Human haematopoietic stem cells remember inflammatory stress

Humans have enormous demand for haematopoietic output that is met by a complex cellular hierarchy with HSCs at its apex¹⁰. Mature blood cells with a finite lifespan are continuously replenished by bone marrow haematopoietic stem and progenitor cells (HSPCs) producing 3 million cells per second in human adults, via a tightly controlled process¹¹. Human HSCs are both genetically and functionally diverse with 50,000–200,000 HSCs contributing to haematopoiesis at any one time¹². With age, HSCs show functional decline^1,9, have a marked decrease in clonal diversity⁷, a consequent increase in incidence of clonal haematopoiesis (CH), and an increased risk for blood malignancy¹³. How daily blood production is coordinated with HSC pool maintenance over a lifespan is unclear, especially when exposed to inflammatory stressors, for example, recurrent infections, which can induce HSC activation^3,4,14,15. In addition, HSC fate transitions from quiescence towards activation are essential for HSC function, but can be dysregulated with ageing and inflammatory stress^{16,17,18,19,20}. Finally, HSCs are not homogeneous; HSCs exhibit transcriptional, epigenetic and functional heterogeneity, where inflammatory pathways are a major source of variation^{14,18,19,20,21,22}. Together, this indicates that heterogeneous cellular responses to inflammation within the human HSC pool might drive ageing phenotypes in response to life-long inflammatory insults^9,23,24.

Human lineage tracking shows that pre-leukaemic CH mutations arise decades before disease detection^25,26,27,28. CH mutation-bearing clones occur ubiquitously in adults older than 60 years of age, although only a fraction expand to reach detectable clone size¹³. Crucially, mechanisms regulating CH clone size are poorly understood. Mouse models have shown that inflammation activates HSCs, consequently impairing differentiation and self-renewal^{1,2,3,4,5,15,29,30}, affecting clonal selection in CH^6,29,31. Some inflammatory responses include activation of target genes downstream of tumour necrosis factor (TNF) via NF-κB, known to regulate HSC survival in both mouse and human settings^3,14. Repeated inflammatory challenges have also been associated with sustained epigenetic changes³⁰ and accelerated ageing of mouse HSCs^1,4, as well as expansion of clones bearing CH-associated mutations^5,6,29,31. However, given the heterogeneity of the HSC compartment, it is unclear whether all HSCs, especially in humans, respond equally to inflammation. If HSCs respond heterogeneously, it is also unclear whether this would affect clonal selection such as in CH.

Human HSC can be deeply quiescent or cell-cycle primed; states tied to variations in regeneration kinetics, 3D chromatin architecture and endolysosomal activity^14,19,20,21. We performed single-cell transcriptional profiling of cord blood (CB) Lin⁻CD34⁺CD38⁻CD45RA⁻CD90⁺CD49f⁺ long-term HSCs (LT-HSCs) and used non-negative matrix factorization (NMF) to explore transcriptional heterogeneity among 3,381 LT-HSCs at homeostasis (Fig. 1a). Programs of quiescence (Fig. 1b,c) and inflammatory signalling (Fig. 1d,e) were identified (Supplementary Tables 1 and 2). These were independently validated in single-nucleus gene expression and chromatin accessibility profiles (hereafter, scMultiome) of 15,590 CB HSCs and multipotent progenitors (HSC–MPPs) as identified using BoneMarrowMap³² (Extended Data Fig. 1a–h and Supplementary Table 3). The inflammatory program was associated with chromatin accessibility of AP-1 and NF-κB transcription factor-binding motifs (Extended Data Fig. 1i–k). To jointly evaluate the relevance of the distinct inflammation and quiescence programs identified by NMF within these datasets, ‘meta-programs’ were defined by consensus; these also showed stronger enrichment for inflammation-specific and quiescence-specific programs, respectively (Supplementary Table 4 and Supplementary Note 1). These data corroborate previous findings^19,20,21,22 of variable levels of transcriptional inflammatory priming within individual CB HSC-MPPs.

The alternative text for this image may have been generated using AI.

Full size image

a, CB LT-HSCs were subjected to scRNA-seq. Consensus NMF (cNMF) was performed on 3,381 cells. b–e, Projection of a cNMF transcriptional program of quiescence (NMF1; b) or inflammation (NMF2; d), as defined by GSEA enrichment (c,e, respectively). aHSC, activated HSC; FDR, false discovery rate; MyMegEry, myeloid-megakaryocyte-erythroid lineage potential; NES, normalized enrichment score; qHSC, quiescent HSC; UMAP, uniform manifold approximation and projection. f–j, NSG mice xenografted with CB-derived CD34⁺CD38⁻ cells were challenged with TNF or LPS at either 6 weeks (g,h; n = 14 for the PBS and n = 15 for the TNF and LPS groups across 3 independent pools) or 20 weeks (i,j; n = 10 for the PBS and n = 12 for the TNF and LPS groups across 2 independent pools) post-transplant. Total human engraftment (g,i) and progenitors (h,j) 16 h after inflammatory challenge or vehicle were measured. IF, injected femur. k, Schematic of the inflammation–recovery xenograft models. l,m, Human engraftment (l) and progenitor composition (m) 18 weeks after a single challenge (n = 15 mice per group across 3 independent CB pools). n,o, Human engraftment (n) and progenitor composition (o) 10 weeks after repeated challenge (n = 13 mice in PBS and LPS groups, and n = 15 in the TNF group across 3 independent CB pools). p, Schematic of the secondary transplantation with limiting dilution of human leukocytes from repeat-challenged xenografts. BM, bone marrow; FACS, fluorescence-activated cell sorting; −Ms, mouse depletion. q,r, Stem cell frequency (f-Stem) estimates for p. Dashed lines and shaded regions show 95% confidence intervals for each group. s, Human engraftment from p transplanted with 300,000 hCD45⁺ cells (n = 12 for PBS and TNF groups and n = 11 for the LPS group across 3 independent pools). Engraftment data are presented as boxplots: quartiles indicate a line at the median; the whiskers show the range; individual points are per mouse; and the symbols reflect independent CB pools. Engraftment data were compared with pairwise two-sided Mann–Whitney tests, whereas chi-squared tests were used for stem cell frequencies; P < 0.1 is shown numerically, whereas P > 0.1 is not shown.

Source data

To investigate how CB HSCs would functionally respond to inflammatory stress, we developed a xenotransplantation model of inflammation and focused on the impact of lipopolysaccharide (LPS), which mimics sepsis³³, and TNF, which has been associated with deleterious effects on human health in the context of advanced age, COVID-19 and sepsis^34,35,