Unique molecular and functional features of extramedullary hematopoietic stem and progenitor cell reservoirs in humans

Abstract

Rare hematopoietic stem and progenitor cell (HSPC) pools outside the bone marrow (BM) contribute to blood production in stress and disease but remain ill-defined. Although non-mobilized peripheral blood (PB) is routinely sampled for clinical management, the diagnosis and monitoring potential of PB HSPCs remains untapped, as no healthy PB HSPC baseline has been reported. Here we comprehensively delineate human extramedullary HSPC compartments comparing spleen, PB and mobilized PB (mPB) to BM using single-cell RNA-seq and/or functional assays. We uncover HSPC features shared by extramedullary tissues and others unique to PB. First, in contrast to actively dividing BM HSPCs, we find no evidence of substantial ongoing hematopoiesis in extramedullary tissues at steady state, but report increased splenic HSPC proliferative output during stress erythropoiesis. Second, extramedullary stem cells/multipotent progenitors (HSC/MPPs) from spleen, PB and mPB share a common transcriptional signature and increased abundance of lineage-primed subsets compared to BM. Third, healthy PB HSPCs display a unique bias towards erythroid-megakaryocytic differentiation. At HSC/MPP level, this is functionally imparted by a subset of phenotypic CD71+ HSC/MPPs, exclusively producing erythrocytes and megakaryocytes, highly abundant in PB but rare in other adult tissues. Finally, the unique erythroid-megakaryocytic-skewing of PB is perturbed with age, in essential thrombocythemia and in beta-thalassemia. Collectively, we identify extramedullary lineage-primed HSPC reservoirs that are non-proliferative in situ and report involvement of splenic HSPCs during demand-adapted hematopoiesis. Our data also establish aberrant composition and function of circulating HSPCs as potential clinical indicators of BM dysfunction.

Postnatal conservation of human blood- and marrow-specific CD34+ hematopoietic phenotypes

“…Given the known adhesion properties of these two components of the cell membrane 33,34 , their higher surface expression on BM versus PB CD49f+ cells might be conjectured to reflect a role in interacting with corresponding receptor-niche elements responsible for the retention of these cells within the BM. This is supported by a recent study reporting altered gene sets associated with cytoskeletal organization, migration, and adhesion between HSCs/MPPs from the BM compared to those isolated from extramedullary sites 35 . The relatively low frequency and somewhat unique phenotype of the CD49f+ cells detected in many of the aged adult Hip BM samples also suggest deviations from normalcy of these samples and/or and their ability to serve as an agerepresentative source of functionally defined HSCs.…”

ClustAssess: tools for assessing the robustness of single-cell clustering

“…A first case study for illustrating the analysis-induced variability is based on the Mende et al data (10x scRNA-seq in-vivo dataset of human hematopoietic stem and progenitor cells originating from bone marrow, peripheral blood, and spleen [27]). As initial QC, the distributions of sequencing depths, number of detected features, proportions of MT and RP genes were assessed (Supplementary Table S1).…”

“…For hematopoietic stem and progenitor cells, characterising differentiation trajectories is of particular interest [27]. Slingshot infers a trajectory on scRNA-seq data using cluster centroids [39].…”

“…Slingshot v2.1.0 [39] was used to perform trajectory inference on the Mende et al dataset [27]. Seurat and Monocle clusterings were used as input with the isolated island corresponding to Seurat cluster 15 removed; Slingshot was applied (with parameters ) to infer cellular trajectories.…”

“…We focus on the evaluation of dataset-specific clustering results, in contrast to the wide-ranging and generic results provided by clustering benchmarks. We illustrate these tools on various partitions driven by choices of features, methods, on both 10x and SMART-seq scRNA-seq datasets [27, 28], as well as a multimodal single-cell RNA and ATAC dataset. We highlight several sources of variation that affect final clustering results, including method parameters, input features and approximation errors.…”