Identification of the metaphyseal skeletal stem cell building trabecular bone

Abstract

Skeletal stem cells (SSCs) that are capable of self-renewal and multipotent differentiation contribute to bone development and homeostasis. Several populations of SSCs at different skeletal sites have been reported. Here, we identify a metaphyseal SSC (mpSSC) population whose transcriptional landscape is distinct from other bone mesenchymal stromal cells (BMSCs). These mpSSCs are marked by Sstr2 or Pdgfrb⁺Kitl⁻, located just underneath the growth plate, and exclusively derived from hypertrophic chondrocytes (HCs). These HC-derived mpSSCs have properties of self-renewal and multipotency in vitro and in vivo, producing most HC offspring postnatally. HC-specific deletion of Hgs, a component of the endosomal sorting complex required for transport, impairs the HC-to-mpSSC conversion and compromises trabecular bone formation. Thus, mpSSC is the major source of BMSCs and osteoblasts in bone marrow, supporting the postnatal trabecular bone formation.

Hypertrophic chondrocytes–derived metaphyseal skeletal stem cells contribute to trabecular osteogenesis.

INTRODUCTION

Skeletal stem cells (SSCs) with the properties of self-renewal and multipotency can generate osteoblast, chondrocytes, adipocytes, and bone mesenchymal stromal cells (BMSCs), thus participating in growth, maintenance, and repair of skeleton (1–3). Through the comprehensive use of single-cell transcriptome, fluorescence-activated cell sorting (FACS) and functional transplantation assays, as well as genetic lineage tracing, several SSC or mesenchymal stem cell (MSC) populations with varied markers, differentiation potential, and active stage have been identified in different anatomic regions, including the periosteum, resting chondrocytes, articular surfaces, bone marrow, as well as mandible and craniofacial suture (4–16). In bone marrow, BMSCs or skeletal stem and progenitor cells (SSPCs) are heterogeneous cell subpopulations that can be labeled by various cell markers, including platelet-derived growth factor receptor β (Pdgfrb), Gli1, leptin receptor (Lepr), early B cell factor 3 (Ebf3), Kitl, and Cxcl12 (17–22). BMSCs exhibit varying progenitor properties, largely due to the fact that BMSCs may contain distinct SSC populations. Recent studies reveal that Fgfr3⁺ bone marrow endosteal stem cells and Lepr⁺Adiponectin⁺ SSPCs contribute to osteogenesis in young and adult stages, respectively (23, 24). However, the expression of most SSC markers spans other cell types. Identifying previously unknown SSC populations with distinct markers, spatiotemporal characteristics, and cellular origin remains a challenge.

A previous report showed that metaphyseal BMSCs have greater potential for growth and multipotential differentiation than diaphyseal BMSCs, raising the possibility that the metaphysis may also harbor SSC populations (18). Grem1 labels some metaphyseal cells that can give rise to growth plate chondrocytes, osteoblasts, and BMSCs (8). Furthermore, growth plate chondrocyte or hypertrophic chondrocytes (HCs) can convert into osteoblasts and BMSCs that mainly occupy the metaphysis (5, 25–33), and a recent report showed that HCs can transform into marrow-associated SSPCs marked by Pdgfra and Lepr (32). In addition, metaphysis also houses Gli1⁺ and Osterix⁺ progenitors, which can produce trabecular osteoblasts and BMSCs (19, 34). However, other metaphyseal SSC (mpSSC) populations and their lineage relationships with known SSC/SSPC populations have not yet been fully identified.

Here, we identify a putative mpSSC population with a distinct transcriptional signature. These mpSSCs emerge at the fetal stage and are restricted to the location immediately underneath the growth plate. These mpSSCs are the first cells that HCs differentiate into when entering the metaphysis. The mpSSCs have properties of SSC, including self-renewal and multilineage differentiation, and serve as a major source of trabecular osteoblasts as well as BMSCs in the lower metaphyseal and diaphyseal bone marrow. We find that hepatocyte growth factor–regulated tyrosine kinase substrate (Hgs), a component of the endosomal sorting complex required for transport, is a critical regulator of HC-to-mpSSC conversion. Deletion of Hgs in HCs impairs HC-to-mpSSC conversion and compromises trabecular bone formation.

RESULTS

Sstr2⁺ or Pdgfrb⁺Kitl⁻ marks a putative SSC population in the metaphysis

In search of putative mpSSCs, we performed single-cell RNA sequencing (scRNA-seq) analysis of CD31⁻CD45⁻Ter119⁻ nonhematopoietic and nonendothelial skeletal cells in the tibia and femur of postnatal day 7.5 (P7.5) Col10a1-Cre;ROSA26^LSL-ZsGreen (Col10a1^ZsGreen) mice (fig. S1A). To ensure that HCs can pass the 10x Genomics microfluidic channels that are ~50 to 60 μm in diameter, we measured the size of HCs in vivo and in vitro using a newly established knock-in mouse line Col10a1-mem-GreenLantern-2A-mito-mScarlet-2A-Rox-ecDHFR-Rox-Cre-Rox12-ecDHFR-Rox12-polyA (Col10a1-memG-mitoS-DD-Cre-DD), where the membrane of HCs was labeled by a fluorescent protein mGreenLantern (fig. S1B). We found that the actual diameter of most HCs in the P7 tibia is less than 25 μm (fig. S1, C to F). ScRNA-seq transcriptomes of 5861 skeletal cells were visualized using a uniform manifold approximation and projection (UMAP). Mapping of historically defined marker genes onto the UMAP representation revealed 14 clusters, generally falling into three major subpopulations according to their gene signatures: (i) BMSCs expressing Pdgfrb, including cluster 1 (mpSSC), cluster 2 [bone mesenchymal progenitor cell (BMPC), marked by Hey1 and Slc20a2], cluster 3 [reticular cell (RetiC), marked by Kitl and Adipoq], cluster 7 [periosteal cell (POC), marked by Aspn and Tnn], and cluster 8 [pericyte (PeriC), marked Rgs5 and Acta2]; (ii) osteo-lineage cells (OLCs) marked by Col1a1, including cluster 4 [pre-osteoblast (preOb)] and clusters 5 and 6 (osteoblast, Ob-1 and Ob-2); (iii) chondrocytes and HCs marked by Acan, including clusters 9 to 11 (HC-1, HC-2, and HC-3), cluster 12 (preHC), and clusters 13 and 14 (chondrocyte, Chon-1 and Chon-2) (Fig. 1, A and B).

Fig. 1. Sstr2⁺ or Pdgfrb⁺Kitl⁻ marks a putative SSC population in the metaphysis.

(A) ScRNA-seq of 5861 skeletal cells from P7.5 hindlimb bones is visualized by UMAP (also see Methods and fig. S1). They are unbiasedly divided into 14 clusters. (B) Dot plot shows the expression level of selected cluster-enriched genes in 14 clusters. Dot size indicates the cell fraction of each cluster that expresses listed genes. Purplish gray color intensity indicates scaled average expression level. The red frame indicates the panel genes that define putative murine SSC. (C) Enriched genes of cluster 1 (mpSSC). Genes with count >1 are identified as positive expressed genes (+); count ≤0 are identified as negative (−). Cells filtered by combined gene expression patterns are labeled with magenta dots and are superimposed on the UMAP plot. (D) Representative multicolor IF staining shows the expression of mpSSC markers in P7.5 femur section. The white framed area is shown adjacently at higher magnifications. The metaphyseal zone (MPZ) right beneath the hypertrophic zone (HZ) is framed by dashed lines. Yellow arrows indicate mpSSCs that are labeled by Sstr2 or by a combined expression of Pdgfrb and CD200/CD51. The mpSSC pointed by the red arrow is shown below at a higher magnification. TZ, trabecular zone. Scale bars, 500 μm.

Next, we examined the expression of CD200/Itgav/Thy1/Enpep/Eng (encoding CD200/CD51/Thy1/Ly51/CD105, respectively) signature, which was used to isolate murine SSCs in long bones (3, 35). CD200⁺CD51⁺Thy1⁻Ly51⁻CD105⁻ signature was enriched in several clusters, including some BMSCs and OLCs as well as HC clusters, suggesting that cells with CD200⁺CD51⁺Thy1⁻Ly51⁻CD105⁻CD45⁻ signature may not represent but rather include possible SSCs (Fig. 1, B and C). Notably, among the four BMSC clusters marked with Pdgfrb (clusters 1, 2, 3, and 8), cluster 1 seemed to share the most enriched gene expression patterns with this CD200⁺Itgav⁺Eng⁻Thy1⁻Enpep⁻ signature (Fig. 1, B and C). The cluster 1 was separated from other skeletal cell clusters by Pdgfrb⁺Kitl⁻, Cd55, Msx1, and Rhoj (Fig. 1, B and C). A single marker somatostatin receptor 2 (Sstr2) could specifically label this subpopulation throughout all clusters (Fig. 1, B and C).

We then examined the location of cluster 1 in femurs of P8 mice using multicolor immunofluorescence (IF) staining. We found that Sstr2⁺ cells resided in the upper trabecular zone and right beneath the hypertrophic cartilage, expressed the BMSC/SSPC marker Pdgfrb, and extensively overlapped with most of the metaphyseal CD200⁺CD51⁺Thy1⁻Ly51⁻CD105⁻ cells (Fig. 1D, yellow arrows). Together, these above data suggest that Sstr2⁺ or Pdgfrb⁺Kitl⁻ cluster 1 would represent a putative SSC population in the metaphysis. We therefore named these cells mpSSCs.

Sstr2⁺ or Pdgfrb⁺Kitl⁻ cells are immediate derivatives of HCs

Previous genetic lineage tracing studies have demonstrated that most HCs can survive at the chondro-osseous junction and further convert into OLCs, adipocytes, or BMSCs (5, 25–32). Given that most HC progeny are enriched in the metaphysis, it is conceivable that HC may be the cellular source of the putative mpSSC in the metaphysis. We therefore performed scRNA-seq analysis of ZsGreen⁺ cells from tibias and femurs of P7.5 Col10a1-Cre;ROSA26^LSL-ZsGreen mice (Col10a1^ZsGreen), which include Col10a1⁺ HCs and their progeny (fig. S1A). A total of 1998 ZsGreen-expressing cells included 1072 cells sorted by FACS and 926 cells algorithmically extracted from the unsorted scRNA-seq data (fig. S1A). ZsGreen⁺ cells were visualized using UMAP, and 11 clusters were divided into three major subpopulations: (i) HCs marked by Col10a1 (clusters 1, 2, 3, HC-1, HC-2, HC-3, respectively); (ii) BMSCs expressing Pdgfrb that located at the trabecular bone and the bone marrow [cluster 4, mpSSC; cluster 5, trabecular BMSCs (tBMSC); cluster 6, RetiC; part of cluster 7, proliferating BMSCs/OLCs (Prol-BMSC/OLC)]; (iii) OLCs marked by Col1a1, residing at the trabecular bone and the endosteum [part of cluster 7; cluster 8 (Ob); cluster 9, mature osteoblast (mOb); cluster 10, osteocyte (Osc); cluster 11, transitional OLC (Trans-OLC)] (Fig. 2, A and B, and fig. S2, A and B). Notably, cluster 4 (mpSSC) was marked by Sstr2 (Fig. 2B), Cd55, Pdgfrb⁺Kitl⁻, Msx1, and Rhoj (fig. S2, A and B). The transcriptional signature of Sstr2⁺ cluster 4 was similar to that of Sstr2⁺ cluster 1 of the scRNA-seq from all skeletal cells (Fig. 1, B and C), indicating that they might be the overlapping cluster. Multiple IF staining of Col10a1-Cre;ROSA26^LSL-tdTomato;Kitl^GFP (Col10a1^tdTomato;Kitl^GFP) mice revealed that all Sstr2⁺ cells were positive for tdTomato, suggesting that HC is the primary source for the Sstr2⁺ cells (Fig. 2C, white arrows, and fig. S2, C and D). Notably, multiple IF staining also confirmed the signature of different populations identified in the scRNA-seq (fig. S2, A and B), including tdTomato⁺Sstr2⁺Pdgfrb⁺Kitl⁻ mpSSCs, tdTomato⁺Sstr2⁻Pdgfrb⁺Kitl⁻ tBMSCs, tdTomato⁺Pdgfrb⁺Kitl⁺ RetiCs, and tdTomato⁺Ncad⁺/Pdpn⁺ OLCs (Fig. 2C, and fig. S2, C and D). Furthermore, we generated an Ihh-Dre knock-in mouse line in which Dre recombinase is restrictively expressed in pre–hypertrophic chondrocytes (pre-HCs) of growth plate (36) (fig. S3, A and B). Again, IF analysis of P8 Ihh-Dre;ROSA26^RSR-tdTomato femurs confirmed that HCs contributed to multiple skeletal lineages under the growth plate, including Sstr2⁺ mpSSCs, Pdgfrb⁺ BMSCs, and Ncad⁺/Pdpn⁺ OLCs, and that Sstr2⁺ cells were exclusively derived from HCs (fig. S3C).

Fig. 2. MpSSCs are immediate derivatives of HCs.

(A) The framed parts of hindlimb bones of P7.5 Col10a1^ZsGreen mice were used for scRNA-seq of HCs and their descendants, and UMAP shows 11 clusters. (B) Selected cluster-enriched genes are superimposed on the UMAP plot. Crimson color intensity indicates the Log₂ expression level of genes. (C) Multicolor IF staining shows the expression of cluster-enriched genes in P7 Col10a1^tdTomato;Kitl^GFP femur. Dashed lines frame the MPZ right beneath the HZ. White arrows, tdTomato⁺Sstr2⁺Pdgfrb⁺Kitl⁻ mpSSCs; green arrows, tdTomato⁺Sstr2⁻Pdgfrb⁺Kitl⁻ tBMSCs; magenta arrows, tdTomato⁺Pdgfrb⁺Kitl⁺ RetiCs; blue arrows, tdTomato⁺Ncad⁺/Pdpn⁺ OLCs; gold arrows, tdTomato⁺Ki67⁺ proliferating cells. (D) Differentiation trajectory was inferred by Slingshot, which starts with the HC-1 cluster. (E) Signaling entropy analysis indicates mpSSCs has a higher differentiation potency than other BMSCs and OLCs. Significant differences between mpSSC and other HC progeny clusters were analyzed using the t test. *P < 0.05, **P < 0.01. (F) Scheme showing the strategy for tracing the fate of an individual growth plate chondrocyte in Col2a1-Cre^ERT;ROSA26^nTnG mouse. Representative multicolor IF staining shows that a clonal column of nGFP⁺ growth plate chondrocyte descendants pass through HZ into MPZ and convert into Sstr2⁺Pdgfrb⁺ mpSSCs (yellow arrows) 5 days after a single dose of tamoxifen (TAM) induction. (G) Scheme showing the strategy for tracing the HC-to-mpSSC conversion in Acan-Cre^ERT2;ROSA26^nTnG mouse. Representative multicolor IF staining shows the conversion of HCs into Sstr2⁺Pdgfrb⁺ mpSSCs (yellow arrows) as well as Ncad⁺ OLCs (red arrows) 4 days after TAM injection. (H) Quantification of each cell compartment within the transited nGFP⁺ cells at MPZ. Data are presented as means ± SD per slide (n = 5). Those framed areas are shown adjacently at higher magnifications. Scale bars, 200 μm in (A), (C), and (G); 50 μm in (F).

The specific location of Sstr2⁺ cells raised a possibility that the putative Sstr2⁺ mpSSCs may be from the direct conversion of HCs. Pseudotime analyses of Slingshot and single-cell entropy (SCENT) algorithms were used to infer the hierarchical relationships between all clusters of HC progeny (37, 38). Slingshot revealed a clear transcriptional continuum from HCs (clusters 1 to 3) to mpSSCs (cluster 4) and then tBMSCs (cluster 5), followed by branching into RetiCs (cluster 6), Prol-BMSCs/OLCs (cluster 7), and Ob (cluster 8) (Fig. 2D). SCENT algorithm, which characterizes the developmental plasticity of multipotent stem and progenitor cells by calculating the signaling entropy rate based on gene and protein signaling diversity, revealed that cluster 4 displays the highest entropy among all HC progeny clusters, indicating that mpSSCs have the higher differentiation potential than other BMSCs and OLCs (Fig. 2E). Therefore, HCs seem to directly dedifferentiate into Sstr2⁺ or Pdgfrb⁺Kitl⁻ cells, which may be a previously unknown SSC population standing at the top of the hierarchy along the HC differentiation trajectory.

Then, we verified this possibility by three independent evidences. First, using the Col2a1-Cre^ERT;ROSA26^nTnG mouse line, which allowed tracing of “clonal” descendants that are originated from a single parent Col2a1⁺ growth plate chondrocyte, we proved that HCs first converted into Sstr2⁺ or Pdgfrb⁺ cells in vivo (Fig. 2F). Second, we quantified the percentage of nuclear-targeted Green Fluorescent Protein (nGFP)-labeled cells that become Sstr2⁺/Pdgfrb⁺ cells at different time points in the inducible Acan-Cre^ERT2;ROSA26^LSL-nTnG mice, whose induction efficiency of tamoxifen was higher than that of Col2a1-Cre^ERT mice. When Acan-Cre^ERT2;ROSA26^LSL-nTnG mice were induced at P7 and P30 and harvested within a short time window, the top population of nGFP⁺ HC progeny was Sstr2⁺ cells (51.4 ± 8.8% at P11, 45.5 ± 11.3% at P40) followed by Ncad⁺ cells (32.9 ± 11.4% at P11, 39.2 ± 8.7% at P40) (Fig. 2, G and H, and fig. S4, A and B). Third, under the renal capsule, transplanted hypertrophic cartilage from Col10a1-Cre;ROSA26^tdTomato;Kitl^GFP mice firstly converted into tdTomato⁺Pdgfrb⁺ cells and tdTomato⁺Ncad⁺ cells and then tdTomato⁺Kitl^GFP cells (fig. S5, A to C). Overall, these above results suggest that Sstr2⁺ or Pdgfrb⁺Kitl⁻ mpSSCs are exclusively originated from HCs and are the direct derivatives of HCs.

Sstr2⁺ cells behave as SSCs in vivo

We next investigated whether Sstr2⁺ cells met the criteria for SSCs in vivo. We generated two independent Sstr2 knock-in mouse lines, where Cre^ERT2 was placed just after the translation initiation codon ATG of Sstr2 locus in one line and before the termination codon TGA in the other line (fig. S6, A and C). To test the clone-formation and self-renewal ability of Sstr2⁺ cells, we performed clonal genetic tracing using Ihh-Dre;Sstr2-Cre^ERT2;ROSA26^Confetti2 mice, where among the Ihh-Dre–driven HC progeny, Sstr2⁺ cells and their progeny can be labeled with one of three fluorescence proteins upon tamoxifen treatment (Fig. 3A). After 2 days upon tamoxifen induction at P2, a few separate Confetti⁺ cells were observed in the metaphysis (Fig. 3B). After 12 days after tamoxifen injection, Confetti⁺ clones were present in the metaphysis and the bone marrow (Fig. 3C and fig. S6B). Furthermore, some Confetti⁺ clones in the metaphysis contained Sstr2⁺ cells, indicating the self-renewal ability of Sstr2⁺ mpSSCs in vivo (Fig. 3C). Notably, some separate, not in clones, Confetti⁺ cells that still remained at the metaphysis expressed Sstr2, indicating that a relatively quiescent portion of Sstr2⁺ mpSSCs remained at least at this stage (Fig. 3C and fig. S6B).

Fig. 3. HC-derived Sstr2⁺ cells behave as SSCs in vivo.

(A) Scheme showing the strategy of clonal genetic lineage tracing using Ihh-Dre;Sstr2-Cre^ERT2;ROSA26^Confetti2 mouse. (B) Representative cryosection showing Confetti2 fluorescent protein-labeled Sstr2⁺ cells at the MPZ of P4 ulna 2 days after TAM administration. (C) Representative cryosections with Sstr2 IF staining showing Confetti2 fluorescent protein-labeled Sstr2⁺ cells and their progeny within the bone marrow of P16 ulna 14 days after TAM administration. Cyan dashed circles indicate clones (≥ 4 continuous cells of a single color). Yellow arrows indicate quiescent Sstr2⁺ cells residing in MPZ. White arrows indicate Sstr2⁺ mpSSCs with clonal progenies extending from the MPZ (shown below at a higher magnification). (D) Scheme showing the strategy for labeling HC-derived Sstr2⁺ mpSSCs and their progeny in Ihh-Dre;Sstr2-Cre^XER;ROSA26^{LSL-EYFP/RSR-tdTomato} mouse. (E) Representative multicolor IF staining of an ulna section from a P14 Ihh-Dre;Sstr2-Cre^XER;ROSA26^{LSL-EYFP/RSR-tdTomato} mouse, which shows Sstr2⁺ mpSSCs (white arrow), Sstr2⁻Pdgfrb⁺ BMSCs (magenta arrow), Ncad⁺ osteoblasts (Ob, cyan arrow), and Pdpn⁺ osteocytes (Osc, yellow arrow) within EYFP⁺ cells. The framed areas are shown adjacently at higher magnifications. (F) Quantification of Sstr2⁺ mpSSCs, Sstr2⁻Pdgfrb⁺ BMSCs, Ncad⁺ osteoblasts (Ob), Pdpn⁺ osteocytes (Osc), and undefined cells within EYFP⁺ cells as shown in (E). Data are presented as means ± SD per slide (n = 4). Dashed lines outline the MPZ beneath the HZ. Scale bars, 100 μm.

To access the multipotency o HC-derived Sstr2⁺ mpSSCs, we used the Ihh-Dre;Sstr2-Cre^XER;ROSA26^{LSL-EYFP/RSR-tdTomato} mouse line to achieve the constitutive expression of Cre in HC-derived Sstr2⁺ cells (Fig. 3D). As expected, many enhanced yellow fluorescent protein–positive (EYFP⁺) cells (indicating a history of Sstr2 expression) were found in the metaphysis and bone marrow of the P14 Ihh-Dre;Sstr2-Cre^XER;ROSA26^{LSL-EYFP/RSR-tdTomato} mice (Fig. 3E and fig. S6D). These EYFP⁺ cells were composed of Sstr2⁺ mpSSCs, Pdgfrb⁺ BMSCs, Ncad⁺ osteoblasts, and Pdpn⁺ osteocytes, reflecting the multiple differentiation potency of Sstr2⁺ mpSSCs in vivo (Fig. 3, E and F). Together, these results demonstrate that HC-derived Sstr2⁺ cells are functional SSCs in vivo.

The in vitro and ex vivo function of mpSSCs

Although Sstr2-Cre^ERT2 or Sstr2-Cre^XER can specifically label HC-derived mpSSCs, the low efficiency of Sstr2-driven Cre (24.9 to 33.6%; fig. S6E) hindered the sufficient recording of HC-to-mpSSC conversion. Therefore, we turned to the Pdgfrb^LSL-Dre mouse line (fig. S7A) and generated Col10a1-Cre;Pdgfrb^LSL-Dre;ROSA26^{LSL-EYFP/RSR-tdTomato} (CPYT) mice, in which two recombination events occurred (Fig. 4A). All HC progeny were labeled with the Cre reporter ROSA26^LSL-EYFP, which was triggered by Col10a1-Cre. Simultaneously, Cre excised the LoxP-floxed STOP sequence before Dre at the Pdgfrb^LSL-Dre locus. Thus, HC progeny that have expressed Pdgfrb could be recorded by the Dre reporter ROSA26^RSR-tdTomato. Therefore, EYFP⁺tdTomato⁺ cells went through two stages, first expressing Col10al and then Pdgfrb, while EYFP⁺tdTomato⁻ cells have not expressed Pdgfrb, although they shared the same HC origin (Fig. 4A). We verified the stringency and efficiency of this sequential genetic lineage tracing system (from 94.0 ± 4.4% to 97.1 ± 6.8% in multiple long bones of CPYT mice; fig. S7, A to D). Until E16.5, few Sstr2⁺ and tdTomato⁺ cells (HC offspring with a Pdgfrb-expression history) emerged in CPYT metaphysis (fig. S7E). At P1, tdTomato⁺ cells rose rapidly and expressed EYFP as expected (fig. S7F). The EYFP⁺tdTomato⁺ cells were positive for Pdgfrb or Ncad, whereas EYFP⁺tdTomato⁻ cells were only positive for Ncad (fig. S7F).

Fig. 4. The in vitro and ex vivo function of mpSSCs.

(A) Schematic of Col10a1-Cre;Pdgfrb^LSL-Dre;ROSA26^{LSL-EYFP/RSR-tdTomato} (CPYT) mice. HCs and their progeny are labeled by Col10a1-Cre-triggered ROSA26^LSL-EYFP, while Pdgfrb⁺ mpSSC and their descendants are also marked by ROSA26^RSR-tdTomato expression. RC, resting chondrocyte; PC, proliferating chondrocyte; BM, bone marrow. (B) The ratio of CFU-Fs generated from P30 CPYT hindlimb bone marrow suspensions. Data are presented as means ± SD per experiment (n = 5). (C and D) In vitro successional CFU-F formation and multilineage differentiation of P30 CPYT hindlimb bone marrow. Representative multilineage differentiation results show the first-generation clone and the third-generation clone that are derived from a single stem cell–like EYFP⁺tdTomato⁺ cell. (E) Scheme showing the functional characterization of HC-derived Pdgfrb⁺Kitl⁻ mpSSCs using renal capsule and bone injury transplantation models. Blue frames indicate the metaphyseal regions of P15 to P30 Col10a1-Cre;Pdgfrb^LSL-Dre;ROSA26^RSR-tdTomato;Kitl^GFP (CPT;Kitl^GFP) bones used for FACS. (F) The FACS plots are representative experimental results for sorting primary tdTomato⁺Pdgfrb⁺Kitl⁻ mpSSCs. (G) Representative wholemount fluorescent image, histology, and multicolor IF staining of bony organoids formed by tdTomato⁺Pdgfrb⁺Kitl⁻ mpSSCs 4 weeks after renal capsule transplantation. Dashed lines indicate the margins of tdTomato⁺ bony organoids. White frames indicate tdTomato⁺Pdgfrb⁺Kitl⁻ cells. (H) The flow cytometry plots are representative experimental results for analyzing secondary tdTomato⁺Pdgfrb⁺Kitl⁻ mpSSCs extracted from callus 14 days after bone injury transplantation. (I and J) Representative multicolor IF staining of recipient tibias 14 days after tdTomato⁺Pdgfrb⁺Kitl⁻ mpSSCs transplantation. Serial sections cut across the callus (white frames) show tdTomato⁺Pdgfrb⁺Kitl⁻ mpSSCs, tdTomato⁺Pdgfrb^+/−Kitl⁺ RetiCs, tdTomato⁺Sox9⁺ chondrocytes (chondro-), and tdTomato⁺Ncad⁺ osteoblasts (osteo-) that are indicated by frames at higher magnifications. (K) Pie chart showing the average ratio of each cell compartment to the tdTomato⁺ cells identified by multicolor IF staining (n = 3). Scale bars, 200 μm in (C), (D), and (G); 500 μm in (I) and (J).

Next, we used an in vitro successional colony forming unit–fibroblastic (CFU-F) assay and multilineage differentiation model to examine whether the HC-derived lineage contained stem cell–like cells (fig. S7G). The CFU-F assay of whole bone marrow cells isolated from CPYT mice revealed that EYFP⁺tdTomato⁺, but not EYFP⁺tdTomato⁻ cells, formed CFU-Fs, suggesting that a history of Pdgfrb expression is associated with the colony-formation ability of HC progeny (Fig. 4B). Furthermore, 9 of 30 clones exhibited trilineage differentiation potential (Fig. 4C). Even after two passage generations, trilineage differentiation potential was retained in 3 of 30 EYFP⁺tdTomato⁺ clones (Fig. 4D).

We further assessed the ex vivo differentiation and self-renewal ability of these tdTomato⁺ mpSSC cells by transplanting them under the renal capsule and into the bone injury sites as well. Because of the lack of a reliable Sstr2 antibody for flow cytometry, a tdTomato⁺Pdgfrb⁺Kitl⁻ panel was used to isolate mpSSCs from the metaphyseal bones of Col10a1-Cre;Pdgfrb^LSL-Dre;ROSA26^RSR-tdTomato;Kitl^GFP (CPT;Kitl^GFP) mice by FACS (19.0 ± 5.2% tdTomato⁺Pdgfrb⁺Kitl⁻ cells within isolated tdTomato⁺ cells, n = 12), followed by transplantation under the kidney capsule of immunodeficient mice (Fig. 4, E and F). After 4 weeks, tdTomato⁺Pdgfrb⁺Kitl⁻ cells formed bony organoids, as evidenced by von Kossa staining and safranin O staining (Fig. 4G). These bony organoids included Sox9⁺ chondrogenic cells, Ncad⁺ OLCs, Pdgfrb⁺ or Kitl⁺ stromal cells, as well as host-derived blood vessels and CD45⁺ or Ter119⁺ hematopoietic cells (Fig. 4G). Notably, a small portion of tdTomato⁺ cells was Pdgfrb⁺Kitl⁻ (Fig. 4G). We further validated the self-renewal ability of mpSSCs using a bone injury transplantation model. The exogenous callus derived from the fractionated tdTomato⁺Pdgfrb⁺Kitl⁻ cells retained the tdTomato⁺Pdgfrb⁺Kitl⁻ mpSSC compartment (Fig. 4H, 16.2 ± 3.2% tdTomato⁺Pdgfrb⁺Kitl⁻ cells within isolated tdTomato⁺ cells, n = 3) and gave rise to the other three compartments, including Ncad⁺ OLCs, Sox9⁺ chondrogenic cells, and Pdgfrb^+/−Kitl⁺ stromal cells (Fig. 4, I to K). In contrast, tdTomato⁺Pdgfrb⁺Kitl⁺ cells, the putative direct descendants of Pdgfrb⁺Kitl⁻ cells, failed to repopulate their own compartments in the grafted bones, possibly because of defects in cell proliferation or survival (fig. S7H). The above results indicate that HC-derived Pdgfrb⁺Kitl⁻ cells have clonal multipotency and self-renewal capability and are functional SSCs in vitro and in the context of cell transplantation.

MpSSCs give rise to most HC-derived BMSCs and OLCs

Given the inherent nature of SSCs, we hypothesized that mpSSCs are likely to produce most of the HC progeny building trabecular bone. We mapped all HC descendants in CPYT mice at different time points as well as at different sites including trabeculae, cortical bone, and bone marrow (Fig. 5, A and B, and fig. S8, A to C). At P1, the ratio of EYFP⁺tdTomato⁺ cells to EYFP⁺tdTomato⁻ cells was 1:2.19 (Fig. 5B). However, the ratio of EYFP⁺tdTomato⁺ cells to EYFP⁺tdTomato⁻ quickly reversed at P7 as shown by EYFP⁺tdTomato⁺ cells accounting for most of total EYFP⁺ HC descendants (77.1 ± 1.7%) at multiple sites, such as 57.4 ± 5.1% in cortical endosteum, 80.0 ± 1.4% in trabeculae, and 97.7 ± 1.7% in bone marrow (Fig. 5, A and B). The number of EYFP⁺tdTomato⁺ cells continued to increase and peaked at P30 (Fig. 5B and fig. S8A). Consistently, EYFP⁺tdTomato⁺ cells accounted for 84.5 ± 4.0%, 96.7 ± 0.8%, and 99.5 ± 0.4% of total EYFP⁺ cells in P30 cortical bone, trabeculae, and bone marrow, respectively (Fig. 5B). Although both the number of EYFP⁺tdTomato⁺ and EYFP⁺tdTomato⁻ cells gradually decreased along with the cessation of endochondral ossification at P180, the percentage of EYFP⁺tdTomato⁺ cells was extremely high over time (91.1 ± 1.8%, 96.7 ± 0.6%, 98.5 ± 0.6%, and 98.8 ± 0.6% at P15, P30, P180, and P480, respectively; Fig. 5B and fig. S8, B and C).

Fig. 5. MpSSCs generate most HC descendants.

(A) Mapping of EYFP⁺tdTomato⁺ cells (red dot) and EYFP⁺tdTomato⁻cells (green dot) in femur sections of P7.5 Col10a1-Cre;Pdgfrb^LSL-Dre;ROSA26^{LSL-EYFP/RSR-tdTomato} mice. White, magenta, and yellow dashed lines outline the HZ, cortical bone (CB), trabecular bone (TB), and bone marrow (BM), respectively. (B) Quantification of EYFP⁺tdTomato⁺ and EYFP⁺tdTomato⁻ cells in femur sections of Col10a1-Cre;Pdgfrb^LSL-Dre;ROSA26^{LSL-EYFP/RSR-tdTomato} mice at different time points as well as at different sites. Data are presented as means ± SD per slide (n = 6). (C) Mapping of EYFP⁺tdTomato⁺ and EYFP⁺tdTomato⁻ HC descendants in femur sections of P7 Col10a1-Cre;Pdgfrb^LSL-Dre;ROSA26^{LSL-EYFP/RSR-tdTomato} mice. Dashed lines outline the MPZ beneath the HZ. Magenta arrows, EYFP⁺tdTomato⁺Pdgfrb⁺ cells; yellow arrows, EYFP⁺tdTomato⁺Ncad⁺/Pdpn⁺ cells; blue arrows, EYFP⁺tdTomato⁻Ncad⁺/Pdpn⁺ cells; white arrows, EYFP⁺tdTomato⁺Pdgfrb⁺Lepr⁻ cells. (D) Color dot-represented cell compartments are superimposed on femur sections of the P7 Col10a1-Cre;Pdgfrb^LSL-Dre;ROSA26^{LSL-EYFP/RSR-tdTomato} mice. Colored pentagons represent Ki67-positive cells from the same cell compartment. (E) Quantification of each cell compartment shown in (D). Data are presented as means ± SD per slide (n = 4). Scale bars, 500 μm in (A) and (D); 50 μm in (C). See also fig. S6.

In accordance with the multipotential ability of mpSSCs, the IF analysis of P7 CPYT femurs revealed that EYFP⁺tdTomato⁺ cells accounted for almost all BMSC lineages, including Sstr2⁺ mpSSCs and their downstream Sstr2⁻Pdgfrb⁺ or Pdgfrb⁺Lepr^+/− BMSCs in metaphysis and bone marrow, as well as a large part of Ncad⁺ or Pdpn⁺ OLCs. In contrast, the EYFP⁺tdTomato⁻ population was largely composed of Ncad⁺ or Pdpn⁺ OLCs, similar to the findings at P1 (Fig. 5, C to E, and fig. S7F). Furthermore, Ki67⁺ proliferative cells were highly enriched in EYFP⁺tdTomato⁺ populations, including Sstr2⁺, Sstr2⁻Pdgfrb⁺, and Ncad⁺ cells (Fig. 5, D and E). This could partially contribute to the robust increase in EYFP⁺tdTomato⁺ cells. However, the percentage of Ki67⁺ cells in the Sstr2⁺ cells gradually decreased, from 38.2 ± 1.8% at P7 to 19.0 ± 1.6% at P30 and 9.0 ± 3.2% at 180, respectively (fig. S8D). Thus, the reduced proliferative capacity might contribute to the gradual decrease in the Sstr2⁺ cells. Even so, EYFP⁺tdTomato⁺Sstr2⁺ cells remained in older mice, reflecting the long-term survival or self-renewal ability (fig. S8B). In addition, a few EYFP⁺tdTomato⁺ adipocytes were observed in the bone marrow of the P180 CPYT mice (fig. S8E). Together, mpSSCs give rise to most HC descendant cells, including BMSCs and OLCs.

Hgs deletion in HCs impairs the conversion of HCs to mpSSCs

The abrupt changes in cellular morphology and transcriptional signatures from HCs to mpSSCs raised the question on how HC acquired mpSSC identity. Next, we used Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) analyses to investigate the biological processes, cellular components, and signaling pathways enriched in HCs, particularly in HC-1, which seemed to be responsible for initiating the HC-to-mpSSC conversion (Fig. 6, A to C). We focused on endosomal transport pathways that were enriched in GO terms (biological processes and cellular components) and KEGG analyses, as the endosomal transport pathway is essential for diverse receptor-mediated signal transduction and is widely implicated in various cellular processes (39) (Fig. 6, A to C). We found that Hgs, a pivotal component of the endosomal sorting complex, was enriched in the HC-1 subset and involved in multiple cellular transport biological processes (Fig. 6D). To evaluate the role of Hgs in regulating the HC-to-mpSSC conversion, we generated Col10a1-Cre;Hgs^fl/fl mice to delete the Hgs gene in HCs (Fig. 6E). Although the trabecular bone appeared normal in Hgs-deficient mice at E18.5, the number of HC progeny and Sstr2⁺ cells decreased by 11.2 ± 11.0% and 22.8 ± 8.0%, respectively (fig. S9, A to C), indicating that the conversion of HCs to mpSSCs might be impaired in Hgs-deficient mice. We further confirmed it using CPYT;Hgs^fl/fl mice. EYFP⁺tdTomato⁺Sstr2⁺ mpSSCs in Hgs knockout mice decreased by 76.4% at P7 (Fig. 6, F and G, and table S1). Furthermore, the number of total EYFP⁺tdTomato⁺ populations, including Sstr2⁻Pdgfrb⁺ BMSCs and Ncad⁺ or Pdpn⁺ OLCs, decreased accordingly (Fig. 6G and table S1). In the renal capsule transplantation model, hypertrophic cartilage from Col10a1-Cre;Hgs^fl/fl;ROSA26^LSL-tdTomato mutant mice showed a notable failure of conversion into tdTomato⁺Pdgfrb⁺ BMSCs (fig. S9, D to F), consistent with the in vivo finding in Hgs-deleted mice (Fig. 6G).

Fig. 6. Hgs deletion in HCs impairs the conversion of HCs to mpSSCs.

(A to D) Single-cell transcriptome analysis reveals that Hgs-mediated endosomal machinery is enriched in HC-1 cells. Dotplot (A) to (C) and cnetplot (D) of remarkable biological pathways enriched by high-expressed genes of HC-1 cells. The Hgs-related biological processes [(A) and (D)], cellular component (B), and KEGG pathways (C) are enriched in HC-1 cells. The Hgs gene and its related biological processes are highlighted in cnetplot (D). (E) Representative safranin O staining and Hgs staining in femurs of P7 control and Col10a1-Cre;Hgs^fl/fl mice. White dashed lines outline the proliferating zone (PZ), HZ, and MPZ. (F) Representative multicolor IF staining of femurs indicating impaired conversion of HC- to-mpSSC in P7 Col10a1-Cre;Pdgfrb^LSL-Dre;ROSA26^{LSL-EYFP/RSR-tdTomato};Hgs^fl/fl (CPYT;Hgs^fl/fl) mice. The framed areas are shown adjacently at higher magnifications. White and gold arrows indicate tdTomato⁺Sstr2⁺ mpSSCs and tdTomato⁺Ki67⁺ proliferative cells, respectively. (G) Quantification of the lineage of two-fated HC descendants and their proliferating status in P7 femurs of Col10a1-Cre;Pdgfrb^LSL-Dre;ROSA26^{LSL-EYFP/RSR-tdTomato};Hgs^+/+ (CPYT;Hgs^+/+) and CPYT;Hgs^fl/fl mice. Data are presented as means ± SD per slide (n = 4). Scale bars, 200 μm.

However, Hgs deletion did not markedly change either the proliferation or the apoptosis of HC descendants (Fig. 6G; fig. S9, G and H; and table S1). Notably, when Hgs was deleted in BMSC cells using Lepr-Cre mice, we found that Hgs deletion did not lead to a decrease of EYFP cells (fig. S10). These data suggested that loss of Hgs in HCs maybe specifically impaired the HC-to-mpSSC conversion. The population of EYFP⁺tdTomato⁻Ncad⁺ OLCs, which was possibly directly transdifferentiated from HCs, increased by 1.36 fold in CPYT;Hgs^fl/fl mice (Fig. 6G). However, the increase in EYFP⁺tdTomato⁻ OLCs was insufficient to rescue the decrease in the overall pool size of EYFP⁺ HC descendants (Fig. 6G), supporting the key role of mpSSCs in generating HC progeny.

Hgs deletion in HCs compromises trabecular bone formation

We further assessed the effect of Hgs loss on bone formation. At P30, the Hgs knockout mice displayed virtually no BMSCs and a remarkable decrease in the OLC population (Fig. 7, A and B). Micro–computed tomography (micro-CT) demonstrated a defect in trabecular bone formation in Hgs-knockout mice, as evidenced by a lower bone mineral density (BMD) and bone mass (BV/TV) in Hgs-deficient trabecular bone (Fig. 7, C and D). Consistently, Hgs-deficient mice displayed a remarkable decrease in the trabecular number (Tb. Nb) and thickness (Tb. Th), resulting in increased trabecular spacing (Tb. Sp) (Fig. 7, C and D). At P90, a lower BMD and bone mass (BV/TV) still remained in the Hgs-deficient trabecular bone (fig. S11). However, when compared with the Hgs-deficient mice at P30 (Fig. 7, C and D), the loss of trabecular bone at P90 was alleviated. This is consistent with the notion that HC-derived mpSSCs are the cellular source of postnatal bone formation and also indicates that other SSC/BMSC populations are responsible for adult bone formation.

Fig. 7. Hgs deletion in HCs compromises trabecular bone formation.

(A) Representative multicolor IF staining showing a significant decrease in HC-derived mpSSCs, BMSCs, and OLCs in femurs of 1-month-old Col10a1^EYFP;Hgs^fl/fl mice. The framed areas are shown adjacently at higher magnifications. MP, metaphysis; BM, bone marrow. (B) Quantification of each cell compartment shown in (A). Data are presented as means ± SD per slide (n = 6). **P < 0.01 as compared with the corresponding cell compartment in Col10a1^EYFP;Hgs^+l+ mice. (C) Representative reconstructed μCT images of femurs showing impaired trabecular bone formation in 1-month-old Col10a1-Cre;Hgs^fl/fl mice. (D) Quantification of metaphyseal parameters of femurs of 1-month-old Col10a1-Cre;Hgs^fl/fl and control mice. Data are presented as means ± SD (n = 6). Tb. BV/TV, trabecular bone surface area/bone volume; Tb. Th., trabecular thickness; Tb. Sp., trabecular spacing; Tb. Nb., trabecular number; Ct.Th., cortical thickness. **P < 0.01, N.S., no significance. (E) Scheme summarizing that HC-derived mpSSCs contribute to trabecular bone formation. Scale bars, 500 μm.

DISCUSSION

Here, we identified a previously unappreciated mpSSC population, which were derived from HCs and involved in trabecular bone formation, establishing a previously unknown cellular paradigm for trabecular bone formation (Fig. 7E). A prominent character of mpSSCs is their exclusive cellular origin. Theoretically, marrow SSCs are thought to be derived from perichondral cells that migrate into marrow or HCs that dedifferentiate at the perinatal stage (1, 2). Our multiple lines of genetic tracing results demonstrated that most, if not all, mpSSCs are the direct descendants of HCs, suggesting that the dedifferentiation of HC into mpSSC is a key process bridging chondrogenesis and osteogenesis during endochondral ossification. MpSSCs emerged at perinatal stage, peaked at the age of 1 month, and declined after the age of 3 months. The temporal and spatial pattern of mpSSCs corresponds to the active HC transition and robust trabecular bone formation in the postnatal stage. Thus, adopting a mpSSC fate not only endows HCs with differentiation potential but also enables HC descendants to survive in the long term and produce progeny that participate in postnatal bone formation. Consistently, the physiological function of mpSSC is critical for trabecular bone formation, as a reduced mpSSC pool caused by loss of Hgs led to remarkably compromised trabecular bone formation. It is worth noting that the defect in trabecular bone formation caused by Hgs deletion was gradually recovered, indicating that the effect of mpSSCs loss might be compensated by non-HC–derived cells such as those originated from the perichondrial cells, another important cellular source of marrow stromal cells and osteoblasts (33, 40, 41), thus supporting the concept that cartilage and perichondrial derived stromal cells functionally complement each other in trabecular osteogenesis (33). Furthermore, we revealed that HC-derived mpSSCs stand at the pinnacle of the HC dedifferentiation trajectory. Previous lineage tracing studies have shown that HCs can give rise to over half of all Lepr⁺ BMSCs as well as marrow-associated SSPCs marked by Pdgfra and Lepr (31, 32). When comparing to the localization and transcriptional profiling of HC progeny defined here, both HC-derived Lepr⁺ BMSCs and SSPCs are possibly downstream offspring of Pdgfrb⁺Kitl⁻ mpSSCs.

MpSSCs serve as a reservoir for previously reported metaphyseal mesenchymal progenitors and diaphyseal BMSCs. In the metaphysis region, Grem1⁺, Gli1⁺, Osterix⁺, and Pdgfra⁺Pdgfrb⁺Hey1⁺ label multiple types of metaphyseal mesenchymal progenitors that contribute to trabecular osteoblasts and bone marrow stromal cells (8, 18, 19, 34). In addition, the lower regions of the metaphysis and diaphyseal marrow harbor several overlapping bone marrow stromal cell populations marked by Lepr, Kitl, and Cxcl12 (20, 22, 42). We found that mpSSCs expressed high levels of Pdgfrb and Hey1 but little or very low levels of Kitl, Lepr, Grem1, Sp7/Osterix, and Gli1, whereas tBMSCs and RetiCs, two mpSSCs downstream populations, expressed Postn, Gli1, Sp7, Lepr, Cxcl12, Kitl, and Grem1 to varying extents. We propose that HC-derived Pdgfrb⁺Kitl⁻ mpSSCs represent a subpopulation of Hey1⁺Pdgfrb⁺ BMSCs and could give rise to Gli1⁺, Sp7⁺, Lepr⁺, Cxcl12⁺, and Grem1⁺ progenitor cells, therefore serving as a key cellular source for skeletal progenitors and bone marrow stromal cells. Notably, mpSSCs give rise to a fraction but not all of the aforementioned BMSCs populations as evidenced by scRNA-seq, multiple IF staining, and CFU-F results. This raised a possibility of whether other SSC populations with a non-HC origin exist in the lower metaphyseal and diaphyseal bone marrow. On the other hand, other SSC populations may not be necessary, as the extraordinary cellular plasticity of BMSCs enable them to reacquire the ability of SSCs as shown in bone regeneration model of mouse or zebrafish (43, 44).

The nature of HC-derived mpSSC provides insight into the heterogeneity and interactions of diverse previously reported SSCs at different sites. Previous studies have shown that epiphyseal growth plate houses the PTHrP⁺ SSCs that reside at resting zone and give rise to HCs, bone marrow stromal cells, and osteoblasts (5, 10). Notably, only a few offspring of PTHrP⁺ SSCs can extend into the metaphysis beyond the age of 2 months (5), whereas the active HC-to-mpSSC transition mainly occurs before P30, suggesting that Sstr2⁺ mpSSC may not be the descendants of PTHrP⁺ SSCs. Together with the fact that the offspring of periosteal SSCs are confined to periosteum, it seems that SSCs at distinct sites exert their distinct functions in a site-specific manner. However, a recent study showed that periosteal SSCs control the function of the growth plate resting PTHrP⁺ SSCs by periosteal SSCs-derived Indian hedgehog (45). Therefore, although there is still a lack of evidence for cellular cross-talk between these SSCs, the functional interaction between mpSSCs and other previously identified SSCs need to be clarified in future studies. Nevertheless, together with murine SSCs identified in the growth plate, periosteum, and articular surface (4–6, 46), these findings support that multiple SSCs, which have distinct temporal and spatial contexts and origins as well as different functional properties, work in concordance to achieve longitudinal bone development.

MATERIALS AND METHODS

Mice

The mouse lines used in this study were as follows: Col10a1-Cre, Kitl^GFP, Ihh-mKate2^tomm20-Dre, Lepr-Cre, CAGG-Cre^ERT, Col2a1-Cre^ERT, Acan-CreERT2, Cre-activated ROSA26^LSL-EYFP, ROSA26^LSL-nTnG, ROSA26^LSL-ZsGreen, ROSA26^LSL-tdTomato, Dre-activated ROSA26^RSR-tdTomato, and Cre/Dre-activated ROSA26^Confetti2 (28, 36, 42, 47–53). Kitl^GFP mice indicated that a green fluorescent protein (GFP) cassette was introduced into the second exon of the Kitl locus. For renal capsule transplantation experiment, 6- to 8-week-old nonobese diabetic (NOD) severe combined immunodeficient (SCID) immunodeficient mice were used as recipient hosts.

We generated Pdgfrb^LSL-Dre, Col10a1-memG-mitoS-DD-Cre-DD, Sstr2-mKate2-2A-Cre^ERT2, and Sstr2-2A-tdKatushka2-2A-Cre^XER knock-in mouse lines by CRISPR-Cas9 mediated genome editing. A LoxP-5 × PloyA-LoxP-Dre cassette was inserted after the translational start codon of the Pdgfrb gene. F₀ adult mice were outcrossed to wild-type mice, and F₁ mice were genotyped to assess germline transmission of inserted fragments.

All mice were maintained in individual ventilated cages. All mouse experiments were performed according to the protocols approved by the Animal Experiment Committee of the Beijing Institute of Lifeomics and conformed to the Regulations of People’s Republic of China on the Administration of Experimental Animals.

Cell preparation for scRNA-seq

P7.5 Col10a1^ZsGreen mice were used for single-cell analysis. The femurs and tibias from 12 mice were dissected and cleaned by brief digestion and brushing in 0.2% collagenase type I (Gibco, 17100017) at 37°C for 30 min. To exclude cell interference from the secondary ossification center and other sites of the growth plate, we only collected the distal part of the femur and proximal part of the tibia, with half of the growth plate cartilage excised (fig. S1). To help release the embedded cells, cortical bone, trabecular bone, and hypertrophic cartilage were stripped and sliced into pieces using a syringe needle under a stereoscope. The samples were then digested in 0.2% collagenase type I and 0.1% collagenase type II (Gibco, 17100015) for 2 to 3 hours at 37°C under gentle agitation. The liberated cells were pipetted, filtered, neutralized, and centrifuged every 20 min. Dead cells were removed using a kit (130090101, STEMCELL Technologies). The collected cells were prepared to establish two types of single-cell transcriptomes. The first was the unsorted scRNA-seq data, which were constitutive from skeletal cells of the whole bone with hematopoietic and endothelial cells briefly depleted by CD31/CD45/Ter119-coated magnetic beads (Biotin anti-mouse CD31 Antibody, BioLegend, 102503; Biotin anti-mouse CD45 Antibody, Biolegend, 103104; Biotin anti-mouse TER-119 Antibody, Biolegend, 116203; EasySep Mouse Streptavidin RapidSpheres Isolation Kit, STEMCELL Technologies, 19860). The other was composed of ZsGreen-positive cells sorted by FACSAria III (BD Bioscience). Single cells were encapsulated in emulsion droplets using a chromium controller (10x Genomics). The scRNA-seq libraries were prepared using Chromium Single Cell 3′ Reagent Kits v3 (10x Genomics), according to the manufacturer’s protocol. The scRNA-seq libraries were evaluated and quantified using an Agilent 2100 Bioanalyzer/Fragment Analyzer 5300 and Qubit HS. The libraries were sequenced on a NovaSeq platform (Illumina) to generate 150-bp paired-end reads, according to the manufacturer’s instructions. Two datasets of unsorted scRNA-seq and one dataset of ZsGreen-sorted scRNA-seq were established. The median genes per cell were 1534, 1601, and 4304, and the median unique molecular identifier (UMI) counts per cell were 5102, 5528, and 20512, respectively.

Processing and analysis of scRNA-seq data

Raw data based on 10x Genomics were processed using the Cell Ranger software suite (v3.1.0) with reference genome mm10, and doublets were removed using Scrublet (v0.2.1) (54). Hematopoietic and endothelial cells were removed from ZsGreen-sorted and unsorted scRNA-seq data. Cells with gene number more than 700 (for the unsorted scRNA-seq data) or 2000 (for the ZsGreen-sorted scRNA-seq data), as well as with a percentage of mitochondrial genes lower than 15%, were kept and projected into two dimensions by UMAP and clustered by K-means algorithm using Scanpy (55) (v1.7.2) in the Python environment (v3.9). ZsGreen-expressing cells (UMI > 0) were extracted from the unsorted scRNA-seq data and integrated with ZsGreen-positive cells sorted by FACS. The batch effect was removed using the cellranger aggr function. Cells were projected into two dimensions by UMAP using a cellranger and clustered by the Leiden algorithm (resolution = 0.3) using Scanpy (55) (v1.7.2) in the Python environment (v3.9). Slingshot trajectory inference was based on UMAP reduction dimensions. The Wilcoxon rank-sum test was performed to identify differentially expressed genes in each cell cluster, and the P values were adjusted using the Benjamini-Hochberg correction. Genes with an adjusted P value <0.05, log (fold change) >0.5, and expressed in more than 30% of the cells in the cluster were selected for GO and KEGG enrichment analyses using the R package ClusterProfiler (v3.14.3) (56). Last, signaling entropy was computed using the R package SCENT (v1.0.2) to infer differentiation potency at the single-cell level (38).

Micro-CT analysis

Femurs were freshly dissected, fixed in 4% paraformaldehyde (PFA), scanned, and analyzed using a micro-CT Skyscan 1076 μCT scanner (Bruker Corporation) and software. The scanner was set at a voltage of 80 kV, a current of 500 mA, and a resolution of 9.2 μm per pixel. A 200-μm-thick slice of the femur was reconstructed through its longitudinal axis. Trabecular bone was reconstructed from 0.5 to 2.3 mm below the lowest part of the metaphyseal growth plate and then measured for the metaphyseal parameters including BMD, trabecular bone surface area/bone volume (Tb. BV/TV), trabecular thickness (Tb. Th.), trabecular spacing (Tb. Sp.), and trabecular number (Tb. Nb.). The cortical bone was reconstructed from 2 to 2.5 mm below the lowest part of the metaphyseal growth plate of the femur and measured for its thickness (Ct. Th.).

Linage tracing performed on inducible mouse models

To evaluate the lineage relationships between growth plate chondrocytes and mpSSCs, we used inducible Col2a1-Cre^ERT;ROSA26^LSL-nTnG and Acan-Cre^ERT2;ROSA26^LSL-nTnG mouse models. To trace the fate of an individual growth plate chondrocyte, a single low dose of tamoxifen (40 μg/g body weight) was administrated at P4. Six days after tamoxifen administration, the hindlimb bones of Col2a1-Cre^ERT;ROSA26^LSL-nTnG mouse mice were harvested. To quantify the transition of HCs at different developmental stages, we triggered nGFP expression in a vast majority of growth plate chondrocytes of Acan-Cre^ERT2;ROSA26^LSL-nTnG mice by two successive high doses of tamoxifen (200 μg/g body) at P7 and P30, respectively. Four or 10 days after tamoxifen administration, the hindlimb bones of Acan-Cre^ERT2;ROSA26^LSL-nTnG mice were dissected and sectioned for immunohistochemical studies.

To test the deletion efficiency of Cre recombinase upon the Loxp-STOP-Loxp cassette within Pdgfrb^LSL-Dre locus, two successive high doses of tamoxifen (200 μg/g body) were administrated at P8 and P9 in CAGG-Cre^ERT;Pdgfrb^LSL-Dre;ROSA26^RSR-tdTomato mice, in which the Cre activity would be induced in widespread cells or tissues. Three days after tamoxifen administration, the skeletal elements of CAGG-Cre^ERT;Pdgfrb^LSL-Dre;ROSA26^RSR-tdTomato mice were dissected and sectioned for immunohistochemical studies. The deletion efficiency could be calculated on the basis of the percentage of tdTomato⁺ cells in Pdgfrb⁺ compartment.

To trace the fate of HC-derived Sstr2⁺ mpSSCs in Ihh-Dre;Sstr2-Cre^ERT2;ROSA26^Confetti2 mice, a more active form of tamoxifen, 4-hydroxytamoxifen, was used at 100 μg/g body to overcome the low induction efficiency of Sstr2-Cre^ERT2 transgene. A single dose of 4-hydroxytamoxifen was injected at P2, and the skeletal elements of Ihh-Dre;Sstr2-Cre^ERT2;ROSA26^Confetti2 mice were harvested at P4 and P16.

Renal capsule transplantation of hypertrophic cartilage

The rib cages dissected from P1 mice were digested with 0.2% collagenase type I at 37°C for 30 min. The connective tissues, including the muscle, fascia, and periosteum, were thoroughly removed by repeated brushing and digestion. Hypertrophic cartilage was then excised from the T6–T8 ribs with a fine blade under a fluorescent stereoscope and transferred underneath the renal capsules of 6- to 8-week-old NOD SCID immunodeficient mice.

Marrow stromal cell isolation, successional clonal formation, and multilineage differentiation

Bone marrow cells from tibias and femurs of P30 Col10a1-Cre;Pdgfrb^LSL-Dre;ROSA26^{LSL-EYFP/RSR-tdTomato} mice were repeatedly flushed out using a 1-ml syringe. To help release the stromal cells residing between the metaphysical trabeculae, the growth plate cartilage was peeled off, and the metaphysis part of bones was cut off. They were gently disrupted using a mortar and pestle and digested in 0.2% collagenase type I for 2 to 4 hours. The liberated primary bone marrow and stromal cells were filtered, collected, isolated, and plated at a density of 2 × 10⁵ cells/10 cm dish to allow the formation of individual colonies after growth in Dulbecco’s Modified Eagle Medium (Gibco, 11885084) containing 12% fetal bovine serum (Gibco, 12664025) and 1% antibiotic-antimycotic (Gibco, 15240-062) at 37°C under a hypoxic condition (5% O₂, 5% CO₂) for 2 weeks. Single clones with ≥600 cells (Ø ≥ 4 mm) were selected and labeled as first clones. They were equally divided into two parts: One was replated in a 10-cm dish to allow the formation of second individual colonies, and the other part was expanded in six-well plates to reach the cell population (≥30,000 cells) for further multilineage differentiation potential test of the first clones. We generated third individual colonies (≥400 cells/Ø ≥ 3 mm) from second colonies (≥400 cells/Ø ≥ 3 mm) via the same replating protocol. For in vitro multilineage differentiation potential test, each expanded first and third clone was split for differentiation into chondrocytes, osteoblasts, and adipocytes using a differentiation kit (STEMCELL Technologies, 05507, 05455, and 05504). The in vitro differentiation reported in this study is clonal. Images were taken using a Nikon ECLIPSE Ti-U inverted microscope system.

Cell isolation, FACS, and renal capsule transplantation

Limb bones were dissected from mice at the indicated age. They were briefly digested with 0.2% collagenase type I at 37°C for 30 min. After the removal of the muscle and ligament, the tibiae and femurs were dissociated by mechanical and enzymatic dissociation according to a published protocol (35). Flow cytometry was performed using FACSAria III. The total dissociated cells were blocked with rat immunoglobulin G and stained with fluorochrome-conjugated antibodies against CD45 (BioLegend, 103154, 1:400), CD31 (BioLegend, 102528, 1:400), Ter119 (BioLegend, 116223, 1:200), and CD140b (Pdgfrb, BioLegend, 136010, 1:400) for fractionation by FACS. Dead cells were excluded using 7-Aminoactinomycin D (7-AAD) staining (54). The 300 to 500 tdTomato⁺Pdgfrb⁺Kitl⁻ mpSSCs could be enriched from the fore/hindlimb bones and scapulae per P10 to P20 CPT;Kitl^GFP mice. Sorted cells (6000 to 10,000) were pelleted, resuspended in 2 to 3 μl of Matrigel (Corning, 356234), and then injected underneath the renal capsules of 6- to 8-week-old NOD SCID immunodeficient mice. After 4 to 6 weeks, the recipient kidneys were dissected for histological and immunohistochemical analysis.

Bone injury and transplantation

To transplant isolated cells to the bone injury site, 6- to 8-week-old NOD SCID immunodeficient mice were used as recipient hosts. First, a 26-gauge syringe needle was placed parallel to the tissue between the cortical bone and periosteum/fascia, making a hole at the metaphyseal medial tibial shaft 1 mm beneath the growth plate, which was covered by the periosteum/fascia. A second Ø 0.5-mm hole passing through the cortical bone and periosteum/fascia was then made by drilling at the diaphyseal medial tibial shaft, 3 mm beneath the growth plate. Using a 31-gauge insulin pen needle and a microinjector, 6000 to 10,000 sorted cells resuspended in 2 to 4 μl of Matrigel were injected into the bone marrow through the metaphyseal hole. After 10 to 14 days, recipient tibiae were dissected for flow cytometry or immunohistochemical studies. Before flow cytometry, nonskeletal cells were briefly depleted by CD31/CD45/Ter119-coated magnetic beads. To identify the cell type within the callus, six 6-μm serial sections with an interval of 60 μm were cut from the recipient tibia and were subject to multicolor IF staining. The number of each cell compartment was counted from these six representative sections.

Multiplex IF assay

Samples of the skeletal elements and kidney were dissected, fixed in 4% PFA, and decalcified in 0.5 M EDTA/phosphate-buffered saline (PBS)/0.1% PFA. Paraffin-embedded samples were cut into 4- to 6-μm sections, followed by deparaffinization and heat-mediated antigen retrieval treatment (Histova Biotechnology, BoneRetrival-M, BRM5L, 72°C for 6 hours). Multiplex IF assay was conducted as previously described (57, 58). Briefly, sections were incubated with primary antibody for 2 hours, followed by detection using HRP-conjugated secondary antibody and Tyramide Signal Amplification (TSA)-fluorophores (Histova Biotechnology, NECC7100). Then, the primary and secondary antibodies were eliminated by heating the slides in retrieval/elution buffer (Histova Biotechnology, ABCFR5L) for 10 s at 95°C using microwaves or by incubating slides in elution buffer (Histova Biotechnology, ABCCC30) for 20 to 30 min at 37°C. Serially, each antigen was labeled with distinct antibodies and TSA fluorophores. The cryosections of Ihh-Dre;Sstr2-Cre^ERT2;ROSA26^Confetti2 bones were cut at 60 μm, treated with 0.25% Triton X-100/PBS for 20 min, and then subject to antibody incubation and IF detection. Multiplex antibody panels applied in this study include the following: GFP (CST, 2956, 1:400), red fluorescent protein (Rockland, 600-401-379, 1:800), Pdgfrb (CST, 3169, 1:400), Emcn (Thermo Fisher Scientific, 14-5851-82, 1:800), Sstr2 (Abcam, ab134152, 1:800), N-cadherin (Abcam, ab76011, 1:500), Pdpn (Sino Biological, 50256-R066, 1:1000), CD200 (Novusbio, BAF3355, 1:400), CD51 (Abcam, ab208012, 1:400), Ly51 (Sino Biological, 50082-T24, 1:1000), Thy1 (Sino Biological, 50461-T44, 1:1000), CD105 (Abcam, ab221675, 1:400), CD31 (CST, 77699, 1:400), CD45 (CST, 70257, 1:400), Ter119 (BioLegend, 116201, 1:1000), Ki67 (CST, 12202, 1:400), Lepr (Novusbio, BAF497, 1:200), Hgs (Abcam, ab155539, 1:300), and Sox9 (Abcam, ab185966, 1:400). Multiplex IF slides were imaged using a confocal laser scanning microscopy platform Zeiss LSM880 equipped with 405-nm/458-nm/488-nm/514-nm/543-nm/594-nm/633-nm lasers. Some data were further processed and statistically analyzed using the Bitplane Imaris software (Bitplane AG) or the HALO image analysis platform (Indica Labs). In situ mapping and quantification were performed on Imaris software (Bitplane AG) or the HALO image analysis platform. Segmented cell compartments were visualized by colored dots and were superimposed on tissue section.

Histological analysis

Decalcified or undecalcified samples were embedded in paraffin, cut into 5-μm sections, and subjected to safranin O and Fast Green staining or von Kossa staining according to the standard protocol.

Statistical analysis

Data are expressed as means ± SD, and significance was set at P < 0.05 (see each figure for details). The sample sizes are shown in the corresponding results and figure legends. Paired or unpaired two-tailed Student’s t test was used for two-group comparisons. One-way analysis of variance (ANOVA) followed by Tukey’s post hoc test was performed for multiple comparisons. All statistical analyses were performed using Microsoft Excel.

Supplementary Materials

This PDF file includes:

Figs. S1 to S11

Table S1