Characterization of naked mole‐rat hematopoiesis reveals unique stem and progenitor cell patterns and neotenic traits

Abstract

Naked mole rats (NMRs) are the longest‐lived rodents yet their stem cell characteristics remain enigmatic. Here, we comprehensively mapped the NMR hematopoietic landscape and identified unique features likely contributing to longevity. Adult NMRs form red blood cells in spleen and marrow, which comprise a myeloid bias toward granulopoiesis together with decreased B‐lymphopoiesis. Remarkably, youthful blood and marrow single‐cell transcriptomes and cell compositions are largely maintained until at least middle age. Similar to primates, the primitive stem and progenitor cell (HSPC) compartment is marked by CD34 and THY1. Stem cell polarity is seen for Tubulin but not CDC42, and is not lost until 12 years of age. HSPC respiration rates are as low as in purified human stem cells, in concert with a strong expression signature for fatty acid metabolism. The pool of quiescent stem cells is higher than in mice, and the cell cycle of hematopoietic cells is prolonged. By characterizing the NMR hematopoietic landscape, we identified resilience phenotypes such as an increased quiescent HSPC compartment, absence of age‐related decline, and neotenic traits likely geared toward longevity.

In‐depth profiling of the naked mole‐rat hematopoietic system by surface marker analysis and single‐cell sequencing uncovers resilience phenotypes and unexpected similarities with humans.

Introduction

The naked mole rat (NMR) (Heterocephalus glaber; hgl) emerged as an animal model of exceptional longevity and resistance to age‐associated diseases (Gorbunova et al, 2014). At the size of a mouse, these rodents reach a lifespan of over 30 years in captivity and do not display increased mortality with aging (Ruby et al, 2018). Despite no demographic aging, there is evidence of epigenetic aging in NMRs (Horvath et al, 2022; Kerepesi et al, 2022). Characteristic molecular traits include but are not limited to higher translation fidelity due to split processing of 28S rRNA (Azpurua et al, 2013), a unique splicing product from the senescence‐inducing INK4/ARF locus and abundant high molecular weight hyaluronic acid (HMW‐HA) responsible for the resistance of NMRs to solid tumors (Tian et al, 2013, 2015).

The blood is the most regenerative tissue, producing > 10¹⁴ cells per year in humans (Dancey et al, 1976). Fostered by recent advances in single‐cell transcriptomics (Paul et al, 2015), hematopoiesis is viewed as a continuum of individual cells that traverse the differentiation process from unprimed hematopoietic stem cells (HSCs) directly into unipotent progenitors (Velten et al, 2017). Studies of the unperturbed hematopoietic system are most advanced in their understanding of HSPC hierarchies and concepts of stemness in mice (Laurenti & Gottgens, 2018). There are, however, fundamental differences in certain aspects of the blood system between mice and humans (Copley & Eaves, 2013). At the genetic level, orthologs for one of the major murine HSC markers, Sca‐1 (Ly6a), are found only in rodents but not in primates, carnivores, birds, or fish. Interestingly, NMRs are among the few rodents without a Sca‐1 ortholog (Table S1).

Here, we developed a flow cytometry (FACS) labeling strategy using cross‐reactive antibodies to sort, culture, and transplant NMR hematopoietic stem and progenitor cells (HSPCs). A panel of six surface markers enabled purification of primitive stem cells with multi‐lineage potential, distinct cell stages during early erythroid, and T‐lymphoid commitment (Emmrich et al, 2021) and distinguished the major blood leukocyte fractions. NMR HSPCs showed striking similarities to human HSPCs, such as a CD34⁺ compartment harboring primitive progenitors, marrow granulopoiesis, and slow cell metabolism. Further adaptations found in NMRs include a prolonged cell cycle duration, splenic erythropoiesis, and retention of youthful platelet and leukocyte counts in middle‐aged animals, revealing systemic deviations from traditional concepts of hematopoiesis to concertedly promote longevity. Our findings provide a comprehensive resource for the studies of immunosenescence, “inflammaging” and stem cell biology in the NMR as a model of exceptional longevity.

Results

The developmental landscape of naked mole‐rat hematopoiesis

To separate NMR hematopoietic cells, we screened 101 commercially available monoclonal antibodies (moAb; Dataset EV1) against human, mouse, rat, and guinea pig CD markers, and identified human CD11b, CD18, CD34 and CD90, mouse CD11b and CD125, rat Thy1.1, and guinea pig CD45 as cross‐reactive to bind distinct subsets of viable NMR bone marrow (BM) cells (Fig 1A). Antibodies were validated by isotypes, Fc blockers, and cross‐species comparison (Fig 1, EV1).

Figure 1. Purification of blood cell types and the developmental hierarchy in the marrow.

• A
  Frequency of naked mole‐rat bone marrow cells stained with cross‐reactive antibodies (n = 101); mock, unstained; isotype, unspecific IgG with same fluorescent conjugate than specific moAb; cpo, guinea pig target host; hsa, human; mmu, mouse; rno, rat. Dotted line, 5% threshold for unspecific binding.
• B–E
  Representative FACS gating of B, blood (PB) stained with Thy1.1 and CD11b or D, marrow (BM) stained with Thy1.1 and CD34. Sorting gates: GC, neutrophil granulocytes; BC, B cells; TC, T cells; MO, monocytes; EO, eosinophils; HSPC, hematopoietic stem and progenitor cells; MEP, megakaryocytic erythroid progenitor; ERY, erythroid cells. May‐Grünwald‐Giemsa staining of sorted C, PB or E, BM cells; Scale bar 20 µm, same magnification for each micrograph.
• F
  Sorted PB (n = 1) and BM (n = 3) were used for CITE‐Seq (G‐L) with antibodies from (B‐E).
• G
  UMAP of Louvain‐clustered single‐cell transcriptomes, color legend is used throughout this dataset; 1799 differentially expressed genes were used for fGSEA‐based cell type annotation. Tile‐stack inset reflects relative cluster frequencies [freq; y‐axis] and sorted library fractions of the dataset [x‐axis] as proportions. HPC, hematopoietic progenitor cell; EB, erythroblast; GMP, granulocytic monocytic progenitor; BM‐GC, marrow neutrophils; PB‐GC, blood neutrophils; DC, dendritic cells; BCP, B cell progenitor; PC, plasma cells.
• H
  Heatmap showing top 25 overexpressed genes by fold‐change of 14 single‐cell clusters from sorted BM and PB randomly downsampled to ≤ 500 cells, curated cell type markers are labeled.
• I, J
  Scaled CITE‐UMI counts per cell as I, Thy1.1 versus CD11b for PB and J, Thy1.1 versus CD34 for BM.
• K
  UMAP‐based Blendplots showing pairs of differentially expressed lineage markers conserved across species; gene1 (red, high expression), gene2 (blue, high expression), and co‐expressing cells (purple). See scale on the right; expression, scaled UMI counts.
• L
  PHATE model of single‐cell transcriptomes, HSPC cluster is highlighted in black; inset depicts model colored by annotation from G, showing position of progeny cell types relative to HSPCs.

Figure EV1. FACS antibody staining pattern validation, transcriptome assembly statistics, and sorting input for naked mole‐rat scRNA‐Seq.

1. Single color flow cytometry histograms of naked mole‐rat BM stained for indicated surface markers with off‐the‐shelf monoclonal FACS antibodies (moAb), conjugates are suffixed as: FITC, fluorescein isothiocyanate; PE, phycoerythrin; APC, allophycocyanin; AH7, APC‐Cy7. Grey line, unstained BM; black line, isotype; colored line, Fc Blocker cocktail + moAb; colored Area‐Under‐Curve (AUC), moAb. Antibodies cross‐reactive with naked mole‐rat cells described previously are indicated by the respective reference in the top left histogram corner.
2. Single color flow cytometry histograms of human (hsa), mouse (mmu), and naked mole‐rat (hgl) BM stained for indicated surface markers; solid line, Fc Blocker cocktail + isotype; shaded AUC, Fc Blocker cocktail + moAb.
3. FACS backgating of indicated populations from Fig 1B into side (SSC) and forward (FSC) scatters. Major blood cell type gates for lymphocytes, monocytes, and granulocytes were gated and colored as shown in Fig 1B.
4. Postsort with 92 viable events of the sorting strategy for PB CITE‐Seq, sorting gates referring to Fig 1B.
5. Postsort of LIN^– BM (left panel) with 242 viable events, sorting gates referring to Fig 2C; LIN⁺ BM (right panel) with 1,027 viable events; sorting gates indicated.
6. FRAMA transcriptome mapped to ENSEMBL genome and transcriptome using cds (left) or full transcript sequence (right). Total, total annotated transcripts; mapped, blastn FRAMA exon contig alignment mapped to genomic coordinates (> 95% identity, e‐value < 1e⁻⁵); mapped 80%, mapping FRAMA exon contig with > 80% mapping coverage (percent of contig length covered by blastn alignments).
7. Structural differences in FRAMA transcripts with a difference in mapping coverage between genome and ENSEMBL transcriptome of > 20%; light blue, transcripts with new exons; darkblue, transcripts with exon extensions.
8. Seurat cell cycle scoring of sorted naked mole‐rat PB and BM (NMR dataset) with clustering from Fig 1G.
9. UMAP‐based hexbin projection of NMR dataset with clustering from Fig 1G; scaled expression as probability for each conserved gene. HOXA9 2.4‐fold up in HSPC, 1.7‐fold up in HPC; TM4SF1 3.7‐fold up in HSPC, 2.6‐fold up in HPC; TM6SF1 2.4‐fold up in BM‐GC, 2‐fold up in PB‐GC, 1.9‐fold up in GMP; MDK 2.6‐fold up in MEP, 2.2‐fold up in HPC, 1.7‐fold up in HSPC.

When we stained red blood cell (RBC)‐depleted NMR peripheral blood (PB) with Thy1.1 and CD11b, the five major blood cell types could be distinguished: neutrophil granulocytes (GC), eosinophil granulocytes (EO), T cells (TC), B cells (BC), and monocytes (MO) (Fig 1B). Cytochemistry of FACS‐purified cells revealed multi‐lobulated nuclei and a pH‐neutral cytoplasm for Thy1.1^hi/CD11b⁺ GCs and bi‐lobulated nuclei with a high density of acidic granulation for Thy1.1^int/CD11b⁺ EOs, with both populations exhibiting granulocytic scatter properties (Figs 1C and EV1C). Two populations with lymphocytic morphology and size are labeled as Thy1.1^int/CD11b^– and Thy1.1^lo/CD11b^–, while Thy1.1^–/CD11b⁺ cells resemble MOs. BM labeling showed a distinct CD34^hi/Thy1.1^int population of monomorphic cells with small cytoplasm that was absent in PB. CD34 and CD90/THY1 are human stem cell markers (Bhatia et al, 1997); we thus hypothesized these cells to contain the HSPC compartment (Fig 1D and E).

To increase cell type resolution, we performed CITE‐Seq (Stoeckius et al, 2017) for CD11b, CD34, and Thy1.1 on sorted NMR PB and BM cell populations (Figs 1F and EV1D and E). A de novo transcriptome assembly from deep sampling of NMR whole marrow was prepared according to the FRAMA pipeline (Bens et al, 2016) and used for transcript annotation, which revealed hundreds of previously unannotated genes as well as thousands of novel transcript isoforms (Dataset EV2, Fig EV1F and G). We referenced quality control data, clustering, cell cycle scoring, and cell type annotation with a human and a murine droplet‐based single‐cell RNA‐Sequencing (scRNA‐Seq) dataset of hematopoietic hierarchies from the literature (Appendix Figs S1A–D and S2A–D, Dataset EV2). Mapping of 11,920 NMR orthologs to 22,561 cells yielded 14 clusters, which displayed a densely interconnected map of hematopoietic development (Fig 1G). NMR hematopoietic cell types expressed canonical lineage markers and are associated with corresponding gene signatures in gene set enrichment analysis (GSEA; Fig 1H, Dataset EV3). The PB or BM fraction CITE counts each confirmed their FACS staining pattern for the indicated clusters (Fig 1I and J). We identified a megakaryocytic–erythroid progenitor (MEP) cluster overexpressing both GATA1 and GATA2, maintaining CD34 and downregulating Thy1.1 levels. While MEPs resemble progenitor morphology, CD34^– ERY cells mostly contain RBCs and reticulocytes (Fig 1D and E). Canonical cell cycle marker expression revealed the HSPC cluster almost exclusively in G₁, consistent with earlier findings of fate decisions uncoupled from cell division in mice (Grinenko et al, 2018). By contrast, hematopoietic progenitor cells (HPC), MEP, erythroblast (EB), B cell progenitor (BCP), and granulocyte monocyte progenitor (GMP) clusters have the highest G₂/M and S phase signatures (Fig EV1H).

To confirm our annotations and to profile expression kinetics across cells and clusters, we highlighted differentially expressed genes with conserved roles during hematopoiesis (Orkin & Zon, 2008). Blended transcript expression showed strict confinement of NMR CYTL1, a bone mass modulator exclusively induced in human CD34⁺ HSPCs (Shin et al, 2019), to HSPC/HPC clusters, whereas CD34 was also expressed in MEPs and BCPs (Fig 1K). Across humans, mouse and NMR HOXA9 expression was found in the most primitive HSPC cluster, with TM4SF1 specific for NMR HSPCs, whereas TM6SF1 as a marker of lymphomyeloid differentiation is conserved in rodents (Fig EV1I, Appendix Figs S1F and S2F). Midkine (MDK), a pleitropic growth factor, emerged as a specific marker of NMR MEPs. Similar to human and mice however, NMR GATA1 determined erythroid commitment by specific expression in MEPs, EBs, and erythroid cells (ERY), while GATA2 was expressed in HSCs and an HPC subset to merge with GATA1 in MEPs and early EBs (Fig 1K, Appendix Figs S1E and S2E). Interestingly, GP9, expressed at the surface of platelets, was specifically produced in MEP cells not expressing GATA2. This suggests NMR MEPs differentiate through a GATA2‐dependent switch between the erythroid route along GATA1⁺/GATA2⁺/EPOR⁺ versus the GATA1⁺/GATA2^–/GP9⁺/CLEC1B⁺ axis toward megakaryoblasts. In the lymphoid branch we found three developmental stages of CD20⁺ B cell clusters, with BCP showing exclusive expression of VPREB1, located on proB and preB cells (Fig 1K). Cluster BC maintained VPREB3 overexpression, suggesting broad species conservation of successive VPREB gene waves as required for BC development. PU.1 (SPI1) is a transcription factor of the myeloid lineage with a role in HSC maintenance (Staber et al, 2013). Likewise, SPI1 was induced in naked‐mole‐rat HPCs in close proximity to GMPs (Fig 1G–K). Its absence in MEPs suggests conservation of the classic GATA1‐PU.1 bi‐modal switch. Thus SPI1 expression in the HSPC compartments marks the onset of myeloid commitment, converging with CEBPE into the granulocytic lineage, a pattern conserved in mouse (Appendix Fig S2E). Strikingly, an algorithm which captures patterns in high‐dimensional data without referral to prior clustering (Moon et al, 2019), projected HSC/HSPC clusters as central hubs connecting between clusters of the three major lineages (erythroid, lymphoid, myeloid) throughout every species and dataset we tested (Fig 1L, Appendix Figs S1G and S2G).

In summary, we established a cross‐reactive FACS antibody panel to purify HSPC and mature blood cell populations and mapped the major hematopoietic lineages in NMRs.

No age‐associated cell type frequency changes in 11‐year‐old naked mole‐rat blood

Next, we ran CITE‐Seq on unfractionated PB from four 3‐year‐old and three 11‐year‐old NMRs (Fig 2A, Appendix Fig S3A). We used canonical correlation analysis to integrate all blood libraries (Stuart et al, 2019). Louvain clustering found 14 communities of 46,107 NMR PB cells (Fig 2B), and cluster annotation was performed based on GSEA (Dataset EV4). The most abundant fractions were GCs and CD4‐TCs, followed by BCs and MOs, expressing conserved cell‐type markers (Appendix Fig S3B). Indeed, using a previous 10× NMR blood scRNA‐Seq dataset (Hilton et al, 2019), the major leukocyte fractions from our dataset are obtained (Appendix Fig S3C–E), although some low‐abundant communities are not found due to lower capture efficiency and sequencing coverage in the earlier report. Specifically, we found a HSPC cluster in NMR PB, expressing CD34, GATA2, TM4SF1, and other markers (Fig 2B, Appendix Fig S3B, Dataset EV4) we found in HSPCs from sorted BM (Fig 1G–K). Using CITE‐Seq, we clearly recovered the PB staining pattern for Thy1.1/CD11b seen with FACS (Fig 1B), with GC as Thy1.1⁺/CD11b⁺, MO as Thy1.1^–/CD11b⁺, TCs Thy1.1⁺/CD11b^–, and BCs Thy1.1^–/CD11b^– (Fig 2C). To determine age‐related changes in NMR blood, we first compared differences in cell type frequencies across 3‐year‐old young (Y) versus 11‐year‐old middle‐aged (MA) animals from FACS measurements, yielding no significant changes between the age groups (Fig 2D). Likewise, cluster abundances between Y‐PB and MA‐PB age groups from scRNA‐Seq data revealed no significant differences (Fig 2E). Next, we ran two murine pan‐tissue aging clocks (Tyshkovskiy et al, 2019), one modeling linear aging and another in logarithmic scale. We selected 3‐year‐old NMRs as young adults, and 11‐year‐old animals as middle‐aged (MA) NMRs, conversely we chose 3‐month‐old mice as young adults and 12‐month‐old mice as counterpart to MA NMRs (Flurkey et al, 2007). These aging groups were used for CITE‐Seq of unfractionated BM and PB for NMRs. Impressively, both clocks showed median transcriptomic age (tAge) increase in mouse (mmu) BM but not NMR (hgl) PB and BM (Fig 2F).

Figure 2. NMRs maintain youthful blood cell composition into midlife.

• A
  Unfractionated PB of 3 year‐old (n = 4) and 11‐year‐old (n = 3) NMRs were subjected to CITE‐Seq. CITE‐moAbs: Thy1.1, CD34, Cd11b, Cd11c, Nk‐1.1, CD90.
• B
  UMAP of the CCA‐integrated dataset with fGSEA‐annotated cell types. MEMP, megakaryocytic erythroid mast cell progenitor; PLT, platelets; Treg, regulatory T cell; γδTC, gamma/delta‐T cell.
• C
  Scaled CITE‐UMI counts per cell as Thy1.1 versus CD11b for PB GC, MO, BC, CD4‐TC and CD8‐TC clusters.
• D
  FACS PB WBCs across age bins; P‐value determined by Sidak’s Two‐way ANOVA comparing 3 year‐old (n = 17) versus 11‐year‐old (n = 13) NMR blood.
• E
  Single cell RNA‐Seq cluster cell frequency across age; P‐value determined by Fisher’s Two‐way ANOVA comparing 3‐year‐old (n = 4) versus 11‐year‐old (n = 3) NMRs.
• F
  Transcriptomic Age (tAge) change coefficients within cell type clusters of unfractionated NMR PB (n = 13), NMR (hgl, n = 14) and mouse BM (mmu, n = 15). Biological replicates derived from age groups for NMR BM as 3‐year‐old (n = 2) versus 11‐year‐old (n = 2), and mouse BM as 3‐month‐old (n = 2) versus 12‐month‐old (n = 2); P‐value determined by one‐sample Wilcoxon signed‐rank test. Central band as median, boxes correspond to the first and third quartiles (the 25^th and 75^th percentiles), whiskers extend to 1.5x(IQR) [inter‐quartile range], outliers beyond whiskers are plotted individually.
• G
  FACS PB WBC frequencies; P‐value determined by Sidak’s Two‐way ANOVA comparing biological replicates of mouse (mmu, n = 29) versus NMR (hgl, n = 34), **P < 0.01.
• H, I
  H, Hemoglobin concentration and I, hematocrit levels between mouse (mmu, n = 83) and naked mole‐rat (hgl, n = 104) blood; P‐values were determined by unpaired Welch’s t‐test, **P < 0.01.
• J, K
  J, Volumetric white blood cell (WBC) and K, platelet (PLT) numbers across animal age. R ², coefficient of determination; P‐values were determined by conventional linear regression fitting both slope and intercept; n = 114.

Using our PB FACS gating (Fig 1B), we compared the major blood cell types to mice and found dramatically increased granulocytes and reduced BCs in NMRs (Fig 2G), confirming the higher myeloid:lymphoid ratio reported by Hilton et al (2019). FACS‐based blood cell quantifications were fortified by hemanalyzer data (Appendix Fig S3F). The complete blood counts of NMRs showed higher hematocrit and RBC hemoglobin compared to mice (Fig 2H and I). Total RBC numbers were lower than in mice and did not change with age (Appendix Fig S3G and H). In contrast to murine blood leukocyte counts (Appendix Fig S3I), there was no increase in WBC with age in an NMR cohort spanning 12 years of age (Fig 2J). Likewise, blood platelet levels increased in mice, but did not increase and were ~2‐fold lower in NMRs (Fig 2K, Appendix Fig S3J). Imaging of longitudinal femur sections showed fewer erythropoietic islets and Megakaryocytes (MKs) for NMR long bones as compared to mice (Appendix Fig S4A and B). Hemanalyzer differential platelet counts between the two species were corroborated with RBC:PLT ratios obtained from Wright‐Giemsa stained blood smears (Appendix Fig S4C).

Collectively, NMR blood features higher amounts of neutrophils and TCs, but fewer BC than mouse PB, whereas NMRs did not display age‐associated increase of blood leukocytes and platelets, pointing toward reduced chronic inflammation and delay of age‐associated thrombosis. We thus conclude that long‐lived NMRs retain a youthful PB composition for the first decade of their lifespan.

The naked mole‐rat spleen is a major site of erythropoiesis

One of the first milestones toward prospective isolation of HSCs was the early notion that the cell fraction with hematopoietic regenerative potential was nearly devoid of markers of mature blood cell types (Muller‐Sieburg et al, 1986). We thus designed an NMR lineage depletion cocktail (LIN) from the validated cross‐reactive antibodies consisting of CD11b/CD18/CD90/CD125, which purified LIN^– HSPCs as demonstrated by significantly higher colony formation than LIN⁺ or total BM (Fig 3A–C, Appendix Fig S4D). The LIN⁺ fraction strongly enriched Thy1.1^hi GCs, and to a lesser extent TCs and BCs according to their Thy1.1 label intensity in PB populations (Fig 1B). CD11b and CD18 both form the integrin Mac‐1, marking myeloid or NKC commitment. Moreover, we found that the anti‐rat Thy1.1 MoAb labels additional cells not commonly stained with anti‐human CD90 MoAb in BM, most likely due to different epitopes, each with a proteoform‐specific label for NMR THY1 (Appendix Fig S4E). Scatter backgating demonstrated that CD90^hi/Thy1.1^hi cells were neutrophils, while dim CD90^lo/Thy1.1^int cells had lymphoid scatter properties. We therefore used CD90‐antigen to deplete committed cells from the Thy1.1 label. CD125, the IL‐5 receptor alpha subunit, is primarily expressed on eosinophils and activated BCs (Huston et al, 1996). Indeed, BCs were the sole fraction positive for CD125 in PB, while TCs, Eos, and GCs were gradually labeled by CD90‐antigen (Appendix Fig S4F). Most cells of the LIN^– fraction were Thy1.1^–/CD34^– (CP7; candidate population), resembling committed cells not covered by the NMR LIN cocktail (Fig 3C). Surprisingly, both LIN⁺ and LIN^– contain a Thy1.1^int/CD34^hi population, which we termed CP1 and CP2, respectively. CP3 is LIN^–/Thy1.1^lo/CD34^hi, CP4‐5 are LIN^–/Thy1.1^–/CD34^hi and LIN^–/Thy1.1^–/CD34^lo, respectively, while LIN^–/Thy1.1^lo/CD34^– is CP6. Checking which LIN factor is differentially expressed on Thy1.1^int/CD34^hi cells, we found CD11b and CD90‐antigen absent in CP7 and CP2 but present in CP1 and in most viable Thy1.1^int/CD34^hi cells, all four being negative for CD125 (Appendix Fig S4G). We subset CP1 and distinct cell populations of the LIN^– fraction and found each of CP1, CP2, and CP3 were < 1% of total BM leukocytes (Fig 3D), the frequency of the mouse LIN^–/Sca‐1⁺/Kit⁺ (LSK) hematopoietic stem and progenitor cell compartment (Appendix Fig S12C) (Morrison & Weissman, 1994).

Figure 3. Normal erythropoiesis predominantly occurs in the spleen.

• A–C
  Sorting strategy for the HSPC compartment with A, lineage (LIN = CD11b/CD18/CD90/CD125) depletion, B, gating of LIN⁺ CP1 and C, gating of LIN^– CP2‐7.
• D
  Frequencies of BM CP cell fractions. P‐value determined by Brown‐Forsythe’s One‐way ANOVA; n = 39; animal age range 1–4 years.
• E
  Representative gating of LIN^– CP2‐7 in spleen.
• F
  Frequencies of spleen CP cell fractions; n = 30. P‐value determined by Sidak’s Two‐way ANOVA comparing BM versus spleen.
• G
  Benzidine staining of whole spleen [top] or marrow [bottom] from 3‐month old mice [left] or 3‐year old NMRs [right]. Scale bar 250 µm, arrows indicate nucleated erythroid progenitors (NEPs).
• H
  Relative counts of Benzidine‐stained cytospins from whole spleen or WBM. P‐value determined by Sidak’s Two‐way ANOVA comparing BM versus spleen between mouse (n = 5) and naked mole‐rat (n = 8).
• I
  Relative counts of Benzidine‐stained cytospins from naked mole‐rat sorted spleen fractions; P‐value determined by Sidak’s Two‐way ANOVA comparing BM versus spleen; n = 4.
• J
  Pulsing of EdU (1 mg/animal; 4 h before culling) in NMRs (n = 2), spleens were harvested, frozen and later co‐stained for NMR FACS markers and EdU.
• K
  Relative counts of Benzidine‐stained cytospins from whole spleen; no P‐value determined due to (n = 2).
• L
  Mean corpuscular Volume (MCV) as RBC size; P‐values were determined by unpaired Welch’s t‐test; n(mmu) = 85, n(hgl) = 71.
• M
  RBC diameter as measured from blood smears; P‐values were determined by unpaired Welch’s t‐test; n(mmu) = 8, n(hgl) = 9.
• N
  Benzidine‐stained colony assays from sorted BM cells, n = 3. Error bars denote SD, P‐value determined by Sidak’s Two‐way ANOVA.

In most mammals, the spleen primarily acts to recycle aged erythrocytes (Mebius & Kraal, 2005). However, species‐specific adaptations have been found, such as the murine spleen acting as a reservoir of MOs or the equine spleen as a storage of up to 30% RBCs (Kunugiyama et al, 1997; Swirski et al, 2009). We observed a drastic difference in the LIN staining pattern as compared to BM with a strongly expanded LIN^dim population corresponding to elevated Thy1.1^–/CD34^–/lo/hi cells in NMR spleens (Appendix Fig S4H). The frequencies of CP4 (2.3‐fold), CP5 (3.5‐fold), and CP6 were significantly increased in spleens relative to BM (Fig 3E and F). Likewise, we found increased RBC content in splenic versus marrow organ sections in NMRs but not in mice (Appendix Fig S4A). Reanalysis of scRNA‐Seq datasets (Hilton et al, 2019) confirmed the progenitors and differentiated cells of the erythroid lineage in NMR spleens, which were absent in mice (Appendix Fig S5, Dataset EV5). Moreover, Benzidine‐stained cytospins of whole spleens revealed significantly more RBCs in 3 year‐old NMRs than 3‐month‐old mice (Fig 3G). We further detected Benzidine⁺ nucleated erythroid precursors in NMR but not mouse spleens. Strikingly, in adult mice where normal erythropoiesis is known to occur in the BM, the number of nucleated erythroid progenitors diminished 19‐fold from BM to Spleen, in contrast to a 2‐fold increase from BM to Spleen in NMRs (Fig 3H), pointing toward shared splenic and medullary erythropoiesis. To link elevated nucleated erythroid progenitor levels with expansion of the Thy1.1^–/CD34^lo/hi compartment, we sorted CP3 and CP4 cells from spleens for Benzidine staining (Fig 3I). This clearly demonstrated an increase of nucleated erythroid progenitors along with a decline of CD34 expression from CP4 to CP5. Next, we labeled cycling cells in vivo by EdU pulsing, and compared a 3‐year‐old with a 9‐year‐old animal. Spleens co‐stained with our FACS panel were gated for Thy1.1^–/CD34⁺ cells (Appendix Fig S4I and J), the population containing CP4/5. We detected a substantial fraction (3 years, 46.5%; 9 years, 50.5%) of gated splenocytes as S‐phasing EdU⁺, confirming actively cycling CD34⁺ progenitors in NMR spleens in both young and middle‐aged animals (Fig 3J). Benzidine stainings of whole spleens did not show differences in RBCs and nucleated erythroid progenitors across age groups (Fig 3K). The latest time point in ontogeny where active erythropoiesis takes place in the murine spleen is 6–7 weeks postnatal (Chen et al, 2021), thus continuous utilization of splenic erythropoiesis throughout life can be considered a neotenic trait in NMRs.

Interestingly, NMRs appear to have larger RBCs, with the mean corpuscular volume (MCV) significantly elevated compared to mice (Fig 3L). The data is fortified by converting pixel width to diameter and measuring RBC size from PB smears (Fig 3M). Finally, Benzidine‐stained colony assays showed an increase in the proportion of hemoglobin‐containing colonies from CP1 (0.08) over CP2 (0.26) and CP3 (0.56) to CP4/5 (0.94/0.79; Fig 3N). Notably CP1 colonies featured fewer mixed Benzidine^+/– colonies than CP2, pointing toward lymphomyeloid lineage bias of CP1. We thus defined erythroid commitment in the LIN^– compartment by a gradual loss of Thy1.1, directly followed by successive downregulation of CD34.

Shared splenic and medullary erythropoiesis may have evolved in NMRs as an adaptation to life in hypoxic conditions (Stutte et al, 1986). Additionally, it also provides an alternative functional HSC niche throughout life, which may benefit longevity by sustaining youthful RBC production and preventing age‐associated anemia.

LTCs are the main source of naked mole‐rat hematopoietic stem cells

We next performed population RNA‐Seq of sorted CP1‐4 fractions to annotate their developmental status. Unsupervised clustering by t‐distributed stochastic neighborhood embedding (t‐SNE) separated transcriptomes in accordance with their immunophenotype (Fig 4A). We named the CP2 fraction LTC (LIN^–/Thy1.1^int/CD34^hi), since this population has low lineage marker together with elevated Thy1.1‐antigen and is CD34⁺, in close resemblance of the human HSPC immunophenotype. Due to the transition from LIN^– to LIN⁺ between CP2 and CP1, we checked which genes were successively downregulated during transition from LTCs to CP1 and CP3/4 (Fig 4B). We retrieved 116 genes showing this expression pattern, of which 40 are found in the LTC RNA‐Seq signature (Dataset EV6). A key finding was high expression of ID2, which blocks BC differentiation, can enhance erythropoiesis and expands HSCs (Ji et al, 2008; van Galen et al, 2014). LTCs also showed high expression of CD81, a tetraspanin which has been shown to maintain self‐renewal in HSCs (Lin et al, 2011). Notably, TM4SF1, the top marker of NMR HSPCs from the scRNA‐Seq atlas (Fig 1H), and the pluripotency marker EPCAM, which facilitates reprogramming (Kuan et al, 2017), are most abundant in LTCs. Next we derived differentially expressed genes specific to each CP1‐4 (Dataset EV6). Strikingly, the strongest GSEA association for CP3 was MEPs, while CP4 is negatively correlated with myeloid, lymphoid, and HSPC signatures, and both share elevated expression of erythroid marker genes GATA1, EPOR, TFRs, KEL, and FECH in CP3/4 (Figs 4C and EV2A). CP1 was enriched with the most HSPC‐associated genesets (Dataset EV6). The human CD34⁺ signature displayed an enrichment gradient from CP1 to CP4, revealing a CD11b⁺ primitive progenitor fraction with a stemness expression profile in adult NMRs. Moreover, a panel of canonical HSPC markers was consistently enriched in CP1/LTCs compared to CP3/4, with the exception of TAL1 increasing toward erythroid commitment, and LMO2 showing no difference in expression levels across all four populations (Fig EV2B–D). Similarly, FST, MECOM, HOXA10, MSI2, and ERG are overexpressed in HSPC/HPC clusters from sorted BM scRNA‐Seq (Fig EV2E and F).

Figure 4. LTCs define the primitive HSPC compartment.

1. BM from 1 to 3 year old NMRs sorted into indicated cell populations for RNA‐Sequencing or xenotransplantations, color legend applied throughout the figure. Unsupervised t‐SNE clustering, effectively separating each CP group; vst‐transformed counts as input.
2. 116 gradually downregulated genes from CP2 to CP1, displayed are 20 genes with known roles in hematopoiesis.
3. GSEA of sorted BM fractions displaying top 10 q‐value terms from a geneset collection of human and mouse HSPCs, derived from (Schwarzer et al, 2017). NES, normalized enrichment score; GeneRatio, (signature ∩ term) / (signature ∩ all terms).
4. Schematic workflow of single cell colony assay (boxed), FACS biplot as representative gating of NMR colony assay clones. H4435, vendor ID for human methylcellulose assay with HSC‐enriched cytokine cocktail (Stemcell Technologies, Vancouver). MoP, monocytic precursor; EP, erythroid progenitor; E, erythroid cell.
5. Scatterplot overlays of the sorted sample (light grey plotting area) and each sorted fraction (CP1, blue; LTC. black; CP3, orange red). CP1 is LIN⁺ (left plot), CP3 and LTC are LIN^–.
6. Representation of cell numbers per clone (bar graph upper row) and population frequencies per clone (stacked bar plots lower row), gating according to Fig 4D (biological n = 2).
7. Quantitation of population frequencies per clone across sorted NMR BM cell fractions from Fig 4F; P‐value determined by Tukey’s Two‐way ANOVA (clones tested: CP1, n = 25; LTC, n = 30; CP3, n = 22), **P < 0.01.
8. Quantitation of cell numbers per clone across sorted NMR BM cell fractions from Fig 4F; P‐value determined by Tukey’s One‐way ANOVA (clones tested: CP1, n = 25; LTC, n = 30; CP3, n = 22).
9. Colony forming capacity (CFC) analyzed through L‐Calc software by determining (x CFC cells)/(n input cells); 1/3, one out of three cells forms a colony. Values calculated from (biological n = 4) experiments with each 1‐, 2‐, and 3‐cell per well sorting inputs, see also Appendix Fig S6G.

Figure EV2. Canonical HSPC markers in NMR population and scRNA‐Seq.

• A
  Top color bar reflects population clustering by Euclidean distance using top 12 MEP/erythroid leading edge genes; color key as in Fig 3A. Genes were obtained by ssGSEA of combining CP1/2 (n = 10) versus CP3/4 (n = 10) populations and performing differential gene expression (DGE) testing, see also Dataset EV6.
• B–D
  Filtered, normalized, log2‐transformed, and median‐centered TPMs [transcripts per million] of naked mole‐rat bulk HSPC transcriptomes for NMR orthologs of B, HOXA10, MEIS1, TIE1, and C, MPL, MSI2, TAL1, BMI1, and D, CXCR4, ELF1, ETS2, LMO2; demonstrating expression of conserved hematopoietic stem and progenitor genes in NMR HSPC fractions.
• E, F
  UMAP‐based hexbin projection of NMR dataset with clustering from Fig 1G; scaled expression as probability for each conserved gene. FST 2.9‐fold up in HSPC; CDC42 1.7‐fold up in PB‐GC; MSI2 1.2‐fold up in PC; MYC 2.3‐fold up in HPC, 1.7‐fold up in MEP. Note that MECOM and HOXA10 are specifically expressed in HSPCs and HPCs, however their expression was close to the detection limit of the 10× droplet‐Seq platform, as seen also for ERG.

The capacity to give rise to several distinct lineages via differentiation, referred to as multipotency, can be assayed through quantitation of progenitor frequencies during colony formation (Roy et al, 2012). We developed a single cell colony assay to quantify colony formation capacity (CFC) and clonal differentiation by FACS (Fig 4D). Using mouse BM‐derived HSPC populations, we validated the in vitro assay (Appendix Fig S6). We sorted LSKs and committed progenitors GMPs and MEPs (Appendix Fig S6A), and scored 4 major immunophenotypes within each clone obtained after 12d incubation (Appendix Fig S6B): Granulocytic Precursors (GP) were Cd11b^hi/Gr‐1^hi/Kit^lo/Cd71^–; monocytic precursors (MoP) were Cd11b^hi/Gr‐1^lo/Kit^lo/Cd71^–/lo; common myeloid progenitors (CMP) were Cd11b^lo/Gr‐1^–/Kit^hi/Cd71^–/lo; and erythroid precursors (EP) were Cd11b^–/Gr‐1^–/Kit^–/lo/Cd71^hi. Since between the three murine BM populations only LSKs harbor HSCs, those consistently gave rise to more clones, which grew larger colonies with a higher variety of clonal immunophenotypes formed (Appendix Fig S6C–E). In line with this the LSK, CFC was markedly higher than for GMPs and MEPs (Appendix Fig S6F). We developed an analogous FACS gating for the NMR colony assay (Fig 4D): clones featuring a Thy1.1⁺/CD34^– subset formed scattered, colorless colonies comprised of small cells positive for alkaline phosphatase (AP), thus termed CFU‐GM (Appendix Fig S7A), while Thy1.1^–/CD34^– CFU‐Ms formed dense, colorless colonies consisting of large AP^– cells (Appendix Fig S7B). Only clones with the combined Thy1.1^–/+/CD34^–/lo/hi immunophenotype gave rise to large colonies with a densely packed, pigmented center and scattered outline, featured both AP^–/+ and Benzidine^–/+ cells and morphologies of macrophages, HPCs and very small erythroid precursors, thereby containing myeloid and erythroid cells as CFU‐GEM (Appendix Fig S7C). Erythroid precursor CFU‐Es comprised dense and pigmented colonies of very small cells overtly Benzidine⁺ in high numbers (Appendix Fig S7D). All colony types were also obtained in the bulk assay format incubated at 32°C in methylcellulose supplemented with human cytokines (Appendix Fig S8A). To functionally classify the most primitive HSPC fractions based on their gene expression (Fig 4C), we sorted CP1, LTC, and CP3 in dilutions of 1, 2, or 3 cell/well (Fig 4E). Strikingly, LTC clones formed most colonies, which also contained more cells than for CP1 and CP3 (Fig 4F–H). Conversely, the highest CFC was obtained with LTCs (Fig 4I), albeit to a lesser extent than mouse LSKs (Appendix Fig S6F), which we partly attribute to the use of human cytokines for NMR cells. For bulk colony assays scoring of CFU types was validated by cytochemistry of single colonies (Appendix Fig S8B and C). Of all NMR BM populations only CP6 and CP7 did not grow in methylcellulose assays (Appendix Fig S8D). Furthermore, the proportion of erythroid over total colonies declined from CP4/5 (0.94/0.81) to CP3 (0.53), and CP2 (0.29) was lowest in CP1 (0.2). Myeloid output was not significantly different between CP1 and LTC but decreased in CP3. Serial replating yielded 1.5‐fold more total colonies for LTC compared to CP1, although no colony type frequency was significantly altered between these two, as seen for original platings (Appendix Fig S8E).

Multipotency can further be assessed by transplantations into preconditioned immunodeficient hosts, through which high levels of sustained primary engraftments could be obtained in a variety of humanized mouse models (Goyama et al, 2015). Given the successful in vitro growth of NMR HSPCs with human cytokines, we reasoned that the NSGS mouse model with constitutive production of human IL‐3, M‐CSF and SCF would render optimal support to NMR xenografts (Wunderlich et al, 2010). We indeed observed robust engraftment rates for LTCs at 4 weeks post transplantation in recipient BM as compared to untransplanted mice (Fig 5A). Xenografts recapitulated the FACS staining pattern of NMR BM origin and could be separated from host cells, which are not labeled by validated NMR Thy1.1 and CD34 moAbs (Appendix Fig S9A–D). At week 2, host BM chimerism resembled colony yields from methylcellulose assays with CP6/7 engraftments being below background, supporting the notion that the NMR HSPC compartment is CD34⁺ as in humans (Fig 5B). All other populations produced clearly detectable engraftment in NSGS BM ranging from 1.6% (CP5) over 4.2% (CP4) and 12.1% (CP3) to 14% (LTC) and 16% (CP1). Repopulation of host spleens was markedly reduced for all engrafted groups; strikingly CP3 was functionally classified as the most primitive committed erythroid progenitor and enriched in host spleens with higher engraftment than LTC (Fig 5C). Though FACS analyses verified xenograft cells in blood for CP1, LTC, and CP3, levels ranged below 1% of viable leukocytes. BM engraftment at week 4 for CP3 (1%) depleted earlier than for CP1 (12.8%, P < 10⁻⁴) and LTC (11.7%, P < 10⁻⁴) (Fig 5D). The early loss of erythroid‐primed CP3 is consistent with higher residual chimerism at week 8 for myeloid‐primed CP1 compared to LTC (5.3% versus 1.3%, P = 0.04). Unexpectedly none of the three most primitive stem and progenitors or whole marrow (WBM) showed sustained BM engraftment past 12 weeks (Appendix Fig S9E), a fact we primarily attribute to the difference in body temperature between NMRs (thermoneutral at ~32°C) and humans or mice, leading to niche stress on the graft and its depletion. Next, we quantified lineage commitment over time by selecting the Thy1.1⁺/CD34⁺ compartment of xenografts (HSPC, Fig 5A). Although this rapidly depleted for all cell types at week 4, the initial replicative burst was greater in CP1 compared to LTC (Fig 5E), suggesting that CP1 cells are activated to a greater extent by the inflammatory host environment that ultimately exhausts engraftment. Xenograft CD34⁺ cells (ERY, Fig 5A) resembling the erythroid lineage decline towards week 4 for CP1 and CP2, whereas CP3 CD34⁺ output remained similar (Fig 5F). Conversely, we used Thy1.1^hi cells as myeloid output (GC, Fig 5A), which revealed most efficient myelopoiesis at week 8 in CP1 compared to LTC and CP3 (Fig 5G). B‐lymphopoiesis in NMR BM is conserved (Fig 1K, Appendix Figs S1E and S2E), and since blood BCs are labeled by Thy1.1^lo/CD11b⁻, we reasoned that CP6 cells would contain marrow and spleen BCs, albeit with less purity. The xenograft lymphomyeloid population (LYMY, Fig 5A) significantly dropped in CP3 cells at weeks 4 and 8 but is more efficiently sustained in CP1 and LTC with higher myeloid potential (Appendix Fig S9F). Since the heterogeneity of this FACS fraction does not provide evidence over definitive B‐lymphoid commitment in xenografts, we performed scRNA‐Seq from week 4 CP1 and LTC grafts (Fig 5H, Appendix Fig S9G). Integrative clustering on 4904 cells resulted in 9 communities containing erythroid cells, HSPCs, and > 80% myelocytes (Fig 5I and J, Appendix Fig S9H, Dataset EV7). Due to the exhaustive effect of the host BM niche, the G₁ fraction of xenograft HSPCs is lowered to < 50% (Fig 5K), whereas NMR BM HSPCs are > 80% in G₁ (Fig EV1H). Lymphocytes were only found upon sub‐clustering a mixed fraction comprised of BCs and DCs (Fig 5L). LTCs produced more BCs (0.033% total xenograft compared to 0.016% for CP1; Fig 5M) and higher BC gene expression (Fig 5N), suggesting that lymphoid commitment within primitive LTCs is diminished upon CD11b/CD90‐antigen expression at the onset of myelopoiesis in CP1. Concordantly, the CITE counts for CD11b corresponded with LIN sorting between CP1 and LTC (Fig 5O). We further confirmed lymphomyeloid potential in LTCs and CP1 using in vitro differentiation through recombinant IL‐7 and qPCR of early factors during B‐development (Appendix Fig 9I and J). In bulk cultures both CP1 and LTC cells expressed CD10 and TCF3, although in single cell clones only LTCs formed multiple clones with detectable EBF2 and PAX5 mRNA (Appendix Fig S9K and L).

Figure 5. Naked mole‐rat hematopoietic xenografts.

• A
  Gating strategy to quantify total engraftment; untransplanted recipient [left] versus CP2 xenograft 4 weeks post Transplantation (Tx) [right]; HSPC, Thy1.1^int/CD34⁺ stem and progenitors; ERY, CD34⁺ erythroid cells; GC, Thy1.1^hi granulocytes; LYMY, Thy1.1^lo/int/CD34^– lymphomyeloid cells.
• B, C
  Recipient chimerism 2 weeks post Tx in B, BM or C, spleen; total recipients from 3 donors for each CP graft.
• D
  BM chimerism over time; 2–3 recipients each from 3–5 donors for weeks 4, 8 and 12. P‐value determined by Fisher’s Two‐way ANOVA; curve‐fitting by cubic polynomial.
• E–G
  Kinetics of engraftment proportions for E, HSPC, F, ERY and G, GC; 2–3 recipients each from 3 to 5 donors for sorted NMR populations and time points. Error bars, SEM, P‐value determined by Tukey’s Two‐way ANOVA, **P < 0.01.
• H
  ScRNA‐Seq of week 4 BM CP1 and LTC xenografts (n = 3); sorting gates excluding the bulk of host cells are depicted.
• I
  T‐SNE of the CCA‐integrated xenograft dataset with fGSEA‐annotated cell types, contaminating mouse cells filtered out. GCP, granulocytic precursor; mix, not assignable to a single cell type.
• J
  Mean frequencies of GSEA‐annotated cell types between graft sources.
• K
  Seurat cell cycle scoring of xenograft cell types.
• L
  UMAP of the mix cluster subset, split by graft populations.
• M
  Quantitation of BC fraction per total xenograft cells (n = 3); P‐value determined by un‐paired, two‐tailed t‐test.
• N
  Top marker expression identifying dendritic cell (DC) and B cell (BC) subset clusters; circle sizes reflect percent of cells of the cluster expressing respective markers.
• O
  CITE‐CD11b levels per cell for the HSPC cluster from the integrated xenograft dataset.

Altogether CP1 likely contains CMPs and BCPs, the lymphoid potential suggests a heterogeneous population of committed progenitors with prevalent myeloid bias and decreased capacity of differentiating towards the erythroid lineage. Furthermore, we functionally defined the most primitive HSPC compartment as LTCs (LIN^–/Thy1.1^int/CD34^hi), exhibiting the highest degree of CFC and multipotency. Our FACS panel effectively subsets the primitive HSPC compartment in NMRs, wherein diminished Thy1.1 expression of CD34⁺ cells correlated with erythroid fate decision along the LTC‐CP3/4 axis, while rising CD11b levels corresponded to myelopoiesis through LTC‐CP1.

Expansion of marrow granulopoiesis and the erythroid lineage but reduced B‐lymphopoiesis

Next, we ran CITE‐Seq on WBM from two 3‐month‐old and two 12‐month‐old mice (mmu) against WBM from two 3‐year‐old and two 11‐year‐old NMRs (hgl; Fig 6A, Appendix Fig S10A). We used canonical correlation analysis to integrate the four marrow libraries separately for each species. Louvain clustering found 15 communities from a total of 19,298 mouse marrow cells, and 14 communities in 21,678 NMR marrow cells (Fig 6B). Cluster annotation based on GSEA (Dataset EV8). In mice, cell types were strongly aligned with the CITE signals (Appendix Fig S10B), for example, a rare HSPC population of < 1% total BM expressed ANGPT1, GATA2, and HOXA9 and had CITE‐LIN^–/Kit⁺/Sca‐1^–/+, co‐clustering myeloid progenitors (LKs) and LSKs as the murine HSPC compartment. Likewise, NMR HSPCs overexpressed TM4SF1, GATA2, and HOXA9, and were CD11b^–/lo/CD34⁺/Thy1.1^–/lo/int, suggesting LTCs, CP1, and CP3 collectively clustered as HSPC. Moreover, we applied SCENIC to map transcription factor (TF) regulons within single cell transcriptomic landscapes of either species (Aibar et al, 2017), which found Cebpe, Erg, Gata1, Meis1, Pbx1, Runx1, Tal1, and T‐bet/Tbx21 regulons across mouse BM clusters (Appendix Fig S11A). For NMRs we created a de novo cisTarget motif database and ran it with pySCENIC (Van de Sande et al, 2020), which yielded regulons of hematopoietic TFs such as Cebpb, Ets1, Gata1, Gata2, Irf7, Meis3, and Tcf3 (Appendix Fig S11B and C). The Hoxa10 regulon was common to both mouse (Fig EV3A–C) and NMR HSPC clusters (Fig EV3D–F), further evidence underscoring stem cell identity of NMR HSPCs.

Figure 6. BM CITE‐Seq comparing mouse and naked mole‐rat.

• A
  Unfractionated BM of perfused mice (n = 4) or NMRs (n = 4) were subjected to CITE‐Seq. mmu CITE‐moAbs: Sca‐1, Kit, Cd11b, Cd11c, Nk‐1.1, Cd4, Cd8a, Cd3e, Cd19, Cd25, Cd44, Gr‐1, Ter119; hgl: Thy1.1, CD34, Cd11b, Cd11c, Nk‐1.1, Gr‐1, CD90.
• B
  UMAP of the CCA‐integrated mouse [left] or naked mole‐rat [right] dataset with fGSEA‐annotated cell types. Bar chart [center] displaying average cluster frequencies across species; APC, antigen‐presenting cell.
• C
  T‐SNE of SCTransform‐integrated mmu and hgl BM, colorbar legend for species‐integrated clusters below. Encircled coordinates for HSPC, LMPP, and MEP clusters; cycling, co‐clustered based on active cell cycle gene expression.
• D, E
  T‐SNE from species‐integration for D, mouse or E, naked mole‐rat partition; Cluster annotation and coloring from the single‐species analysis in (B).
• F
  Differential cell type abundance across species; dotted lines, 2‐fold change. MLP, P < 10⁻¹⁰; ProBC, PreBC, MEP, P < 10⁻⁷; BC, ERY, P < 10⁻⁵; PC, P = 0.0017; MO, P = 0.0088; GCP, P = 0.027. **P < 0.01, *P < 0.05.
• G, H
  Differential cell type abundance across age for G, mouse or H, naked mole‐rat single‐species analysis. hgl CD8‐TC, P = 0.002; hgl BCP, P = 0.013, **P < 0.01, *P < 0.05. Dotted lines, 2‐fold‐change.
• I–K
  BM cell frequencies of I, LTC/CP2, J, CP1 or K, CP3 across age; n = 60, linear regression with 95% CI as trend line; P < 0.05, significance.

Figure EV3. SCENIC HOXA10 regulon analysis in mouse and naked mole‐rat BM.

• A–F
  Aggregated averaged expression of genes associated with the HOXA10 regulon in A, mouse and D, NMR BM. Circled in red are the HSPC clusters of either species. Expression of HOXA10 in B, mouse and E, NMR BM. UMAP dimensionality reduction computed by the SCENIC workflow for C, mouse and F, NMR BM; Seurat clustering for each species was taken from Fig 6A and B.

Next, we quantified cell types across species and grouped them into major branches (Fig 6B). The most abundant fractions were GCs and granulocytic precursors (GCP), expressing conserved cell type markers, which account for ~50% WBM in mice and ~75% in NMRs. By contrast, the BC compartment with > 26% in mice is reduced 4‐fold to < 7% in NMRs. As reported previously for spleen (Hilton et al, 2019), there are no NKCs detected in NMR BM. Surprisingly, although the TC partition has the same frequency in both species, mice show a 2:3 ratio of CD4‐ versus CD8‐TCs, NMRs a ratio of 5:1. The pattern is exacerbated in blood, 3‐year‐old NMRs had 15.4‐fold more PB‐CD4‐TCs than PB‐CD8‐TCs, whereas 11‐year‐old NMRs featured a ratio of 9.5:1 (Fig 2E). In fact, by absolute copy number quantitation of sorted PB‐TCs, we determined a CD4:CD8 mRNA ratio of 1:3.5 in mice, which was inverted to 2.5:1 in NMRs (Appendix Fig S10C). This pattern was confirmed in whole lymph nodes with mature peripheral TCs as the primary source of CD4/CD8 expression, showing a 1:2.2 ratio in mice and 5.8:1 in NMRs (Appendix Fig S10D). Hereby NMRs resemble humans with a ratio > 1 (Bruno et al, 2017), rather than mice. The CD4:CD8 ratio is used to discriminate the risk of disease progression in HIV/AIDS and decreases with age in patients (Castilho et al, 2016), a low CD4:CD8 ratio indicates immunosenescence and is associated with wide‐ranging pathology (Huppert et al, 2003). The unusually high CD4:CD8 ratio in NMRs suggests a reduced dependence on cell‐mediated immunity.

Next, we integrated the species data using SCTransform (Hafemeister & Satija, 2019), and overlaid species annotations on integrated clusters (Fig 6C–F). Surprisingly, there were no NMR counterparts co‐clustering with murine multipotent lymphoid progenitors (MLPs) within the integrated MLP community, thus mouse BM strongly enriched for MLPs (59‐fold, q < 10⁻⁹; Fig 6F). Mouse MLPs had lower levels of stem cell factors Lmo2 and Pbx1 than HSPCs, and instead shared markers upregulated throughout the B‐lineage such as Ets1, Fli1, Cd48, Il7r, Tcf3, and showed specific overexpression of Tcf4, Irf8, and Flt3 (Fig EV4A). NMR cells mapping to the integrated MLP population were annotated as MOs in the single‐species clustering (Fig 6E) and overexpressed IRF8 and TCF4 but not IL7R or LMO2. As the earliest HSPC‐to‐B‐transition intermediate MLPs were accumulated along with ProBCs, PreBCs (both 21‐fold, q < 10⁻⁶), and BCs (10‐fold, q < 10⁻⁵) in mouse BM (Fig 6F). Evidently the significant B‐lineage reduction in NMRs manifests from primitive progenitors to mature BCs. However, in NMR BM a rare population with high expression of JCHAIN, MZB1, XBP, and EAF2 likely comprised germinal center B or plasma cells (PC; Figs 1G and H, and EV4A, Appendix Figs S3B–E, S5 and S10B). Intriguingly, there are more PCs in NMR BM than in mice (Fig 6F). The higher CD4‐TC abundance combined with a compressed BC compartment leads to a higher CD4:APC (antigen presenting cell) ratio and could more efficiently activate BCs, relative to their total frequency, resulting in more plasmablastoid differentiation. On the other hand, increasing the amount of the terminal effector cell, for example, through lower cell turnover, could have evolved to compensate for a less abundant BC compartment. Mouse megakaryocytic/erythroid/mast cell progenitors (MEMP) form a distinct cluster in mouse whole marrow, which is reformed upon integration of both species (MEP). Within this common MEP cluster, the NMR fraction is derived from NMR‐clustered HSPCs (Fig 6B–E). We observed splenic erythropoiesis as the primary route to NMR RBC production; however, the increase of erythroid cells in BM (11‐fold, q < 10⁻⁵) suggested a higher prevalence of erythroid commitment (Fig 6F). Surprisingly mouse BM contained 22‐fold more MEPs (q < 10⁻⁶), supported by GATA1/GATA2 co‐expression common in MEPs of both species (Fig EV4B). Mouse early megakaryoblasts are GATA2⁺/GATA1^lo (Appendix Fig S2E), as were integrated mouse MEPs, whereas integrated NMR MEPs expressed GATA2⁺/GATA1^hi (Fig EV4B). A potential erythroid fate bias is further in line with reduced BM MKs and PB platelets in NMRs.

Figure EV4. Unfractionated BM cross‐species markers and NMR HSPC GSEA of age groups.

• A, B
  Cross‐species expression of selected A, differentially regulated or B, conserved markers between cell types. SCT‐UMI, sctransform‐scaled UMI counts; % exprsd, percentage of cells/cluster with UMI ≥ 1. GSEA using
• C, D
  C, The MSigDB hallmark geneset collection or D, the MSigDB C5 GO subset Biological Process (BP) geneset collection; NES, normalized enrichment score. The HSPC cluster from the unfractionated NMR BM dataset was split by age, and 11‐year‐old HSPCs were run against 3‐year‐old HSPCs for differentially expressed genes, thus genesets with positive NES were enriched and negative NES genesets were depleted in 11‐year‐old HSPCs.

Assuming a linear relationship between age groups across species, we compared differences in gene expression and cell type abundances of 12‐month‐old versus 3‐month‐old mouse BM with 11‐year‐old versus 3‐year‐old NMR BM. In mice no cluster abundance was significantly altered across age groups (Fig 6G); however, GSEA on 760 top differentially expressed genes across all clusters between age groups revealed 3 upregulated terms related to proliferation and growth in older mice (Appendix Fig S10E). Only CD8‐TCs were significantly elevated in older NMRs to 1% of total BM cells (11‐fold, q < 0.033; Fig 6H), which most likely reflected memory cell acquisition over age. We found 91.4% of mouse HSPCs and 85.8% of NMR HSPCs mapping to the same integrated HSPC cluster, which commonly expressed stem cell markers such as HOXA9, KIT, and ANGPT1 (Fig EV4B). This suggested a higher HSPC abundance in NMRs (0.54% mmu versus 2% hgl; Fig 6B); however, myeloid‐biased CP1 LIN⁺/Thy1.1^int/CD34^hi progenitors co‐cluster with the primitive HSPC compartment of both species, supported by transcriptional signatures of human HSC and CD34⁺ cells associated with CP1 (Fig 4C). Conversely, CITE‐CD11b⁺ HSPC cluster cells increased in older animals (Appendix Fig S10F), and the HSPC cluster of older NMRs was expanded by ~2‐fold, albeit not significantly (Fig 6H). Further functional interrogation is mandated to arrive at a more precise picture of the role of CP1 cells in NMR hematopoiesis. However, we compared gene expression profiles of HSPC clusters from older NMRs to that of 3‐year‐old animals and performed GSEA using the MSigDB Hallmark geneset collection on a BM HSPC aging signature (Fig EV4C). Consistent with HSPC cluster expansion during aging, we found MYC targets, MTORC1 signaling, and Unfolded Protein Response genesets enriched and Apoptosis depleted in 11‐year‐old HSPCs. Remarkably, four high‐level genesets related to Inflammaging were negatively correlated with increased age in NMR HSPCs, while DNA repair was enriched with age (Fig EV4C). In addition, the GO term Bio‐Process geneset compendium yielded confirmative GSEA results, showing enriched Nucleotide Excision Repair and Cellular Oxidant Detoxification pathways along with depleted Senescence gene expression in older NMR HSPCs (Fig EV4D). Strikingly, both CP1 and CP2/LTC BM frequencies significantly increased with age, whereas erythroid progenitors remained constant (Fig 6I–K, Appendix Fig S10G). Interestingly, an age‐associated expansion does not seem to translate into the myeloid differentiation bias progressing with age, as seen for human HSCs (Pang et al, 2011). Instead, 11‐year‐old BM has slightly increased lymphoid progenitor frequencies (BCPs; Fig 6H), and the aged HSPC signature is negatively associated with Myeloid Cell Differentiation (Fig EV4D).

We next checked if sex had an impact on cell type frequencies within the BM scRNA‐Seq datasets. We did not see any significant differences in cell type abundance between ♂ and ♀ mouse BM (Appendix Fig S10H). In ♀ NMRs, the HPC cluster was increased (Appendix Fig S10I), comprised of heterogeneous progenitor subsets and cycling cells, requiring more cell captures and donor animals to characterize the sex differences in this context. Notably, total marrow cellularity increased with similar regression trends independent of sex, at least until 12 years of age (Appendix Fig S10J).

In summary, we have shown that NMR BM maintains a myeloid bias toward granulopoiesis, accompanied by reduced B‐lineage commitment. Erythropoiesis is favored over megakaryopoiesis in NMR marrow, contributing to maintenance of low platelet levels in PB. A stem cell state was portrayed by pervasive expression of TM4SF1, the top HSPC marker by fold‐change (Fig 1H), with highest expression in LTCs (Dataset EV6) and specific for NMR HSPCs (Fig EV1I, Appendix Figs S1F, S2F and S10B). Moreover, NMR HSPCs express a conserved Hoxa10 TF regulon next to multiple other stem cell genes and expand with age. Remarkably, the NMR HSPC aging signature is depleted for inflammation‐related genesets and is enriched for several cellular resilience pathways, likely contributing to NMR longevity.

Naked mole‐rat HSPCs display low metabolism signature and stem cell polarity

Since CITE‐Seq‐driven cell‐type annotations matched HSPC FACS populations, we examined sorted population level transcriptomes of corresponding developmental stages across species (Fig 7A, Appendix Fig S12A–E). A comprehensive collection of distinct human and murine HSPC stages was retrieved from GEO and integrated with bulk RNA‐Seq data from human and NMR. The dataset of 9,422 orthologs across 218 transcriptomes was segregated into three groups: primitive (LT‐HSC, MPP), lymphomyeloid, and erythroid progenitor. Using GSEA with MSigDB hallmark genesets, we found that mouse cells through all stages were enriched in mitotic and pro‐proliferative pathways (Fig 7A). Interestingly, functional annotation of scRNA‐Seq cluster signatures across species showed mouse HSPCs and several more committed cells enriched in pro‐proliferative pathways compared to their NMR counterparts (Appendix Fig S12F). Human HSPCs scored high for apoptosis, glycolysis, and OXPHOS pathways, whereas NMR HSPCs strongly enriched for adipogenesis, cholesterol homeostasis, and fatty acid oxidation (FAO)‐related terms. Indeed, a plasma metabolite signature of multiple upregulated lipid sub‐classes has been reported earlier (Lewis et al, 2018). FAO provides the substrates for OXPHOS, while Aldehyde dehydrogenases (ALDH) neutralize aldehydes arising from processes such as lipid peroxidation. Notably, ALDH staining revealed 2‐fold higher levels in LTCs compared to LSKs (Appendix Fig S12G), indicating a countermeasure against elevated FAO activity.

Figure 7. Low metabolism and cell polarity of naked mole‐rat HSPCs.

• A
  Cross‐species integration of bulk RNA‐Seq datasets. Naked mole‐rat BM populations CP1‐4 were matched to human and murine counterparts for GSEA with MSigDB hallmark genesets. Primitive stem and progenitor [left], hsa: LT‐HSC, LIN^–/CD34⁺/CD38^lo/CD90⁺/CD45RA^– (n = 7); HSPC, LIN^–/CD34⁺/CD38^lo (n = 11); MPP, LIN^–/CD34⁺/CD38^lo/CD90^–/CD45RA^– (n = 4); mmu: LT‐HSC, LIN^–/Sca‐1⁺/Kit⁺/CD150⁺/CD48^– (n = 25); MPP, LIN^–/Sca‐1⁺/Kit⁺/CD150^–/CD48⁺ (n = 6); hgl: LTC, LIN^–/Thy1.1^int/CD34⁺ (n = 5). Lymphomyeloid [center], hsa: LMPP, LIN^–/CD34⁺/CD38^lo/CD123^lo/CD45RA⁺ (n = 9); CLP, LIN^–/CD34⁺/CD38^hi/CD10⁺/CD45RA⁺ (n = 6); MLP, LIN^–/CD34⁺/CD38^lo/CD90^–/CD45RA⁺/CD71^– (n = 4); CMP, LIN^–/CD34⁺/CD38^hi/CD123^lo/CD45RA^– (n = 26); GMP, LIN^–/CD34⁺/CD38^lo/CD123^lo/CD45RA⁺ (n = 18); mmu: LMPP, LIN^–/Sca‐1⁺/Kit⁺/Flt3^hi (n = 5); CMP, LIN^–/Sca‐1^–/Kit⁺/CD16/32^lo/CD34⁺ (n = 8); GMP, LIN^–/Sca‐1^–/Kit⁺/CD16/32^hi/CD34⁺ (n = 16); hgl: CP1, LIN⁺/Thy1.1^int/CD34⁺ (n = 5). Erythroid [right], hsa: MEP, LIN^–/CD34⁺/CD38^hi/CD123^–/CD45RA^– (n = 23); MKP, LIN^–/CD41a⁺/CD42b⁺ (n = 3); ERY, LIN^–/CD34^lo/–/CD36⁺/GYPA⁺/CD71⁺ (n = 12); mmu: MEP, LIN^–/Sca‐1^–/Kit⁺/CD16/32^–/CD34^– (n = 15); hgl: CP3, LIN^–/Thy1.1^lo/CD34⁺ (n = 5); CP4, LIN^–/Thy1.1^–/CD34⁺ (n = 5). Within each group, cell types were pooled for each species. Shown are pathways related to proliferation and metabolism, full results see Dataset EV9. NES, normalized enrichment score; GeneRatio, (signature ∩ term)/(signature ∩ all terms); FDR, false discovery rate.
• B
  Seahorse assay with cell types sorted according to (A); hOPP, human oligopotent progenitors LIN^–/CD34⁺/CD38⁺; mOPP, LIN^–/Sca‐1^–/Kit⁺. ECAR, extracellular acidification rate; OCR, oxygen consumption rate. Error bars, SEM; Human cell types, n = 4; mouse, n = 3, naked mole‐rat, n = 3.
• C
  Mitotracker Red staining in mouse (n = 4), human (n = 4), and naked mole‐rat (n = 5) BM, histogram of merged per‐species data. unstained, BM from mouse [solid], human [dotted] or NMR [area under curve (AUC)]; yLSK, LIN^–/Sca‐1⁺/Kit⁺, 3‐month old; oLSK, 24 month; yLTC, LIN^–/Thy1.1^int/CD34⁺, 3‐year old; hHSC, human CD34⁺/CD38^lo; quantitation see Appendix Fig S12C.
• D
  Mean fluorescence intensities of TMRE stainings; P‐values were determined by Tukey’s One‐way ANOVA. LSK, n = 8; LTC, n = 7. FCCP, negative control.
• E
  Schematic drawing illustrating the stem cell polarity concept during aging.
• F
  Confocal fluorescence imaging of sorted mouse LT‐HSCs (LIN^–/Sca‐1⁺/Kit⁺/CD150⁺/CD48^–) and NMR LTCs across age; Y, young (mouse, 4‐month‐old; NMR, 3‐year‐old); O, old (mouse, 24‐month‐old); MA, middle‐aged (NMR, 11‐year‐old). Scale bar 10 µm.
• G, H
  Polarization quantification for G, mouse LT‐HSC (biological n = 3); and H, NMR LTC (Y, n = 5; MA, n = 4). Line at mean; error bars, SEM; P‐value determined by Sidak’s Two‐way ANOVA.

Metabolic paradigms of HSCs are their reliance on glycolysis and low mitochondrial activity (Simsek et al, 2010; Vannini et al, 2016). We measured mitochondrial respiration and glycolysis in sorted HSPCs from three species by the Seahorse assay (Fig 7B, Appendix Fig S12H). We found that human long‐term (LT) HSCs and LTCs had the lowest metabolic profile, resembling quiescent cells. Mouse cells showed 2–4‐fold higher respiration (OCR) and glycolysis‐driven acidification than their human and NMR counterparts, suggesting that quiescent mouse LT‐HSCs have a higher basal metabolic rate. The mitochondrial mass of young mouse and NMR HSPCs was similar, whereas human HSPCs had less mitochondria (Fig 7C, Appendix Fig S12I). The mitochondrial membrane potential (MMP) as a resultant of OXPHOS and FAO is an indicator of mitochondrial activity. We reproduced (Ho et al, 2017) in that old LSKs feature an increased fraction of cells with low MMP compared to young LSKs (Appendix Fig S12J and K). Increased MMP in LTCs compared to LSKs was shown using Tetramethylrhodamine (TMRE) sequestration (Fig 7D). Superoxide levels in LTCs were lower than those of LSKs (Appendix Fig S12L). In line with Seahorse metabolic profiles, intracellular reactive oxygen species (ROS) levels were higher in LSKs than in LTCs and human HSCs (Appendix Fig S12M).

A key feature of homeostatic young adult HSCs is cell polarity governed by CDC42 (Florian & Geiger, 2010). The small Rho‐GTPase regulating cytoskeletal organization is elevated in and de‐polarizes old mouse HSCs (Fig 7E), and pharmacological CDC42 inhibition re‐polarized and rejuvenated aged HSCs (Florian et al, 2012). Importantly, CDC42 activity increases with age in human blood cells and is linked to cardiovascular disease and myocardial infarction in the elderly (Florian et al, 2017). Moreover, treatment with the GTPase inhibitor CASIN resulted in CDC42 re‐polarization of aged human HSCs (Amoah et al, 2022). Using confocal 3D imaging, we reproduced CDC42 and Tubulin polarity in young LT‐HSCs (Fig 7F) as reported in Florian et al (2018), showing a 2.4‐fold decrease in CDC42 polar HSCs in 24‐month old animals (Fig 7G). CDC42 was detected well over background in LTCs, although a polarity pattern was less emergent than in mouse cells, where also the LSK fraction showed clear polarity (Appendix Fig S13A–C). We however observed a strong polar distribution of Tubulin in NMR LTCs (Fig 7F), which displayed a higher fraction of Tubulin polarization than mouse LT‐HSCs and intriguingly, a significant increase of polar cells in middle‐aged (MA) compared to young LTCs (Fig 7H). It is possible that rather than CDC42, one or more of its homologs play the polarity orchestrating role in NMRs, warranting further investigation of stem cell polarity patterns in this longevity model organism. We conclude that the strong signals of Tubulin polarization in LTCs confirm the identity of NMR HSPCs and are retained up to 12 years of age.

Taken together, these data suggest that NMR HSPCs have evolved a mechanism of stem cell homeostasis involving elevated MMP and ALDH activity while maintaining very low respiration rates. OXPHOS through FAO is more energy‐efficient and prevents lactate‐caused cytoplasmic acidification, which could contribute to preservation of quiescence and tissue homeostasis during aging. Furthermore, LTCs feature HSC polarity for Tubulin, which does not de‐polarize in middle‐aged animals, pointing toward increased resilience against aging phenotypes in NMR HSPCs.

Naked mole‐rat bone marrow features slow cell cycling and hypersensitivity to 5‐FU

Within NMR HSPCs, a signature of actively cycling cells was enriched for CP1 and depleted for LTC (Fig 4C). Pyronin Y staining confirmed more G₀ LTCs than CP1 (1.5‐fold; Fig 8B). Cell cycle scoring of BM scRNA‐Seq clusters revealed 2‐fold more mouse HSPCs in S phase compared to NMR HSPCs (Fig 8A). Conversely, Ki67 staining showed a higher LTC G₀ fraction as compared to CP1 (3.4‐fold) and mouse HSPCs (Fig 8B). Next, we performed Dual‐Pulse labeling (Akinduro et al, 2018) by successive injection of EdU and BrdU to compare cell cycle kinetics in vivo (Fig 8C). Using the EdU label together with DNA content staining, we found a 3‐fold increase in S‐Phase LSKs over LTCs (P < 10⁻⁴; Fig 8D, Appendix Fig S14A and B). Committed progenitors of either species did not differ in their cell cycle properties. Combined subsequent use of 2 label incorporations allows quantitation of the cells entering S phase, excluding cells retaining the 1st label to purify cells in early S phase via the 2^nd label (Weber et al, 2014). Cells in early S phase incorporate only BrdU (EdU⁻BrdU⁺). Cells in mid/late S phase are at DNA synthesis during both label administrations (EdU⁺BrdU⁺). Cells post S phase between the two labels incorporated only the first label (EdU⁺BrdU⁻). As expected, the CP3/4 progenitor partition showed markedly more mid/late and post S phase cells than the LTC HSPC compartment (Fig 8E). Accordingly, the same pattern can be seen for mouse myeloid progenitor LKs versus LSK HSPCs. However, virtually all NMR cells did not show EdU⁻BrdU⁺ early S phase cells (Appendix Fig S14C and D). We thus conclude that the S/G₂/M period in NMR cells extends beyond the typical 4 h in mice (Dowling et al, 2014). Consequently, even the highly proliferative CP3/4 fraction did not feature early S cells with the 2 h between‐label interval as compared to LKs (Fig 8F), thus showing prolonged G₁‐S progression in NMRs.

Figure 8. Slow cell cycle in naked mole‐rat HSPCs.

1. Seurat (Satija et al, 2015) cell cycle scoring of mouse [top] and naked mole‐rat [bottom] whole BM scRNA‐Seq.
2. Cell cycle staining with Ki67 [left] (CP1/LTC, biological n = 12); young mouse BM (n = 4) was used for LK and LSK. Error bars, SD; P‐value determined by Tukey’s Two‐way ANOVA. Note that CP1 cells do not differ in any cell cycle stage from LKs. Cell cycle staining with Pyronin Y [right] (biological n = 6). Error bars, SD; P‐value determined by Sidak’s Two‐way ANOVA.
3. Dual‐Pulse labeling in vivo with 3–4 month‐old mice (n = 7) or 2–3‐year‐old NMRs (n = 5), see D‐E. EdU, 5‐ethynyl‐2′‐deoxyuridine; BrdU, 5‐bromo‐2´‐deoxyuridine.
4. Cell cycle analysis of BM populations by EdU versus DNA‐content. LK, LIN^–/Sca‐1^–/Kit⁺; CP3/4, LIN^–/Thy1.1^lo/–/CD34⁺. Error bars, SD; P‐values obtained from Tukey’s Two‐way ANOVA.
5. Dual‐Pulse analysis of BM populations by EdU versus BrdU. Error bars, SD; p‐values obtained from Tukey’s Two‐way ANOVA.
6. Representative quantitation of merged per‐species Dual‐Pulse measurements for mouse LK [top] and naked mole‐rat CP3/4 [bottom].
7. 5‐FU administration into 6‐month‐old mice (n = 4, bodyweight 25 ± 2 g) or 2–3‐year‐old NMRs (n = 5, bodyweight 29 ± 4 g); i.p., intraperitoneal; PS, probability of survival.
8. Naked mole‐rat BM untreated or treated with 5‐FU (n = 5), no LIN gating.

Using the bulk transcriptomes of mouse and human HSCs, we compared their expression of a curated list of dormancy genes from (Cabezas‐Wallscheid et al, 2017) to NMR LTCs and CP1 (Appendix Fig S15A). Unsupervised hierarchical clustering grouped LTCs with human HSCs and CP1 with murine HSCs, in line with the cell cycle activities above. Expression of major HSC quiescence genes CDKN2A/p16 and CDKN2B/p15 was below the detection limit for all three species, while CDKN2C/p18 was upregulated in CP1 and mouse HSCs compared to LTC and human HSCs. CDKN1A/p21 expression was not different between groups, whereas CDKN1B/p27 was again downregulated in LTCs and humans HSCs; however, CDKN1C/p57 was overexpressed in CP1 and LTCs. Interestingly, NMR cells showed the lowest expression levels of CDCD42, CLK1, and MECOM. For genes upregulated in activated HSCs, CDK6 showed highest expression in mouse, followed by NMR and human cells, whereas PF4 was not detected in NMR cells and overexpressed in mouse HSCs; however, NMR cells had highest transcript levels of MYC, MCM7, and CDK1 (Appendix Fig S15B). Sorted NMR BM single‐cell transcriptomes revealed widespread expression profiles for CDC42 and CDK6 in erythroid, lymphoid, and myeloid progenitors, while MECOM expression was specific to HSPCs (Fig EV2E). The dormancy genes described by Cabezas‐Wallscheid et al (2017) and others, therefore show, to some degree, species‐specific regulatory circuits to maintain the corresponding cell cycle activities and quiescence patterns of human, murine, and NMR HSCs.

Given their unique quiescence properties, we aimed to engage NMR HSPCs through 5‐Fluorouracil (5‐FU), which eliminates cycling hematopoietic cells and activates the dormant HSC fraction to repopulate the BM (Lerner & Harrison, 1990). We treated NMRs with 150 mg/kg 5‐FU, a dose causing sublethal myeloablation in mice. In same‐sized NMRs however, this dose led to 100% mortality before day 15 post administration (Fig 8G). At day 9 when BM is almost completely reconstituted in mice (Venezia et al, 2004), the entire CD34⁺ compartment in NMRs was lost leaving an aberrant LIN⁺/Thy1.1^hi/CD34^hi marrow GCP fraction (Fig 8H). In terminal anemic animals, erythroid Thy1.1^–/lo/CD34⁺ fractions in the spleen were not regenerated (Appendix Fig S14E), strongly supporting that LTCs, which are not restored upon ablation, contain bona fide HSCs. LTCs displayed a stronger Rhodamine 123 (Rho) efflux compared to LSKs, which functionally enriches human and mouse HSCs (Uchida et al, 1996, 2003) (Appendix Fig S14F), indicating that the sensitivity to 5‐FU was not caused by impaired drug transporter systems.

Overall, by comparing HSPC single‐cell transcriptomes from three species and using in vivo pulse labeling, we demonstrate that NMR HSPCs contain a higher fraction of mitotically inactive or quiescent cells than mice, and prolonged duration of the cell cycle is common to all NMR hematopoietic cells. Moreover, NMR BM and spleen are completely myeloablated through 5‐FU dosing sublethal in mice.

Naked mole‐rat hematopoietic stem and progenitor cells do not respond to poly(I:C) and are activated through irradiation and fighting‐related injury

We next aimed to activate NMR HSPCs through a nontoxic stimulus, and opted for polyinosinic‐polycytidylic acid or poly(I:C), a double‐stranded RNA mimetic and TLR3 ligand inducing type I Interferon (IFN) signaling. In response to IFNα through its cognate receptor IFNAR1, HSCs exit dormancy and become activated (Essers et al, 2009), while the transcriptional suppressor of type I IFN signaling IRF2 prevents stem cell exhaustion (Sato et al, 2009). We setup an experiment combining poly(I:C) stimulation with the dual‐pulse labeling, which we extended to 4 h total pulse length to obtain more NMR cells in the EdU^–/BrdU⁺ fraction (Fig 9A). As expected, upon poly(I:C) treatment mouse BM LSKs were expanded by 2.7‐fold (Fig 9B and C). However, no NMR HSPC population in BM was significantly changed by poly(I:C) (Fig 9D and E). Murine HSPC activation was accompanied by stark increase of EdU^–/BrdU⁺ and EdU⁺/BrdU^– fractions, which concurrently diminished unlabeled cells, in both LKs and LSKs (Figs 9F and G, and EV5A). In contrast, no pulse‐label population was significantly altered in NMR LTCs, although a trend could be seen toward increased EdU^–/BrdU⁺ cells in early S‐Phase (Fig 9H and I). Likewise, CP3/4 increased the EdU^–/BrdU⁺ population from 1.6 ± 0.8% to 4.8 ± 2.9% post poly(I:C), and CP1 elevated this fraction from 1.8 ± 0.8% to 5 ± 3.1% (Fig EV5B–D), yet it was not statistically significant (Fig 9I). Clearly, the efficacy of poly(I:C) in activating NMR HSPCs was close to the lower detection limit in our assay, whereas mouse HSPCs were overtly responsive. By analyzing the mouse BM scRNA‐Seq data, we observed that the Kit⁺ HSPC cluster, MLPs T cells and monocytes overexpressed the IFNα receptor, and other cell types ranged between moderate and low IFNAR1 levels (Fig EV5E–G). Strikingly, in the NMR BM scRNA‐Seq dataset IFNAR1 was not detected in CD34+ HSPCs, but spuriously expressed in NMR GCs and MOs (Fig EV5H–J). Thus the lack of IFN‐induced activation is largely caused by very low levels of IFNAR1 on NMR HSPCs, however a slight increase in cycling cells of LTC, CP1 and CP3/4 subsets upon poly(I:C) treatment suggests that the conserved TLR3‐IFN axis is modified or less responsive in the NMR HSPC compartment, rather than completely absent.

Figure 9. Response to poly(I:C) stimulation in mouse and naked mole‐rat BM.

1. Experimental outline: Animals received 5 mg/kg poly(I:C) or H₂O in physiological saline solution for i.p. injection. 20 h later animals were injected with 1 mg EdU, another 3 h later animals were given 2 mg BrdU, finally 1 h later animals were culled for tissue extraction, with total poly(I:C) exposure time of 24 h.
2. FACS gating of 3–4‐month‐old mouse BM LIN^– HSPC fractions upon poly(I:C) stimulation.
3. Quantification of mouse BM HSPC fractions (biological n = 3); P‐value was determined by Sidak’s Two‐way ANOVA.
4. FACS gating of 2–4‐year‐old NMR BM LIN^– HSPC fractions upon poly(I:C) stimulation.
5. Quantification of NMR BM HSPC fractions (H₂O, biological n = 3; poly(I:C), n = 4); P‐value was determined by Sidak’s Two‐way ANOVA.
6. Representative gating of merged per‐condition Dual‐Pulse measurements for mouse BM LSKs.
7. Dual‐Pulse analysis of mouse BM HSPC populations by EdU versus BrdU (biological n = 3). Error bars, SD; P‐values obtained from Tukey’s Two‐way ANOVA.
8. Representative gating of merged per‐condition Dual‐Pulse measurements for NMR BM LTCs.
9. Dual‐Pulse analysis of NMR BM HSPC populations by EdU versus BrdU (H₂O, biological n = 3; poly(I:C), n = 4). Error bars, SD; P‐values obtained from Tukey’s Two‐way ANOVA.
10. NMR BM FACS of 2 healthy, nonaggressive animals; no LIN marking, gated population contains CP1/LTC.
11. NMR BM FACS of 2 animals involved in infighting from the same colony.

Figure EV5. Poly(I:C) response and IFNAR1 expression in mouse and NMR progenitors.

For each species individual sample data was combined to one merged dataset to illustrate gating and population patterns; mouse, n = 3; NMR H₂O, n = 3; NMR poly(I:C), n = 4.

• A
  Gating strategy for EdU/BrdU analysis of mouse LKs (LIN^–/Kit⁺/Sca‐1^–).
• B
  Comparison of CP1 frequency between H2O‐ and poly(I:C)‐treated NMRs.
• C, D
  Gating strategy for NMR BM EdU/BrdU analysis of C, combined CP3/4 or D, CP1.
• E–J
  UMAP analysis as shown in Fig 6B for E, mouse and H, NMR BM; HSPC clusters circled in black. UMAP‐based hexbin projection of mouse BM for F, KIT and G, IFNAR1 or NMR BM showing expression of I, CD34 and J, IFNAR1; scaled expression as probability for each conserved gene.

The major cause of death for NMRs from laboratory colonies is infighting due to challenging the breeding hierarchy. Here, we provide a case report highlighting the response to acute body injury in BM and PB. A moribund 7.2‐year‐old female covered with deep biting marks lost 5‐fold of its marrow cellularity compared to uninjured same‐aged animals (Fig 9J and K). Two days later a heavily wounded 7.5‐year‐old male was euthanized from the same colony, which stopped aggressions. Both animals showed several fold increases of Thy1.1^int/CD34⁺ cells comprising LTC/CP1 populations in BM. Interestingly, the mortally wounded female animal showed a very strong CP1/LTC increase and a TC expansion in the periphery (Appendix Fig S16A and B), whereas the male animal, euthanized due to unhealed wounds and weight loss, displayed neutrophil hyperproliferation and a reduction of the lymphoid compartment, a common stress reaction in NMR blood as shown below.

The first in vivo assay for stem‐cell function was based on the rescue of lethal irradiation by bone marrow transplantation (JACOBSON et al, 1951). Moreover, radiation is commonly used as a pre‐conditioning regimen for stem cell transplantations (Pinnix, 2019). Exposure to moderate doses of total body irradiation (TBI) causes long‐term hematopoietic injury (Testa et al, 1985). In mice, sublethal TBI dosed at 6.5 Gy results in rapid decrease of BM‐WBCs, LSKs, and their CFC, with the nadir of LSK frequency observed at day 3 post irradiation (Wang et al, 2006). Here, we irradiated NMRs at 5 Gy and quantified their stem and progenitor frequencies in BM (Fig 10A). The entire CD34⁺ compartment was strongly decreased in animals receiving TBI, specifically Thy1.1^int/CD34⁺ cells comprising LTC/CP1 were reduced by 3.1‐fold, and Thy1.1^–/CD34⁺ erythroid precursors CP4/5 diminished by 6.6‐fold compared to controls (Fig 10B). However, using AnnexinV staining we only saw a limited increase of dead AnnexinV⁺/PI⁺ cells in BM, whereas spleen and cervical thymus showed significantly more dead and fewer viable cells (Fig 10C, Appendix Fig S16C). Likewise, we detected decreased thymic in CD34⁺ ETPs and CD4^–/CD8^– TC precursors (Appendix Fig S16D) (Emmrich et al, 2021). BM apoptosis in response to TBI could be less severe in NMRs compared to mice since the majority of erythropoiesis takes place in spleens (Fig 3A). At 30 days post TBI, mouse LSKs recovered to 20% abundance of uninjured LSK levels, the low HSPC frequencies persisting as long‐term residual BM injury (Shao et al, 2014). Surprisingly, the CP1/LTC combined HSPC compartment was highly elevated, 3.7‐fold upregulated in BM and 10.4‐fold in spleens (Fig 10D and E, Appendix Fig S16E). Moreover, PB CD34⁺ cells were increased by 12.3‐fold to 1.8 ± 0.7% of total WBCs (Fig 10F, Appendix Fig S16F), accompanied by sustained loss of TCs and BCs together with prevalent GC expansion 4 weeks post TBI (Fig 10G, Appendix Fig S16G). These results suggest that myeloid skewing as functional consequence of irradiated HSCs is conserved in NMRs, however, this is accompanied by an overcompensating increase in HSPC abundance unique to NMRs.

Figure 10. Response to irradiation of naked mole‐rat HSPCs.

1. Marrow FACS gating 48 h post sham‐ [no TBI] and 5 Gy irradiated NMRs. No LIN gate was applied; Thy1.1^int/CD34⁺, contains CP1/LTC; Thy1.1^–/CD34⁺, contains CP4/5.
2. Quantification of indicated BM fractions 48 h post TBI; P‐value was obtained from Sidak’s Two‐way ANOVA.
3. AnnexinV staining quantification across indicated NMR tissues of sham‐irradiated (n = 3) or 5 Gy irradiated (n = 4) animals. AnnexinV^–/PI^–, viable cells; AnnexinV⁺/PI^–, pro‐apoptotic cells; AnnexinV⁺/PI⁺, dead cells. Error bars, SD; P‐value derived from Two‐stage step‐up method of Benjamini, Krieger, and Yekutieli Two‐way ANOVA.
4. Marrow FACS gating 30 days post sham‐ [no TBI] and 5 Gy irradiated NMRs. No LIN gate was applied; Thy1.1^int/CD34⁺, contains CP1/LTC.
5. Quantification of HSPCs 30 days post TBI; P‐value was obtained from Sidak’s Two‐way ANOVA.
6. Quantitation of CD34⁺ WBCs; P‐value determined by Welch’s t‐test.
7. PB WBC subset quantitation according to Fig 1B on PB collected from sham‐irradiated NMRs (n = 6) and animals 30 days post TBI (5 Gy, n = 5); P‐value determined by Sidak’s Two‐way ANOVA.
8. Summary of the main findings including comparisons of blood and marrow compositions, sites of TC maturation (Emmrich et al, 2021) and erythropoiesis, HSPC markers and phenotypes between mouse, human, and NMR. Cell type proportion in bone marrow of mouse and NMR based on Fig 6B.

In conclusion, IFN‐mediated activation of NMR cells is largely ineffective due to low levels of IFNAR1 in HSPCs, which may be related to a reduced inflammatory signature. Severe BM injury and blood loss due to infighting caused an acute HSPC expansion, whereas recovery from sublethal TBI caused a persistent expansion of the NMR HSPC compartment.

Discussion

NMRs are the longest‐lived rodents that remain healthy until the end of their lives. Adult stem cells are essential for maintenance and repair of tissues, thus NMR stem cell biology is of immediate interest to biomedical research.

Here, we present a comprehensive analysis of the blood system in > 100 NMRs including functional and molecular characterization of stem and progenitor subtypes and map a landscape of the hematopoietic hierarchy. While NMRs are rodents and thus phylogenetically closer to mice than to humans, a common trait NMRs share with humans is longevity. Surprisingly, many characteristics of the NMR hematopoietic system showed higher similarity to humans than to mice (Fig 10H).

It had been proposed that long‐lived NMRs, as well as humans, display neotenic traits compared to their short‐lived relatives (Skulachev et al, 2017). Neoteny is the preservation of juvenile characteristics in adulthood (Bufill et al, 2011). Species displaying neoteny are the Axolotl which remains in its highly regenerative larval stage unless ambient water supply ceases (Safi et al, 2004), the cave‐dwelling Olm which never leaves its larval stage and is predicted to live > 175 years (Voituron et al, 2011), and “immortal jellyfish” of the genus Turritopsis which can revert from a sexually mature medusa stage into the budding polyp (Bavestrello et al, 1992). Humans are considered to be neotenic apes due to traits such as orthognathy, near absence of body hair, high relative brain weight, and prolonged growth periodNeotenic traits have also been described for NMR including small size, lack of hair, and slow brain maturation reviewed in (Skulachev et al, 2017). Splenic erythropoiesis and expansion of medullary granulopoiesis along with compression of the BC compartment are clear neotenic traits seen in NMRs. Neoteny has been linked to longevity in Axolotl, Olm, and human (Skulachev et al, 2017). Hence, the multiple neotenic features of the hematopoietic system we identified are likely related to NMR longevity.

A peculiar phenomenon our study uncovered is the TC pool composition in adult NMRs throughout all hematopoietic tissues. Thymic TC development revealed fewer CD4^–/CD8^– double‐negative progenitors than in mice, conversely NMR thymi contain more mature TCs (Emmrich et al, 2021). Here, we found an unusually high CD4:CD8 TC ratio in blood, marrow, and lymph nodes, also seen in NMR thymi as we reported earlier, and evident in NMR spleens (Hilton et al, 2019). We further revealed very low IFNAR1 expression in NMR BM cells, leading to almost no proliferative response in HSPCs upon poly(I:C) stimulation. NMRs are extremely susceptible to Herpes Simplex Virus infections (Artwohl et al, 2009) and Coronavirus infections (Ross‐Gillespie et al, 2007). The low HSPC IFN signaling together with largely diminished cytotoxic CD8‐TCs may explain the insufficient anti‐viral response of NMRs. We also showed that total thymic cellularity was markedly lower in NMRs compared to mice (Emmrich et al, 2021). NMRs may not have evolved to deal with viral challenges because subterranean lifestyle with limited contact outside the native colony limits viral spread. Indeed, as reported earlier by (Hilton et al, 2019), the splenic myeloid compartment is enlarged alongside a compressed BC lineage, with complete absence of NK cells. We found the same pattern in NMR BM, suggesting a reduced reliance on both humoral and cell‐mediated adaptive immune responses toward a prevalence of innate immune effector cells. The intriguing implication of diminished antiviral responses is a reduction in interferon, TNFα and potentially retrotransposon‐mediated pro‐inflammatory signaling, limiting chronic sterile inflammation associated with age. While inflammaging is considered a ubiquitous feature of aging in mice and humans, we found several inflammaging‐related pathways to be down‐regulated in aged NMR HSPCs. It would be important to further investigate the mechanisms leading to reduced inflammatory signature aged NMR.

A striking feature of NMR stem cells was a higher proportion of cells in quiescence. The dynamic equilibrium between quiescence and specific cell cycle kinetics is a hallmark of adult stem cells, hence mouse HSCs have been shown to contain a dormant fraction of ~20% (Foudi et al, 2009). Remarkably we found no significant differences in the frequencies of LTCs, LSKs, and human CD34⁺/CD38^lo HSPCs in normal young adult BM (Appendix Fig S12C), suggesting conserved stem cell pool frequencies. Our data suggest an expanded quiescent HSC pool in NMRs and rigid control of cell cycle genes at the transcriptional level (Fig EV1H, Appendix Figs S1D and S2D). The enlarged quiescent HSC pool would benefit longevity by minimizing damage to stem cells and decelerating clonal expansion, which is a key feature of an aged hematopoietic system (Zink et al, 2017). Strikingly, LTCs comprised less actively cycling cells than CP1 (Fig 8B), which was also evident from pulse‐labeling (Fig 9I). While CP1 and LTC are capable of B‐lymphoid commitment in vivo and in vitro, xenograft CITE‐CD11b levels demonstrate lack of CP1 to form CD11b^lo LTCs (Fig 5O). This supports the conclusion that LTCs transition into CD11b⁺/Thy1.1^int/CD34^hi myeloid‐primed CP1 with less lymphoid and erythroid potential than LTCs (Figs 4F and G, and 5L–N, Appendix Figs S8D and S9K and L). We note that colony assays and liquid cultures with human cytokines as well as the humanized mouse models used here, are limited in their capacity as functional readouts for NMR HSC self‐renewal and multipotency. The optimal test would be transplantations into NMR hosts. However, intravenous injections are complicated due to absence of a sizeable tail vein, and the retro‐orbital plexus behind the NMR degenerated eye is inaccessible without surgery. Intrafemural delivery is challenged by thicker and stronger ossification and the tight medullary cavity. Importantly, NMRs are not inbred, thus the allografts would require immunosuppression and pre‐conditioning regimen. Our 5‐FU and TBI experiments demonstrate the sensitivity of NMRs to BM injury, whether this is HSC intrinsic or caused by the BM microenvironment remains to be established.

FACS quantitation revealed > 4‐fold increase of CP1 over LTCs in marrow and > 3‐fold in spleen (Fig 3B–F), underscoring a myeloid differentiation bias in NMR hematopoiesis. BM scRNA‐Seq revealed 2.5‐fold increase of HSPCs in 11‐year‐old versus 3‐year‐old NMRs. The common HSPC cluster comprised LTC/CP1/CP3, and we showed that myeloid CP1 progenitors and LTCs increase with age. However, we observed no age‐associated increase in NMR PB‐WBCs. The aging phenotype of NMR HSPCs was less evident than in mice, that is, transcriptomic aging clocks showed a significant tAge increase in 3‐month‐old versus 12‐month‐old mouse BM, whereas NMR PB and BM with an 8‐year age range did not show tAge increase. In conjunction, we found GSEA enrichment of pro‐mitotic pathways in older mouse BM and NMR HSPCs. However, several inflammatory signaling modules were downregulated and cytoprotective pathways such as DNA Repair were upregulated in aged NMR HSPCs. Moreover, stem cell polarity defined by Tubulin did not disperse with age in NMRs. Hence, the age‐associated HSPC expansion is uncoupled from multiple other HSC aging phenotypes in NMRs, suggesting a compound mechanism that delays age‐related decline in stem cell function. Similarly, the megakaryocytic differentiation bias, a common hallmark of aging hematopoietic lineage trajectories (Sanjuan‐Pla et al, 2013), was not as evident in 11‐year‐old NMRs as in 12‐month‐old mice (Fig 6G and H). Consequently, platelets were elevated significantly in aged NMRs. Maintenance of youthful effector cell compositions despite HSPC expansion in middle‐aged animals could be a direct result of an enlarged and better preserved quiescent HSPC fraction in concert with a prolonged cell cycle, delaying peripheral manifestation of clonal hematopoiesis.

In summary, the entire hematopoietic system of NMRs evolved a combination of unique or neotenic adaptations to an extended healthspan and lifespan, such as diminished platelets and delay of age‐associated leukocytosis, active hematopoiesis in the spleen, and as described earlier, additional cervical thymi and absence of thymic involution (Emmrich et al, 2021). Furthermore, NMR hematopoietic cells feature a slower G₁‐S‐transition, stem and progenitors are less metabolically active than those from short‐lived mice and the HSC compartment contains a higher fraction of quiescent cells. NMRs have evolved extreme longevity and resistance to a range of age‐related diseases. Understanding the molecular mechanisms of these evolutionary adaptations can lead to novel strategies for improving human health. Our scRNA‐Seq resource is available to browse as an online data resource tool, together this tool set provides a platform for using NMRs as a research model in stem cell biology, immunology, inflammation, and the studies of systemic factors in aging.

Materials and Methods

Animals

All animal experiments were approved and performed in accordance with guidelines instructed by the University of Rochester Committee on Animal Resources with protocol numbers 2009‐054 (naked mole rat) and 2017‐033 (mouse). Naked mole rats were from the University of Rochester colonies, housing conditions as described (Ke et al, 2014). C57BL/6 mice were obtained from NIA. In comparative flow cytometric and polarity assays, young LSK were sorted from 3 to 4 month and old LSK from 25‐month‐old mice. For xenografts, the immunodeficient strain NSGS [NOD. Cg‐Prkdc^scid   Il2rg^tm1Wjl  Tg(CMV‐IL3,CSF2,KITLG) 1Eav/MloySzJ] was purchased from JAX.

Primary cell isolation

Marrow from mice and NMRs was extracted from femora, tibiae, humeri, iliaci, and vertebrae by crushing. Spleen, liver, thymus, and lymph nodes were minced over a 70‐µm strainer and resuspended in FACS buffer. Blood from mice was drawn via retroorbital capillary bleeding, NMR blood was obtained via heart puncture. Human marrow was obtained from the URMC Pathology and Laboratory Medicine in accordance to RSRB STUDY00006161. Human BM cell fractions shown in Fig 8 were based on Wintrobe’s monograph (Greer et al, 2019) and cross referenced with Osgood and Seaman (1944); human marrow HSC fraction was approximated accordingly (Pang et al, 2011).

Hematology analyzer

Peripheral blood (PB) parameters were measured with a Vet ABC Plus⁺ (scil) Analyzer. Specifically, NMR and mouse samples were measured with the “mouse_research” protocol (scil Tech Support, available upon request), which provides a 3‐part differential in 17 parameters.

Histology

Imaging and analysis was performed using a using a Nikon Eclipse Ti‐S microscope. Coverslips were applied with DEPEX Mounting media (Electron Microscopy Sciences), except for Alkaline Phosphatase staining where Vectashield HardSet Mounting Medium for Fluorescence (Vector) or Vectashield Vibrance Antifade Mounting Medium (Vector) was applied. Femur bones were decalcified with 14% EDTA for a minimum of 2 weeks and stored in 10% neutral buffered formalin. Soft tissues were stored in 10% neutral buffered formalin, processing was done using a Sakura Tissue‐Tek VIP 6 automated histoprocessor, paraffin embedding was done using a Sakura Tissue‐Tek TEC 5 paraffin embedding center. A Microm HM315 microtome was used to section tissues at a thickness of 5 µm, which then were floated onto a slide with a water bath at a temperature between 45 and 55°C. Sections were deparaffinized and rehydrated to distilled water through xylene and graded ethanol (100–70%).

May‐Grünwald‐Giemsa

Cytospins of whole spleen or WBM or sorted cells were prepared using a Rotofix 32A (Hettich) and stained at room temperature with May‐Grünwald solution (Sigma) for 5 min, washed in phosphate buffer pH 7.2 (Sigma) for 1.5 min, and counterstained in 4.8% Modified Giemsa (Sigma) for 13–15 min.

Alkaline phosphatase

Cytospins were stained with the Alkaline Phosphatase kit (Sigma). Cytospins of sorted PB cells were stained according to manufacturer’s instructions with the exception of combining FBB‐Alkaline Solution with Hematoxylin Solution, Gill No. 3 as counterstain. Cytospins of picked colonies were stained according to manufacturer’s instructions using the suggested stain‐counterstain combination of FRV‐Alkaline Solution with Hematoxylin Solution, Gill No. 3.

Benzidine mayer’s hematoxylin

Slides were fixed at room temperature with methanol for 30 s, incubated with o‐Dianisidine (Sigma) 1% in methanol for 1 min and stained with H₂O₂ 2.5% in ethanol for 30 s before rinsing for 15 s in water and counterstaining with Mayer’s Hematoxylin Solution (Sigma) for 2 min.

Wright Giemsa

Blood films were incubated with Wright‐Giemsa Stain (Electron Microscopy Sciences) for 1 min, rinsed briefly with water, and developed in phosphate buffer pH 7.2 for 2 min, then rinsed again. Slides were scored by taking three random micrographs of monolayers from the feathered edge of each sample to count both RBCs and platelets and average the technical replicates. Then mean RBC levels from the bloodcounter measurements (mouse, 9.1e¹²/l; naked mole‐rat 5.4e¹²/l) were used to convert PLT/RBC ratios to a volumetric PLT count via bloodcounter by

$$\frac{(RBC\times {10}^{12})\times (PL{T}_{\mathit{count}})}{l\times (RB{C}_{\mathit{count}})}$$

Hematoxylin and Eosin

Sections were stained with Mayers Hematoxylin (Sigma) for 1 min and washed with tap water to remove excess blue coloring. Soft tissue sections were further decolorized with three dips in 0.5% acid alcohol and washed in distilled water. The nuclei of sections were blued in 1× PBS for 1 min and washed again in distilled water. An Alcoholic‐Eosin counterstain was applied for 30 s before slides were immediately dehydrated and cleared through three changes of 95% ethanol, two changes of 100% ethanol, and three changes of Xylene for 1 min each.

Microwave Giemsa for plastic marrow sections

Paraffin‐embedded femora were subjected to the microwave modification of a conventional Giemsa stain, which we found to produce clearer contrast of megakaryocytic cells as distinguished by their pale purple cytoplasm and abundant nuclear chromatin staining due to polyploidy. The stain was performed as described in www.urmc.rochester.edu/urmc‐labs/pathology. Slides were scored by taking three random micrographs of marrow from medullary canal for each sample to count polyploidy giant Megakaryocytes, average the technical replicates, and convert micrograph pixel size via magnification to bone area in mm².

Measurement of red blood cell size

Whole PB cytospins were stained with Wright Giemsa and imaged. RBCs in the image with high circularity were manually highlighted in a photo‐editing software (paint.net) that counted the number of pixels enclosed in the highlighted region. Image resolution was normalized as necessary to a standard size of 1,636 × 1,088 pixels. The average diameter for each RBC was calculated by $ter=2\ast \sqrt{\frac{surfacearea}{\pi }}$. The RBC diameter was then converted from pixels to microns using a scale factor encoded in the metadata of calibrated images taken at 60× and 100× with a resolution of 1,636 × 1,088. For 100×, 113,611 pixels equates to 1 µm and for 60×, 68,167 pixels equates to 1 µm. A total of 400 red blood cells from eight mice and 366 red blood cells from 13 NMRs were measured.

Methylcellulose colony assays

Bulk colony assays

Fresh sorted or whole BM NMR cells were tested to grow in mouse (M3434, SCT), rat (R3774, SCT), or human (HSC005, RnD Systems) methylcellulose formulations to show the highest colony numbers, colony sizes, and cell viability with human cytokine cocktails. Either 1 × 10⁴ whole marrow or 1 × 10³ sorted NMR cells were added to 3 ml of HSC005 supplemented with 1% Penicillin/Streptomycin and 1× GlutaMAX (both Thermo Fisher), equally divided into two 35 mm dishes, grown for 21 days at 32°C, 5% CO_2, and 3% O_2, and scored. Although hematopoietic NMR cells will grow at 37°C, the total number as well as colony and cell type diversity is strongly enhanced at 32°C (data not shown). Colony assays grown at 37°C give rise to two types of colonies (erythroid versus myeloid), which are notably smaller than each of the 4 colony types we can distinguish at 32°C (Appendix Fig S8A). Replatings were done by resuspending scored dishes at day 21 in FACS buffer, counting cells and replating 1 × 10⁴ cells into the above‐mentioned growth conditions. CP3 cells did not grow substantially in the first replating, and a second replating showed no sizable colonies for all NMR HSPC types. Benzidine staining of NMR methylcellulose assays was done as described (Murphy, 1978). Briefly, a 0.2% benzidine dihydrochloride (Sigma) solution in 0.5 M acetic acid was prepared, which was supplemented with 0.2% of 50% H₂O₂. One milliliter of this solution was layered carefully over each dish, and after 5 min, colonies were scored for the proportion of colonies which are uniformly benzidine‐unreactive (color‐less), uniformly benzidine‐reactive (blue), and colonies containing both reactive and unreactive cells (mixed colonies containing both differentiated, hemoglobin containing, and nonerythroid cells).

Single cell colony assays

For each 96‐well plate all outer rim wells were filled with 2× Antibiotic‐Antimycotic (Gibco, Thermo Fisher) in 200 µl sterile H₂O to support humidification of the inner 60 wells filled with 100 µl methylcellulose medium. Frozen BM was sorted for mice into Mouse Methylcellulose Enriched Media (HSC007, RnD Systems). For NMRs, frozen BM was sorted into Human Methocult Enriched (H4435, SCT) for myelo‐erythroid differentiation and CFC analysis, or sorted into Human Metylcellulose Base Media (HSC002, RnD Systems) supplemented with human recombinant FLT3L 25 ng/ml, SCF 25 ng/ml, and IL‐7 20 ng/ml (all Preprotech). Mouse colonies were grown at 37°C for 12 days before colony scoring and flow cytemetric analysis, NMR colonies were grown at 32°C for 18 days before imaging, scoring, FACS or cytochemistry or RNA‐extraction.

Seahorse assay

Sorting for human (Appendix Fig S12A), mouse (Appendix Fig S12B), and NMR (Fig 2A–C) marrow stem and progenitor population was used to purify live cells. Sorted cells were collected in Seahorse XF96 Cell Culture Microplates (Agilent Technologies) with 50–250 × 10³ cells per 200 µl of the following culture media: Human LT‐HSCs (LIN^–/CD38^lo/CD34⁺/CD45RA^–/CD90⁺) in StemSpan Serum‐free expansion medium (SFEM; Stemcell Technologies) supplemented with 100 ng/ml of human SCF, 100 ng/ml human of FLT3L, 20 ng/ml of human IL‐6, 50 ng/ml of human TPO (all Peprotech), and 0.75 µM Stemregenin (SR‐1; Stemcell Technologies Cat# 72342), modified as described (Genovese et al, 2014). Human MPPs (LIN^–/CD38^lo/CD34⁺/CD45RA^–/CD90^–) in StemSpan supplemented with 50 ng/ml of hSCF, 50 ng/ml of hFLT3L, 10 ng/ml of hIL‐3, 10 ng/ml of hIL‐6, 20 ng/ml of hTPO (all Peprotech), and 0.25% Chemically defined lipid concentrate (CDLC; ThermoFisher Cat# 11905031). Human oligopotent progenitors (hOPP; LIN^–/CD38^hi/CD34⁺) in RPMI with 10% FBS (both Gibco), 1% GlutaMAX^TM (Thermo Fisher), 5 ng/ml of hSCF, 5 ng/ml of hGM‐CSF, and 5 ng/ml of hIL‐3. Mouse LT‐HSCs (LIN^–/Sca‐1⁺/Kit⁺/CD48^–/SLAM⁺) in the long‐term HSC expansion cocktail (Wilkinson et al, 2020). Mouse MPPs (LIN^–/Sca‐1⁺/Kit⁺/CD48⁺/SLAM^–) in StemSpan with 1% GlutaMAX^TM, 10 ng/ml of mSCF, 20 ng/ml of mTPO (all Peprotech), 10 ng/ml of mFGF1 (all Peprotech), and 20 ng/ml of mIGF2 (BioLegend Cat# 588204). Mouse LIN^–/Kit⁺/Sca‐1^– (LK; mOPP) in StemSpan with 10% FBS, 10 ng/ml of mSCF, 10 ng/ml of mIL‐3, and 10 ng/ml of mIL‐6 (all Peprotech). NMR LTC (LIN^–/Thy1.1^int/CD34^hi; CP2) in StemSpan with 1% GlutaMAX^TM, 1% CDLC, 100 ng/ml of hSCF, 100 ng/ml of hFLT3L, 20 ng/ml of hIL‐6, 50 ng/ml of hTPO, 1 µM SR‐1, and 0.1 µM of UM‐171 (Selleck Chemicals Cat# S7608). NMR CP1 (LIN⁺/Thy1.1^int/CD34^hi) in StemSpan with 1% GlutaMAX^TM, 1% CDLC, 50 ng/ml of hSCF, 50 ng/ml oif hFLT3L, 20 ng/ml of hIL‐6, 10 ng/ml of hGM‐CSF, and 1 µM of UM‐729 (Stemcell Technologies Cat# 72332). NMR CP3 (LIN^–/Thy1.1^lo/CD34^hi; MEP) in StemSpan with 1% GlutaMAX^TM, 1% CDLC, 50 ng/ml of hSCF, 50 ng/ml of hTPO, 1 U/ml of hEPO (all Peprotech). All expansion cocktails were supplemented with 1% Penicillin‐Streptomycin (Thermo Fisher Cat# 15140163). Cells were allowed to settle for 16–20 h at 37°C (32°C for NMR), 5% CO_2, and 0.5% O₂. We used Corning Cell‐Tak Cell and Tissue Adhesive (Thermo Fisher Cat# CB‐40240) at 22.4 µg/ml concentration per well to prepare coated XF96 microplates according to the manufacturers’ guidelines (Agilent Technologies). Cells were seeded into the coated microplates immediately before the assay by centrifugation with 200 g for 1 min without brake. Subsequently, we strictly adhered to the Seahorse XF Cell Mito Stress Test Kit protocol (Agilent Technologies Cat# 103015‐100). Cells were counted before and after the assay using a Celigo S Image cytometer (Nexcelom Biosciences) with automated 96‐well Brightfield imaging at the URMC Flow core. All cells were assayed in Seahorse XF RPMI medium (Agilent Technologies Cat# 103681‐100). Final well concentrations were 1.5 µM Oligomycin, 1 µM FCCP, and 0.5 µM Rotenone/Antimycin A. Measurements were taken on a Seahorse XFe96 Analyzer in the URMC Flow core using Wave 2.6.1 software (Agilent Technologies).

NMR HSPC in vitro liquid culture

For BC differentiation 96‐well plates were coated with ICAM‐1 (#552906 BioLegend) using 5 µg/ml in PBS 0.1% BSA at 4°C over night. NMR sorted CP1 or LTCs were plated by a brief pulse spin to 300 g in weeks 1–2 medium: StemSpan SFEM (SCT) with 1% GlutaMAX^TM, 1% Penicillin‐Streptomycin (Thermo Fisher Cat# 15140163), 10% ESC FBS (#S10250 RnD Systems), 1% nonessential amino acids (Gibco), 0.25% CDLC, 25 ng/ml of hSCF, 25 ng/ml of hFLT3L, and 20 ng/ml of hIL‐6 (all Preprotech). Media was replaced every 3^rd day, from weeks 3 to 4 we used the same formulation except exchanging hIL‐6 to 20 ng/ml hIL‐7.

Confocal microscopy for stem cell polarity

Cells were sorted (HSC population in mice and LTC population in NMR) and seeded on fibronectin‐coated (Millipore Sigma, cat. F1141‐2MG) IBIDI chambers (IBIDI, cat. 80296). Cultures were incubated at corresponding temperatures (37°C for mouse and 32°C for NMR, 5% CO2, 3% O₂) for 32–34 h in IMDM plus 100 ng/ml TPO, G‐CSF, and SCF containing media. Cells were fixed with BD Cytofix Fixation Buffer (Fischer Scientific, cat. BDB554655) for 20 min at RT. Then, cells were gently washed with PBS. Permeabilization with 0.2% Triton X‐100 (Sigma‐Aldrich, cat. X100‐500 ml) in PBS was done at RT for 20 min. Solution was aspirated and blocking was performed with 10% Normal Goat Serum (Millipore Sigma, cat. S26‐100 ml) in PBS for 1 h at RT. Incubation with primary (Rabbit anti‐CDC42 antibody, Abcam, ab155940, 1:100 dilution; Rat anti‐tubulin antibody, Abcam, ab6160, 1:1,000 dilution) and secondary antibodies (Goat anti‐rabbit antibody, Alexa Fluor 568, A‐11011, 1:1,000 dilution; goat anti‐rat antibody, Alexa Fluor 647, A‐21247, 1:1,000 dilution) in blocking buffer for 1 h at room temperature. Cells were coated with ProLong Gold Antifade Reagent (Thermo Fischer Scientific, P36934) and imaged on Nikon C2 confocal system with 60× objective, 15–45 cells per group, biological n = 2–5. Polarization quantification was performed using Nikon NIS‐Elements AR software for intensity profile derivation. Briefly, cells were assessed in 3D rendition, pole axis was identified, and intensity profile line was delineated. Cell was separated into two halves along the polarization axis; ratio of intensity in polar part was derived by dividing signal intensity in polar half by total intensity in the whole cell. Polarization threshold was set at 0.65. Percentage of polarized cells per animal was derived, averages per group were used for statistical analysis using Graph‐Pad Prism 8.0.1.

Flow cytometry

Flow cytometry analysis was performed at the URMC Flow Core on a LSR II or LSRFortessa (both BD), or on our CytoFlex S (Beckman Coulter). Kaluza 2.1 (Beckman Coulter) was used for data analysis. Staining and measurement were done using standard protocols. Red blood cell lysis was done by resuspending marrow pellets in 4 ml, spleen pellets in 1 ml, and up to 500 µl of blood in 20 ml of RBC lysis buffer, prepared by dissolving 4.1 g NH₄Cl and 0.5 g KHCO₃ ^– into 500 ml of double‐distilled H₂O and adding 200 µl of 0.5 M EDTA. Marrow and spleen were incubated for 2 min on ice, blood was lysed for 30 min at room temperature. Cells were resuspended in FACS buffer (DPBS, 2 mM EDTA, 2% FBS [Gibco]) at 1 × 10⁷ cells/ml, antibodies were added at 1 µl/10⁷ cells, vortex‐mixed, and incubated for 30 min at 4°C in the dark. DAPI (Thermo Fisher) at a concentration of 1 µg/ml was used as viability stain. The primary gating path for all unfixed samples was: scatter‐gated WBC (FSC‐A versus SSC‐A) => singlets1 (SSC‐W versus SSC‐H) => singlets2 (FSC‐W versus FSC‐H) => viable cells (SSC versus DAPI) => proceed with specific markers/probes. Compensation was performed using fluorescence minus one (FMO) controls for each described panel. For antibody validation, we incubated 1 × 10⁶ cells in 100 µl of Cell Staining Buffer (BioLegend; Cat# 420201) and added 5 µl of Human TrueStain FcX™ and 0.5 µl TruStain FcX™ PLUS, followed by incubation for 10 min at 4°C. We then proceeded with fluorescent antibody staining as stated above. All clones, conjugates, isotypes, and CITE‐Seq antibodies are listed in Dataset EV1.

Immunophenotyping of NMR BM, spleen, thymus, PB, and lymph nodes: CD90 FITC; CD125 PE; Thy1.1 PE‐Cy7; CD34 APC, CD11b APC‐Cy7. Quantification of murine BM SLAM HSCs was performed using mouse LIN Pacific Blue; Sca‐1 BUV395; CD150 PE; Kit PE‐Cy7; CD48 APC‐Cy7. Quantification of human BM LT‐HSCs was performed using human LIN Pacific Blue; CD34 APC; CD38 APC‐Cy7; CD45RA FITC; CD90 PE‐Cy7. Fluorescence minus one (FMO) controls were applied for fluorescent spillover compensations for each species and tissue used. All antibodies can be found in Dataset EV1.

Sorting was performed at the URMC Flow Core on a FACSAria (BD) using an 85‐μm nozzle, staining was done as described. Human HSCs were sorted for population RNA‐Seq as LIN^–/CD34⁺/CD38^Lo/CD45RA^–/CD90^Dim (Appendix Fig S12A). NMR HSPC populations were sorted as described with a lineage cocktail comprised of CD11b, CD18, CD90, and CD125 (NMR LIN). NMR marrow and spleen sorting panel was: NMR LIN Pacific Blue; Thy1.1 PE‐Cy7; CD34 APC. NMR blood sorting panel was: Thy1.1 PE‐Cy7; CD11b APC‐Cy7.

Molecular probing was performed on frozen aliquots from mouse and NMR BM. For each probe, cells were diluted in 1 ml pre‐warmed DMEM⁺ at 1 × 10⁶ cells/ml. All stainings were performed simultaneously for 4–6 NMRs, 2–4 old mice, 2–4 young mice, and 4 human biological replicates. ALDEFLUOR (SCT) reagent was added at 0.5 µl/ml, mixed and incubated for 15 min at 37°C in a water bath. MitoStatus TMRE (BD) was added to 0.5 × 10⁶ cells/ml at 25 nM and incubated for 10 min at room temperature in the dark. FCCP (Trifluoromethoxy carbonylcyanide phenylhydrazone) was added to negative controls at 5 µM during TMRE staining. JC‐1 (Thermo Fisher) was added at 1 µM with or without 5 µM FCCP and incubated for 15 min at 37°C. MitoTracker Orange CMTMRos (Thermo Fisher) was added at 10 nM and incubated for 45 min at 37°C. MitoSOX red (Thermo Fisher) was added at 5 µM and incubated for 30 min at 37°C. CellROX Orange (Thermo Fisher) was added at 5 µM and incubated for 60 min at 37°C. The subsequent antibody staining was performed as above with 30 min incubation on ice, the panel was Sca‐1 (mouse) or CD34 (naked mole‐rat) APC; Kit (mouse) or Thy1.1 (naked mole‐rat) PE‐Cy7; Lineage Cocktail V450. Rhodamine 123 staining was performed by incubating 1 × 10⁶ cells for 30 min with 1 µg/ml Rho in HBSS⁺ (HBSS, 2% FBS, 10 mM HEPES; all Gibco) at 37°C, then cells were washed with 2 ml HBSS⁺, spun down and reincubated for 15 min at 37°C.

Pyronin Y staining

Mouse and NMR BM cells from frozen aliquots were count‐adjusted to 1 × 10⁶ cells/ml and resuspended into 1 ml of DMEM⁺ (DMEM high Glucose, 2% FBS, 10 mM HEPES; all Gibco). Upon addition of 50 μg/ml of Verapamil (Sigma) and 5 μM DyeCycle Violet (Thermo Fisher), cells were incubated for 45 min at 37°C in a water bath, vortex‐mixed every 15 min. Past 45 min, 0.1 μg/ml Pyronin Y was added to the reaction and incubated an additional 15 min at 37°C, then washed with 3 ml ice‐cold Staining buffer (HBSS [Gibco], 0.33 M HEPES, 3.5% FBS, 0.02% NaN₃ [Sigma]). A subsequent antibody staining was performed as above with incubation on ice, the panel was Sca‐1 (mouse) or CD34 (naked mole‐rat) APC; Kit (mouse) or Thy1.1 (naked mole‐rat) APC‐Cy7; Lineage Cocktail FITC; 500 nM SYTOX Green (Thermo Fisher) was used as viability stain.

Ki67 staining

Mouse and NMR BM cells from frozen aliquots were count‐adjusted to 1 × 10⁷ cells/ml and antibody staining was performed as described, panel was Sca‐1 (mouse) or CD34 (naked mole‐rat) APC; Kit (mouse) or Thy1.1 (naked mole‐rat) APC‐Cy7; Lineage Cocktail FITC. For fixation and permeabilization, we used the buffers from the BrdU Flow Kit (BD). Briefly, cells were fixed for 30 min in Cytofix/Cytoperm on ice at 100 µl/1 × 10⁶ cells, permeabilized for 10 min in CytopermPlus on ice at 100 µl/1 × 10⁶ cells, refixed for 5 min in Cytofix/Cytoperm on ice at 100 µl/1 × 10⁶ cells, all washes done with 1× Perm/Wash. Cells were resuspended in Staining buffer at 1 × 10⁷ cells/ml, Ki67 antibodies (mouse: clone 16A8; naked mole‐rat: clone Ki‐67; both PE‐conjugated, BioLegend) were added at 5 µl/1 × 10⁶ cells and incubated for 30 min at room temperature in the dark, 1 µg/ml DAPI was used as DNA stain.

EdU‐BrdU dual‐pulse labeling

Mice aged 6 months or NMRs aged 2–4 years were intraperitoneally (i.p.) injected with 1 mg (2′S)‐2′‐Deoxy‐2′‐fluoro‐5‐ethynyluridine (F‐ara‐EdU; Sigma) from a 10 mg/ml stock in DMSO diluted with sterile 0.9% sodium chloride solution (Sigma). Exactly 2 h later animals were i.p. injected with 2 mg 5‐Bromo‐2′‐deoxyuridine (BrdU; Sigma) from a 20 mg/ml stock in DMSO diluted with sterile 0.9% sodium chloride solution (Sigma). Animals were euthanized for tissue harvest 30 min post BrdU administration. Mouse and NMR BM cells from frozen aliquots were count‐adjusted to 1 × 10⁷ cells/ml. Antibody staining was performed as described before fixation, panel was LIN‐V450/BV421, Sca‐1 (mouse) or CD34 (naked mole‐rat) APC, Kit (mouse) or Thy1.1 (naked mole‐rat) APC‐Cy7. We used the fixing and permeabilization buffers from the Click‐iT EdU Plus Kit (Thermo Fisher). Antibody‐stained cells were washed twice in PBS 1% BSA (Cell Signaling Technology), then resuspended with 100 µl/1 × 10⁶ cells Fixative and incubated for 15 min at room temperature (RT) in the dark. cells were washed twice in PBS 1% BSA (Cell Signaling Technology), then resuspended with 100 µl/1 × 10⁶cells in Perm/Wash buffer and incubated for 15 min at RT in the dark. Click‐iT Plus reaction cocktail was prepared according to the Kit (Thermo Fischer), directly added to the permeabilization mix and incubated for 30 min at RT in the dark. Cells were washed two times with Perm/Wash and resuspended in 100 µl of 300 µg/ml DNAse1 into 30 µg/1 × 10⁶ cells, and incubated for 1 h at 37°C in a waterbath. Cells were washed with Perm/Wash and stained with 1 µl/1 × 10⁶ cells anti‐BrdU from the FITC BrdU Flow Kit (BD Biosciences) for 20 min at RT in the dark. For EdU cell cycle measurements no DNAse1 digestion and BrdU labeling was performed, instead cells were stained with 500 nM SYTOX Green.

Xenotransplantations

NMR BM and/or spleen cells were extracted, sorted, and directly transplanted into 2.5 Gy‐irradiated (24 h pre Tx) NSGS recipients between 5 and 9 weeks of age at cell doses between 5 and 10 × 10⁴ sorted or 1–5 × 10⁶ whole marrow NMR cells. Injections were done via the retroorbital sinus, blood sampling was performed via maxillary vein or retroorbital plexus at weeks 4, 8, and 12. Hosts were culled at 2, 4, 8, or 12 weeks and engraftment frequencies were estimated by flow cytometry using only NMR markers not cross‐reactive with mouse cells and CD45.1 (A20, BioLegend). Engraftment rates were adjusted for input cell dose to 100k/Tx. Gating path was WBC (FSC‐A versus SSC‐A) => singlets1 (SSC‐W versus SSC‐H) => singlets2 (FSC‐W versus FSC‐H) => viable cells (SSC versus DAPI) => NOT Thy1.1^–/CD34^– (CD34 versus Thy1.1) == engrafted NMR cells (Fig 5A). One limitation for quantifying engraftment levels is that NMR BM features cells negative for the above markers, which can arise from transplanted HSPCs as xenogenic CP7 (Fig 5A). A cross‐reactive guinea pig CD45 antibody does not stain > 80% of NMR WBM cells and exhibits notable cross‐reactivity with BM from NSGS recipients (Appendix Fig S9B–D). We further detected cells double‐positive for guinea pig CD45 and CD45.1. Cells stained as Thy1.1⁺ and/or CD34⁺ are clearly originated by the xenograft as untransplanted NSGS BM does not feature any Thy1.1 of CD34 labeled cells (Fig 5A). Since all three different cell populations from guinea pig CD45 versus CD45.1 staining (DN, CD45.1⁺, CD45⁺/CD45.1^–) contain a different pattern of cells stained as Thy1.1⁺ and/or CD34⁺, we considered any cell positive for one or both markers as xenograft. We reasoned that due to the in vitro cross‐reactivity of human SCF engraftment would be supported when using NSGS hosts. However, when we compared the engraftment efficiency for ~1 × 10⁵ LTCs transplanted into NSGB (NOD. Cg‐B2m^tm1Unc Prkdc^scid Il2rg^tm1Wjl /SzJ) or NSGS mice at 4 weeks and same cell dose between NSG (NOD. Cg‐Prkdc^scid Il2rg^tm1Wjl /SzJ) and NSGS at 8 weeks, we found no significant differences between the strains.

5‐FU treatments

Mice aged 6 months or NMRs aged 2–4 years were given intraperitoneal (i.p.) injections with 150 mg/kg 5‐Fluorouracil (5‐FU; Sigma) from a 50 mg/ml stock in DMSO diluted with sterile 0.9% sodium chloride solution (Sigma). Animals were monitored daily and euthanized when found moribund.

poly(I:C) treatments

Mice aged 4 months or NMRs aged 3 years were given intraperitoneal (i.p.) injections with 5 mg/kg Polyinosinic:polycytidylic acid [poly(I:C), HMW, Invivogen] in sterile H₂O, controls were injected an equivalent amount of sterile H₂O based on their body weight. For EdU/BrdU pulsing, we varied the intervals into 20 h post poly(I:C) 1 mg EdU, followed by 2 mg BrdU 3 h later, then animals were euthanized 1 h post BrdU injection.

Irradiation treatments

NMRs aged 2–4 years were gamma irradiated with one dose of 5 Gy per animal. For 2 weeks animals were monitored daily, then twice weekly before culling 30 days post total body irradiation (TBI).

Quantitative PCR

Mouse and NMR sorted PB‐TCs and lymph nodes were used for RNA extraction by Trizol (Thermo Fisher). NMR PB‐WBCs, BM‐WBCs, in vitro clonogenic colonies and liquid cultures were used for RNA extraction with RNeasy mini or micro kit, respectively (Qiagen). RNA was quantified using a NanoDrop One (Thermo Fisher), and 100 ng was used as input for the High Capacity cDNA Reverse Transcription Kit (Thermo Fisher). RT reaction was performed according to instructions and the 20 µl reaction diluted to 200 µl, of which 5 µl were used per qPCR reaction. We used iTaq Universal SYBR Green Supermix (Bio‐Rad) on a CFX Connect® RealTime System (Bio‐Rad) with a three‐step cycling of 10 s 95°C, 20 s 60°C, and 30 s 72°C for 40 cycles. All primers (IDTDNA) were validated to amplify a single amplicon at the above PCR conditions by gel electrophoresis. Gene sequences for primer design by Primer3Plus were retrieved from ENSEMBL. Absolute copy number quantitation was done as described (Emmrich et al, 2021), amplicon gel images, amplicon plasmids, and standard curve data is available upon request. Relative quantification for NMR B‐lineage genes was performed using B2 M as housekeeping gene to calculate the $\mathrm{\Delta }Ct=2^{‐{\overline{\mathit{Ct}}}_{\mathit{GOI}}‐\overline{C{t}_{B2M}}}$. All Primers can be found in Dataset EV1.

Transcriptome assembly

All NMR RNA‐Seq was performed with the GRC URMC Rochester. RNA from whole bone marrow (WBM) was sequenced with ~230 million reads on a HiSeq2500v4 (Illumina). Raw Illumina paired‐end sequencing reads where assessed with FastQC. Rcorrector (Song & Florea, 2015) was used to correct sequencing errors and read pairs with uncorrectable errors were removed using a custom python script (GRC URMC Rochester). Adapter and base quality trimming was performed using Trim Galore and Cutadapt (Martin, 2011) resulting in high quality reads that were used as input to Trinity (Grabherr et al, 2011) for assembly. FRAMA (Bens et al, 2016) was used to postprocess the de novo assembly, including reduction of contig redundancy, ortholog assignment using human as a reference, correction of misassembled transcripts, scaffolding of fragmented transcripts, and coding sequence identification. Quality assessment of the final FRAMA transcriptome was performed using BUSCO and TransRate (Simao et al, 2015; Smith‐Unna et al, 2016). The transcriptome was mapped by blastn to the NMR genome (hetgla_female_1.0) or to transcript sequences annotated in ENSEMBL97. Mapped genomic coordinates of transcripts were thus compared to those of annotated genes using a custom python script. We found that 512 nonoverlapping FRAMA transcripts (i.e., gene loci) were absent from the annotation, and another 5,281 had > 20% transcript length mapped to the genome but not matching annotated isoforms (Dataset EV2).

Population RNA‐Seq

RNA from sorted human and NMR populations was sequenced at ~100 million reads on a HiSeq2500v4 (Illumina). We used the SMARTer® Ultra® Low RNA Kit (Takara) for library preparation. All GEO datasets for human and mouse HSPC populations were acquired with SRA toolkit and processed from raw fastq files. Raw Illumina paired‐end sequencing reads where subjected to base quality trimming using Trimmomatic (Bolger et al, 2014) and were assessed with FastQC. RSEM v1.3.0 with the STAR aligner option was used to calculate expected counts and TPMs (Li & Dewey, 2011). We used a customized perl script to run RSEM with the FRAMA transcriptome as reference using the bowtie2 aligner option. We also ran RSEM with the ENSEMBL94 hetgla_female_1.0 annotation using the STAR aligner option to confirm all clusterings and differential gene expression signatures for all NMR samples, results from which were almost identical to those obtained with FRAMA.

Subsequent analysis was done with R 4.0.2 and Bioconductor (Gentleman et al, 2004). Expected counts from different transcript isoforms of the same gene were added up to one unique identifier (uniquefy) using ddply and numcolwise functions of the plyr package (Wickham, 2011), edgeR was used to calculate size factors with method=”RLE” and to compute CPMs (Robinson et al, 2010). We applied genefilter to calculate the interquartile range (IQR) of CPMs with IQR(x) > 1 to filter unexpressed and outlier genes; library‐size normalized, IQR‐filtered log2‐transformed CPMs were vst‐transformed by DESeq2 (Love et al, 2014), then a PCA embedding from the stats package was used as input for Rtsne (van der Maaten & Hinton, 2008). We applied limma to perform voom‐transformation and select for differentially expressed genes (DEGs) with P < 0.05 and log‐fold‐change 1 (Ritchie et al, 2015).

GSEA was performed using the gsva package with method=”ssGSEA” using either the hematopoietic stem and progenitor geneset collection modified from Schwarzer et al (2017) in Dataset EV1 or the MSigDB v6.0 hallmark genesets with a P‐value threshold of 0.05 (Subramanian et al, 2005; Barbie et al, 2009; Hanzelmann et al, 2013; Liberzon et al, 2015). All GSEA calculations were performed on the combined up‐ and downregulated DEG signature for each group, see Dataset EV6. The fGSEA package was used to retrieve leading edge genes after reperforming GSEA under default conditions (Sergushichev, 2016), with the required rank metric generated after (Plaisier et al, 2010). All NMR population RNA‐Seq DEG signatures (Dataset EV6) were used to create genesets and were added to Table S1.

The expression gradient in Fig 4B was calculated by a customized R function, which ordered the log2‐transformed CPMs for each gene along their numeric value, allowing to filter out the genes subsequently changing expression from one group to another, see Dataset EV6.

For the 3‐species comparison uniquefied human, mouse, and NMR TPM datasets (Fig 7A, Appendix Fig S12) were merged based on HGNC symbols, then genefilter was used to calculate IQR of TPMs with IQR(x) > 1 to filter unexpressed and outlier genes; The TCC package was used to calculate TMM‐based size‐factors (Sun et al, 2013). The function betweenLaneNormalization with median scaling from the EDASeq package was used to normalize for sequencing batch effects (Risso et al, 2011). We used the RUVSeq package to normalize TPMs for batch effects across datasets (Risso et al, 2014). The limma package was used to plotMDS of the full 3‐species dataset (Appendix Fig S12D), see Dataset EV9 (sheet “metadata.population.RNA‐Seq”). Next, we split the dataset collection into three subsets based on developmental stage of each population. TPMs were vst‐transformed by DESeq2, then a PCA embedding from the stats package was used as input for Rtsne (Appendix Fig S12E); DGE and GSEA were performed as above using the population species as contrast and the MSigDB v6.0 hallmark genesets.

Single cell RNA‐Seq

NMR sorted CITE‐Seq data (Figs 1 and EV1)

Marrow, blood, and thymus cells from 2 animals aged 11 months (male and female) were enriched by flow sorting. For BM we sorted CP1 3 × 10³, LTC 3 × 10³, CP3 2 × 10³, CP4 2 × 10³, CP5 2 × 10³, CP6 2 × 10³, CP7 3 × 10³, LIN⁺/CD34^– 1.5 × 10³, and LIN⁺/CD34⁺ 1.5 × 10³ for a total of 40,000 marrow cells from 2 animals as three 10× v2 chemistry libraries (2 replicates LIN^– pooled, one replicate LIN⁺ pooled; Fig EV1E). For PB we sorted GC 1.5 × 10³, MO 1 × 10³, BC 1 × 10³, and TC 1.5 × 10³ for a total of 10,000 peripheral blood leukocytes from the same animals as above into one pooled 10× v2 chemistry library (Fig EV1D). Cells were pooled according to their tissue origins and processed for CITE‐Seq using a protocol from the Stoeckius lab and the Chromium Single‐Cell 3′ Library & Gel Bead Kit v2 (10× Genomics) (Stoeckius et al, 2017). Raw reads generated on the Illumina NovaSeq6000 sequencer were demultiplexed using Cellranger 3.0.2 software in conjunction with Illumina’s bcl2fastq 2.19.0. Cellranger was also used to align the read data to the FRAMA de novo transcriptome assembly and ENSEMBL94 hetgla_female_1.0, barcode count, UMI compress, and filter for “true” cells. CITE‐Seq data for each capture was also demultiplexed using bcl2fastq and processed with CITE‐seq‐Count 1.4.2 (Roelli et al, 2019) given the antibody barcode sequences, a white list of filtered cell barcodes from the matching Cell Ranger “count” run, and parameters: “‐cbf 1 ‐cbl 16 ‐umif 17 ‐umil 26”.

Subsequent analysis was done with R 4.0.2 and Bioconductor. The marrow and blood libraries were merged and FRAMA Trinity isoforms were uniquefied by row‐wise addition of UMI‐counts for each isoform of the same gene using a data.table snippet.10× files were assigned to a SingleCellExperiment S4 class (Lun, 2019), and each gene without any counts in any cell was removed. We converted the S4 class into a Seurat 3.1 object (Butler et al, 2018) and added the CITE‐signals in form of an independent “assay,” barcodes were quality filtered to keep cells between 200 and 5,000 detected genes/cell and < 25,000 counts per cell. RNA assay was log‐normalized with “scale.factor = 1e4,” CITE assay was “CLR” normalized. Variable features were detected with arguments selection.method = "vst", nfeatures = 3000. Scores for G2 M and S phases were obtained using Seurat CellCycleScoring as described in the respective Seurat vignette. Clustering was done using Seurat’s FindClusters function with resolution = 0.5. Next, we used the doublet detection and removal workflow as suggested in the Bioconductor OSCA vignette. Briefly, we run findDoubletCluster from the scDblFinder package, followed by in silico simulation of doublets from the single‐cell expression profiles (Dahlin et al, 2018) using computeDoubletDensity from BiocSingular package, and excluded any cluster which was identified in both methods. The DEGs for each cluster were detected by FindAllMarkers function with arguments test.use = "MAST," logfc.threshold = log(2), min.pct = 0.25, return.thresh = 0.05. Hematopoietic cell type annotation was done through fGSEA using the modified HSPC geneset collection (Schwarzer et al, 2017), extended with the upregulated DEGs from the NMR population RNA‐Seq analysis, upregulated DEGs from joint analysis of murine HSPCs from multiple studies (Sun et al, 2014; Behrens et al, 2016; George et al, 2016; Luis et al, 2016; Dong et al, 2019; Han et al, 2019; Bernitz et al, 2020), upregulated DEGs from human HSPC population RNA‐Seq datasets (Chen et al, 2014; Corces et al, 2016; Amon et al, 2019; Drissen et al, 2019), selected genesets from MSigDB and Immgen databases and example genesets from the SingCellaR software (Dataset EV2). To determine the rank metrics for fGSEA, the q‐value requires to be transformed by ‐log10(q‐value) (Plaisier et al, 2010). Seurat’s FindAllMarkers function can generate 0 q‐values (P_val_adj, Dataset EV3) for high confidence hits, thus for any 0 we added the lowest q‐value > 0 of the entire marker list for the group to test to each marker with q‐value = 0. This generates ties in the pval ranking by fGSEA for the genes with modified 0 q‐values, which are automatically resolved by retaining their order according to their fold‐change of expression. A custom script was generated to pipe fGSEA with our HSPC geneset collection through each clusters marker genes, results are deposited in Dataset EV3. The entire process was done in an iterative manner to condense multiple clusters of the same overabundant cell type (e.g., neutrophil granulocytes) into one partition, while maintaining distinctive low abundance clusters. Single cell expression maps (Fig EV1I, Appendix Figs S1F and S2F) were done with schex package using nbins = dim(Seurat.object)[2]/200. PhateR was used as suggested by running an initial graph imputation, and obtaining the final graph with parameters knn=8, decay=100, t = 25 (Moon et al, 2019).

Human cell atlas data (Appendix Fig S1)

The original data comprising 380,000 marrow cells from 8 human donors are available from the HCA data portal or as the HCAData R package. We used a subset of this dataset available through the SeuratData package, randomly downsampled to 40,000 cells. The cell type annotation was obtained by reference mapping according to the Seurat vignette. The respective reference was created by weighted nearest neighbor analysis of CITE‐Seq data from human marrow according to the Seurat vignette (Stuart et al, 2019).

Variable features were detected with arguments selection.method = "vst", nfeatures = 10000. Cell cycle scoring and doublet detection were performed as described above. Clustering and marker gene identification was done using the same parameters as for NMR sorted CITE‐Seq data. Hematopoietic cell type annotation was done as described above (Dataset EV3). PhateR was run with parameters knn=3, decay=100, t = 12.

RNAMagnet data (Appendix Fig S2)

We used the processed main dataset together with the prior cell type annotation (Baccin et al, 2020). Barcodes of tissue type “bone” were excluded, leaving Kit⁺ HSPCs, WBM, and CD45^– cells in the dataset. Gene features were uniquefied with data.table, and variable features were detected with arguments selection.method = "vst", nfeatures = 2000. Cell cycle scoring and doublet detection were performed as described above. Clustering and marker gene identification was done using the same parameters as for NMR sorted CITE‐Seq data. Hematopoietic cell type annotation was done as described above (Dataset EV3). PhateR was run with parameters knn=3, decay=100, t = 28.

Calico data (Appendix Fig S5)

We downloaded the raw fastq files for mouse and NMR scRNA‐Seq from spleen (Hilton et al, 2019) using SRA toolkit. Cellranger 3.1.0 (10× Genomics) was used to generate reference and count matrices for mouse data from ENSEMBL99 or from FRAMA for naked mole‐rat. Barcodes were quality filtered to keep cells between 200 and 2,500 detected genes/cell and < 10,000 counts per cell. Gene features were uniquefied with data.table and variable features were detected with arguments selection.method = "vst", nfeatures = 2000. Cell cycle scoring and doublet detection were performed as described above. Clustering and marker gene identification was done using the same parameters as for NMR sorted CITE‐Seq data. Hematopoietic cell type annotation was done as described above (Dataset EV5).

Xenograft data (Fig 5, Appendix Fig S9)

Host WBM from frozen stocks were subjected to CITE‐Seq using the Chromium Single‐Cell 3′ Library & Gel Bead Kit v3 (10× Genomics). Cells were processed for TotalSeq™ CITE reagents according to the manufacturers’ instructions (BioLegend), using both human and mouse Fc blocking reagents (BioLegend). Following fluorescent antibody staining, samples were sorted for xenograft cells by Thy1.1/CD34 staining and excluding the “mouse” gate as shown in Fig 5H. For library preparations see below for 10× v3 chemistry. Cellranger 3.1.0 was used to generate count matrices from both an ENSEMBL94 mouse reference and FRAMA for the same library. Gene features were uniquefied with data.table, barcodes were quality filtered to keep cells between 200 and 10,000 detected genes/cell and < 20,000 counts per cell. RNA assay was log‐normalized with “scale.factor = 1e4”, CITE assay was “CLR” normalized. Variable features were detected with arguments selection.method = "vst", nfeatures = 3000. Libraries were integrated using FindIntegrationAnchors with dims = 1:50, anchor.features = 3000, reduction = "cca". Mouse cells were removed by subsetting the integrated Seurat object. Cell cycle scoring and doublet detection were performed as described above. Clustering and marker gene identification was done using the same parameters as for NMR sorted CITE‐Seq data. CITE feature/antibody marker detection was done as described for transcript cluster markers with the exception of test.use = “wilcox”. Hematopoietic cell type annotation was done as described above (Dataset EV7).

Unfractionated BM data (Fig 6, Appendix Fig S10)

10,000 DAPI^– BM cells from 2 mice aged 3 months (male and female) and 2 mice aged 12 months (male and female), or 2 NMRs aged 3 years (male and female) and 3 NMRs aged 11 years (male and female), were subjected to CITE‐Seq using Chromium Single‐Cell 3′ Library & Gel Bead Kit v3 (10× Genomics). Cells were processed for TotalSeq™ CITE reagents according to the manufacturers’ instructions (BioLegend), using both human and mouse Fc blocking reagents (BioLegend). Cellular suspensions were loaded on a Chromium Single‐Cell Instrument (10× Genomics, Pleasanton, CA, USA) to generate single‐cell Gel Bead‐in‐Emulsions (GEMs). Single‐cell RNA‐Seq libraries were prepared using Chromium Next GEM Single Cell 3′ GEM, Library & Gel Bead Kit v3.1 (10× Genomics). The beads were dissolved and cells were lysed per manufacturer’s recommendations. GEM reverse transcription (GEM‐RT) was performed to produce a barcoded, full‐length cDNA from poly‐adenylated mRNA. After incubation, GEMs were broken and the pooled post‐GEM‐RT reaction mixtures were recovered and cDNA was purified with silane magnetic beads (DynaBeads MyOne Silane Beads, PN37002D, ThermoFisher Scientific). The entire purified post GEM‐RT product was amplified by PCR. This amplification reaction generated sufficient material to construct a 3′ cDNA library. Enzymatic fragmentation and size selection was used to optimize the cDNA amplicon size and indexed sequencing libraries were constructed by End Repair, A‐tailing, Adaptor Ligation, and PCR. Final libraries contain the P5 and P7 priming sites used in Illumina bridge amplification. In parallel, CITE‐seq library amplification was performed following SPRI bead purification of CITE‐seq cDNA using Q5 Hot Start HiFi Master Mix (New England Biolabs, Ipswich, MA), SI PCR primer (IDT, Coralville, IA), and indexed TruSeq Small RNA PCR primers (Illumina, San Diego, CA) as specified (Stoeckius et al, 2017). Amplified CITE‐seq libraries were purified using AMPure XP (Beckman Coulter, Indianapolis, IN) beads and quantified by Qubit dsDNA assay (ThermoFisher, Waltham, MA) and Bioanalyzer HSDNA (Agilent, Santa Clara, CA) analysis. CITE‐seq libraries were pooled with 10× Genomics gene expression libraries for sequencing on Illumina’s NovaSeq 6000. Barcodes were quality filtered to keep cells between 200 and 5,000 detected genes/cell and < 20,000 counts per cell. RNA assay was log‐normalized with “scale.factor = 1e4”, CITE assay was “CLR” normalized. Variable features were detected with arguments selection.method = "vst", nfeatures = 3000. Canonical correlation analysis (CCA) was used to integrate libraries (Stuart et al, 2019) from either species with FindIntegrationAnchors with dims = 1:50, anchor.features = 3000, reduction = "cca". Cell cycle scoring and doublet detection were performed as described above. Clustering and marker gene identification was done using the same parameters as for NMR sorted CITE‐Seq data. CITE feature/antibody marker detection was done as described for transcript cluster markers with the exception of test.use = “wilcox”. Hematopoietic cell type annotation was done as described above (Dataset EV8). Differentially expressed markers between age groups for mouse used FindMarkers with test.use = "MAST", logfc.threshold = log(2), min.pct = 0.1. Next we ran fGSEA with the MSigDb hallmark geneset as mentioned in the population RNA‐Seq section, and plotted any pathway with FDR < 0.05 (Appendix Fig S10E). Differentially expressed markers between age groups for NMR HSPC cluster used FindMarkers with test.use = "MAST", logfc.threshold = 0; fGSEA with MSigDb hallmark of C5 GO Bio‐Process genesets was done as above (Fig EV4C and D). Differential abundance (DA), testing the cell abundances for clusters across conditions, was performed as described (Lun et al, 2017). Briefly, edgeR was used to apply negative binomial generalized linear model dispersion to each library as outlined in the OSCA Bioconductor collection. SCTransform was used to integrate scaled, clustered, and annotated mouse and NMR unfractionated BM datasets with the following arguments (Hafemeister & Satija, 2019): SelectIntegrationFeatures with nfeatures = 3000, FindIntegrationAnchors with normalization.method = "SCT". Cell cycle scoring, clustering, and marker detection performed as described above. Conserved markers were identified by running FindConservedMarkers for each cluster across species of the SCT‐integrated dataset with test.use = "MAST", logfc.threshold = log(2), min.pct = 0.25. Differentially expressed markers per cluster between species run FindMarkers with test.use = "MAST", logfc.threshold = log(2), min.pct = 0.1. We performed fGSEA with the MSigDb hallmark geneset and plot any pathway with FDR < 0.05 (Differentially expressed markers between age groups for mouse used FindMarkers with test.use = "MAST", logfc.threshold = log(2), min.pct = 0.1. Next, we run fGSEA with the MSigDb hallmark geneset as mentioned in population RNA‐Seq, and plot any pathway with FDR < 0.05 (Appendix Fig S12F).

In order to infer gene regulatory networks and reconstruct regulons at the single‐cell level, we used pySCENIC v0.9.18, an improved implementation of the original SCENIC framework (Aibar et al, 2017, Van de Sande et al, 2020). In the case of mouse, we used the BM dataset generated in this study containing 19,298 cells from two 3‐month‐old and two 12‐month‐old mice. In order to properly run pySCENIC, we filtered the gene count matrix to the 10,000 most highly‐variable genes using the seurat_v3 method implemented in scanpy. To run the gene regulatory network step, we first downloaded the motifs‐v9‐nr.mgi‐m0.001–o0.0.tbl file from the cisTarget databases website (https://resources.aertslab.org/cistarget/). From this file, we isolated 1,721 mouse transcription factors. Using these inputs and the single‐cell data, we created an adjacency matrix with the pySCENIC grn function running with default parameters. We also downloaded the mm10__refseq‐r80__10kb_up_and_down_tss.mc9nr.feather database from the cisTarget database. Using the previously generated adjacency matrix (the output from the grn function), the downloaded feather database, the set of transcription factor annotations, and the mouse single‐cell BM data, we generated a motif matrix using the pySCENIC ctx function with default parameters. Finally, we calculated cellular enrichment of regulons using the pySCENIC aucell function and the previously generated motifs matrix. In the case of naked mole‐rat, no cisTarget database was available for download. To remedy this, we created an NMR specific cisTarget database, using the repository at https://github.com/aertslab/create_cisTarget_databases. For this, we defined genomic regions of 10 kb before and after the TSS for 9,867 genes on the NMR genome (hetgla_female_1.0), matching the methodology employed for the existing mouse database. We also downloaded and used the nonredundant JASPAR2022 set of conserved vertebrate motifs. Once an NMR cisTarget database was generated, the same steps as described above for the mouse BM dataset were used to calculate regulon enrichment in 21,678 single‐cells from two 33‐month‐old and two 137‐month‐old NMRs. In addition to running the pySCENIC workflow with the NMR cisTarget database generated here, we complemented our analyses by also running the NMR single‐cell data with the existing mouse databases (modified so mouse genes were mapped to NMR orthologues).

Unfractionated PB data (Fig 2, Appendix Fig S3)

10,000 DAPI^– BM cells from 4 NMRs aged 3 years (one female and three males) and 3 NMRs aged 11 years (three males), were subjected to CITE‐Seq using Chromium Single‐Cell 3′ Library & Gel Bead Kit v3 (10× Genomics) as described for unfractionated BM. Barcodes were quality filtered to keep cells between 200 and 5,000 detected genes/cell and < 20,000 counts per cell. RNA assay was log‐normalized with “scale.factor = 1e4”, CITE assay was “CLR” normalized. Variable features were detected with arguments selection.method = "vst", nfeatures = 3000. Canonical correlation analysis (CCA) was used to integrate libraries (Stuart et al, 2019) from either species with FindIntegrationAnchors with dims = 1:50, anchor.features = 3000, reduction = "cca". Cell cycle scoring and doublet detection were performed as described above. Clustering and marker gene identification was done using the same parameters as for NMR sorted CITE‐Seq data. CITE feature/antibody marker detection was done as described for transcript cluster markers with the exception of test.use = “wilcox”. Hematopoietic cell type annotation was done as described above (Dataset EV4). Differential abundance (DA) testing was performed using the edgeR method as above.

To assess how observed age‐related changes in mouse and NMR cells affect their transcriptomic age, we applied mouse multi‐tissue transcriptomic clocks based on gene expression signatures of murine aging across 17 different tissues identified as explained (Tyshkovskiy et al, 2019). For the current analysis, we utilized two clocks based on elastic net linear models that were designed to predict chronological age in linear and logarithmic scale. Using the clocks, we calculated the change in transcriptomic age (tAge) between old and young cells separately for each individual cell type in mouse bone marrow (BM), NMR bone marrow, and NMR peripheral blood (PB). The distribution of change in tAge across different cell types was visualized with boxplots. Statistical significance of median tAge increase in each model (mouse BM, NMR BM and NMR PB) was assessed using one‐sample Wilcoxon signed‐rank test. Difference between median tAge changes between the models was assessed for each pair of models using Mann–Whitney U test. Adjustment for multiple comparisons was performed using Benjamini‐Hochberg correction method. Age‐related increase of tAge or difference in tAge predictions between the models was considered significant if corresponding adjusted P‐value < 0.05.

Quantification and statistical analysis

Data are presented as the mean ± SD. Statistical tests performed can be found in the figure legends. P < 0.05 were considered statistically significant. Statistical analyses were carried out using Prism 9 software (GraphPad) unless otherwise stated.

Disclosure and competing interests statement

The authors declare that they have no conflict of interest.

Author contributions

Stephan Emmrich: Conceptualization; Resources; Data curation; Software; Formal analysis; Supervision; Validation; Investigation; Visualization; Methodology; Writing—original draft; Project administration; Writing—review & editing. Alexandre Trapp: Resources; Data curation; Software; Validation; Visualization; Methodology; Writing—original draft; Writing—review & editing. Frances Tolibzoda Zakusilo: Resources; Software; Formal analysis; Validation; Investigation; Visualization; Methodology; Writing—original draft. Maggie E Straight: Formal analysis; Visualization; Methodology; Writing—original draft; Writing—review & editing. Albert K Ying: Resources; Data curation; Software; Visualization. Alexander Tyshkovskiy: Software; Validation; Methodology. Marco Mariotti: Resources; Software; Validation; Investigation. Spencer Gray: Resources; Software; Methodology. Zhihui Zhang: Investigation; Methodology. Michael G Drage: Investigation; Methodology. Masaki Takasugi: Investigation; Methodology. Jan‐Henning Klusmann: Funding acquisition; Investigation. Vadim N Gladyshev: Supervision; Funding acquisition; Project administration. Andrei Seluanov: Supervision; Funding acquisition; Writing—original draft; Project administration; Writing—review & editing. Vera Gorbunova: Conceptualization; Supervision; Funding acquisition; Writing—original draft; Project administration; Writing—review & editing.

In addition to the CRediT author contributions listed above, the contributions in detail are:

SE designed and supervised research, performed most experiments, and analyzed data; ATr wrote scripts for single‐cell fGSEA and contributed to bioinformatics analyses and FACS; FTZ performed histology quantifications, animal perfusions and data analysis; MM performed FRAMA‐genome alignments; MES performed histology and most histochemistry stainings; AKY developed the online data resource tool; ATy performed tAge computations; ZZ performed mouse blood analysis; MGD performed histology and provided human BM specimen; SG, MT, J‐HK, ZZ, and VNG contributed to data analysis; AS and VG supervised research; SE, AS and VG wrote the manuscript with input from all authors.

Supporting information

Appendix

Expanded View Figures PDF

Dataset EV1

Dataset EV2

Dataset EV3

Dataset EV4

Dataset EV5

Dataset EV6

Dataset EV7

Dataset EV8

Dataset EV9

The EMBO Journal (2022) 41: e109694

Data availability

The raw single cell RNA‐Sequencing data that support the findings of this study are available in GEO as the superseries GSE185724 (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE185724); processed data is available in figshare.com with the identifiers 10.6084/m9.figshare.c.5472735, 10.6084/m9.figshare.c.5470587, 10.6084/m9.figshare.c.5472684, 10.6084/m9.figshare.c.5474256. The online data resource tool is linked through http://gladyshevlab.org:3838/scNMR, a repository can be found at our GitHub pages at https://github.com/albert‐ying/NMR_scRNAseq_browser, the custom code to run GSEA for scRNA‐Seq clusters for cell type annotation was deposited at https://github.com/alex‐trapp/sc‐fgsea.