Main

Blood vessels are necessary for the development and maintenance of organs throughout our bodies. They provide oxygen and nutrients to parenchymal cells in each tissue and secrete paracrine molecules, known as angiocrine factors, which drive cell differentiation and tissue morphogenesis1,2,3. This vascular function is mediated by organ-specific endothelial cells (ECs), the cellular and molecular heterogeneity of which have been intensively studied using single-cell transcriptomic technologies4,5.

Bone marrow endothelial cells (BMECs) play a central role in the spatiotemporal relationship between angiogenesis and osteogenesis, referred to as angiogenic–osteogenic coupling6,7,8,9,10. They also serve as a stem cell niche component that supports the self-renewal and differentiation of haematopoietic stem and progenitor cells (HSPCs)11,12,13,14. Recent studies utilizing single-cell RNA sequencing (scRNA-seq) uncovered the cellular taxonomy of stromal compartments in the bone marrow (BM)15,16, which enhanced our understanding of cell-to-cell interactions governing bone metabolism as well as life-long haematopoiesis. However, information about the spatiotemporal and functional composition of BMECs is limited and this must be analysed in more detail.

In the present study, we profiled BMECs isolated from mouse femurs using scRNA-seq and defined their heterogeneity. The data uncovered an unrecognized capillary subtype, termed type S (secondary ossification) ECs, exclusively existing in the epiphysis, the end of long bones17,18. Type S ECs possess unique phenotypic characteristics, which contributed to the maintenance of bone strength and constituted a distinct HSPC niche. Our findings provided the landscape for the cellular architecture in the BM vasculature and enhanced our understanding of the fundamental mechanism regulating osteogenesis and haematopoiesis.

Results

Spatial heterogeneity of transcriptomes in BMECs

As an initial step to explore the heterogeneity of BMECs, we conducted droplet-based scRNA-seq of CD31+CD45− BM cells isolated from three surgically separated fragments of femurs, diaphyses, metaphyses and epiphyses of postnatal day 11 (P11) mice (Fig. 1a and Extended Data Fig. 1a–c). Using the Uniform Manifold Approximation and Projection (UMAP) of cells, we defined 11 EC clusters, including arterial, venous and capillary ECs, with few contaminations of osteoblasts (OBs)/mesenchymal stromal cells (MSCs)15,19, erythroid cells and myeloid cells (Fig. 1b–d) based on known markers4,5,20,21. Type H and type L ECs have been histologically characterized as specialized capillary populations predominantly existing in the metaphysis and diaphysis, respectively22,23. For capillary EC clusters, based on the spatial origin and known markers14,24, we assigned one cluster to type H ECs and five clusters to type L ECs (Fig. 1b–e). Type L ECs were divided into sinusoidal (SEC) and non-sinusoidal (non-SEC) ECs based on known SEC markers such as Stab2 and Vcam1. Although transcription of Pecam1 and Emcn, which are immunohistochemical markers of type H ECs, did not strikingly differ among capillary subtypes, expression of type H markers such as Efnb2 and Sox17 (ref. 14) was highest in the capillary cluster annotated as type H ECs (Fig. 1e). Intriguingly, we identified a unique non-SEC capillary population, distinct from any other ECs, which exclusively originated from the epiphysis. The development of this area is called the secondary ossification centre (SOC), and is the reason why we named this EC population type S ECs (Fig. 1c, cluster 8). By extracting genes upregulated in type S ECs compared with all EC clusters, Ly6a and Ly6c1, which are highly expressed in arterial ECs (AECs)16 and type H ECs14, were the no. 1 and no. 8 ranked upregulated genes in terms of the average log(fold-change) (log(FC)) (Fig. 1f and Supplementary Table 1). Intriguingly, Col1a1 and Col1a2, both of which encode type I collagen, were also upregulated in type S ECs (Fig. 1f,g, Extended Data Fig. 1d and Supplementary Table 1), although feature plots of these genes showed that they were most highly expressed in OBs/MSCs (Fig. 1g). This atypical expression of type I collagen genes suggested that type S ECs play a distinct angiocrine role during osteogenesis. The other collagen genes were diversely expressed in type S, type H and type L ECs (Fig. 1h). Ingenuity Pathway Analysis (IPA) uncovered the top four networks represented by differentially expressed genes in type S ECs, including growth factor signalling (network 2) and integrin–extracellular matrix interaction (network 4) (Extended Data Fig. 1e). Flow cytometric analysis of isolated CD31+CD45− BM cells confirmed that type S ECs were characterized by Ly6ahighLy6chigh expression on their surface (Fig. 1i). Immunohistochemistry confirmed that Ly6a expression was higher in type S ECs than in type H ECs (Fig. 1j). These data demonstrate that type S ECs possess unique phenotypic characteristics, although some of them overlap with those of type H ECs and AECs.

Fig. 1: Spatial heterogeneity of transcriptomes in BMECs.
figure 1

a, Schematic diagram depicting the strategy for scRNA-seq of BMECs. b–d, Feature plots of BMECs of three bone fragments and clustering based on the expression of canonical endothelial subtype markers. Dia, diaphysis; Epi, epiphysis; Meta, metaphysis. b, Feature plot showing all cells derived from three datasets (epiphysis, metaphysis and diaphysis). c, Feature plot showing all cells derived from three datasets (epiphysis, metaphysis and diaphysis) with annotation for each cluster. d, Feature plots showing expression of marker genes with enriched expression for each endothelial cluster. The arrowheads in d indicate clusters represented by each marker. e, Violin plots showing the expression of pan-endothelial and type H markers. f, Violin plots showing the expression of genes enriched in type S ECs. g, Feature plots of BMECs. Col1a1 and Col1a2 genes are highly expressed in type S ECs (arrowheads), although their expression is highest in OBs/MSCs (arrows). h, A heatmap showing expression of various collagen genes. i, Flow cytometric analysis of isolated CD31+CD45− BM cells at P11 (Ly6ahighLy6chigh surface protein levels) (n = 4, biologically independent experiments). The P value for the epiphysis versus the metaphysis is 0.000012. j, Immunohistochemistry of femur sections at P11. Ly6a is abundantly expressed in type S ECs (closed arrowheads) and AECs (asterisks), but weakly in type H ECs (open arrowheads). In the box plots of e and f the lower and upper bounds of box represent the 25th and 75th percentiles, respectively; the box represents the middle 50% of the data, namely the interquartile range (IQR); the horizontal line within the box represents the median; and the lower and upper bounds of the whiskers represent the minimum and maximum values within 1.5× the IQR below the 25th percentile and above the 75th percentile, respectively. The black dots represent outliers. Scale bar, 200 µm. Data presented are the mean ± s.d. The comparisons between the averages of two groups were evaluated using the two-tailed Student’s t-test.

Source data

Full size image

Molecular and morphological features of type S ECs

Recent advances in histological techniques enabled clear visualization of angiogenic–osteogenic coupling, focusing on the metaphysis22,23. Considering the scRNA-seq data described above and the existence of a unique niche maintaining skeletal stem cells in the epiphysis17,18, which suggests that there is a differential mode of angiogenic–osteogenic coupling in this area, we thoroughly examined the spatiotemporal transition of angiogenesis and osteogenesis during postnatal development from P7 to P28 (Fig. 2a–l). ECs first invaded the epiphysis at P7 from two entry points on the anterior side (Fig. 2a,h,j,m and Supplementary Video 1) and rapidly expanded their network until weaning (Fig. 2c–g). Ossification proceeded synchronously (Fig. 2a–g). The overall shape of blood vessels in this area was tree like or dendritic, in contrast with the columnar morphology of those vessels in the metaphysis (Fig. 2h–l,n). The growing edge of type S vessels highly expressed CD31 and projected short filopodia towards avascular areas such as type H vessels (Fig. 2o and Extended Data Fig. 2a). Expression of Vegfr1 strikingly differed between type H and type S cells: type H ECs highly expressed Vegfr1, whereas Vegfr1 expression was barely detectable in type S ECs in Vegfr1-BAC-RFP mice (Fig. 2o and Extended Data Fig. 2b). Skeletal growth in mice is largely completed by 10 weeks after birth. Therefore, we examined Vegfr1 expression and found that it was retained even at 6 months of age (Fig. 2p). Expression of other vascular endothelial growth factor (VEGF) receptors, Vegfr2 and Vegfr3, did not apparently differ between type S and type H ECs (Fig. 2o and Extended Data Fig. 2c,d); these receptors were both highly expressed in type L ECs and were strikingly expressed in the growing edges of type S and type H ECs.

Fig. 2: Molecular and morphological features of type S ECs.
figure 2

a–f, Whole-mount Alizarin Red staining (b at P7, (d) at P11 and f at P28) and section immunostaining of femurs (a at P7, c at P11 and e at P28). Vessel invasion of the epiphysis (open arrowheads) and SOC formation (arrows) occur at P7 and spread three-dimensionally during the second week after birth. g, Quantification of vessels and calcified areas in the epiphysis (P3, P7 and P11: n = 4; P28: n = 5, biologically independent experiments). h–l, Immunohistochemistry of femurs (vertical (h at P7 and i at P11) or horizontal sections j at P7 (epiphysis), k at P11 (epiphysis) and l at P11 (metaphysis)). Open arrowheads indicate vessel invasion points. m, Micro-CT angiography at P11 showing the heatmap of vessel perfusion of the knee joint. Open arrowheads indicate vessel invasion points. n, Schematic diagram showing the structural difference between type H (columnar) and type S (dendritic) vessels. o,p, Immunohistochemistry of femur sections at P11 (o) or 6 months after birth (p). The growing edges of type S and type H vessels highly express CD31 (open arrowheads) and Vegfr3 (arrows), and project short filopodia towards avascular areas (insets in the leftmost panels). Vegfr1 is specifically expressed in type H ECs (closed arrowheads). R3, Vegfr3; R2-GFP, GFP in Vegfr2-GFP transgenic mice; R1-RFP, red fluorescent protein (RFP) in Vegfr1-RFP transgenic mice. Scale bars, 1 mm (a–f, h–l and m); 200 µm (o and p). Data presented are the mean ± s.d.

Source data

Full size image

Type S vessels are crucial for epiphyseal osteogenesis

Considering the lack of Vegfr1, a negative regulator buffering VEGF signalling1, we suspected that type S ECs were more vulnerable to deletion of Vegfr2 than type H ECs. Endothelial-specific Vegfr2-knockout (Cdh5-BAC-CreERT2+Vegfr2flox/flox, hereafter referred to as Vegfr2iΔEC) mice showed severely impaired angiogenesis and osteogenesis in the epiphysis during the early postnatal weeks (Fig. 3a–d and Extended Data Fig. 2e). In agreement with a previous report25, the morphology of type H vessels was abnormal in Vegfr2iΔEC mice, but the impairment was milder than that of type S vessels (Fig. 3b and Extended Data Fig. 2f). Accordingly, metaphyseal ossification was not apparently delayed in Vegfr2iΔEC mice (Fig. 3c,d). The numbers of Runx2+ pre-osteoprogenitors26 and osterix (Osx)+ osteoprogenitors27, which usually exist around blood vessels10, were significantly reduced in the epiphysis of Vegfr2iΔEC mice (Fig. 3e–k). Mice with endothelial-specific deletion of Dll4 (Cdh5-BAC-CreERT2+Dll4flox/flox, hereafter referred to as Dll4iΔEC), the predominant Notch ligand in ECs, showed a dense but reduced vascularized area in the epiphysis (Fig. 3l,m and Extended Data Fig. 2e). Accordingly, Dll4iΔEC mice showed reduced osteogenesis as well as decreased numbers of pre-osteoprogenitors and osteoprogenitors (Fig. 3n–v). Notch signalling in osteogenic progenitors is reportedly critical for bone fracture repair28; therefore, we deleted Rbpj, the principal effector of Notch signalling, in Osx-expressing mesenchymal progenitors (Osx–CreERT2+Rbpjflox/flox, hereafter referred to as RbpjiΔosx). However, RbpjiΔosx mice had no apparent defect in osteogenesis (Extended Data Fig. 3a–h). In retinal angiogenesis, Dll4 deletion was reported to overcome the anti-angiogenesis effects of Vegfr2 deletion29. By contrast, in vascularization of the epiphysis, endothelial ablation of Vegfr2 prevented the hypervascularity induced by Dll4 deletion (Extended Data Fig. 3i–m). Csf1op/op mutant mice, in which osteoclasts are almost completely depleted30, did not show apparent vascular defects in the epiphysis, suggesting that osteoclasts are not relevant to vascularization of this area (Extended Data Fig. 4). Taken together, these data show that type S vessels are highly plastic during postnatal development, even more plastic than type H vessels, and robustly contribute to epiphyseal osteogenesis.

Fig. 3: Type S vessels are highly plastic and crucial for osteogenesis.
figure 3

a, Protocol for 4OHT injection in neonates. b–d, Femur sections (b) and whole-mount Alizarin Red staining (control (c) and Vegfr2iΔEC (d)) at P11. Vegfr2iΔEC mice show severely impaired angiogenesis (open arrowheads) and osteogenesis (arrows) in the epiphysis, although metaphyseal ossification is largely intact (asterisks). e–j, Immunohistochemistry of femur sections stained with Osx and DAPI (e,f) or Runx2 and Emcn (g–j). Panels i and j how the high magnification views of the area indicated in g and h. Panels e, g and i are from control, and panels f, h and j are from Vegfr2iΔEC mice. Osx+ and Runx2+ cells are found around vessels, their numbers are reduced (arrows) and they are detected only around surviving vessels (arrowhead) in Vegfr2iΔEC mice. k, Quantification of the SOC area (n = 3, biologically independent experiments) and Osx+ cells (n = 4, biologically independent experiments) in the epiphysis. l, Protocol for 4OHT injection in neonates. m–o, Femur sections (m) and whole-mount Alizarin Red staining (control (n) and Dll4iΔEC (o)) at P11. Dll4iΔEC mice have an increased vessel density but a reduced vascularized area (open arrowheads) and reduced osteogenesis (arrows) in the epiphysis, although metaphyseal ossification is largely intact (asterisks). p–u, Immunohistochemistry of femur sections stained with Osx and DAPI (p,q) or Runx2 and Emcn (r–u). Panels t and u show the high magnification views of the area indicated in r and s. Panels p, r and t are from control, and panels q, s and u are from Dll4iΔEC mice. Osx+ and Runx2+ cells are found around vessels and their numbers are reduced (arrows) in Dll4iΔEC mice. v, Quantification of the SOC area (n = 3, biologically independent experiments) and Osx+ cells (control: n = 4, Dll4iΔEC: n = 6, biologically independent experiments) in the epiphysis. Data are presented as the mean ± s.d. ***P < 0.001; two-tailed Student’s t-test. Scale bars, 1 mm (c–h and n–s); 200 µm (b,i,j,m,t,u). The comparisons between the averages of two groups were evaluated using the two-tailed Student’s t-test.

Source data

Full size image

Type S vessels regulate osteogenesis by secreting collagen I

Considering the data described above, we were interested in how type S vessels contribute to epiphyseal osteogenesis. Type S ECs did not apparently preferentially express canonical osteogenic factors such as bone morphogenetic proteins (Bmps), fibroblast growth factors (Fgfs) and insulin-like growth factors (Igfs), in contrast with their specific expression of Col1a1 and Col1a2 (Fig. 4a). This expression pattern of Col1a1 and Col1a2 was confirmed by quantitative (q)PCR analysis of sorted BMECs (Fig. 4b). Bone is a hybrid tissue with two main components: collagen fibril frameworks and calcium phosphate minerals. Type I collagen is the most abundant protein constituting collagen fibrils in bones. Therefore, we generated endothelial-specific Col1a1-knockout (Cdh5-BAC-CreERT2+Col1a1flox/flox, hereafter referred to as Col1a1iΔEC) mice and analysed the collagen pattern in epiphyseal bone tissues by transmission electron microscopy (TEM) at P11 (Fig. 4c). As expected, the bands of collagen fibres around blood vessels were thin and sparse in Col1a1iΔEC mice (Fig. 4d–i). It is interesting that mineral deposition evaluated by Alizarin Red staining was significantly reduced in the epiphysis, but not in other bone portions, of Col1a1iΔEC mice (Fig. 4j,k). The numbers of pre-osteoprogenitors and osteoprogenitors were not significantly affected in Col1a1iΔEC mice (Fig. 4l–p); therefore, insufficient collagen frameworks might impair mineral deposition or stability31. At 5 weeks after birth, Col1a1iΔEC mice showed reduced X-ray intensity (Fig. 4q), reflecting impaired mineral deposition. Accordingly, bone strength evaluated by the compression test was significantly reduced in the epiphysis, but not in the diaphysis, of Col1a1iΔEC mice (Fig. 4r–t). These data suggest that mesenchymal cell-derived type I collagen is insufficient for epiphyseal osteogenesis, but ECs complement this insufficiency by secreting type I collagen.

Fig. 4: Type S vessels contribute to osteogenesis by secreting type I collagen.
figure 4

a, A heatmap showing osteogenic factors based on the scRNA-seq data shown in Fig. 1. b, Quantitative PCR analysis of CD31+CD45− cells derived from femurs of P11 mice (n = 3, biologically independent experiments). c, Protocol for 4OHT injection in neonates. d–i, TEM of epiphyseal sections of P11 mice. Panels d, f and h are from control, and e, g and i are from Col1a1iΔEC mice. Panels f and g show the area of perivascular collagen fibres with dotted area of d and e. Panels h and i show high magnification views of the area indicated in d and e, respectively. The bands of collagen fibres around blood vessels (dotted area) are thin and sparse in Col1a1iΔEC mice (asterisks). j,k, Whole-mount Alizarin Red staining of femurs at P11 (j) and quantification (k) (control: n = 7; Col1a1iΔEC: n = 6, biologically independent experiments). Arrows indicate reduced mineralization in the epiphyses and the asterisk indicates intact ossification in the diaphyses and metaphyses of Col1a1iΔEC mice. l–p, Immunohistochemistry of femur sections at P11 stained with Runx2, Emcn and DAPI (l,m) or Osx, Emcn and DAPI (n,o). Panels l and n are from control, and m and o are from Col1a1iΔEC mice. Quantification of Osx+ cells (n = 4, biologically independent experiments) is shown in p. q, X-ray images of adult mice. Col1a1iΔEC mice show reduced X-ray intensity (asterisks) in the epiphysis. r–t, Mechanical strength of the femoral distal epiphysis or diaphysis measured by a compression test. Panel r shows the scheme of the test, and quantification of the maximum load of the epiphysis (s) and the metaphysis (t) is shown individually (control: n = 4; Col1a1iΔEC: n = 3, biologically independent experiments). The epiphysis has inferior biomechanical properties in Col1a1iΔEC mice. Scale bars, 1 mm (j and q); 200 μm (l–o); 10 μm (d–g); and 1 μm (h and i). *P < 0.05; data presented are the mean ± s.d. The comparisons between the averages of two groups were evaluated using the two-tailed Student’s t-test.

Source data

Full size image

Type S vessels establish a satellite niche for HSPCs

In haematopoietic analysis of BM cells, epiphyses are usually discarded during the process of flushing out the BM cavity. Therefore, the emergence and residence of HSPCs in epiphyses have not been intensively studied. Although Sca1 staining only marked endothelial cells (Fig. 1j), visualization of putative HSPCs by c-Kit immunostaining (Fig. 5a–g,n) showed that these cells started to appear in the entry point of type S vessels around P11 (Fig. 5d,e). Thereafter, they expanded throughout the BM cavity in the epiphysis until weaning (Fig. 5c,f,g). Next, we investigated MSCs, one of the most important components of the HSPC niche11,19, using Cxcl12+/DsRed mice. Although DsRed expression in these mice was detected in only a few AECs before P7, it was robustly detected in perivascular reticular cells after P13, suggesting that the emergence of MSCs synchronizes with the growth of type S vessels (Fig. 5h–o and Extended Data Fig. 5a–g). HSPCs were located in close proximity to type S ECs or MSCs (Fig. 5m), suggesting the existence of vascular and mesenchymal HSPC niches11. Lineage tracing using Cdh5−BAC-CreERT2 mice showed that c-Kit+ HSPCs were not labelled, excluding the possibility of their de novo specification from haemogenic endothelium (Fig. 5p–r). Rather, they were supposedly derived from circulating HSPCs in peripheral blood. Reanalysis of scRNA-seq data found that type S ECs showed moderate but higher expression of some HSPC niche factors such as Cxcl12, Tgfb1, Tgfb2, Spp1 (osteopontin) and TnC than type H ECs (Fig. 5s). We also visualized haematopoietic stem cells (HSCs) by performing CD150 staining via modification of a protocol reported by other groups25,32,33, and found that the distance between HSCs and ECs was smaller in epiphyses than in metaphyses (Fig. 5t,u). Consistently, the number of c-Kit+ HSPCs in epiphyses was significantly reduced in BM sections of Vegfr2iΔEC mice at P18 (Fig. 6a,b). It is of interest that MSCs were associated with type S ECs and thus the number of these cells was also decreased in Vegfr2iΔEC mice (Fig. 6b). Flow cytometric analysis showed that the numbers of lineage-Sca1+c-Kit+ (LSK) HSPCs and CD34lowLSK HSCs were significantly decreased in adult Vegfr2iΔEC mice (Fig. 6c,d). Taken together, these data demonstrate that type S vessels postnatally establish a satellite niche for HSPCs that is distant from the main BM cavity.

Fig. 5: Type S vessels postnatally establish a satellite niche for HSPCs.
figure 5

a–g, Immunohistochemistry of femur sections at 3 months after birth (a–c), P7 (d), P11 (e), P13 (f) and P18 (g). Panels b and c show the high magnification views of the area indicated in a. The c-Kit+ cells start to appear in the entry point of type S vessels around P11 (arrows) and gradually spread in the epiphysis thereafter (arrowheads). h–m, Immunohistochemistry of femur sections from Cxcl12+/DsRed mice at P7 (h), P11 (i), P13 (j) and P18 (k), and 3 months after birth (l,m). Panel m shows the high magnification view co-stained with cKit and Emcn. The arrows indicate DsRed expression detected in arteries. MSCs appear at P11 and spread throughout the epiphysis thereafter (open arrowheads). Most c-Kit+ cells are located in contact with MSCs (closed arrowhead) or ECs (open arrow). n,o, Quantification of c-Kit+ cells (n) and Cxcl12-DsRed+ cells (o) in BM sections at various stages (n = 3, biologically independent experiments). p, Protocol for 4OHT injection in neonates. q,r, Immunohistochemistry of femur sections. Panel r shows the high magnification view of the area indicated in q. Lineage tracing using Cdh5-BAC-CreERT2 mice shows that c-Kit+ cells (arrowheads) do not originate from haemogenic endothelium in type S vessels. s, A heatmap showing HSPC niche factors based on the scRNA-seq data shown in Fig. 1. t,u, Immunohistochemistry of femur sections from a mouse at 3 months (t) and quantification (u) (n = 8, biologically independent experiments). The distance between HSCs (open arrowheads) and ECs is smaller in epiphyses than in metaphyses. Asterisks are CD31+ megakaryocytes. Scale bars, 1 mm (a and q); 200 µm (b–g and h–l,r); 50 µm (m); and 10 µm (t). Data presented are the mean ± s.d. *P < 0.05. The comparisons between the averages of two groups were evaluated using the two-tailed Student’s t-test.

Source data

Full size image
Fig. 6: Type S vessels harbour HSCs.
figure 6

a, Protocol for 4OHT injection during the second week after birth. b, Immunohistochemistry of femur sections at P18 and quantification (vessel area: n = 3, c-Kit+ cells per FOV: n = 4, biologically independent experiments). Vegfr2iΔEC mice lack c-Kit+ cells especially in the avascular area (asterisks). c, Protocol for tamoxifen injection in adult mice. d, Flow cytometric analysis of BM cells isolated from the epiphysis of 12-week-old mice and quantification (n = 3, biologically independent experiments). Scale bar, 200 µm. **P < 0.01; *P < 0.05. Data presented are the mean ± s.d. The comparisons between the averages of two groups were evaluated using the two-tailed Student’s t-test.

Source data

Full size image

Epiphyseal HSCs reconstitute long-term haematopoiesis

Stromal cells in the metaphysis, including osteogenic cells and type H ECs, provide a favourable microenvironment for dormant HSCs14,24,34. This prompted us to compare the ability of metaphyses and epiphyses to function as an HSPC niche. Flow cytometric analysis showed that there were comparable numbers of HSPCs and HSCs in metaphyses and epiphyses (Fig. 7a,b). Next, we conducted a bone marrow transplantation (BMT) and compared the haematopoietic reconstitution activity of epiphyseal and metaphyseal HSCs. First, donor cells (4 × 105 BM mononuclear cells (BMMNCs)) isolated from the epiphysis or metaphysis of Ly5.2 mice were transplanted into lethally irradiated Ly5.1 mice (n = 6 per group). Although there was no difference in chimaerism in peripheral blood between metaphyseal and epiphyseal cells during the first month, epiphyseal cells showed higher reconstitution activity than metaphyseal cells at 2 months after transplantation (Fig. 7c,d). The frequency of donor-derived HSPCs in the BM was also higher in mice transplanted with epiphyseal cells than in mice transplanted with metaphyseal cells (Fig. 7e). This difference in reconstitution activity between epiphyseal and metaphyseal cells became much more evident when BMMNCs were serially transplanted into secondary recipient mice (Fig. 7f,g). Next, we evaluated the cell-cycle state in HSPCs and HSCs (Fig. 7h,i). More HSPCs and HSCs were in G0 phase in epiphyses than in metaphyses, suggesting that epiphyseal HSCs are relatively dormant, which probably accounts for their superior BM reconstitution ability without a difference in phenotypic characteristics in flow cytometric analysis. Moreover, we assessed the functional capacities of single HSCs by performing limiting dilution analysis of donor HSCs from metaphyses and epiphyses. At 3 months after transplantation, the engraftment rate of donor cells from epiphyses was 2.5× higher than that of donor cells from metaphyses, although the difference was not significant (P = 0.19; n = 8 recipients per group) (Extended Data Fig. 6). These data suggest that type S vessels constitute a distinct and efficient reservoir for HSCs separate from the main BM cavity.

Fig. 7: Epiphysial HSCs reconstitute long-term multilineage haematopoiesis.
figure 7

a,b, Flow cytometric analysis of BM cells isolated from the epiphysis and metaphysis of 12-week-old mice (a) and quantification (b) (n = 3, biologically independent experiments). c,d, Quantification of donor-derived cells (c) and lineage differentiation in peripheral blood samples (d) after the first BMT (n = 6, biologically independent experiments). e, Chimaerism of donor-derived cells in the BM (n = 6, biologically independent experiments) at 4 months after the first BMT. f,g, Quantification of donor-derived cells (f) and lineage differentiation in peripheral blood samples (g) after the second BMT (n = 6, biologically independent experiments). h,i, Flow cytometric analysis of BM cells isolated from the epiphysis and metaphysis of 12-week-old mice (h) and quantification (i) (n = 4, biologically independent experiments). The cell-cycle state was determined by DAPI and Ki67 staining. ***P < 0.001; **P < 0.01; *P < 0.05. Data presented are the mean ± s.d. The comparisons between the averages of two groups were evaluated using the two-tailed Student’s t-test.

Source data

Full size image

Discussion

In the present study, using scRNA-seq, we thoroughly characterized a spatial heterogeneity of transcriptomes in BMECs. During the analysis, we identified a capillary subtype found exclusively in the epiphysis. These cells possessed unique structural and gene expression features and robustly contributed to bone strength as well as haematopoietic maintenance.

Previous studies analysing BM cells by scRNA-seq uncovered the cellular taxonomy of stromal compartments such as mesenchymal stem/stromal cells, chondrocytes, fibroblasts, BMECs and pericytes15,16. In these studies, BMECs were largely divided into two populations: AECs and SECs. Our current analysis annotated 11 EC subclusters. These previous studies did not recognize some EC populations, in particular type H ECs, which are probably included in the AEC cluster because they used Ly6a expression as a hallmark of AECs. It should be noted that the BM was flushed and enzymatically digested after the epiphyses were cut out and therefore type S ECs were inevitably discarded in these studies.

Recent discoveries in the field of bone metabolism greatly improved our understanding of angiogenic–osteogenic coupling, especially in endochondral ossification of long bones6,7,8,9,22,23. Type H vessels promote bone formation and contribute to the maintenance of HSPCs14,23. Pharmacological augmentation of type H vessels enhances bone fracture healing and counteracts bone loss in ovariectomy-induced osteoporotic mice35. Our current study also underscores the importance of type S vessels, especially in the ossification of epiphyses. In terms of physical activity, type S ECs are critically important because epiphyses sustain the stability of joints, which often becomes problematic in ageing and various bone diseases such as osteoporosis and osteonecrosis. On the other hand, from the viewpoint of systemic haematopoietic maintenance, the contribution of type S ECs might be minimal considering the small size of epiphyseal BM. Although our experiments showed that type S ECs harbour transplantable HSCs, epiphyses probably act as a satellite niche, which might be utilized in an emergency such as extensive loss of the main BM cavity. In general, BMECs have two functional aspects: haematopoiesis and osteogenesis. Type S ECs may prioritize the latter compared with BMECs in other fragments, although it is unclear which biological factor regulates their prioritization of osteogenesis.

Type I collagen is produced by mesenchymal cells such as OBs and fibroblasts. Our data showing that ECs are an important source of type I collagen specifically in epiphyses raise the following question: why do epiphyses employ such a unique ossification system? We speculate that the mode of epiphyseal growth answers this question. Epiphyses undergo rapid ossification and angiogenesis synchronously in the postnatal stage, when the joints are already used daily and therefore epiphyses are exposed to friction forces. In this complicated and emergent process, it is speculated that type S ECs participate in matrix deposition, which is conducted by OBs alone in other fragments. Such a function of ECs could be a unique angiocrine mode3, that is, constructing the framework required for organ functions. However, it should be mentioned that the phenotype of Vegfr2iΔEC mice was much more severe than that of Col1a1iΔEC mice, suggesting that the contribution of type S ECs to osteogenesis cannot be solely attributed to type I collagen secretion. Other angiocrine factors regulating osteogenesis36 or delivery of oxygen and nutrients might account for the phenotypic gap between these two mutant lines.

It is unclear how type S ECs preferentially produce type I collagen compared with other collagens and what makes type S ECs distinct from other EC subtypes. Epigenetic regulation might account for this unique expression in type S ECs. Moreover, type S ECs might have some characteristic features in multiple steps of collagen synthesis such as posttranslational modification in the endoplasmic reticulum, release from the ribosome, glycosylation of procollagens and secretion37.

In the future, it is important to determine whether type S ECs are present in humans because epiphyseal blood vessels are critically involved in human diseases such as Legg–Calvé–Perthes disease (LCPD), a type of femoral head necrosis that affects children. There is no known cause of impaired blood supply to the femoral head in LCPD38.

Overall, our current study provides new insight into the landscape for the cellular architecture and heterogeneity of BMECs and uncovers the importance of epiphyseal ECs, which has been overlooked so far. In the future, our findings may lead to the discovery of new therapeutic options for bone diseases such as LCPD and osteoporosis.

Methods

Mice

Animal experiments were approved by the Institutional Animal Care and Use Committee of Keio University (protocol no. A2021-003) and were performed in accordance with the Guidelines of Keio University for Animal and Recombinant DNA Experiments (protocol no. D2005-044). Cdh5-BAC-CreERT2 (ref. 39), Osx-CreERT2 (ref. 8), Vegfr2-flox40, Dll4-flox41, Col1a1-flox42, CAG-LSL-EGFP43, Vegfr1-BAC-DsRed44, Vegfr2-BAC-GFP44 and Rbpj-flox45 mice had been developed previously. All mice were crossed with C57BL/6J mice more than 8× and maintained. For analyses of phenotypes, mice of both sexes were used. Csf1op/op (stock no. 000231) and Cxcl12+/DsRed (stock no. 0225458) mice were obtained from the Jackson Laboratory. For neonatal experiments of tamoxifen-inducible expression of Cre, 40 μg of 4-hydroxytamoxifen (4OHT, Sigma-Aldrich, catalogue no. H6278) was subcutaneously injected at the indicated timepoints. For adult experiments of tamoxifen-inducible expression of Cre, 1 mg of tamoxifen (Sigma-Aldrich, catalogue no. H5648) was subcutaneously injected at the indicated timepoints.

Preparation of tissue sections

Samples were collected and placed in 4% paraformaldehyde at 4 °C for 16–24 h under rotation. Decalcification was performed with 0.5 M EDTA for 16–24 h at 4 °C under rotation. Decalcified bone was infiltrated with 30% sucrose for 2 d under rotation, embedded with optimal cutting temperature compound and sectioned at a thickness of 100 μm using tungsten carbide knives (Thermo Fisher Scientific).

Immunostaining

Sections were air-dried at room temperature, rehydrated in phosphate-buffered saline (PBS) and then blocked using 2% bovine serum albumin (BSA) overnight at 4 °C. Sections were incubated with primary antibodies overnight at 4 °C in a humidified chamber and then with secondary antibodies overnight at 4 °C in a humidified chamber. After thorough washing, sections were mounted. The following primary monoclonal antibodies were used: anti-CD31 (Abcam, catalogue no. ab119341; 1:1,000), anti-Runx2 (Abcam, catalogue no. ab192256, 1:500), anti-ASMA (Sigma-Aldrich, catalogue no. A5228; 1:200), anti-Ly6a (BD Biosciences, catalogue no. 553333; 1:500), anti-endomucin (Santa Cruz, catalogue no. sc-65495; 1:500) and anti-CD150 (BioLegend, catalogue no. 115905; 1:200). The following primary polyclonal antibodies were used: anti-green fluorescent protein (GFP)-Alexa Fluor-488 (Molecular Probes, catalogue no. A21311, 1:500), anti-c-Kit (R&D Systems, catalogue no. AF1356, 1:500), anti-Sp7/Osx (Abcam, catalogue no. ab22552, 1:500), anti-cathepsin K (Abcam, catalogue no. ab19027, 1:500), anti-VEGFR3 (R&D Systems, catalogue no. AF743, 1:1000), anti-aggrecan (Sigma-Aldrich, catalogue no. AB1031, 1:500), anti-CGRP (Sigma-Aldrich, catalogue no. C8198, 1:8000) and anti-Erg (Abcam, catalogue no. ab92513, 1:2,000). The secondary antibodies used were Alexa Fluor-488-conjugated immunoglobulin (Ig)G (Molecular Probes, catalogue nos. A11034, A11006, and A11055; 1:500) and Cy3/Cy5 DyLight549/DyLight649-conjugated IgG (Jackson ImmunoResearch, catalogue nos. 711-165-152, 112-165-167, 127-165-160, 711-605-152, 112-605-167 and 127-605-160; 1:500). For nuclear staining, specimens were treated with DAPI (Molecular Probes, catalogue no. D-1306). For immunostaining of both intracellular and extracellular aggrecan, slides were pretreated with hyaluronidase (Sigma-Aldrich, catalogue no. H4272, 2 mg ml−1) at 37 °C for 30 min before primary antibody staining.

Confocal microscopy

Fluorescent images were obtained using a confocal laser scanning microscope (Olympus, FV3000). Obtained images were analysed using FV10-ASW 3.0 Viewer (Olympus). Cells or substances of interest were quantified in the entire area of the epiphysis, in a 600 × 600 µm2 field of view (FOV; to count OSX+ and c-Kit+ cells and to quantify relative intensity of immunofluorescence), or in a 1-mm length of the vascular front (to count tip-cell filopodia). To measure the distance between HSCs and ECs, two HSCs were randomly picked per scanned image. ImageJ software (the US National Institutes of Health) was used to quantify the indicated areas in scanned images.

Flow cytometric analysis of BMECs

Femurs of wild-type mice were collected and incubated for 30 min at 37 °C in Dulbecco’s modified Eagle’s medium (DMEM) containing 1% collagenase D (from Clostridium histolyticum, Sigma-Aldrich, catalogue no. C5138-1G), 1 U ml−1 of dispase (Thermo Fisher Scientific, catalogue no. 17105-041) and 1 U ml−1 of DNase (Invitrogen) before cells were dissociated by trituration. Cells were dissociated to generate a single-cell suspension by filtering through a 40-μm nylon mesh. RBCs were eliminated using RBC Lysis Buffer (pluriSelect, Deutscher, catalogue no. 60-00050-11). Cells were subjected to Fc blocking (BioLegend, catalogue no. 101320, 1:200) and stained with FITC-conjugated anti-CD31 (BioLegend, catalogue no. 102506, 1:200), allophycocyanin (APC)-conjugated anti-CD45 (BioLegend, catalogue no. 103112, 1:200), phycoerythrin (PE)-conjugated anti-Ly6a (BioLegend, catalogue no. 122507, 1:200) and APC/Cy7-conjugated anti-Ly6c (BioLegend, catalogue no. 128025, 1:200) antibodies for 40 min. Analysis was performed with an SH800S Cell Sorter (SONY) and FlowJo software (Tree Star).

ScRNA-seq

ScRNA-seq was performed using a Chromium instrument (10x Genomics). Single-cell suspensions from femurs were generated as described above. Cells were subjected to Fc blocking and stained with FITC-conjugated anti-mouse CD31 (BioLegend, catalogue no. 102506, 1:200) and APC-conjugated anti-mouse CD45 (BioLegend, catalogue no. 103112, 1:200) antibodies for 40 min. CD45−CD31+ BMECs were sorted with an SH800S Cell Sorter. Libraries were constructed using the commercial microdroplet-based platform Chromium Single Cell 3′ v.3.1 (10x Genomics) according to the manufacturer’s instructions. A DNBSEQ-G400 platform (MGI) was used to sequence 28 and 91 bp-end reads. Sequenced reads were aligned to the reference genome (mm10) using Cell Ranger v.3.0.2 software (10x Genomics). The data were processed and analysed using Seurat package v.3.5.2 43. UMAP was used for dimensionality reduction that captures both local and global structures in scRNA-seq data. Clusters were visualized in two dimensions as a feature or violin plot and annotated based on expression of canonical endothelial subtype markers. To construct the lists of upregulated genes in type S ECs, differentially expressed genes were extracted using Wilcoxon’s rank-sum test in the Seurat package. IPA (Ingenuity Systems) was performed to identify the most significant pathways and to construct molecular interaction networks of differentially expressed genes (log(FC) > 0.5 or <−0.5) in type S ECs. In brief, the gene list file containing gene symbols, FCs and P values was uploaded to IPA and core analysis was performed.

Flow cytometric analysis of HSPCs and HSCs

Femurs of 12-week-old wild-type mice were collected in PBS. The epiphyses and metaphyses were crushed separately and BM cells were collected for analysis and isolation of HSCs and HSPCs. After haemolysis, surface markers were stained with fluorophore-conjugated antibodies. HSCs and HPCs were phenotypically defined based on the presence and absence of surface markers, as follows: HSCs, lineage−Sca-1+c-Kit+Flt3−CD150+CD48−; HPCs, lineage−Sca-1+c-Kit+Flt3+. The methods used to identify and isolate HSCs and other haematopoietic cells have been described previously32,46,47. Cell sorting was performed using a FACS Aria II or FACS Aria IIIu instrument (BD Biosciences). Data were analysed with FlowJo software (Tree Star). Anti-lineage marker (CD4, CD8a, Gr-1, Mac-1, Ter-119, B220)-PerCP-Cy5.5, anti-c-Kit-APC-Cy7, anti-Sca-1-PE-Cy7, anti-CD150-PE, anti-CD48-FITC and anti-Flt3-APC antibodies were used. These antibodies were purchased from BD Biosciences, BioLegend, TONBO Biosciences and eBioscience.

Quantitative PCR analysis

Total RNA was prepared from isolated cells, and reverse transcription was performed using Superscript III (Invitrogen). The qPCR assays were performed with the ABI 7500 Fast Real-Time PCR System using TaqMan Fast Universal PCR Master Mix (Applied Biosystems) and TaqMan Gene Expression Assay Mix with Col1a1 (Mm01302043_g1). A mouse β-actin (Mm00607939_s1) assay mix served as an endogenous control. Data were analysed using 7500 Fast System SDS software v.1.3.1.

Cell-cycle analysis of HSPCs

Femurs from wild-type mice at 12 weeks postnatally were collected in PBS containing 2% fetal calf serum (FCS). The epiphyses and metaphyses were crushed separately using a mortar and pestle on ice followed by filtration with a 70-μm cell strainer (Corning), the diaphyses were flushed with PBS containing 2% FCS and BM cells were collected. After lysis of RBCs, 1–4 million BMMNCs were aliquoted into Eppendorf tubes and 1 µl of Fc-block was added to each sample and incubated for 10 min at 4 °C. Anti-CD150-BV421, anti-CD48-FITC, anti-lineage (CD4, CD8a, Gr-1, Mac-1, Ter-119, B220)-PerCP-Cy5.5, anti-c-Kit-APC-Cy7 and anti-Sca-1-PE-Cy7 antibodies were used to detect surface antigens. Stained samples were centrifuged at 340g (2,000 r.p.m.) and 4 °C for 5 min and then 250 µl of BD Cytofix/Cytoperm (BD Biosciences) was added and incubated for 20 min at room temperature for fixation. Fixed cells were washed twice at 2,000 r.p.m. and 4 °C with 1 ml of BD Perm/Wash buffer (BD Biosciences). Each sample was stained with 1 µl of Ki67-eFluor660 (eBioscience) for 30 min at room temperature in the dark. Ki67-stained cells were washed once at 2,000 r.p.m. and 4 °C with BD Perm/Wash buffer. Samples were resuspended in 500 µl of PBS containing 2% FCS and 10 µg ml−1 of Hoechst 33432, filtered and then analysed with a BD FACS Aria IIIu instrument.

BMT

Femurs of wild-type mice at 12 weeks postnatally were collected in PBS containing 2% FCS. The epiphyses and metaphyses were crushed separately and the diaphyses were flushed to collect BM cells for BMT. A total of 4 × 105 BMMNCs from C57BL/6J-Ly5.2 and 4 × 105 BMMNCs from a competitor recipient (C57BL/6J-Ly5.1) were transplanted retro-orbitally into lethally irradiated (9.5 Gy) recipients (Ly5.1). At 1, 2, 3 and 4 months after BMT, peripheral blood was collected and the percentage of donor-derived cells and their differentiation status were determined using MACSQuant v.2.1.1 (Miltenyi). Then, 40–80 μl of peripheral blood was sampled from the retro-orbital plexus using heparinized glass capillary tubes (Drummond Scientific) and suspended in 1 ml of PBS containing heparin. The blood suspension was centrifuged at 340g for 3 min. The supernatant was discarded and the pellet was resuspended in 1 ml of PBS containing 1.2% (w:v) dextran (200 kDa; Nacalai Tesque) for 45 min at room temperature. The supernatant was centrifuged at 340g for 3 min and pellets were resuspended in 0.17 M NH4Cl solution to lyse residual RBCs for 5–10 min until the suspension became clear. Cells were resuspended in 50 μl of PBS containing 2% FCS with 0.3 μl of Fc-block. Surface antigens were stained using the following antibody panel: Gr1-PE-Cy7, Mac-1-PE-Cy7, B220-APC, CD4-PerCP-Cy5.5, CD8a-PerCP-Cy5.5, CD45.1-PE and CD45.2-FITC; 0.3 μl of each antibody was added per sample. The frequency (%) of donor-derived cells was calculated as follows:

100 × donor-derived (GFP+ or Ly5.2+Ly5.1−) cells (%)/(donor-derived cells (%) + competitor- or recipient-derived (Ly5.2-Ly5.1+ or GFP−) cells (%)).

Myeloid cells, B cells and T cells were marked by Gr-1+ or Mac-1+, B220+ and CD4+ or CD8+, respectively. Total cell chimaerism represents the frequency of donor-derived Ly5.2+Ly5.1− cells over the frequency of Ly5.2−Ly5.1+ cells among mononuclear cells. At 4 months after BMT, the frequency of donor-derived cells in BM was determined using one femur and tibia per recipient. After counting BM cells using a TC10 automated counter (BioRad), equal volumes of cell suspensions (20–30% of total volume) from each recipient were pooled and 2 × 106 cells were resuspended in SF-O3 medium containing 0.1% BSA. The cell suspension (200 μl per recipient) was transplanted by retro-orbital injection into lethally (9.5 Gy) irradiated Ly5.1+ recipients with a 1-ml syringe and 27-gauge needle. The remaining cells were stained to assess BM chimaerism. Anti-CD150-BV421, anti-CD48-PE, anti-Flt3-APC, anti-lineage (CD4, CD8a, Gr-1, Mac-1, Ter-119, B220)-PerCP-Cy5.5, anti-c-Kit-APC-Cy7, anti-Sca-1-PE-Cy7, anti-Ly5.1-Alexa Fluor-700 and anti-Ly5.2-FITC antibodies were used to detect surface antigens. Then, 1 μl of each antibody was added per sample.

For limiting dilution analysis to estimate the frequency of HSCs, BM cells were collected from the metaphyseal or diaphyseal region of a cohort of four mice, as described above. HSCs were sorted at a density of 1, 3 or 10 cells per well directly into a 96-well, non-treated, U-bottomed plate (Greiner) containing 300 μl of SF-O3 medium with 8 × 105 Ly5.1+ whole BM cells. Individual cell suspensions were carefully collected from each well using a 27-gauge, 1-ml syringe (Terumo) and subsequently administered via retro-orbital injection into lethally irradiated (8.5 Gy) Ly5.1+ recipient mice. At 3 months posttransplantation, the frequency of donor-derived cells was assessed as detailed above. Mice exhibiting donor-derived cell frequencies >0.1% and the existence of donor-derived B cells, T cells and myeloid cells were considered to be successfully repopulated. The HSC frequency was estimated using the elda package of R software v.1.5.0 (R-project.org).

TEM analysis

Bone samples were prepared for TEM observation as described previously48. Briefly, the tissues were dissected and fixed with 2.5% glutaraldehyde (TAAB Laboratories Equipment Ltd) diluted in 30 mM HEPES buffer (pH 7.4, Sigma-Aldrich) for 16–24 h under rotation at 4 °C. Decalcification was performed with 0.5 M EDTA for 3 weeks at 4 °C under rotation. After 2 h of postfixation with 1.0% OsO4 (TAAB Laboratories Equipment Ltd), samples were dehydrated in a series of increasing concentrations of ethanol (70–100%) and then in acetone (Sigma-Aldrich), n-butyl glycidyl ether (QY1; Oken-shoji Co. Ltd) and graded concentrations of Epon prepared in QY-1. Samples were then incubated in 100% Epon (100 g of Epon contains 27.0 g of 1-methylnicotinamide, 51.3 g of EPOK-812, 21.9 g of dodecenyl succinic anhydride and 1.1 ml of 2,4,6-tris(dimethylaminomethyl)phenol-30, all from Oken-shoji Co. Ltd) for 72 h at 4 °C to enhance infiltration of the resin. After 72 h of polymerization in 100% Epon at 60 °C, semithin sections (1-μm thick) were prepared with a glass knife and stained with toluidine blue. Ultrathin sections (70-nm thick) were prepared using an ultramicrotome (Leica UC7, Leica Biosystems) with a diamond knife. A slice every ten slices was collected on a copper grid and silicon wafer. Ultrathin sections were stained with uranyl acetate and subsequently with lead citrate for 10 min. Dried sections on copper grids were examined with a transmission electron microscope (JEM-1400plus, JEOL).

X-ray and micro-CT analyses

Soft X-rays were obtained with Softex type CMB2 (Softex Co.) and micro-computed tomography (CT) images were obtained using a micro-CT apparatus (R_mCT2; Rigaku) under 90 kV and 160 μA with 512 projections per 360°. Micro-CT images were analysed using Tri/3DBON software (Ratoc Systems Engineering).

Whole-mount Alizarin Red staining

Femurs were collected and incubated overnight in both 100% ethanol and 100% acetone at room temperature. Next, samples were placed in 1% KOH overnight with shaking and in 0.005% Alizarin Red (Sigma-Aldrich, A5533) diluted in 1% KOH for 4 h. Finally, samples were placed in 1% KOH containing 50% glycerol overnight with shaking, imaged and stored in 100% glycerol.

Mechanical testing

Femurs were obtained immediately after animals were killed, and the femoral epiphyses were cut out. A femoral epiphysis compression test was conducted using a mechanical testing machine (Model TK-252C, Muromachi Kikai Co.). A load was placed at the centre of the bone at a rate of 2 mm min−1 until the bone broke. The parameters analysed included stiffness and ultimate force.

Statistics and reproducibility

Results are expressed as the mean ± s.d. The comparisons between the averages of two groups were evaluated using two-tailed Student’s t-test. P values < 0.05 were considered statistically significant. For histological analyses, at least three but typically more independent samples were quantified or qualitatively analysed, with each experimental repeat yielding highly similar results. No statistical method was used to predetermine sample size. No data were excluded from the analyses. The experiments were not randomized. PCR genotyping determined group allocation and the investigators were not blinded in this process. Data analysis was performed by different lab members involved in each analysis, to avoid conscious and unconscious bias. No statistical methods were used to predetermine sample sizes, but our sample sizes are similar to those reported in previous publications21,39. Data distribution was assumed to be normal but this was not formally tested.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.