Stress-induced changes in bone marrow stromal cell populations revealed through single-cell protein expression mapping

Summary:

Stromal cell populations that maintain hematopoietic stem and progenitor cells (HSPCs) are generally characterized in steady-state conditions. Here we report a comprehensive atlas of bone marrow stromal cell subpopulations under homeostatic and stress conditions using mass cytometry (CyTOF)-based single-cell protein analysis. We identified 28 subsets of nonhematopoietic cells during homeostasis, 14 of which expressed hematopoietic regulatory factors. Irradiation-based conditioning for HSPC transplantation led to the loss of most of these populations, including the LeptinR+ and Nestin+ subsets. In contrast, a subset expressing Ecto-5’-Nucleotidase (CD73) was retained and a specific CD73+NGFR(hi) population expresses high levels of cytokines during homeostasis and stress. Genetic ablation of CD73 compromised HSPC transplantation in an acute setting without long-term changes in bone marrow HSPCs. Thus, this protein-based expression mapping reveals distinct sets of stromal cells in the bone marrow and how they change in clinically-relevant stress settings to contribute to early stages of hematopoietic regeneration.

Graphical Abstract

Severe and colleagues used single cell mass cytometry to define 28 subsets of bone marrow stromal cells based on protein expression. Cytokine production and response to stress functionally selected niche candidates. While LeptinR+ and nestin+ cells were lost with conditioning, CD73+ subpopulations contribute to HSPC engraftment and acute hematopoietic recovery.

Introduction:

The relationship between bone and blood is ancient. All vertebrates since the divergence of fish that have bones and blood, make blood in their bones. This long co-evolution engendered complex interrelationships including the first proposed and first experimentally defined niche for stem cells in mammals (Calvi et al., 2003; Schofield, 1978; Zhang et al., 2003). Multiple bone marrow stromal cell types serve as regulators of hematopoiesis (Kfoury and Scadden, 2015) and dysfunction of some enable myelodysplasia and leukemia (Dong et al., 2016; Kode et al., 2014; Raaijmakers et al., 2010). Thus, the microenvironment serves the dual purpose of supporting but also preserving the integrity of hematopoiesis. However, these studies have mainly focused on a specific stromal cell type while comprehensive atlases of stromal subpopulations are just being generated. We sought to provide one focused particularly on protein-based analyses and in the setting of bone marrow transplantation to further understand marrow changes under stress and possible discovery of stroma-produced regulators of HSPC regeneration.

Bone marrow transplant is an extreme example of organismal stress in which otherwise lethal cytotoxic injury provides a potentially curative opportunity to replace malignant or damaged bone marrow cells with healthy ones. It currently requires genotoxic conditioning (radiation or cytotoxic chemotherapy) to eliminate the resident bone marrow HSPC rendering patients’ immune deficient before transplanted cells reconstitute hematopoiesis. Reconstituting host immune function is a matter of existential importance to patients and understanding how transplanted HSPC establish themselves in a damaged microenvironment is important for improving medical care.

Mass Cytometry (CyTOF) single-cell proteomic analysis is advantageous compared to transcriptome based approaches as there is significant disparity between mRNA expression and actual protein level (Frei et al., 2016; Vogel and Marcotte, 2012). CyTOF has previously been used to study hematopoietic cells (Bendall et al., 2011; Lavin et al., 2017; See et al., 2017) and we applied it to assess bone marrow stroma under homeostasis and after genotoxic injury from radiation, to roughly mimic clinical transplantation conditioning. This type of ‘stress selection’ differential analysis enabled identification of novel subsets of bone marrow stromal cells. We reasoned that a step-wise examination of cell numbers and cytokine production during homeostasis and post-irradiation, we could identify novel candidate modulators of hematopoietic engraftment and regeneration.

Using this strategy, we report: 1) 28 distinct clusters of bone marrow stromal cells under homeostatic conditions, 2) 14 of these subsets express cytokines relevant for hematopoiesis, 3) diverse subsets of stromal cells shown highly disparate sensitivity to irradiation insult, and among those LeptinR and Nestin expressing cells are lost whereas only 3 subsets are maintained and 4) one subset had significantly higher hematopoietic cytokine levels after irradiation.

The family of cells producing the enzyme Ecto-5’-nucleotidase (CD73), an enzyme that converts AMP to adenosine (Allard et al., 2017), distinctively persisted after irradiation and expressed hematopoietic niche factors enabling engraftment of hematopoietic stem and progenitor cells and blood regeneration. Furthermore, depletion of CD73 in stromal cells demonstrated that the enzyme itself contributes to hematopoietic reconstitution after the extreme stress of radiation injury.

Results:

Multi-Dimensional Single-Cell Mass Cytometry analysis reveals distinguishable clusters of bone marrow stromal cells

CyTOF permits multidimensional relative protein quantitation in single-cells (Bendall et al., 2011; See et al., 2017) beyond that possible with flow cytometry and we applied it to mouse bone marrow stromal cells using a customized CyTOF panel (32 antibodies) (Fig.1A, Star Methods Table). Because there are not readily available mouse osteocalcin antibodies that are sensitive and specific for osteoblastic cells, we used Osteocalcin reporter mice (OCN-GFP^topaz) (Bilic-Curcic et al., 2005), so we could use an anti-GFP antibody to detect Osteocalcin positive osteoblastic cells.

Figure 1: Mass Cytometry analysis reveals 28 subsets of bone marrow stromal cells.

(A) Bone stromal cells were enriched by digesting bones and flow sorting the CD45⁻ and Ter119⁻ negative fraction of cells. Mouse samples were individually barcoded, pooled together and stained with the CyTOF antibody panel. Single-cell data analysis clustered cells with a similar protein profile in specific populations. Clusters were then isolated, and functional distinctions evaluated. (B) tSNE plot of 191,610 single-stromal cells (n=20 mice). Cells are colored by families of clusters. (C) Force directed layout showing 23,060 single stromal cells clustered by X-Shift (n=20 mice) (D) Heat map representation of 28 subsets of stromal cells defined by 16 phenotypic markers (2 independent experiments, n=8 mice, homeostatic conditions). (E) Single-cells were FACS sorted from the different stromal populations and cultured in vitro for 10 days under hypoxic condition (2% O₂) to evaluate the percentage of adherent colonies (F) The percentage of stromal cells enzymatically isolated from the bone (crushed fraction) or the bone marrow plug (flushed fraction) was evaluated by staining both fractions separated and analyzed by CyTOF (n=7 mice).

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
CD45 Monoclonal Antibody (30-F11)	Thermo Fisher Scientific	Cat# 14–0451-82, RRID:AB_467251
Purified anti-mouse TER-119/Erythroid Cells (MaxPar® Ready) antibody	Biolegend	Cat# 116241, RRID:AB_2563789
Anti-Mouse CD31/PECAM-1 (390)-165Ho	Fluidigm	Cat# 3165013B RRID:AB_2801434
Purified anti-GFP antibody	Biolegend	Cat# 338002, RRID:AB_1279414
PDGF Receptor alpha antibody [APA5]	Abcam	Cat# ab90967, RRID:AB_2049372
Rat Anti-CD51 Monoclonal Antibody, Unconjugated, Clone RMV-7	BD Biosciences	Cat# 550024, RRID:AB_393537
Purified anti-mouse CD105 antibody	Biolegend	Cat# 120402, RRID:AB_961070
Mouse VCAM-1/CD106 MAb (Clone 112734) antibody	R and D Systems	Cat# MAB6432, RRID:AB_2214051
Anti-Mouse CD90.2 (30-H12)-156Gd	Fluidigm	Cat# 3156006B RRID:AB_2801433
CD73 antibody	BD Biosciences	Cat# 550738, RRID:AB_393857
Anti-Mouse CD117/c-kit (2B8)-166Er	Fluidigm	Cat# 3166004B RRID:AB_2801435
Purified anti-mouse CD200 (OX2) antibody	Biolegend	Cat# 123802, RRID:AB_1236498
NGFR / CD271 / TNR16 Antibody (clone 25–8)	Lifespan Biosciences	Cat# LS-C179536
Anti-Mouse Ly-6A/E (Sca-1) (D7)-164Dy	Fluidigm	Cat# 3164005B RRID:AB_2801436
Nestin antibody [2Q178]	abcam	Cat# ab6142, RRID:AB_305313
Mouse Leptin R Affinity Purified Polyclonal Ab antibody	R and D Systems	Cat# AF497, RRID:AB_2281270
RUNX2 (D1L7F) antibody	Cell Signaling	Cat# 12556, RRID:AB_2732805
Anti-Mouse Embigin Purified 50 ug antibody	Thermo Fisher Scientific	Cat# 14–5839-81, RRID:AB_2016582
LEAF™ Purified anti-mouse IL-3 antibody	Biolegend	Cat# 503906, RRID:AB_2280178
Anti-Mouse TNFa (MP6-XT22)-162Dy	Fluidigm	Cat# 3162002B RRID:AB_2801437
IL7 Antibody IHC-plus™	Lifespan Biosciences	Cat# LS-B14351
KITLG / SCF Antibody	Lifespan Biosciences	Cat# LS-C203250
SDF1 alpha antibody	Abcam	Cat# ab25117, RRID:AB_2088164
Goat Anti-Mouse Osteopontin Polyclonal antibody, Unconjugated	R and D Systems	Cat# AF808, RRID:AB_2194992
LEAF™ Purified anti-mouse CD254 (TRANCE, RANKL) antibody	Biolegend	Cat# 510008, RRID:AB_2287603
Angiopoietin 1 antibody	Abcam	Cat# ab95230, RRID:AB_10862531
G-CSF Monoclonal Antibody (9B4CSF), Functional Grade, eBioscience(TM)	Thermo Fisher Scientific	Cat# 16–7353-85, RRID:AB_11039531
LEAF™ Purified anti-mouse GM-CSF antibody	Biolegend	Cat# 505408, RRID:AB_315384
Anti-Ki-67 (B56)-168Er	Fluidigm	Cat# 3168007B RRID:AB_2800467
Rabbit Anti-Histone H2A.X, phospho (Ser139) Monoclonal Antibody, Unconjugated, Clone 20E3	Cell Signaling	Cat# 9718, RRID:AB_2118009
APC/Cyanine7 anti-mouse CD45 antibody	Biolegend	Cat# 103116, RRID:AB_312981
Anti-Mouse TER-119 Antibody, APC-Cy7	BD Biosciences	Cat# 560509, RRID:AB_1645230
CD31 (PECAM-1) antibody	BD Biosciences	Cat# 565097, RRID:AB_2739066
NGFR / CD271 / TNR16 Antibody (clone 25–8)	Lifespan Biosciences	Cat# LS-C179536
PE/Cy7 anti-mouse CD105 antibody	Biolegend	Cat# 120410, RRID:AB_1027700
Embigin Monoclonal Antibody (G7.43.1), PE	Thermo Fisher Scientific	Cat# 12–5839-82, RRID:AB_2016706
Rat Anti-CD73 Monoclonal Antibody, Phycoerythrin Conjugated, Clone TY/23	BD Biosciences	Cat# 550741, RRID:AB_393860
Alexa Fluor® 488 Rat Anti-Mouse CD73 Clone  TY/23	BD Biosciences	Cat# 561545, RRID:AB_10714516
Brilliant Violet 785™ anti-mouse CD90.2 antibody,	Biolegend	Cat# 105331, RRID:AB_2562900
Brilliant Violet 711™ anti-mouse CD117 (c-Kit) antibody	Biolegend	Cat# 105835, RRID:AB_2565956
Brilliant Violet 421™ anti-mouse Ly-6A/E (Sca-1) antibody	Biolegend	Cat# 108127, RRID:AB_10898327
Mouse Leptin R Biotinylated Affinity Purified PAb antibody	R and D Systems	Cat# BAF497, RRID:AB_2296953
CD140a (PDGFRA) Monoclonal Antibody (APA5), APC,	Thermo Fisher Scientific	Cat# 17–1401-81, RRID:AB_529482
Alexa Fluor® 488 anti-mouse CD106 antibody	Biolegend	Cat# 105710, RRID:AB_493427
Alexa Fluor® 700 anti-mouse CD45.1 antibody	Biolegend	Cat# 110724, RRID:AB_493733
Brilliant Violet 650™ anti-mouse CD45.2 antibody	Biolegend	Cat# 109836, RRID:AB_2563065
CD45R/B220 antibody	BD Biosciences	Cat# 553088, RRID:AB_394618
Rat Anti-CD19 Monoclonal Antibody, FITC Conjugated, Clone 1D3	BD Biosciences	Cat# 557398, RRID:AB_396681
PE/Cy7 anti-mouse/human CD11b antibody	Biolegend	Cat# 101216, RRID:AB_312799
Ly-6G (Gr-1) Monoclonal Antibody (RB6–8C5), PE-Cyanine7	Thermo Fisher Scientific	Cat# 25–5931-82, RRID:AB_469663
CD3 antibody	BD Biosciences	Cat# 563565, RRID:AB_2738278
CD4 Monoclonal Antibody (RM4–5), APC	Thermo Fisher Scientific	Cat# 17–0042-83, RRID:AB_469324
Brilliant Violet 570™ anti-mouse CD8a antibody	Biolegend	Cat# 100740, RRID:AB_2563055
Rat Anti-CD8a Monoclonal Antibody, Biotin Conjugated, Clone 53–6.7	BD Biosciences	Cat# 553029, RRID:AB_394567
CD3e biotin antibody	BD Biosciences	Cat# 553060, RRID:AB_394593
Rat Anti-CD4 Monoclonal Antibody, Biotin Conjugated, Clone GK1.5	BD Biosciences	Cat# 553728, RRID:AB_395012
Rat Anti-CD45R / B220 Monoclonal Antibody, Biotin Conjugated, Clone RA3–6B2	BD Biosciences	Cat# 553086, RRID:AB_394616
Rat Anti-TER-119 / Erythroid cells Monoclonal Antibody, Biotin Conjugated, Clone TER-119	BD Biosciences	Cat# 553672, RRID:AB_394985
Rat Anti-CD11b Monoclonal Antibody, Biotin Conjugated, Clone M1/70	BD Biosciences	Cat# 553309, RRID:AB_394773
Rat Anti-Ly-6G, Ly-6C Monoclonal Antibody, Biotin Conjugated, Clone RB6–8C5	BD Biosciences	Cat# 553125, RRID:AB_394641
Streptavidin BUV395	BD Biosciences	Cat# 564176
APC/Cyanine7 anti-mouse CD48 antibody	Biolegend	Cat# 103432, RRID:AB_2561463
PE/Cy7 anti-mouse CD150 (SLAM) antibody	Biolegend	Cat# 115914, RRID:AB_439797
BV605 CD16/CD32 antibody	BD Biosciences	Cat# 563006, RRID:AB_2737947
CD34 Monoclonal Antibody (RAM34), FITC	Thermo Fisher Scientific	Cat# 11–0341-85, RRID:AB_465022
PE anti-mouse CD117 (c-kit) antibody	Biolegend	Cat# 105808, RRID:AB_313217
Mouse Leptin R Affinity Purified Polyclonal Ab antibody	R and D Systems	Cat# AF497, RRID:AB_2281270
Purified anti-mouse Ly-6A/E (Sca-1) antibody	Biolegend	Cat# 122502, RRID:AB_756187
Rabbit Anti-Mouse Collagen Type I Polyclonal antibody, Unconjugated	CEDARLANE Laboratories Limited	Cat# CL50151AP, RRID:AB_10061240
Green Fluorescent Protein (GFP) Antibody	Aves Labs	Cat# GFP-1020, RRID:AB_10000240
Alexa Fluor® 647 anti-mouse CD144 (VE-cadherin) antibody	Biolegend	Cat# 138006, RRID:AB_10569114
Alexa Fluor(R) 647 anti-mouse CD31 antibody	Biolegend	Cat# 102516, RRID:AB_2161029
Ly-6A/E (Sca-1) Monoclonal Antibody (D7), APC, eBioscience(TM)	Thermo Fisher Scientific	Cat# 17–5981-83, RRID:AB_469488
Chemicals, Peptides, and Recombinant Proteins
Paraformaldehyde 20%	Electron Microscopy Sciences	Cat# 15713
Methanol	Sigma	Cat# A412–4
7AAD	BD Biosciences	Cat# 559925
ACK Lysing Buffer, Quality Biological	VWR	Cat# 10128–802
DPBS 1X	Corning	Cat# 21–031-CV
HBSS 1X	Gibco	Cat# 14025–092
MEM Alpha modification 1X	GE Healthcare life science	Cat# SH30265.01
FBS	Gibco	Cat# A31605–01
Pen Strep	Gibco	Cat# 15140–122
Methocult™ GF	Stem Cell Technology	Cat# 03434
Bovine Serum Albumin	Sigma	Cat# A9647–100G
Human TGF-beta1	Peprotech
Ascorbic Acid	Sigma	Cat# A4544
Beta-Glycerophosphate	SIgma	Cat# G9422
Rock inhibitor	Axon MedChem	Cat# Y-27632
Dexamethasone	Sigma	Cat# D4902
Insulin, human recombinant zinc	Gibco	Cat# 12585–014
Alcian Blue	Sigma	Cat# A5268
Alizarin Red	Sigma	Cat# 5533
Oil Red O	Sigma	Cat# O0625
OneComp eBeads	Thermo Fischer Scientific	Cat#2020–06-30
Count Bright™ Absolute counting	Thermofischer scientific	Cat# C36950
Critical Commercial Assays
APC Conjugation Kit	Abcam	Cat# Ab201807
Maxpar® X8 Multimetal Labeling Kit—40 Rxn	Fluidigm	Cat# 201300
Cell-ID™ 20-Plex Pd Barcoding Kit	Fluidigm	Cat# 201060
Cell-ID™ Intercalator-Ir	Fluidigm	Cat# 201192B
Cell-ID™ Intercalator-Rh	Fluidigm	Cat# 201103A
NucleoSpin® RNA XS RNA isolation kit	Macherey-Nagel	Cat# 740902.50
High capacity RNA-to-cDNA kit	Applied Biosystems	Cat# 4387406
RNeasy Mini Kit	Qiagen	Cat# 74104
RNA library preparation kit	NEB	Cat# E7490
Vybrant™ DiD Cell-Labeling Solution	Thermo Fisher Scientific	Cat# V22887
OsteoSense 750EX Fluorescent Imaging Agent	Perkin Elmer	Cat# NEV10053EX
Deposited Data
Gene Expression Omnibus (GSE131305)		RRID:SCR_005012
Mendeley data (http://dx.doi.org/10.17632/y6stzwzs39.1)		RRID:SCR_015671
Experimental Models: Organisms/Strains
B6.129(Cg)-Leprtm2(cre)Rck/J	The Jackson Laboratory	RRID:IMSR_JAX:008320
B6.129S1-Nt5etm1Lft/J	The Jackson Laboratory	RRID:IMSR_JAX:018986
C57BL/6J	The Jackson Laboratory	RRID:IMSR_JAX:000664
β-actinDsRED (Tg(CAG-DsRed*MST)1Nagy/J	The Jackson Laboratory	RRID:IMSR_JAX:005441
B6.129X1-Gt(ROSA)26Sortm1(EYFP)Cos/J	The Jackson Laboratory	RRID:IMSR_JAX:006148
Software and Algorithms
Cytobank	Cytobank Inc.	RRID:SCR_014043
JMP	SAS	RRID:SCR_014242
ImageJ	Schneider CA et al., 2012	RRID:SCR_003070
R Project for Statistical Computing	R Core Team	RRID:SCR_001905
FlowJo software	BD	RRID:SCR_008520
Java Treeview	Saldanha A, 2004	RRID:SCR_016916

We enriched stromal cells by FACS sorting CD45⁻ and Ter119⁻ cells (Fig.S1A) and by excluding contaminant hematopoietic cells using the Cytobank analysis software (Fig.S1B). t Distributed stochastic neighbor embedding (t-SNE) representation (Amir et al., 2013) or the more recent graphical tool Xshift (Samusik et al., 2016) of all bone marrow stromal cells showed distinct groups of cells (Fig.1B and Fig.1C) to which we applied k-means clustering to evaluate heterogeneity. This resulted in 6 families of clusters (Fig.1D): an endothelial population expressing CD31 (Blue Box); a mature osteoblast population expressing osteocalcin (GFP) (Green Box); a stromal progenitor population expressing CD105 or LeptinR (Red Box); a second stromal progenitor family expressing PDGFRα, CD90, Sca-1 (Black Box) and a third stromal progenitor subset expressing CD73 or Embigin (Orange Box). More subpopulations were evident by adding the markers NGFR, c-Kit, Nestin, Runx2, CD51, CD106 and CD200. Cubic clustering criteria defined 28 groups as a group number beyond which meaningful differences could not be discerned (Fig.S1C). To note, one cluster of cells (Grey Box) was negative for all markers in our panel. Cells expressing CD105 were the most abundant (Red clusters). Endothelial cells were also well represented (Blue clusters). In contrast, CD73⁺ and Embigin⁺ clusters (Orange clusters), CD90⁺/PDGFRα⁺/Sca1⁺ stromal cells (Black clusters) and Osteocalcin osteoblast subsets (Green clusters) were rarer (Fig.S1D). While eleven color flow cytometry analysis (Fig.S1E, Star Method Table) could not discriminate all clusters, we could use it to isolate some clusters. Single bone marrow stromal cells sorted from these groups were cultured under hypoxic condition (2% O₂) to assess their capacity to adhere to plastic, a common criterion for mesenchymal stromal cells (Fig.1E, S1F). The quantification showed distinctive percentages of adherent colonies, and as expected, the more mature subsets of cells (endothelial and osteoblastic cells) showed the lowest number of adherent colonies compared to mesenchymal stromal cell subsets except for LeptinR/CD106 stromal cells that did not form adherent colonies (Fig.1E). To confirm that the isolated cells were stromal and not hematopoietic, we plated cells expressing similar phenotypic markers but from the CD45+ compartment and could not detect adherent colonies in vitro (data not shown). However, this does not exclude the possibility that some of our clusters contain CD45⁻ hematopoietic cells as recently shown (Boulais et al., 2018).

Stromal cells were isolated by using different enzymatic protocols to digest the flushed bone marrow plug and the crushed bone fraction separately (Fig. 1F). Notably, clusters #1, #2 and #4 that express CD73 were enriched in the bone (crush) fraction. OCN-GFP⁺ cells associated with clusters #5 and #6 were detected in both the cortical bone from the crushed bone fraction and the trabecular bone from the bone marrow plug. Previously described LeptinR⁺ stromal cells, cluster #7 and cluster #22, were observed only in the bone marrow (flush) fraction indicating their presence in the central rather than periosteal region of the bone marrow. Nestin⁺ stromal cells were detected only in the bone fraction using our antibody-based approach. Furthermore, 60% of CD31⁺ and Sca1⁺ endothelial cells (cluster #28) were observed within the bone fraction, supporting the data from Kusumbe et al., that CD31^high arterioles are located close to bone-forming cells (Kusumbe, Nature 2014). In contrast CD31+ and CD106+ endothelial cells (cluster #27) were enriched in the bone marrow flush fraction. Therefore, CyTOF identifies 28 distinct stromal cell subsets simultaneously and these subsets are differentially represented in preparations of bone fractions and central bone marrow fractions.

Simultaneous hematopoietic niche factor profiling identifies bone stromal cell niche candidates

To define the potential hematopoietic relevance of specific stromal cell clusters, we complemented our antibody panel with 12 regulatory factors implicated in hematopoietic niche function (G-CSF, GM-CSF, Rank-L, TGF-β1, SDF-1, Kit-L, Angiopoietin-1, Osteopontin, IL-3, IL-6, IL-7 and TNF-α). Currently, protein analysis allows the study of cytokine expression only at the bulk level while CyTOF permits single-cell resolution. Therefore, we defined the cytokine profile of the 28 clusters (Fig.S2). Under homeostatic condition, the cytokine profile is heterogeneous with particular cytokines restricted to specific clusters of stromal cells. For instance, in the osteoblastic cluster #6, Osteopontin is significantly higher compared to all clusters except cluster #4 and cluster #23. The expression of G-CSF is statistically higher in cluster #4 defined by CD73 and NGFR markers. Cluster #10, defined by the progenitor markers CD51/PDGFRa/Sca1, produces significantly higher IL-6 compared to all clusters except clusters #2, #4 and #8. Similarly, cluster #8, defined by CD90/CD51/PDGFRα/Sca1 expresses significantly higher IL-7 compared to all clusters except clusters #4, #10, #23 and #26. Kit-L, SDF-1 and TGF-β1, three factors essential for HSPCs display a widespread expression pattern. Finally, the endothelial cell subsets (cluster #27 and #28) express lower levels of all the cytokines evaluated. The cytokine profile of all stromal cell clusters is summarized in the heat map (Fig.2A).

Figure 2: Bone marrow niche cytokine profiling defines 14 stromal clusters producing hematopoiesis regulators under homeostatic conditions.

(A) Heat map representation of the relative cytokine expression within the 28-distinct bone marrow stromal cell subsets. Each row represents the relative mean intensity of a cytokine per cluster (2 independent experiments, n=8 mice). (B) Cytokine index within the 28 clusters revealed candidate of stromal subsets susceptible to regulate hematopoiesis (Filled dots/circles represent cluster of stromal cells with cytokine index above/below the mean respectively). (C) We ranked the 14-top cluster of stromal cells based on the expression level of each cytokine.

Based on the assumption that the cells most likely to participate in hematopoietic control produce hematopoietic relevant cytokines, we calculated the median of total cytokine index among the 28 clusters of stromal cells (Fig.2B). This strategy selected 14 candidate clusters (filled dots) with the potential to regulate hematopoiesis. Some stromal subsets may of course express a single factor and still be physiologically relevant. Yet, the remaining 14 clusters of stromal cells (circles) were regarded as less likely to participate in hematopoietic regulation due to their relatively low production of hematopoietic niche factors.

We ranked the 14-top clusters of stromal cells based on the expression level of each cytokine (Fig.2C). Strikingly, cluster #4 was always ranked in the top four clusters except for the cytokines IL-3 (6^th rank) and TNFa (7^th rank). These data suggest that cytokine profiling may provide a selective basis for categorization of clusters of stromal cells with hematopoietic niche potential.

By this approach, LeptinR^High/CD106⁺ stromal cells (cluster #7) and Nestin^high stromal cells (cluster #12), two previously described important mesenchymal stromal cells (MSC) niche populations (Ding et al., 2012; Méndez-Ferrer et al., 2010), dropped out of the most relevant hematopoietic niche cell candidates. However, LeptinR^Low cells (cluster #22) and Nestin^Low cells (cluster#21) express more cytokines and were ranked as the 7^th and 12^th best producers of hematopoietic niche factors. Therefore, CyTOF allowed us to further define two distinct subsets of LeptinR and Nestin cells based on marker and cytokine expression which unveil different results than using the LeptinRcre and the Nestin-GFP mouse models.

Limited correspondence of LeptinR and Nestin production to reporter gene expression.

To validate our findings, we performed CyTOF analysis on LeptinRcre-R26-EYFP and Nestin-GFP mice (Fig.3). In the LeptinR mouse model, we found three subsets of LeptinR cells: one subset expressing the YFP reporter and positive for the LeptinR protein by antibody based analysis (average of 361 cells per 10,000 CD45-Ter119- cells), one subset only expressing the YFP reporter without any detectable LeptinR protein by CyTOF (average of 223 cells per 10,000 CD45-Ter119- cells) and another rarer subset of cells only positive for the LeptinR protein by antibody based analysis (average of 112 cells per 10,000 CD45-Ter119- cells) (Fig.3A–3D). These data demonstrate that ~40% of cells that score positive for LeptinR (and would be flow sorted for analysis) using the LepRcre-EYFP mouse have no detectable LeptinR protein (Fig.3B). Further, there are cells with detectable LeptinR protein that do not express YFP raising issues of whether regulation of the LeptinR locus may be perturbed in the engineered mouse (Fig.3A–3D). In all, only approximately half of cells identified and therefore analyzed using the LepRcre animal correspond with cells producing the LeptinR protein.

Figure 3: Characterization of the LeptinRcre-EYFP and Nestin-EGFP mouse models by Mass Cytometry.

Using Mass Cytometry, we analyzed bone marrow stromal cells isolated from the LeptinRcre-EYFP mouse models (n=2 mice) and the Nestin-EGFP mouse models (n=2 mice). (A) Venn diagram representation of the overlap between the LeptinR⁺ antibody cells and the YFP+ cells. (B) Quantification of the number of LeptinR⁺ antibody cells, YFP⁺ cells and the double positive LeptinR and YFP cells per 10,000 CD45⁻Ter119⁻ cells. (C) Median intensity of LeptinR antigen expression within the 3 LeptinR populations and CD73⁺NGFR^high stromal cells (cluster#4). (D) Median intensity of YFP antigen expression within the 3 leptinR populations and CD73⁺NGFR^high stromal cells (cluster#4). (E) Median intensity of the SDF1 antigen expression within the 3 leptinR populations and CD73⁺NGFR^high stromal cells (cluster#4) (*p<0.05). (F) Overlaps between the different populations of LeptinR and Nestin⁺ cells are shown using Venn diagram representation in the LeptinRcre-EYFP mouse model. (G) Venn diagram representation of the overlap between the Nestin⁺ antibody cells and the GFP⁺ cells. (H) Quantification of the number of Nestin+ antibody cells, GFP⁺ cells and the double positive Nestin and GFP cells per 10,000 CD45⁻Ter119⁻ cells. (I) Median intensity of Nestin antigen expression within the 3 Nestin population of cells and CD73+NGFR^high stromal cells (cluster#4). (J) Median intensity of the GFP antigen expression within the 3 Nestin populations and CD73⁺NGFR^high stromal cells (cluster#4). (K) Median intensity of the SDF1 antigen expression within the 3 Nestin populations and CD73⁺NGFR^high stromal cells (cluster#4) (*p<0.05). (L) Overlaps between the different populations of Nestin and LeptinR⁺ cells are shown using Venn diagram representation in the Nestin-EGFP mouse models. (M) Full-bone longitudinal section of Nestin-EGFP femur stained with GFP, LepR, Sca-1 and Collagen.1 antibodies (Stitch of individual images). (N) Nestin-expressing cells under the growth plate show no overlap with LepR+ except in rare cells (i, white arrowheads, 40 μm-thick projection). (O) Sca-1-expressing diaphyseal artery with Nestin-GFP expression in arteriolar endothelium (ii, white arrowheads), peri-arteriolar cells (ii, white arrows) and neuronal structures (ii, black arrows, 12.5 μm-tick projection). LepR+ peri-vascular cells are surrounding Nestin-GFP+ cells (ii, black arrowheads). (P) Most diaphyseal LepR-expressing cells lack detectable GFP expression (12.5μm-tick projection). Scale bars in M: 1000 m; N, O: 10 m; P: 20 μm.

Of note, CD73⁺NGFR^high stromal cells (cluster #4) do not express LeptinR or YFP markers (Fig.3C–D). The cytokine profiling revealed that LeptinR⁺ antibody cells or YFP⁺ cells express more SDF1 compared to the double positive LeptinR⁺/YFP⁺. Interestingly, the expression of SDF1 was still higher in the cluster #4 defined by CD73⁺ and NGFR^high (Fig.3E). Further, the overlap between LeptinR stromal cell subsets and Nestin⁺ antibody stromal cells by CyTOF is very small (~3%) in the LeptinRcre-EYFP mouse model (Fig.3F).

Analyzing Nestin-EGFP mice by CyTOF, we found a very limited overlap of GFP⁺ cells that had detectable Nestin protein by antibody based CyTOF analysis (average of only 4 cells per 10,000 CD45-Ter119- cells). These data confirm a previous report that the Nestin-GFP⁺ cells do not express the Nestin transcript by microarray or RNA sequencing data (Zhou et al., 2014). We did detect Nestin expressing cells (average of 41 cells per 10,000 CD45-Ter119- cells), but they were not the same as those labeled by the Nestin-GFP mouse (average of 32 cells per 10,000 CD45-Ter119- cells) (Fig.3G–J). CD73⁺NGFR^high stromal cells (cluster #4) express low level of the Nestin protein but not of the GFP protein (Fig.3I–J). We found that SDF1 expression was higher in the Nestin⁺ antibody cells compared to the GFP⁺ cells. The Nestin/GFP double positive cells express more SDF1 but their number is extremely low. Interestingly, SDF1 expression was higher in cluster #4 defined by CD73⁺ and NGFR^high compared to the Nestin or GFP cells (Fig.3K).

We could detect only minimum overlap between Nestin and LeptinR⁺ cells by CyTOF (Fig.3L). Because this finding contrasts with prior reports, we did multidimensional full bone imaging, but again only very rare LeptinR+ and Nestin-EGFP+ overlapping cells were observed (Fig 3M–O). Anti-GFP antibody was used in both settings to facilitate signal amplification. However, we were unable to detect GFP dim cells above background levels. Technical differences may account for the disparate results with prior studies, but it is also possible that the engineered mice do not correspond to endogenous gene expression.

The issue of whether LeptinRcre mice and Nestin-GFP mice accurately reflect LeptinR and Nestin protein expressing cell behaviors are raised by our data. We suggest that the results using the LeptinRcre or Nestin-GFP mice should be interpreted with caution when the goal is to evaluate the activities of stromal cells producing those proteins.

Differential sensitivity of bone marrow stromal cells to whole body irradiation as a means of stress selection

To gauge the likely functional relevance of particular bone marrow stromal cell subsets, we assessed their responses under a clinically relevant stress condition. Patients undergoing bone marrow transplantation are conditioned with myeloablative radiation or chemotherapy. In mice however, this is best modeled by a lethal dose of irradiation. Since the HSPC graft is infused one day after radiation, we assessed stromal populations at that time. It should be noted that irradiation and sample preparation may alter protein expression and our results should be viewed with that in mind. However, using multiplexed protein analysis does mitigate this issue.

Significantly smaller number of stromal cells were evident one day post-irradiation injury (median= 1719 +/− 1096 cells) compared to homeostatic conditions (median= 17811 +/− 8482 cells) (p=0.0002). However, differential cluster specific sensitivities were evident (Fig.4A and Fig.S3A). For example, the stromal cells (Red clusters) were markedly decreased one day post-radiation, including the Nestin^high and Nestin^lowcells (cluster #12 and #21), the LeptinR^high and LeptinR^low cells (cluster #7 and #22) and all CD31⁻CD105⁺ cells. Furthermore, the relative number of OCN-GFP⁺ cells are also significantly decreased one day post-radiation (Green clusters). Interestingly, within the 14 stromal cell populations selected as high cytokine producers, only 3 clusters (cluster #1, #2 and #4) preserved their cell number despite irradiation (Fig.4B and Fig.S3B). Notably, these radiation tolerant stromal cells can be defined by significantly higher expression levels of the Ecto-5’-Nucleotidase, CD73 (Fig.4C). We did not detect a significant decrease in their Ki67 expression, indicating no major change in proliferative activity one day post-irradiation, in contrast to CD105⁺ stromal cells (Fig.4D and Fig.S3C). Of note, p-γH2AX was significantly elevated in clusters #1, #2 and #4 one-hour post-irradiation, reverting to baseline within a day (Fig.4E and Fig.S3D), suggesting that these cells are more tolerant of radiation injury. The preservation of CD105⁻CD73⁺ stromal cells in irradiated mice was further confirmed by flow cytometry (Fig.4F–H). Notably, irradiation did not have an impact on CD73 expression on hematopoietic cells (data not shown). Thus, CD73 expressing stromal cells persist post-irradiation and are candidate mediators of HSPC engraftment and hematopoietic regeneration.

Figure 4: Heterogeneity in radiation sensitivity distinguishes persisting CD73⁺ bone marrow stromal cells.

(A) t-SNE representation of bone marrow stromal cells at homeostatic state (146,394 single-cells from 8 mice) and one day post-irradiation (15,036 single-cells from 8 mice). (B) Fold change number of cells within the 14 stromal niche candidates (2 independent experiments, n=8 mice). (C) Median intensity of CD73 antigen expression within the 14 stromal niche candidates. Cluster #1, #2 and #4 are defined by high level of CD73 expression compared to other stromal niche candidates (* p<0.05) (D) Mean intensity of Ki67 within the stromal cluster #1, #2 and #4 at homeostatic state and one day post-irradiation (n=8 mice per group) (E) Mean intensity of p-ƔH2AX within the stromal clusters #1, #2 and #4 at homeostatic state, one hour and one day post-irradiation (n=4 to 8 mice per group, *p<0.05). (F) Bone marrow stromal cells were isolated from control and irradiated mice (n=7 mice per group) and flow cytometry analysis was done to (G) quantify the CD105⁻CD73⁺ stromal cell percentage of live cells changes one day post-irradiation (n=7 mice, *p<0.05). (H) The absolute number of CD45⁻Ter119⁻CD31⁻CD105⁻CD73⁺ stromal cell was assessed at homeostasis and one day post-irradiation (n=3 mice).

Ecto-5’-nucleotidase in the bone marrow stromal niche affects hematopoietic stem and progenitor cell transplantation and acute peripheral blood regeneration

CD73 has been previously characterized as a marker for human and mouse MSCs (Breitbach et al., 2018; Dominici et al., 2006). Intraosseous injection of CD73⁺ stromal cells has been reported to enhance bone marrow transplantation though without clear evidence that CD73 itself played a role (Abbuehl et al., 2017). To further explore the function of CD73 and indirectly CD73⁺ stromal cells, we used a CD73-knockout mouse (CD73^−/−) (Thompson et al., 2004). At steady-state hematopoiesis, we did not detect a significant change in the peripheral blood cell numbers or BM cellularity, but found a modest decrease in MEPs and SLAMs in CD73 deficient mice (Fig S4A–D).

To explore the function of CD73 in the context of transplantation, wild-type labeled bone marrow LKS cells (Lin⁻c-Kit⁺Sca1⁺) were injected into lethally irradiated wild-type control and CD73^−/− recipient mice. 48 hours post-transplantation, we measured the number of labeled LKS (Red dots) in recipient mice by live animal imaging of the skull bone marrow (Fig.5A). We found a 2-fold decrease in the number of DiD⁺ LKS cells in the CD73^−/− mice demonstrating faulty homing (Fig.5B). However, CyTOF analysis comparing wild-type and CD73^−/− stromal cells showed no significant differences in SDF1 or KitL levels in any stromal cell families, indicating that the engraftment phenotype is unlikely to be due SDF1 and Kit-L (FigS5A–B). Lethally irradiated CD73^−/− and wild-type mice were transplanted with total bone marrow cells from CD45.1STEM donor mice (Fig.5C) (Mercier et al., 2016) and followed over time. This revealed decreased acute blood regeneration in CD73-deficient mice compared to control recipient mice post-transplantation (Fig.5D–E). By characterizing each blood compartment in more detail, we found that myeloid cells were significantly decreased (Fig.5F), B-lymphocytes were more modestly decreased in the peripheral blood of CD73^−/− mice (Fig.5G) whereas T-lymphocytes were unaffected (Fig.5H). Together, these data demonstrate that CD73 enzyme in the host environment modulates the reconstitution of myeloid cells in particular and transiently B cells in the acute setting of bone marrow transplantation. However, we did not detect significant changes in HSPC numbers after 16 weeks post transplantation (Fig S4I–R) indicating that CD73 alters hematopoiesis in the short-term post-injury but not long term. Whether the impact of CD73 on short-term stress responses are on cell localization versus cell production or clearance cannot be discerned from our data.

Figure 5: CD73 depletion in the bone marrow microenvironment decreases acute HSPC transplantation efficiency and blood regeneration.

(A) Labeled wild-type Lineage⁻cKit⁺Sca1⁺ (DiD⁺ LKS) cells were injected into lethally irradiated control and CD73^−/− mice. Two days post LKS transplantation, engrafted DiD⁺ LKS cells were imaged in the calvarium of recipient mice by intravital microscopy and (B) the number of DiD⁺ cells were enumerated (n=4 mice per group, 4 independent experiments, *** p<0.001). (C) Control and CD73^−/− recipient mice were lethally irradiated and transplanted with 1 × 10⁶ wild-type BM cells from CD45.1STEM donor mice and peripheral blood regeneration was analyzed by quantification of the absolute number per μl of blood of (D) CD45.1STEM donor cells; (E) Blood chimerism in transplanted mice for CD45.1STEM (F) Myeloid cells within the CD45.1STEM⁺ cells (G) B cells within the CD45.1STEM⁺ (H) T cells within the CD45.1STEM⁺ cells by flow cytometry at 2, 4, 8, 12 and 16 weeks post bone marrow transplantation (n=13 to 16 mice per group from 3 independent experiments (***p<0.001, **p<0.01, *p<0.05).

To further assess CD73 as a mediator of an acute stress response, a sublethal dose of irradiation without bone marrow transplantation was conducted. Only myeloid cells were diminished in the blood of CD73^−/− mice 4 weeks post radiation (Fig.S4E–H). Of note, B and T cell regeneration in the peripheral blood were unchanged. We also conducted RNAseq of donor LKS cells in the CD73^−/− mice at 8 weeks post-transplant and while the data are preliminary, they suggest that stromal CD73 deficiency reduces HSPC cell cycle gene expression and shifts metabolic processes (Fig S5C). The data are consistent with CD73 being both a marker of regulatory stromal cells and of being a regulator of HSPC engraftment and acute blood regeneration after genotoxic stress.

Cytokine profiling on resistant CD73⁺ stromal cells reveals a unique subset of cells defined by NGFR^high expression

To characterize the CD73⁺ cell populations further, cytokine production was assessed and revealed higher levels in Cluster #4 compared to Clusters #1 or #2 one day post-irradiation (Fig.6A and Fig.S6A). The cytokine index at homeostatic conditions as well as post-irradiation was the highest in cluster #4 (Fig.6B and Fig.S6B–C). Cluster #4 also produces high levels of the stromal marker NGFR, allowing us to isolate the cells by flow cytometry for further functional assessment (Fig.6C and Fig.S6D). Isolated CD105⁻NGFR^high, CD105⁻NGFR^low and CD105-NGFR^negative stromal cells (from the CD45-Ter119-CD31- cells) (Fig.S6E), demonstrated corresponding NGFR mRNA expression level (Fig.S6F) which validated our gating strategy. Moreover, by Mass Cytometry, we could identify that 66% (+/− 12%) of the NGFR^high stromal cells are CD73⁺ under homeostasis, the same conditions used for the assay in Figure 6G. We demonstrated by flow cytometry that the percentage of CD105⁻NGFR^high cells is significantly increased one day post-irradiation (Fig.6D and E). No significant change was detected in the absolute number of CD73⁺NGFR^high stromal cells indicating their persistence post-irradiation (Fig.6F). Notably, the NGFR^high cells do not appear to depend on CD73 for radiation tolerance as they neither deplete or alter cytokine levels with irradiation in the CD73^−/− mouse (FigS6G–S5A–B). To better assess the NGFR^high cells, we quantified them during development and found a significantly increased percentage of these cells in pups compared with adult mice (Fig.S6H). Moreover, isolated NGFR^high cells were able to differentiate into the three lineages as shown by Alcian blue, Oil red and Alizarin red staining (Fig.S6I). Since genetic tools do not yet exist to selectively alter the specific cell subset identified here, stromal cell function in hematopoiesis was evaluated by co-culturing stromal cell populations with LKS cells. CFU assays demonstrated significantly better support of hematopoietic colonies when co-cultured with CD105⁻ stromal cells that were NGFR^high compared to NGFR^Neg cells (Fig.6G). Furthermore, the proportion of these cells among CD73+ cells and their expression of cytokines does increase following irradiation (Fig.6E and Fig.S6C). These data indicate that multipotent NGFR^high mesenchymal stromal cells express hematopoietic niche factors and are capable of regulating hematopoiesis in vitro including improving acute hematopoietic regeneration after irradiation stress. To summarize, these data demonstrate that using Mass Cytometry with sequential steps of filtering based on cytokine profile and tolerance to a stress enables identification of a subset of stromal cells defined as CD105⁻CD73⁺NGFR^high and the Ecto-5’-nucleotidase (CD73) as regulators of HSPC engraftment and acute blood regeneration relevant for bone marrow transplantation.

Figure 6: Mass Cytometry defines a unique subset of CD105⁻CD73⁺NGFR^high cells with a specific cytokine signature, capable of regulating hematopoiesis.

(A) Heat-map representation of the cytokine level within radio-resistant cluster of cells #1, #2 and #4 (2 independent experiments, n=8 mice) (see Fig.S8A for all clusters). (B) Cytokine index post-irradiation within cluster #1, cluster #2 and cluster #4 revealed significant higher cytokine levels in the cluster #4 (** p<0.01, * p<0.05) (see Fig.S8B–C for all clusters). (C) Median intensity of NGFR within the 3 clusters of stromal cells (2 independent experiments, n=8 mice) (* p<0.05). (D) Gating strategy for CD105⁻NGFR^high stromal cells within the CD45-Ter119-Cd31- compartment of cells. (E) Bone marrow stromal cells percent change of CD105⁻NGFR^high cells one day post-irradiation (*p<0.05). (F) CD73⁺NGFR^high absolute cell number one day post-irradiation (in the CD45-Ter119-CD31-CD105- compartment). (G) Hematopoietic colony forming units (CFU) produced by LKS cells co-cultured with CD105⁻NGFR^{high or negative} stromal cells (*** p<0.0001).

Discussion:

The bone marrow is a complex tissue comprised of multiple stromal subset of cells to which multiplex approaches are beginning to define stromal cell heterogeneity. Recent work showed that multiplex imaging of the bone marrow can be used to define the microanatomic localization of various stromal cell populations (Coutu et al., 2017). These techniques identify subsets based on parameters of protein or RNA levels though the data are limited in defining those with physiologic importance. Adding the filters of either known function mediators such as cytokines and requiring changes in the context of stresses such as irradiation (stress selection), can provide a rational basis for winnowing subsets to focus on those most likely to regulate physiologic processes like regeneration. In the experiments here, we identified 28 bone marrow stromal subsets with multiple stromal cells having potential mesenchymal and niche activities. Only three stromal subsets persisted post irradiation and showed high hematopoietic niche factor expression profiles. These were defined by CD73 expression. In a recent study, Breitbach et al., have generated a CD73-EGFP mouse that labels multipotent stromal cells and endothelial cells (Breitbach et al., 2018). Using our antibody approach CD73⁺ stromal cells do not express the CD31 endothelial marker suggesting these cells are mesenchymal stromal cells. Also, they do not express CD45 or form hematopoietic colonies but we cannot exclude they may contain CD45- hematopoietic repopulating cells reported by Boulais et al. (Boulais et al., 2018).

Some stromal cell populations that have been previously reported to be relevant for hematopoiesis were lost following irradiation or expressed low levels of hematopoietic cytokines. Notable among these were the LeptinR⁺ and Nestin⁺ cells. These cells have mainly been characterized under homeostatic conditions; our focus was organismal stress and these populations do not appear likely to play a role in that setting.

Focusing on the LeptinR and Nestin positive cells, we found that there was generally poor correspondence of cells with detectable protein by CyTOF and those marked by LeptinRcre-EYFP or Nestin-EGFP. Only ~60% of LeptinRcre-EYFP and almost no cells marked by Nestin-EGFP had detectable protein that putatively defines them. There may be technical variables that contribute to these discrepancies, but detectable protein is arguably a powerful tool to demonstrate the bona fide identity of cells. We regard these data as a cautionary note in interpreting data from these genetically modified mice.

We also found little overlap of the LeptinR and Nestin expressing cells. Prior studies demonstrated overlap between the Nestin-EGFP dim cells and LepR-cre-Tdtomato+ cells in transgenic mice by flow cytometry and imaging (Kunisaki et al., 2013) (Zhou et al., 2014). We found only ~3% of cells with both LeptinR and Nestin. Since our data were different than prior work, high sensitivity whole bone marrow imaging was performed and confirmed the CyTOF results. We note that other studies also have similar results using either imaging (Coutu et al., 2017) or scRNA sequencing (Tikhonova et al., 2019) (Baryawno et al., 2019). While we cannot exclude technical differences resulting in different findings from the transgenic mice bearing fluorescent tags, the possibility that those models do not accurately portray endogenous gene expression should be considered.

Among the CD73 expressing cells, we identified a CD105⁻CD73⁺/NGFR^high population that maintained high levels of cytokines during homeostasis and at the time critical for HSPCs engraftment (one day post-radiation). These cells increased their expression of hematopoietic cytokines following irradiation stress. Interestingly, NGFR is a common marker for human MSCs (Mabuchi et al., 2013; Tormin et al., 2011) and point toward the contribution of MSCs in the hematopoietic niche. Since genetic tools do not yet exist to selectively alter cell subsets based on multiple variables, we required isolation and in vitro testing of the specific subset. It demonstrated that the CD105⁻CD73⁺NGFR^high stromal cells have an increased capacity to support hematopoiesis ex-vivo. Further, we could test the broad loss of CD73 in the microenvironment by transplanting wild-type hematopoietic cells into a CD73^−/− host. We also could determine the role of CD73 in non-transplant setting, however it is important to note that CD73 is also expressed on T cells (specially Tregs) and granulocytes (gene expression common) in this context. Our data indicate that CD73, an adenosine producing Ecto-5’-nucleotidase, expressed by stromal cells participates in early stage of hematopoietic (particularly myeloid cell) regeneration following genotoxic conditioning. Prior studies showed adenosine enhancing HSPC cell cycling in mice (Pospísil et al., 2001) and HSPC emergence in zebrafish (Jing et al., 2015). More recently, Tregs have been shown to also promote HSC engraftment and quiescence via the CD39 ectoenzyme and adenosine (Hirata et al., 2018). Together, these data are consistent with participation of CD73 in the bone marrow stromal niche in the acute setting following bone marrow transplantation. They do not indicate an effect of CD73 on long term hematopoiesis. It suggests that adenosine could regulate other mechanisms as cell migration or egress that could also support our phenotype.

Overall, protein based multiplexed single-cell analytics can add productive means of defining subsets of regulatory cells in complex tissue environments. Applying differentiating filters such as stress or specific product changes as performed here can reveal previously unrecognized functional cell subsets such as CD73⁺NGFR^high cells.

The approach also raises notes of caution about the dominant role proclaimed for some stromal subsets previously defined only through activation of Cre throughout development and points to how distinct populations may be relevant for stress versus steady-state hematopoietic regulation. Further, new regulatory molecules such as CD73 can be revealed by this approach and identified to have therapeutic potential in blood regeneration post-bone marrow transplantation. Based on our studies, we propose that the CD73⁺NGFR⁺ stromal cell subset serves as a niche component during extreme cytotoxic stress and that the CD73 Ecto-5’-nucleotidase may participate in early stage of hematopoietic recovery post-irradiation.

STAR Methods

Contact for reagent and resource sharing

Further information and request for resources and reagents should be directed to and will be fulfilled by the Lead Contact, David T. Scadden (david_scadden@harvard.edu).

Experimental model and subject details

Animals

12 weeks old OCN^GFP-topaz (C57BL6/Tg(BGLAP-Topaz)1Rowe/J) mice (Bilic-Curcic et al., 2005) were used in all CyTOF experiments. In addition, we used LeptinRcre mice (B6.129(Cg)-Lepr^tm2(cre)Rck/J) crossed with the Rosa26-Loxp-stop-loxp-EYFP mice (B6.129X1-Gt(ROSA)26Sor^tm1(EYFP)Cos/J) from the Jackson laboratory (Zhou et al., 2014), and the Nestin-EGFP mice (Mignone et al., 2004).

For flow cytometry validation and stromal cells isolation, C57BL6/J mice (#000664) or OCN^GFP-topaz mice were used as indicated. For the transplantation experiments, we used CD73 knock-out transgenic mice CD73^−/− (Thompson et al., 2004) (B6.129S1-Nt5e^tm1Lft/J, # 018986), compared to aged match control mice C57BL6/J mice (#000664) from The Jackson Laboratory. Bone marrow donor cells for the transplantation were isolated from CD45.1STEM mice established in our laboratory (Mercier et al., 2016). Male and female mice were used in all experiments.

The OCN^GFP-topaz and the Nestin-EGFP mice were genotyped by using the following primers: GFP-Forward primer: 5’-CTGGTCGAGCTGGACGGCGACGTAAC-3’; GFP-Reverse primer: 5’-ATTGATCGCGCTTCTCGTTGGGG-3’.

The CD73^−/− were genotyped using the following primers: Mutant forward: 5’-GCTACTTCCATTTGTCACGTCC-3’; Wild-type forward: 5’-GTTTTGATGCGTTCTGCAAG-3’; common: 5’-TACCGTTGGCTGACCTTTGT-3’ (The Jackson laboratory protocol).

LeptinRcre-EYFP mice were genotyped by using the following primers: Generic-Cre-Forward primer: 5’-GACCAGGTTCGTTCACTCATGG-3’; Generic-Cre-Reverse primer: 5’-AGGCTAAGTGCCTTCTCTCTACAC-3’; R26-YFP-WT: 5’-GGAGCGGGAGAAATGGATATG-3’; R26-YFP-Common: 5’-AAAGTCGCTCTGAGTTGTTAT-3’; R26-YFP-Mutant: 5’-AAGACCGCGAAGAGTTTGTC-3’.

All mice were bred and maintained in pathogen-free conditions and all procedures performed were approved by the Institutional Animal Care and Use Committee of Massachusetts General Hospital.

Bone marrow stromal cells isolation

Mice were euthanized via CO2 asphyxia, followed by cervical dislocation according to the approved IACUC protocol. Bones (femurs, tibias, pelvis and humerus) were harvested from mice, muscle and tendon tissue were removed using a scalpel and kimwipes. The bone marrow fraction was flushed out using a syringe containing PBS complemented with 2% Fetal Bovine Serum. The bone marrow fraction was digested by using Collagenase IV at 1mg/ml (Sigma Aldrich, C5138) and Dispase at 2mg/ml (Gibco by Life technology,17105–041) suspended in HBSS (Ca2+ and Mg2+) (Gibco by Life Technology, 14025–092) 3 × 15 minutes at 37C degree. The bone fraction was crushed and chopped into small pieces before to be digested by using Collagenase I 0.25% (Stem Cell Technologies, 07902), during 45 minutes at 37°C degree under agitation (120rpm). Both buffers were supplemented with DNAse (25 Units/ml) (Thermo scientific, 90083). Both fractions were combined after the digestion step except for the bone vs BM plug fractions analysis. Bone marrow stromal cells were first stained with fluorescently conjugated antibodies against CD45-PE (Biolegend, 103106), Ter119-APC (Biolegend, 116212) and the metal-conjugated antibodies against CD51 and CD31 for 30min at 4°C. Cells were washed and stained with a fixable viability dye 450 (BD Bioscience, 562247) as well as the Rhintercalator (Fluidigm Sciences) for discriminating dead cells by flow cytometry and mass cytometry respectively. At the end, cells are fixed in PFA 4% for 10min at room temperature. Bone marrow stromal cells were filtered using cell strainers and enriched by FAC sorting using a FACS Aria II sorter (BD Bioscience) by excluding dead cells and blood cells CD45⁺ and Ter119⁺.

Bone marrow transplantation

One day prior transplantation, recipient mice were equally subjected to whole body irradiation (1 dose of 9.5Gy) from a ¹³⁷Cs source. Bone marrow cells from CD45.1STEM mice were isolated from bones, stained with Acridine Orange and DAPI solution (Chemometec, 910–3013) and counted using a Cellometer (Nexcelom). Then, recipient mice were administered by retro-orbital injection of 1 × 10⁶ CD45.1STEM donor bone marrow cells.

For non-lethal myeloablation (without transplantation), mice were equally subjected to whole body irradiation (1 dose of 4.75Gy) from a ¹³⁷Cs source.

LKS (Lin⁻cKit⁺Sca1⁺) isolation and transplantation

CD45.1STEM mice were used as donor mice. Long bones, pelvis and spines were harvested and muscle tissue was removed. Bones were crushed in PBS complemented with 2% FBS and bone marrow cells in suspension were filtered on a 70μm cell strainer. Bone marrow cells were stained by a biotinylated anti-mouse lineage cocktail of antibodies (B220, CD3, CD4, CD8, Mac1, Gr1, Ter119), Sca-1 BV421 and c-Kit BV711 for 45 minutes at 4 degrees. Red blood cells were lysed using the ACK lysis buffer (VWR, 118-156-721) for 5 minutes on ice, cells were washed and then stained with the Streptavidin BUV395 for 15 min at 4 degrees. Cells were washed and suspended in PBS 2% FBS for FACS sorting on the FACS Aria II sorter. 7AAD live/dead dye (BD Pharmingen, 559925) was added just prior sorting to the cells. 3 × 10⁴ LKS were injected into lethally irradiated control and CD73^−/− mice with 0.2 × 10⁶ supporting unlabeled wild-type BM cells.

The same protocol was applied to sort LKS cells from β-actinDsRED (Tg(CAGDsRed*MST)1Nagy/J, The Jackson laboratory #005441) mice for the co-culture assay.

Live in-vivo imaging

LKS cells were isolated from CD45.1STEM donor mice as described in the previous section. Freshly isolated LSK cells were labeled with a Vibrant™ DiD cell-labeling solution (1,1′-dioctadecyl-3,3,3′,3′-tetramethylindodicarbocyanine perchlorate) (Thermo Fischer Scientific, V22887) for 15 min at 37degree in the dark according to manufacturer instructions. 3 × 10⁴ LKS cells and 0.2 × 10⁶ supportive bone marrow cells were transplanted to lethally irradiated (9.5Gy) recipient mice via retro-orbital injection. 48hours after LKS cells injection, mice were anesthetized by isoflurane inhalation with 2% O_2. To visualize the bone structure, mice were previously administered intravenously with Osteosense® 750EX (4nmol/mouse, Perkin Elmer). To visualize the vasculature, mice were injected intravenously with anti-mouse CD31-AF647 (5μg/mouse), anti-mouse VE-Cadherin-AF647 (5μg/mouse) and anti-mouse Sca1-APC (2μg/mouse) antibodies (Key resource table) in 100ul of sterile PBS. The calvarium of the mice was imaged using an Olympus IV100 microscope (4X and 20X objective, Z-Stack of 2um). All images were analyzed and stitched using ImageJ software and Did+ cells were enumerated.

Co-culture assay

We co-cultured freshly isolated 1.5 × 10³ stromal cells from pups and 1.5 × 10³ LKS cells from adult donor mice as indicated above. Cells were co-cultured for 4 days under hypoxic condition 2% O₂ in a 96 well plate format, in 100ml of α-MEM (GE Healthcare Life Sciences, SH30265.01) complemented with 10% FBS (Gibco by Life Technologies, A31605–01), 1% Penicillin/Streptomycin (Gibco by Life Technologies, 15140–122) with no supplemental cytokines. The stromal and LKS cells were then plated in methylcellulose with cytokines (Stem Cell Technology, Methocult™ GF M3434) for 14 days and the hematopoietic colonies forming units (CFU) were quantified by two independent blinded evaluators.

Method details

Mass Cytometry Antibodies

Metal-conjugated antibodies used in this study are summarized in the key resource table. Except commercially available pre-conjugated antibodies (Fluidigm Sciences), all antibodies were conjugated to isotopically enriched lanthanide metals using the MaxPAR antibody conjugation kit (Fluidigm Sciences), according to the manufacturer’s recommended protocol. Labeled antibodies were stored at 4°C in PBS supplemented with glycerol, 0.05% BSA and 0.05% sodium azide. All antibodies were tested with control beads as well as positive and negative control cells. Because there are not readily available mouse osteocalcin antibodies that are sensitive and specific for osteoblastic cells, we used Osteocalcin reporter mice (OCN-GFP^topaz), so we could use an anti-GFP antibody to detect Osteocalcin positive osteoblastic cells.

Barcoding and cell preservation

Each cell in suspension was barcoded using a unique combination of palladium isotopes, allowing us to combine the samples in a single tube before the antibody staining step. To do so, sorted cells were washed twice with permeabilization buffer containing saponin (Fluidigm Sciences), and resuspended in 100 μl permeabilization buffer. 900 μl freshly prepared barcoding solution (Fluidigm Sciences) was added to each sample and incubated in room temperature for 30 minutes. Reaction was quenched with addition of 1mL 0.5% BSA in PBS and samples were washed twice using PBS. Barcoded samples from different mice were then mixed together and centrifuged. 100% −20C Methanol was added dropwise until sample methanol concentration was 90% and stored in −80C.

Flow cytometry analysis

For flow cytometry analysis, bone marrow stromal cells were isolated as describe above. However, bone marrow stromal cells were stained with a cocktail of 10 antibodies (Key resource table) in PBS 2% FBS for 30 minutes at 4 degrees. RBC cell lysis was done after the staining using ACK lysis buffer for 5 minutes on ice. Then, cells were washed and suspended in PBS 2% FBS with 7AAD (1/100 dilution) before being analyzed on a FACS Aria II (BD Biosciences). For compensation, all antibodies were conjugated to OneComp eBeads (Thermo Fischer Scientific, #2020-06-30). Unstained and GFP⁺ control samples were used for compensation and control the gating strategy. Fluorescent Minus One (FMO) samples were used for gating strategy controls. Flow cytometry data were analyzed by FlowJo software.

To count the absolute number of stromal cells, we used Count Bright™ Absolute counting beads for flow cytometry (Thermofischer scientific, #C36950). Briefly, 10,000 beads were added per sample and analyzed by flow cytometry. The absolute number of cells in the sample can be calculated by comparing the ratio of bead events to cell events.

NGFR-APC conjugation

We could not find an anti-mouse NGFR antibody already conjugated with a fluorochrome so we conjugated in the laboratory the NGFR purified antibody (Lifespan Biosciences, LS-C179536) using the APC conjugation kit (Abcam, ab201807) according to the manufacturer instructions.

Peripheral blood analysis

To assess the regeneration of the blood compartment, the peripheral blood of recipient mice was collected using heparinized capillaries and stored into EDTA containing tubes at 2 weeks, 4 weeks, 8 weeks, 12 weeks and 16 weeks after bone marrow cells transplantation. Complete Blood Count was performed using the Element Ht5 Auto Hematology analyzer. Then, Red Blood Cells were lysed using ACK lysis buffer for 5 minutes on ice and blood cells were stained in PBS 2% FBS using the following antibodies: CD45.1-AF700, CD45.2-BV650, CD3e-BUV395, CD4-APC, CD8-BV570, Mac1-PeCy7, Gr1-PeCy7, B220-FITC, CD19-FITC (Key resource tabel). In addition, 7AAD dye was used to analyze viable cells only. We analyzed the CD45.1STEM⁺ donor cells and the descendant mature blood cell comparing wild-type and CD73^−/− recipient mice to assess the function of CD73 in bone marrow stromal cells in bone marrow transplantation regulation. Similarly, the peripheral blood of mice was collected and analyzed at 2 weeks and 4 weeks post sublethal irradiation.

Immunostaining and confocal imaging of bone marrow femoral sections

Femoral sections were prepared, immunostained and imaged as previously described (Coutu et al., 2018). Bones were fixed in 4% paraformaldehyde overnight, and decalcified using 10% ethylenediaminetetraacetic acid (EDTA, pH=8) for two weeks. Longitudinal bone sections were stained overnight at RT with primary [anti-LepR (R&D systems, AF497), anti-Sca-1 (Biolegend, 122502), anti-collagen type I (Cedarlene, CL50151AP) and anti-GFP (Aves, GFP-1020)], secondary antibodies (CF488, CF633 and CF680, Biotium) and DAPI. Detection of LepR was performed by biotin-streptavidin CF555 amplification. Finally, sections were optically cleared using graded series of 2,2-thiodiethanol (TDE, Sigma). Three-dimensional full-bone imaging was performed using type F immersion liquid (RI: 1.518) and 20X multiple immersion objective (NA 0.75, FWD 0.680 mm) on Leica TCS SP8 confocal microscope equipped with two photomultiplier tubes, three HyD detectors and three laser lines (405-nm blue diode, argon and white light). 8-bit images were acquired at 400 Hz and 1024 × 1024 resolution (2.49 μm z-spacing) and stitched together.

Reverse transcription Quantitative real-time polymerase chain reaction (RT-QPCR)

To validate our flow cytometry gating strategy, we performed RT-QPCR analysis to assess the mRNA expression level of NGFR comparing the CD105⁻NGFR^high, NGFR^low and NGFR^neg. To do so, 2.0 × 10³ stromal cells form each group were sorted in lysis buffer and the RNA was extracted using the NucleoSpin® RNA XS RNA isolation kit (Macherey-Nagel). Total RNA was then converted to cDNA using the High capacity RNA-to-cDNA kit (Applied Biosystems, 4387406). Then, QPCR analysis was realized using Taqman probes (Thermo Fischer Scientific) specific for NGFR (Mm00446296_m1) and the housekeeping genes HPRT (Mm01545399_m1) and β-Actin (Mm00607939_s1). The relative amount of RNA was calculated by the 2^−ΔΔCt method.

RNA-sequencing analysis

Libraries were prepared with an RNA library preparation kit (E7490, NEB) using RNA obtained by the RNeasy Mini Kit (Qiagen). RNA-seq libraries were sequenced with a 1X50bp strand-specific protocol on Illumina HiSeq 2500. Data were analyzed using a high-throughput next-generation sequencing analysis pipeline: FASTQ files were aligned to the mouse genome (mm10, NCBI Build 38) using STAR. Gene expression profile for the individual samples was calculated as RPKM values. DAVID 6.7 (Huang et al., 2009) was used for analysis of differentially expressed genes obtained from the expression profile data. Gene ontology options GOTERM_BM_ALL (p-value<10⁻⁷) and GOTERM_KEGG PATHWAY were selected, and a functional annotation heat map was generated using JavaTreeview. The mean of WT samples were compared to the KO sample. The raw data are available on GEO (GSE131305).

Chondrocyte, adipocyte and osteoblast differentiation

CD105⁻NGFR^high stromal cells were cultured under hypoxic condition 2% O₂ in a 96 well plate format, in 100ml of α-MEM (GE Healthcare Life Sciences, SH30265.01) complemented with 10% FBS (Gibco by Life Technologies, A31605–01), 1% Penicillin/Streptomycin (Gibco by Life Technologies, 15140–122). Cells were passaged up to three times before to use them for differentiation assays. For the chondrogenic differentiation assay, 1.25 × 10⁵ NGFR^high stromal cells were plated in 10μl of culture medium in a 24 well plate. Cells were allowed to attach to the plastic dish for 2 hours, then 0.5 ml of differentiation medium was added to the cells (α-MEM+10% FBS, 1%P/S+ 10ng/ml hTGF- β1 (Peprotech)+ 50μg/ml ascorbic acid (Sigma Aldrich) + 20μM ROCK inhibitor (Stem cell technologies). The differentiation medium was changes every 3 days and at day 9, cells were fix in PFA 4% for 30 minutes, then wash with 0.1HCl for 5 minutes and stain for 48 hours with 1% Alcian Blue 8GS in 0.1N HCl (pH1.3). For adipocyte differentiation, 1 × 10⁵ NGFR^high stromal cells were plated in a 6 well plate in differentiation medium (α-MEM+10% FBS, 1%P/S+ 100nM dexamethasone (Sigma Aldrich) and 10mg/ml human insulin). The differentiation medium was changed twice a week, and at day 14 to 21, cells were fixed in PFA4% for 10 minutes, wash with 60% isopropanol, and stain with Oil Red O for 10 minutes and immediately wash with distillated water.

For osteoblast differentiation, 3 × 10⁴ NGFR^high stromal cells were plated per cm² in a 6 well plate in differentiation medium (α-MEM+10% FBS, 1%P/S+ 10mM β-Glycerophosphate (Sigma Aldrich) + 50μg/ml Ascorbic Acid (Sigma Aldrich). The differentiation medium was changed twice a week, and at day 21, cells were fixed in cold 100% methanol for 20 minutes at −20 degrees. Then cells were rehydrated for 5 minutes in water at room temperature and stain with Alizarin Red solution (1mg/ml, pH=6.2) for 30 minutes, immediately wash cells with distillated water 3 × 15 minutes.

Quantification and statistical analysis

Mass cytometry analysis and data processing

Fcs files were normalized using internal control four element beads, concatenated and debarcoded using software from Nolan lab and Cytobank (Amir et al., 2013). Debarcoding used cutoffs of 30 for both Mahalanobis distance and separation. The resulting fcs files were uploaded to Cytobank, and stromal cells were manually selected based on negativity for CD45 and Ter119 hematopoietic markers using non-sorted bone marrow samples as control for CD45+ and Ter119+ cells. viSNE analysis was performed with the following parameters: %100 of events, 5000 iterations, 100 perplexities and 0.5 theta. Data was then exported to JMP (JMP® Pro Version 13, SAS Institute Inc., Cary, NC, 1989–2007), PRISM (GraphPad Prism version 6.00, La Jolla California USA) and R software (R Core Team (2017), R Foundation for Statistical Computing, Vienna, Austria. URL: https://www.R-project.org/.) where further analyses were performed and illustrations were prepared. We performed clustering of the single-cell data into 28 subsets (see Fig.S2 for determination of k=28 using cubic clustering criterion) and ordered generated clusters based upon cluster distances (using squared Euclidean distances of normalized cluster means) (Fig.1.C). We obtained identical optimum number of clusters using X-shift algorithm (Vortex clustering environment) using elbow method (Samusik, 2016).

Cytokine index (TCi) for each cell was calculated by:

$$TCi=\sum \frac{intensity(cytokine)}{\text{ max }intensity(cytokine)}$$

Median values of TCi (mTCi) in individual clusters were evaluated in Fig.2.B. We selected clusters of stromal cells above mean mTCi.

Statistical methods

For comparing of stromal cell numbers obtained from mass cytometry before or after radiation injury we used two-tailed Mann-Whitney test. For comparing relative cell number changes within individual clusters, we used Multiple t-tests with false discovery rate Q=0.05. For comparing marker antigen levels across clusters under homeostatic conditions we performed Kruskall–Wallis test with Dunn’s correction for multiple hypothesis testing, P values <0.05. The comparison of cytokine index between cluster #1, #2 and #4 (n=8 mice per group) was evaluated by Kruskall–Wallis test with Dunn’s correction for multiple hypothesis testing P values <0.05. The comparison of stromal cell population expression by flow cytometry under homeostatic condition and post-irradiation was evaluated by unpaired t-test p<0.05. The LKS engraftment differences were evaluated by unpaired t-test p<0.05. The significant differences in peripheral blood reconstitution was calculated by Multiple T-test with Bonferroni correction. The significant changes in hematopoietic Colony forming Units was assessed by using unpaired t-test P values <0.05. The comparisons of cytokine indices of cells from individual clusters 1 day after irradiation and under homeostatic conditions was performed by Mann-Whitney test p<0.01, corrected by Bonferroni method (Fig.S7C).

Data and software availability

Mass Cytometry data is deposited in the Mendeley database under DOI: http://dx.doi.org/10.17632/y6stzwzs39.1

RNA sequencing data is deposited on Gene Expression Common under accession number GSE131305

Highlights:

• Single cell mass cytometry defines 28 subsets of bone marrow stromal cells (BMSCs)

• Cytokine profiling and persistence to radiation reveal BMSCs niche candidates

• LeptinR+ and Nestin+ putative niche cells are lost with radiation conditioning

• CD73+ BMSCs contribute to HSPC engraftment and acute hematopoietic recovery