated congenic CD45.1⁺ B6.SJL mice, respectively. 6 wk after BM transplantation, CD45.2⁺ BM cells from sham- or CL&P-treated mice generated an equivalent number of granulocytes and monocytes, as well as lymphocytes in the recipient mice (Fig. 4 A), demonstrating that the cell-intrinsic developmental potential of HSPCs from CL&P mice is normal. To determine whether the post-infection microenvironment influenced myeloid differentiation, we transferred 2 × 10⁶ BM cells from naïve CD45.1⁺ B6.SJL mice into irradiated B6 mice 12 wk after sham or CL&P surgery. 6 wk later, we found that CL&P recipients possessed significantly more myeloid cells relative to sham hosts whereas lymphocyte counts were comparable in both groups of hosts (Fig. 4 B). Since both sham and CL&P hosts received the same dose of irradiation, the effect of irradiation on the BM niche cannot account for the increased myeloid output in CL&P recipients. These results thus support that severe infection causes durable alterations in the host hematopoietic microenvironment that favor myelopoiesis.

Based on our scRNA-seq data suggesting increased lineage plasticity of HSPCs in CL&P mice and the reciprocal transplantation results highlighting the host’s impact on enhanced myelopoiesis, we hypothesized that the post-infection microenvironment redirects lymphoid precursors to the myeloid cell fate. To test this, we isolated lymphoid-primed MPP4 cells from naïve mice and examined their differentiation potential in sham or CL&P hosts. As previously reported, MPP4 cells primarily gave rise to B cells when transplanted into naive hosts, while MPP3 cells predominantly produced myeloid cells (Fig. 4 C). In striking contrast to the dominant lymphoid potential of MPP4 cells in sham hosts, MPP4 cells in CL&P recipient mice generated a significant number of myeloid cells (Fig. 4 D), demonstrating that the microenvironment in the post-infection mice indeed drives differentiation of lymphoid-primed precursors to the myeloid lineage.

During acute infection, proinflammatory cytokines, including IL-1, IL-6, G-CSF, and TNF, activate HSPCs to proliferate and generate more myeloid cells (Hormaechea-Agulla et al., 2020). However, these cytokines had returned to the preinfection baseline levels by 10 wk after CL&P surgery (Fig. 5 A), suggesting there must be other factors responsible for sustaining myeloid-biased hematopoiesis in mice after infection. We used a cytokine array to identify soluble factors that were elevated in CL&P mice compared with sham controls 10 wk after surgery. A striking difference between sham and CL&P samples was the presence of abundant TIMP1 in the latter (Fig. 5 A). Post-surgery kinetics analysis revealed that TIMP1 was produced at the highest level in the acute phase of infection, gradually decreased over time but still maintained at a higher level in CL&P mice relative to sham mice for at least 12 wk after surgery (Fig. 5 B). 12 wk after surgery, serum TIMP1 levels strongly correlated with granulocyte counts in the BM of sham and CL&P mice (Fig. 5 C).

We next investigated whether increased levels of TIMP1 in mice surviving severe infection played any role in myeloid-biased hematopoiesis. We cultured WT BM cells in methylcellulose medium with a suboptimal dose of cytokines and 10% serum from mice at 12 wk after sham or CL&P surgery. Significantly, more colonies were formed in the culture supplemented with serum from CL&P mice than with serum from sham mice (Fig. 5 D). Flow cytometry analyses showed an increased frequency of Ly6G⁺ granulocytes harvested from the CL&P serum culture compared with the sham control (Fig. 5 E), suggesting that soluble factors in CL&P serum stimulate myeloid differentiation. Treatment of cultures with an anti-TIMP1 antibody diminished the generation of Ly6G⁺ cells, indicating that TIMP1 is a cytokine in the CL&P serum that enhances myelopoiesis (Fig. 5, D and E). To explore the impact of TIMP1-upregulation on hematopoiesis in vivo, we transduced BM LSK cells with a TIMP1-expressing retroviral vector and transferred these cells into irradiated Timp1^−/− recipient mice. Overexpression of TIMP1 was only detected in mice reconstituted with TIMP1-expressing BM cells but not those received control viral vector–transduced BM cells (Fig. S4 A). Consistent with previous reports (Kobuch et al., 2015; Rossi et al., 2011), we found that TIMP1 deficiency did not affect myelopoiesis; however, overexpression of TIMP1 led to an expansion of CD11b⁺ myeloid cells (Fig. 5 F).

TIMP1 is produced by both myeloid and stromal cells (Boyd et al., 2020; Guccini et al., 2021; Kuroda et al., 2019; Wynn and Barron, 2010; Zhao et al., 2021). To determine the cellular contribution to the enhanced durable expression of TIMP1 after infection, we generated BM chimeras by transferring WT or Timp1^−/− (KO) BM cells into irradiated CD45.1⁺ B6.SJL recipients and then performed sham or CL&P surgery 6 wk after BM engraftment (Fig. S4 B). Host mice receiving WT BM cells or KO BM cells manifested identical septic symptoms, and 50% WT and KO BM chimeras surviving sepsis had a comparable increase in granulocytes (Fig. S4 C). Of note, mice reconstituted with Timp1^−/− BM cells had the same basal levels of TIMP1 after sham surgery and the same elevated amounts of TIMP1 after CL&P surgery as those seen in WT BM chimeras (Fig. S4 D), indicating that tissue stromal cells rather than hematopoietic cells are the main producers of TIMP1 in sepsis surviving mice. To examine cellular sources of TIMP1 within the BM compartment, we isolated CD45⁺ hematopoietic cells, CD45⁻ CD31⁺ endothelial cells, and CD45⁻ CD31⁻ stromal cells from sham and CL&P mice (Fig. S4 E) and determined Timp1 transcript levels by RT-PCR. In line with the aforementioned BM reconstitution experiments, CD45⁺ cells expressed little Timp1 in both sham and CL&P mice. Endothelial cells expressed basal levels of Timp1 which were not upregulated by infection. Of note, sham BM stromal cells contained 3.5 times higher Timp1 transcripts than sham endothelial cells, and there was twofold upregulation of Timp1 transcripts in CL&P BM stromal cells relative to the sham counterpart (Fig. 5 G). To further identify the cellular sources of TIMP1 at steady state, we examined published datasets of BM stromal cells (Dolgalev and Tikhonova, 2021). Our analysis revealed that fibroblasts, chondrocytes, and osteogenic cells contain TIMP1 producers (Fig. S4, F and G).

Since the main function of TIMP1 is to inhibit metalloproteinases, we hypothesized that TIMP1 may promote myelopoiesis through the regulation of metalloproteinases that are involved in HSC fate decision. ADAM10 appears to be a putative TIMP1 target since it is the essential sheddase activating Notch, which in turn governs lineage specification during hematopoiesis (Ng et al., 2021; Stier et al., 2002; Tian et al., 2008). To test this hypothesis, we first determined the potency of TIMP1 on inhibiting ADAM10 proteolytic activity. Using an in vitro colorimetric assay, we found that TIMP1 inhibited ADAM10 activity as effectively as the ADAM10 inhibitor GI254023X (Fig. S5 A). To test whether TIMP1 was able to suppress Notch signaling, we measured the expression of the Notch target gene Hes1 in LSK cells stimulated with the Notch ligand DLL4. As expected, the expression of Hes1 expression was reduced to baseline levels in LSK cells treated with the ADAM10 inhibitor due to a blockade of Notch activation in the absence of ADAM10 cleavage. In the presence of TIMP1, DLL4-stimulated LSK cells expressed as little Hes1 as the ADAM10 inhibitor-treated cells (Fig. 6 A), indicating that TIMP1 efficiently inhibits Notch activation in LSK cells. Consistent with this in vitro observation, we found that Hes1 expression was considerably lower in LSK cells from mice at week 12 after CL&P surgery compared with that from sham controls (Fig. 6 B). scRNA-seq analyses showed that while both sham and CL&P LSK cells expressed comparable levels of Notch receptors, ADAM10, γ-secretase complex subunits (Presenilin 1) and Notch co-activators (Mastermind-like [MAML], RBPJ) (Fig. S5 B), there were fewer CL&P LSK cells expressing Notch target genes and at much lower levels relative to sham LSK cells (Fig. 6 C). The gene sets of positive regulation of Notch signaling were also found to be enriched in sham LSK cells as compared with CL&P LSK cells (Fig. S5 C). Together, these results indicate that while the expression of Notch signaling components is not altered in CL&P LSK cells, Notch signaling in these cells is suppressed by TIMP1-mediated inhibition of ADAM10 proteolytic activity.

To further determine whether inhibition of Notch signaling in LSK cells was relevant to the myeloid-biased HSPC differentiation, we cultured LSK cells from B6 mice with cytokines (SCF and Flt3L) that support LSK cell proliferation and myeloid differentiation (Varnum-Finney et al., 2011). Notch signaling activated by DLL4 significantly suppressed the development of CD11b⁺ myeloid cells in a 5-day culture, and the addition of TIMP1 in the culture antagonized the suppressive activity of DLL4 on myeloid differentiation (Fig. 6 D). Moreover, in vivo inhibition of ADAM10 by treating mice with GI254023X led to an increase in CD11b⁺ myeloid cells (Fig. 6 E). These findings thus demonstrate that high levels of TIMP1 can promote myelopoiesis through inhibition of the ADAM10-Notch signaling axis in mouse HSPCs.

To investigate whether TIMP1 has a relevant role in hematopoiesis in human sepsis, we determined the plasma levels of TIMP1 in sepsis patients and in healthy age-matched persons. Plasma samples were collected from patients at the time of sepsis diagnosis (n = 22) or from sepsis survivors at 6 mo after hospital discharge (n = 4). TIMP1 levels were 10-fold higher in the plasma of patients with sepsis than in healthy people (Fig. 7 A). A high mortality rate within the first 28 days of hospitalization was seen in patients with a high level of circulating TIMP1 (Fig. 7 B), indicating that TIMP1 levels correlate to the severity of sepsis. Furthermore, TIMP1 levels were approximately threefold higher in sepsis survivors 6 mo after hospital discharge compared with age-matched healthy controls (Fig. 7 C). Taken together, these findings indicate that, as observed in mouse sepsis model, severe infection in humans also induces rapid upregulation of TIMP1 expression, which is maintained at a higher level even after infection has resolved.

Next, we sought to determine whether TIMP1 also influenced human HSPC differentiation by regulating the Notch signaling pathway. Human HSPCs with >90% of CD45⁺ CD34⁺ cells were obtained from the BM of healthy donors (Table S4). Upon stimulation with DLL4, human CD34⁺ HSPCs significantly upregulated Hes1 expression, and the addition of either the ADAM10 inhibitor GI254023X or TIMP1 suppressed Hes1 expression (Fig. 7 D). To examine whether TIMP1 promoted myelopoiesis of human CD34⁺ HSPCs, we monitored CD11b⁺ myeloid cells developed from the cultured human HSPCs under different stimulation conditions. We found that activation of Notch signaling in HSPCs suppressed myeloid differentiation, as significantly less CD11b⁺ cells were generated in the culture in the presence of DLL4 (Fig. 7 E). The addition of TIMP1 in the culture reversed DLL4-mediated suppression of myelopoiesis in vitro (Fig. 7 E). Thus, these results demonstrate that, similar to what we observed in our mouse sepsis model, infection also induces TIMP1 upregulation in human sepsis patients, which is maintained for a long time after the infection is resolved. In addition, our data support the hypothesis that TIMP1 could also promote persistent post-infection myelopoiesis by suppressing Notch signaling in human HSPCs.

Upon infection, HSCs rapidly shift from a homeostatic to an inflammatory state and increase the output of immune cells of mostly myeloid lineages to boost innate immunity (Baldridge et al., 2011; Caiado et al., 2021; Chavakis et al., 2019). Since a balanced production of multi-hematopoietic lineage cells is crucial to the immune system function, it is important to understand whether hematopoiesis could return to the preinfection state, and if not, what is the mechanism that perturbs HSC function after severe infection. In this report, we used a mouse CL&P polymicrobial peritonitis (sepsis) model as a paradigm of hematopoiesis during and after severe infection. Data about hematopoiesis after sepsis are scanty, possibly due to an assumption that hematopoiesis returns to the steady state, since the proinflammatory cytokines known to influence hematopoiesis have decreased to the basal level. In contrast to this assumption, our results reveal that at least up to 3 mo after surgery, HSPCs have not returned to the preinfection homeostasis. There are still significantly more HSPCs in the cell cycle in CL&P mice as compared with that in sham mice. Unlike steady-state HSPCs that mainly rely on glycolysis (Ito and Suda, 2014; Takubo et al., 2013), CL&P HSPCs preferentially use mitochondrial biosynthesis and respiration metabolic pathways. Metabolic regulators such as Sirt7 and Foxo3, known to suppress myelopoiesis (Kaiser et al., 2020; Miyamoto et al., 2007; Mohrin et al., 2015), are downregulated in CL&P HSPCs as compared with sham HSPCs. Concordant with this molecular signature, CL&P HSPCs generate significantly more myeloid cells relative to sham HSPCs. Thus, emergency myelopoiesis induced by infection persists after the known proinflammatory cytokines wane and the infection has been cleared.

It has been reported that HSPCs not only sense and respond to but also remember the infection so that trained immunity can be acquired following infection as infection-experienced HSPCs preferentially produce more innate immune cells relative to naïve HSPCs (Bekkering et al., 2021). Microbial components, such as Candida albicans β-glucan and Bacillus Calmette–Guérin vaccine can induce HSPC-intrinsic metabolic and epigenetic alterations, thereby establishing an inflammatory memory in HSPCs after infection (Kaufmann et al., 2018; Mitroulis et al., 2018). In our experimental setting, we find that CL&P HSPCs function similarly to sham HSPCs after being transferred into a non-inflammatory environment. Importantly, naïve lymphoid-primed MMP4 cells without prior experience of inflammation began to produce a significantly high number of myeloid cells after being transferred into CL&P mice. Our results show that infection could upregulate novel mediators, such as TIMP1, that bias HSPC differential potential to sustain myeloid-biased hematopoiesis beyond the acute phase of severe infection. This mechanism is likely conserved in mice and humans, given our observation that TIMP1 is also upregulated in human patients surviving sepsis and that TIMP1 promotes myelopoiesis of human HSPCs in vitro. TIMP1 has been shown to be upregulated in BM stromal cells of aged mice, as well as by infection-mimicking agents, such as lipopolysaccharide and polyinosinic–polycytydilic acid (Helbling et al., 2019). Further studies are needed to elucidate whether TIMP1 exhibits a similar myeloid-boosting effect in milder infection conditions and inflammaging. It will also be important to determine whether TIMP1 can cooperate with somatic mutations associated with clonal hematopoiesis of indeterminate potential to further enhance myeloid-biased hematopoietic output.

The BM hematopoietic niche contains various types of cells that foster and educate residing HSCs through cell–cell interactions and secreted mediators (Morrison and Scadden, 2014). Our studies show that infection induces BM stromal cells to produce excessive TIMP1. TIMP1 has not been considered a conventional proinflammatory factor, despite previous literature showing elevated circulating TIMP1 levels in patients with septic shock (Lorente et al., 2014), acute respiratory distress syndrome (Boyd et al., 2020), and inflammatory malignancy (Zhao et al., 2021). TIMP1 is also nonessential for steady-state hematopoiesis (Rossi et al., 2011); however, overexpression of TIMP1 enhances granulopoiesis (Kobuch et al., 2015). Mechanistically, our studies show that TIMP1 inhibits ADAM10-mediated proteolytic activation of Notch. Notch signaling has a complex role in adult hematopoiesis (Bigas and Espinosa, 2012; Lampreia et al., 2017). For example, deletion of Rbpj, the transcription factor of canonic Notch signaling, has no effect on hematopoiesis under both steady-state and stress conditions nor does it affect the expression of Notch target genes (Duarte et al., 2018). On the other hand, Notch signaling and its target Hes1 have been shown to govern the HSC cell fate by suppressing the expression of myelomonocytic transcriptomes (Klinakis et al., 2011; Varnum-Finney et al., 2011). Defective Notch signaling leads to myeloproliferative disorder (Kim et al., 2008; Klinakis et al., 2011), which also phenocopies that of ADAM10 knockout mice (Yoda et al., 2011). Here, we have observed that the generation of myeloid cells is drastically inhibited when both mouse and human HSPCs are cultured in the presence of the Notch ligand DLL4 and that the addition of TIMP1 counteracts the inhibitory effects of DLL4. In agreement with the functional assays, our scRNA-seq analyses reveal that Notch target genes are downregulated in CL&P HSPCs relative to sham cells, which is accompanied by the expansion of an aberrant population of progenitor cells that co-expressed both myeloid and lymphoid commitment genes. These results thus support a model that under the steady state MPPs may commit to the lymphoid cell fate by upregulating the lymphoid transcriptomes and simultaneously downregulating the myelomonocytic transcriptomes by Notch signaling. However, in the post-infection host that overexpresses TIMP1, lymphoid cell-fate specified MPPs cannot complete the cell-fate determination because Notch signaling attenuated by TIMP1 is not sufficient to repress the myelomonocytic gene expression program, thus re-directing the cell fate toward the myeloid trajectory.

Taken together, our work shows that hematopoiesis has not returned to the homeostatic state even after severe infection has resolved. The persistent inflammatory state of HSPCs is supported by an altered hematopoietic microenvironment in animals surviving severe infection. Our studies identify TIMP1 as a previously undescribed inflammatory mediator that regulates HSPC differentiation toward the myeloid lineage via the ADAM10-Notch signaling axis. It is likely that simultaneous derangement of multiple pathways, rather than a single mediator, is involved in maintaining the inflammatory state of hematopoiesis after severe infection. Future studies are needed to identify other novel factors and the biological processes that contribute to enhanced myelopoiesis after infection.

Plasma samples were obtained from patients enrolled in a prospective, non-therapeutic study of adults 21 years of age or older who had a clinical diagnosis of sepsis with known or suspected infection and who were treated at North Shore University Hospital between March 14, 2008 and January 25, 2010. Sepsis was defined as the presence of any two of the following clinical features: core body temperature ≥38°C or ≤36°C, heart rate ≥ 90, respiratory rate ≥ 20, total ventilation >10 L/min or a PaCo₂ of <32 mmHg, WBC ≥ 12,000 or ≤ 4,000, or a differential count showing >10% bands. Patients were excluded from the study if they were pregnant or breast feeding, hospitalized <10 days prior to the current visit, had a history of BM or solid organ transplant, alcoholism, i.v. drug use, chemotherapy, or neutropenia. Plasma samples were obtained at the time of sepsis diagnosis and for a subset of the patients who survived sepsis, at 6 mo after hospital discharge. Blood was collected from sepsis patients with routine morning samples or from a central line access when available to decrease the need for additional needle sticks. As controls, blood was collected from healthy age-matched subjects. Blood processing: 1 ml of EDTA anti-coagulated blood from each study subject was processed for plasma by two rounds of centrifugation (1,000 rpm followed by 10,000 rpm) and then filtration through a 0.22-μm filter. Plasma samples were stored frozen at −70°C until used for analysis. All procedures were approved by The Feinstein Institutes for Medical Research/Northwell Health Office of the Institutional Review Board (IRB# 05-091). All study subjects provided informed written consent.

CD34⁺ HSPCs were purchased from StemCell Technologies and HemaCare, respectively (Table S4).

C57BL/6J (B6), B6.SJL-Ptprc^aPepc^b/BoyJ (B6.SJL), and B6.129S4-Timp1^tm1Pds/J (Timp1^−/−) mice were purchased from the Jackson Laboratory. Mice were housed under specific pathogen–free conditions. Rodent chow and water were provided ad libitum. Mice were at the age of 8–12 wk when experiments started. Animal experiments were approved by the Institutional Animal Care and Use Committee of the Feinstein Institutes of Medical Research and were performed in accordance with the institutional, state, and federal guidelines.

CL&P was performed to induce sepsis in accordance with an established protocol (Cuenca et al., 2010). 8-wk-old male mice were anesthetized with 3% isoflurane and with O₂ flow at 1 L/min. A midline incision was made at the abdomen to exteriorize the cecum. Approximately, one-third of the cecum from its distal end was ligated using 3-0 polydioxanone (PDS II) sutures (Ethicon). The ligated cecum was then punctured once with a 22-gauge needle (BD Biosciences) and a small amount of feces (about 1 mm³) was extruded. After replacing the cecum into the abdominal cavity, the abdominal wall was closed with 4-0 PDS II sutures (Ethicon) followed by a skin closure with medical surgical clips (Braintree Scientific). Sham mice underwent the same surgery except for the ligation and puncture of the cecum. All mice were resuscitated by subcutaneous injection of 1 ml of PBS supplemented with antibiotics (imipenem/cilastatin, 0.5 mg/kg body weight).

Mouse Timp1 cDNA was obtained by PCR with Timp1-specific primers and cloned into the retroviral expressing vector MSCV-GFP to generate the MSCV-TIMP1 construct. Retrovirus was produced by calcium phosphate transfection of 293FT cells (Invitrogen) with MSCV-TIMP1 and a helper plasmid pCL-Eco. Retroviral supernatants were collected twice at 36 and 60 h after transfection.

BM cells were harvested at day 5 after mice received i.p. injection of 3 mg/kg 5-flurouracil (5-FU; Sigma-Aldrich) and cultured in DMEM supplemented with 10% FCS (Hyclone), 10 ng/ml IL-3 (PeproTech), 10 ng/ml IL-6 (PeproTech), and 50 ng/ml SCF (PeproTech). Retroviral infection of BM cells was achieved by spin infection using retroviral supernatants according to a standard protocol.

For MPP cell transplantation, sham and CL&P mice (12 wk after surgery) were sublethally irradiated (7 Gy) and i.v. injected 3,000 MPP3 or MPP4 cells sorted from B6.SJL mice.

All recipient mice in the following BM reconstitution experiments received a total of 10 Gy irradiation delivered in two split doses with a 3-h interval.

For the reciprocal reconstitution studies, 2 × 10⁶ total BM cells collected from B6 mice (CD45.2⁺) at 12 wk after sham or CL&P surgery were i.v. injected into B6.SJL recipients (CD45.1⁺), respectively. Reciprocally, 2 × 10⁶ BM cells were collected from B6.SJL mice (CD45.1⁺) and i.v. injected into irradiated B6 mice (CD45.2⁺) at 12 wk after sham or CL&P surgery.

For transplantation of TIMP1-overexpressing HSCs, 2 × 10⁵ retroviral transduced BM cells were transplanted into irradiated Timp1^−/− mice.

To assess whether TIMP1 was derived from hematopoietic or stromal cells, 2 × 10⁶ total BM cells from Timp1^−/− mice or WT littermates were i.v. injected into irradiated B6.SJL recipients (CD45.1⁺) to generate the BM chimeras.

The ADAM10 inhibitor GI254023X (MedChemExpress) was dissolved in 0.1 M carbonate buffer containing 20% DMSO at 10 mg/ml. 10- to 12-wk-old B6 mice were treated with GI254023X via daily i.p. injection at a dose of 100 mg/kg body weight for 7 days. Control mice received an identical volume of vehicle following the same administration scheme.

Splenocytes or BM cells from two pairs of tibias and femurs per mouse were harvested and red blood cells were lysed using ACK (ammonium-chloride-potassium) lysing buffer (150 mM NH₄Cl, 10 mM KHCO₃, 0.1 mM EDTA). Cells were resuspended in FACS buffer (PBS supplemented with 2 mM EDTA and 0.5% BSA), stained for different cell surface markers, and then subjected to flow cytometric analysis or cell sorting.

For isolation of stromal cells, tibias and femurs were dissected and cleaned to remove attached tissues and then cut into ∼1-mm³ pieces with scissors in digestion buffer (DMEM supplemented with 2% FBS [Gibco], 10 mM HEPES [Lonza], 1 mg/ml Collagenase P [#11249002001; Roche], 1 mg/ml Dispase II [#4942078001; Roche] and 66 μg/ml DNase I [Sigma-Aldrich]). Samples were incubated in a 37°C oven with constant rotation for 15 min. Cell suspensions were removed and fresh digestion buffer was added to the remaining bone fragments for an additional 15 min digestion. Cells from the two-step digestions were combined and red blood cells were lysed with ACK lysing buffer.

Flow cytometric analysis was performed on an LSR II or a Fortessa flow cytometer (BD Biosciences) and analyzed with FlowJo software (v.10.0; TreeStar). All antibodies used are listed in Table S5. To analyze cell populations in the spleen, B220, TCR, CD11b, and Gr1 were used to detect B and T lymphocytes, monocytes, and granulocytes, respectively. To examine the BM cell compartment, a cocktail containing antibodies for lineage markers (Lin) was used: B220, CD19, TCR, CD3, CD4, CD8, CD11c, CD11b, Gr1, Ly6C, F4/80, TER119, and NK1.1 (clone PK136). Other antibodies specific to the following markers were also used for the study: c-Kit, Sca1, CD34, CD150, CD48, and Flt3. Gating definition for HSPCs were as follows: LSK, Lin⁻Sca1⁺c-Kit⁺; LT-HSC, CD150⁺CD48⁻CD34^lowFlt3⁻ LSK; ST-HSC, CD150⁻CD48⁻CD34^hiFlt3⁻ LSK; MPP2, CD150⁺CD48⁺Flt3⁻ LSK; MPP3, CD150⁻CD48⁺Flt3⁻ LSK; and MPP4, CD150⁻CD48⁺Flt3⁺ LSK. To analyze human HSPC differentiation in vitro, we used anti-human CD34 and anti-human CD11b antibodies.

For cell cycle analysis, following staining for LSK cell surface markers, cells were fixed and permeabilized with BD Cytofix/Cytoperm buffer (BD Biosciences). Intracellular staining for Ki67 was done using PermWash solution (BD Biosciences). DAPI (Sigma-Aldrich) was then added for DNA content determination.

Cell sorting was performed on a FACSAria II instrument (BD Biosciences). Lin⁻ cells were enriched from BM cells prior to sorting using EasySep Mouse Hematopoietic Progenitor Cell Isolation Kit (StemCell Technologies). Lin⁻ enriched cells were then stained with antibodies specific for Lin, c-Kit, and Sca1 for sorting LSK cells. To sort BM endothelial cells and stromal cells, as well as CD45⁺ hematopoietic cells, enzyme-digested cells were separated into Lin⁻ and Lin⁺ populations using EasySep Mouse Hematopoietic Progenitor Cell Isolation Kit (StemCell Technologies). CD45⁺ cells were sorted by staining Lin⁺ cells with anti-CD45. Lin⁻ cells were stained with anti-CD45, anti-TER119, anti-CD71, and anti-CD31. Endothelial cells were sorted as CD45⁻ TER119⁻ CD71⁻ CD31⁺ and stromal cells were sorted as CD45⁻ TER119⁻ CD71⁻ CD31⁻ cells.

For activation of Notch signaling, the Notch ligand DLL4 (5 μg/ml of human or mouse DLL4 [R&D]) was coated to U-bottom wells (high-binding 96-well plate; Greiner Bio-One) that had been pretreated with RetroNectin (Takara) following the manufacturer’s instruction. 24 h later, wells were washed with PBS, and human or mouse HSPCs were added to the wells and cultured at 37°C for 5 h to induce Notch activation.

LSK cells were sorted from the BM of B6 mice. To determine the expression of the Notch target gene Hes1 by RT-PCR, 2 × 10⁴ LSK cells were seeded into DLL4-coated wells in DMEM supplemented with 10% FCS (Hyclone), 10 ng/ml SCF (PeproTech), and 2 ng/ml Flt3L (PeproTech) in the presence or absence of 100 ng/ml mouse TIMP1 (R&D) or 1 μM GI254023X (R&D) for 5 h. To examine HSPC differentiation, 4,000–10,000 LSK cells per well were cultured in a previously described medium (DMEM with 10% FCS [Hyclone], 50 ng/ml SCF [PeproTech], and 100 ng/ml Flt3L [PeproTech]) (Varnum-Finney et al., 2011) in the presence or absence of 100 ng/ml of mouse TIMP1 for 5 days.

Human HSPCs (Table S4) were cultured in StemSpan SFEM II medium with StemSpan CD34⁺ Expansion Supplement (StemCell Technologies). To determine the expression of human Hes1 by RT-PCR, human HSPCs were seeded into DLL4-precoated wells at 2 × 10⁴ cells/well and cultured for 5 h ±100 ng/ml of human TIMP1 (R&D) or 1 μM GI254023X (R&D). To examine HSPC differentiation, 5 × 10³–10⁴ human HSPCs were cultured with plate-bound DLL4 with or without 100 ng/ml of human TIMP1 (R&D) for 5 days before being subjected to FACS analysis.

For colony formation assays, a suboptimal conditioned methylcellulose medium was made by mixing M3434 (StemCell Technologies) with M3134 (StemCell Technologies) at a ratio of 1:10. Serum pooled from five mice at 10 wk after sham or CL&P surgery were added to the suboptimal condition to a final concentration of 10%. In some experiments, CL&P serum was pretreated with anti-TIMP1 antibody. BM cells from B6 mice were plated at 2 × 10⁴ cells/3 ml/plate in 35-mm Petri dishes in the suboptimal conditioned methylcellulose medium with sham or CL&P serum. After a 10-day culture, colonies in each plate were counted and cells were then harvested for flow cytometric analysis.

For examination of cytokines in the serum of sham and CL&P mice, serum samples collected from five mice at 10 wk after sham or CL&P surgery were pooled, respectively. Cytokines were detected using Proteome Profiler Mouse Cytokine Array Panel A according to the manufacturer’s instructions (R&D).

For measurement of circulating TIMP1 levels, mouse serum was collected at indicated time points and human samples were obtained from septic patients and septic survivors. The amounts of TIMP1 were measured using a mouse or human TIMP1 Quantikine ELISA Kit according to the manufacturer’s instructions (R&D).

For evaluation of the inhibitory activity of TIMP1 on ADAM10, a colorimetric assay was performed to determine ADAM10-mediated cleavage of a fluorogenic peptide substrate Mca-PLAQAV-Dpa-RSSSR-NH2 (R&D) in the presence or absence of TIMP1.

38 μM of recombinant mouse ADAM10 (R&D) was mixed with 20 μM of Mca-PLAQAV-Dpa-RSSSR-NH2 with or without 45 μM recombinant mouse TIMP1 (BioLegend) or 1 μM of the ADAM10 inhibitor GI 254023X (R&D). The reactions were incubated at 37°C for 30 min, and the amount of cleaved peptide substrate was measured by OD₄₀₅ using a Synergy H1 Hybrid reader (BioTek) and was calculated as the relative unit of ADAM10 proteolytic activity.

Total RNAs were isolated from FACS sorted or cultured cells using RNeasy Micro Kit (for cells <10⁵) or RNeasy Mini Kit (QIAGEN) and reverse-transcribed into cDNA using a SuperScript III cDNA Synthesis Kit (Invitrogen) according to the manufacturers’ instructions. qRT-PCR using cDNA, gene-specific primers, and SYBR Green Master Mix (Thermo Fisher Scientific) was performed on a LightCycler 480 II instrument (Roche). Relative expression was quantified using the ΔΔ Ct methods and normalized to Actb.

Gene-specific primers are as follows:

• Mouse Timp1: Forward 5′-CTG​GCA​TCT​GGC​ATC​CTC​TTG-3′ and Reverse 5′-GTG​GTC​TCG​TTG​ATT​TCT​GG-3′;

• Mouse Hes1: Forward 5′-TCC​AAG​CTA​GAG​AAG​GCA​GAC-3′ and Reverse 5′-TGA​TCT​GGG​TCA​TGC​AGT​TG-3′;

• Mouse Actb: Forward 5′-GCT​CTT​TTC​CAG​CCT​TCC​TTC-3′ and Reverse 5′-CTC​CTT​CTG​CAT​CCT​GTC​AG-3′;

• Human Hes1: Forward 5′-TTG​GAT​GCT​CTG​AAG​AAA​G-3′ and Reverse 5′-TTC​TCC​AGC​TTG​GAA​TG-3′;

• Human Actb: Forward 5′-AAG​GCC​AAC​CGC​GAG​AAG​ATG-3′ and Reverse 5′-ATC​ACG​ATG​CCA​GTG​GTA​C-3′.

LSK cells were sorted from three mice at 12 wk after sham and CL&P surgery, respectively. Cells were washed and resuspended in PBS containing 0.04% BSA at 800 cells/μl for the targeted recovery of 10,000 cells. Single-cell partitioning and cDNA library construction were done using Chromium Single Cell 3′ Kit v3 per the manufacturer’s instruction (10X Genomics). Library sequencing was performed using NovaSeq S4 (Illumina) at Yale Center for Genome Analysis.

Cell Ranger (10X Genomics) was used to process raw sequencing data. For the sham sample, the mean reads per cell was 115,132 and the median genes per cell was 2,881. For CL&P sample, the mean reads per cell was 98,713 and the median genes per cell was 3,202. Subsequent data filtering, normalization, and PC analysis (PCA) were conducted with PartekFlow (Partek). Cells that passed the following criteria were used for further analysis: (1) the percentage of mitochondrial genes per cell ≤10%; (2) the percentage of ribosome genes per cell was between 25% and 50%; and (3) the number of detected genes is between 700 and 5,500. Cells that passed the quality control (88%) were then filtered by lineage-associated marker genes (Table S1), resulting in an LSK cell dataset containing 3,021 sham cells and 3,843 CL&P cells. The data were log₂ transformed and normalized. After PCA, the first 20 PCs were selected for the UMAP dimension reduction and clustering using the SLM modularity-based clustering algorithm. The seven unsupervised clusters were annotated by reference-based classification using previously published datasets (Pei et al., 2020; Rodriguez-Fraticelli et al., 2020). Trajectory analysis with Monocle 3, GO, and gene set enrichment analyses was carried out on the platform of PartekFlow (Partek).

All statistics were performed using Prism software (version 6; GraphPad). Two-tailed Student’s t tests were used, with P values <0.05 considered to be statistically significant.

Fig. S1 shows the mortality rate caused by CL&P surgery, the hematopoietic changes in the periphery following CL&P surgery, and the gating strategy of HSPCs. Fig. S2 provides the procedure for scRNA-seq and the classification of HSPC cell clusters. Fig. S3 shows additional scRNA-seq data for changed biological processes in the post-infection HSPCs. Fig. S4 shows the gating strategy for BM stromal cells and the BM cellular sources of TIMP1. Fig. S5 shows the inhibition of ADAM10 cleavage activity by TIMP1 and Notch signaling in the post-infection HSPCs. Table S1 contains lineage-associated marker genes used to filter contaminated cells in scRNA-seq data sets. Table S2 lists top expressed genes of LT-HSCs and MPPs. Table S3 contains a list of core signature genes of the five trajectory leaf nodes. Table S4 lists the sources and information of human HSPCs used in this study. Table S5 contains a list of antibodies and reagents used for flow cytometry analysis.

The scRNA-seq dataset is available in the Gene Expression Omnibus database under accession number GSE205728. This study also uses a published dataset under accession number GSE145491 (Klein et al., 2022) and data from “Study: Resolving the bone marrow niche heterogeneity” (https://singlecell.broadinstitute.org/single_cell/study/SCP1248).

We thank Pinguang Yang and HaoTing Yen for technical assistance. Some figures were created using BioRender.com.

This work was supported by a Feinstein Institutional Fund (to Y.-R. Zou), National Institutes of Health grants (HL144436 and HL152099 to L. Blanc, R01 AI59742 to B. Sherry, P01 AI102852 to B. Sherry and B. Diamond), and in part by grants from the Canadian Institutes of Health Research (PJT-418606 and PJT-462997 to H. Gu).

Author contributions: T. Song designed and performed research, analyzed data, and contributed to writing; Y. Yao performed experiments; J. Papoin and L. Blanc helped with scRNA-seq experiments; B. Sherry and B. Diamond provided critical reagents; H. Gu interpreted data and edited the manuscript; L. Blanc contributed to writing; Y. Zou conceived and designed the study, supervised research, interpreted data, and wrote the manuscript.