To validate the osteogenic fate of hypertrophic chondrocytes, we first utilized the Col10a1CreERT2 mouse line crossed with a Rosa26^fs-tdTomato reporter (Col10a1CreERT2; Rosa26^fs-tdTomato). These mice were harvested at E16.5 following tamoxifen injections from E13.5-E15.5. Hypertrophic chondrocytes and their descendants found within the newly formed bone marrow cavity were tdTOMATO⁺ (Figure 1a). A number of these descendants co-expressed the osteoblast protein OCN (Figure 1b–c, white arrow); however, OCN-, tdTOMATO⁺ cells were also apparent within the marrow (Figure 1b–c, orange arrow). Additionally, we injected Col10a1CreERT2; Rosa26^fs-tdTomato mice with a single dose of tamoxifen (TM) at postnatal day 6 (P6) and harvested at 12 hr, 24 hr, and 2 weeks following TM (Figure 1d–g). At 12 hr following TM administration, specific expression of the tdTOMATO reporter was observed within hypertrophic chondrocytes and no cells within the marrow or surrounding tissues (Figure 1d). At 24 hr following TM administration, the tdTOMATO reporter could now be observed in a limited number of marrow-associated cells directly below the cartilage growth plate, as well as within hypertrophic chondrocytes (Figure 1e). At 2 weeks following TM administration, the tdTOMATO reporter had cleared from the hypertrophic chondrocytes and was now exclusively present in a number marrow-associated cells; some of those being at a significant distance from the growth plate (Figure 1f). In this distal region of the marrow, we also observed trabecular bone lining cells that co-labeled for tdTOMATO and OCN (Figure 1g; white arrows), suggesting that these cells had transitioned from hypertrophic chondrocytes to osteoblasts. Interestingly, other tdTOMATO-positive marrow-associated cells were not directly in contact with bone matrix and did not co-label with OCN (Figure 1g; orange arrow), suggesting the presence of another cell type or cell stage within the bone marrow that may be derived from hypertrophic chondrocytes. Additionally, by inducing recombination at both 2 months and 4 months of age and chasing for 2 and 4 weeks, respectively, we continued to observe contribution of tdTOMATO⁺ cells to the bone marrow (Figure 1—figure supplement 1a, b).

To get a better understanding of the proportionality of these hypertrophic chondrocyte-derived marrow and osteogenic cells, we performed similar analyses on constitutively active Col10a1Cre; Rosa26^fs-tdTomato mice. While tdTOMATO reporter expression was observed within hypertrophic chondrocytes and osteogenic cells associated with trabecular and endocortical bone of mice at P0 (Figure 1—figure supplement 1d-f) and 2 months (2 M) of age (Figure 1h–j), CRE recombinase could only be detected within hypertrophic chondrocytes (Figure 1—figure supplement 1c). The vast majority of the trabecular and endocortical bone was lined with tdTOMATO reporter-positive cells, and 69.4% ± 14.8% of osteocytes were reporter positive (Figure 1k and l – white arrows). We also observed that 83.3% ± 9.7% of all OCN-positive osteoblasts were tdTOMATO reporter positive, confirming that hypertrophic chondrocytes can give rise to the majority of intramedullary osteoblasts during bone development (Figure 1k and l – green arrows). Of note, a number of tdTOMATO negative and OCN positive cells were also identified lining portions of the trabecular and endocortical bone, indicating that some of these osteoblasts originate from an alternate source such as the perichondrium/periosteum (Figure 1k). Finally, utilizing this model, we were able to confirm the existence of tdTOMATO reporter expressing cells within the marrow that were not associated with bone matrix (white arrow); many of which were localized to vessel-like structures (yellow arrow) (Figure 1j).

Following the observation that tdTOMATO reporter-positive cells were both marrow- and bone-associated (Figure 1), we hypothesized that the hypertrophic chondrocytes may give rise to multiple cell fates. We therefore took an unbiased approach by performing single cell RNA-sequencing (scRNA-seq) on cells derived from Col10a1Cre; Rosa26^fs-tdTomato mice at E16.5, a timepoint coincident with bone marrow cavity initiation and the potential transition of hypertrophic chondrocytes to alternative fates. To isolate the hypertrophic chondrocytes and their descendants we liberated the femur, tibia, and fibula from soft tissues and performed collagenase digestions of the skeletal rudiments. Following matrix digestion, red blood cells were lysed and the remaining cell populations were subjected to FACS. We selected tdTOMATO reporter positive cells that demonstrated high levels of tdTOMATO fluorescence for further analysis (Figure 2—figure supplement 1a). Using the 10 x Genomics platform, we performed scRNA-seq and further analyzed 2,273 tdTOMATO reporter positive sorted cells following quality control processing and elimination of a small number of cells lacking tdTomato transcripts. We then performed a Seurat-based unsupervised clustering of gene expression profiles from the tdTOMATO positive cells, which ultimately yielded eight distinct cell clusters (Figure 2a; see Materials and methods for details). Initial observations demonstrated that both Cre and Col10a1 transcripts were detected in a portion of the cells, but we also observed cells expressing tdTomato without expression of either Cre or Col10a1, indicating the presence of both hypertrophic chondrocytes and their descendants within these clusters (Figure 2—figure supplement 2a, i).

Next we analyzed expression of chondrogenic genes. Differential gene expression analysis revealed the presence of Ihh and Col10a1 expressing chondrocytes (most notably in clusters 1, 3, and 5), identifying cells in transition from pre-hypertrophy to hypertrophy (Figure 2b and i; St-Jacques et al., 1999). Col10a1 expression was observed throughout many clusters; however was most highly enriched in clusters 1–5 and 7 (Figure 2c and i). Moderate expression of Sox9 and Aggrecan were observed throughout all clusters, but primarily enriched in clusters 1–5 (Figure 2—figure supplement 2b-c, i). Mmp13, noted for its role in terminal hypertrophic chondrocytes, reached highest expression in clusters 4, 5, and 6 (Figure 3d and i). The osteogenic genes, Sp7, Clec11a, Col1a1, Bglap, Runx2, Spp1, Alpl, and Ibsp, were most highly expressed in clusters 7 and 8, which represent two potential osteoblast clusters (Figure 2e–h and i and Figure 2—figure supplement 2d-i), while some were also detected at much lower levels in pre-hypertrophic/hypertrophic chondrocyte clusters.

Differential gene expression analysis also identified the SSPC-associated genes, Pdgfra, Ly6a (encoding SCA1), Lepr, Cxcl12, to be expressed within clusters 4, 5, and 6, with some expression in cluster 8 (Figure 3a–e). Each of these genes and/or their regulatory elements genetically identify SSPCs capable of becoming osteoblasts in vivo (Crisan et al., 2008; Morikawa et al., 2009; Omatsu et al., 2010; Worthley et al., 2015; Zhou et al., 2014a). We did not observe Adipoq expression in any clusters, another gene previously shown to be co-expressed with Lepr in some SSPC populations (Figure 2—figure supplement 2h-i; Zhong et al., 2020). To further understand the cluster hierarchy of these hypertrophic chondrocytes and their descendants, we performed pseudotime analysis of the scRNA-seq data. This analysis indicated a singular trajectory linking hypertrophic chondrocytes (clusters 1–5) to osteoblasts (clusters 7 and 8) by way of an intermediate SSPC-like population (cluster 6) (Figure 3f). To validate this transition of hypertrophic chondrocytes to a marrow-associated SSPC fate, we generated Col10a1CreERT2; Rosa26^fs-tdTomato; PDGFRa^H2B-GFP mice and performed additional in vivo lineage-tracing analysis at E16.5. Fluorescent reporter analysis identified a number of tdTOMATO and PDGFRa^H2B-GFP double positive cells within the marrow space that were either OCN⁺ or OCN^- (Figure 3g–h), indicating that tdTOMATO⁺, GFP⁺, and OCN^- cells likely represent a more primitive hypertrophic chondrocyte-derived SSPC. To further validate that hypertrophic chondrocytes are capable of giving rise to marrow-associated SSPCs expressing Lepr, Pdgfra, and Pdgfrb in vivo throughout development, we performed IF co-labeling assays on bone sections from Col10a1CreERT2; Rosa26-tdTomato ^f/+ mice at 3 W of age (TM delivered at P6). Co-labeling of cells with tdTOMATO fluorescence and antibodies for LEPR, PDGFRα, and PDGFRβ demonstrate that hypertrophic chondrocytes indeed give rise to marrow associated SSPCs during skeletal development (Figure 3—figure supplement 1a).

After identifying Col10a1CreERT2; Rosa26^fs-tdTomato reporter positive SSPCs at early stages of development, we hypothesized that these cells may still arise from hypertrophic chondrocytes during later stages of skeletal development. Therefore, we performed scRNA-seq of tdTOMATO⁺ sorted marrow cells from Col10a1Cre; Rosa26^fs-tdTomato mice at 2 months of age (2 M). To isolate and specifically enrich hypertrophic chondrocyte-derived marrow-associated cells, we performed centrifugation of the marrow followed by a gentle flush of femurs and tibias and subjected these marrow plugs to a brief collagenase digestion. This short digestion allowed for potential isolation of a limited number of osteoblasts, since our focus was centered on the analysis of novel non-osteoblast/osteocyte populations. Following the digest, red blood cells were lysed, and the remaining cell populations were subjected to FACS. We selected tdTOMATO reporter positive cells that demonstrated high levels of unambiguous tdTOMATO fluorescence for further analysis (Figure 2—figure supplement 1b). Using the 10 x Genomics platform, we performed scRNA-seq and further analyzed 1,066 tdTOMATO reporter positive sorted cells following quality control processing and elimination of any cells lacking tdTomato transcripts. Since neither Cre nor Col10a1 expression could be detected in any of these cells, they clearly represented descendants of Col10a1Cre-expressing hypertrophic chondrocytes. Finally, we performed a Seurat-based unsupervised clustering of gene expression profiles from the tdTOMATO-positive cells, which ultimately yielded five distinct cell clusters (Figure 4a; see Materials and methods for details).

Differential gene expression analyses further demonstrated that cells associated with clusters 1 and 2 expressed several known osteogenic genes (Figure 4b–h). Many cells in cluster one expressed genes such as Mepe, Dmp1, and Fam20c (Figure 4b–d and h), known for their roles in osteocytes and/or the bone mineralization process (Robling and Bonewald, 2020; Sapir-Koren and Livshits, 2014). Some cells in cluster 2 also expressed Dmp1 and Fam20c; however, their expression was detected at much lower levels. Most cells in cluster two expressed other osteoblast associated genes, such as Bglap and Col1a1 (Figure 4e–f and h), critical components of the organic bone matrix. In addition to the few genes highlighted here, additional osteoblast-associated genes were also observed in cells within clusters 1 and 2 (Figure 4—figure supplement 1a-h). Expression analyses further determined that the master transcriptional regulator of osteoblast differentiation and known osteoprogenitor cell marker, Sp7 (encoding the protein OSTERIX) (Maes et al., 2010; Mizoguchi et al., 2014), not only exhibited expression within osteoblast clusters 1 and 2, but also within a limited number of cells associated with clusters 3, 4, and 5 (Figure 4g–h). These data suggest that the majority of marrow-associated cells derived from the hypertrophic chondrocytes localized to clusters 3, 4, and 5 are not osteoblasts or osteoprogenitors, but rather may be more primitive cells within the osteoblast lineage or distinct cell types altogether. We next utilized differential gene expression analyses to identify genes that were uniquely and broadly expressed across the cells of clusters 3, 4, 5, with limited expression in the more differentiated osteoblasts of clusters 1 and 2. This analysis identified a number of genes utilized for identifying SSPC populations. Specifically, Lepr, Grem1, Pdgfra (but not Ly6a – SCA1), Cxcl12, and Pdgfrb were highly expressed within cells across all three clusters (Figure 5a–f and k). Each gene and/or their regulatory elements genetically identify or can be utilized to label SSPCs with the potential of becoming osteoblasts in vivo (Crisan et al., 2008; Morikawa et al., 2009; Omatsu et al., 2010; Worthley et al., 2015; Zhou et al., 2014a). To confirm that hypertrophic chondrocytes are capable of giving rise to marrow associated SSPCs expressing Lepr, Pdgfra, and Pdgfrb in vivo at this later developmental time point, we performed IF co-labeling assays on bone sections from Col10a1Cre; Rosa26^fs-tdTomato mice at 2months of age. Co-labeling of cells with tdTOMATO fluorescence and antibodies for LEPR, PDGFRα, and PDGFRβ further demonstrate and validate that hypertrophic chondrocytes indeed give rise to marrow-associated SSPCs during postnatal skeletal development (Figure 3—figure supplement 1b). Quantification of Col10a1Cre; Rosa26^fs-tdTomato for LEPR demonstrates that 74.21%±13.92% of all tdTOMATO+ cells were LEPR+ (Figure 3—figure supplement 1c). We additionally performed flow cytometry of marrow cells from Col10a1Cre; Rosa26^fs-tdTomato; PDGFRa^H2B-GFP mice at 1 month of age and observed 73.6% ± 2.83% of tdTOMATO⁺ cells were PDGFRa-GFP+ (Figure 3—figure supplement 1c). At 2 M of age, flow cytometry of marrow cells from Col10a1Cre; Rosa26^fs-tdTomato mice demonstrated that 64.33% ± 3.48% of tdTOMATO⁺ cells were PDGFRB+ (Figure 3—figure supplement 1c) and that 37.4% ± 6.2% of the total LEPR⁺ population were also tdTOMATO⁺ as per quantitation from stained bone sections (Figure 5—figure supplement 1a). This was further corroborated by utilizing flow cytometry of marrow cells from Col10a1Cre; Rosa26^fs-tdTomato; PDGFRa^H2B-GFP mice, which demonstrates that a similar 26.93%±3.79% of all PDGFRa^H2B-GFP+ cells were also tdTOMATO reporter positive (Figure 5—figure supplement 1b). These data indicate that hypertrophic chondrocyte-derived SSPCs constitute nearly 25–45% of all marrow SSPCs at these stages, and that a significant portion of the marrow-associated SSPCs are likely derived from other origins such as the perichondrium and/or periosteum in mice 2 months of age and younger.

FACS has been highly utilized to identify, isolate, and enrich populations of SSPCs from bone marrow and periosteal tissues in mice and/or humans based on the expression of specific cell surface proteins (Chan et al., 2015; Chan et al., 2018; Zhou et al., 2014a). FACS assays utilizing antibodies against Integrin Subunit Alpha V (ITGAV) have been used for the identification, isolation, and enrichment of bone marrow stromal cell populations in mice, which are noted to be largely quiescent and capable of generating most of the colonies within a colony-forming unit – fibroblastic (CFU-F) assay; a technique utilized to identify heterogeneous, non-hematopoietic, marrow stromal cells with progenitor cell characteristics (Boulais et al., 2018; Robey et al., 2014). While we identified Itgav expression sporadically throughout cells of all clusters, it is enriched in the SSPC clusters 3, 4, and 5 (Figure 5g and k). Recently, CD200, ITGAV, and CD105 antibodies have been used in conjunction with antibodies detecting the hematopoietic and endothelial lineages (CD45, CD31, and Ter119) for flow cytometric evaluation and identification of unique subsets of SSPCs (Gulati et al., 2018). These studies indicate that lineage negative (CD45^-,CD31^-,Ter119^-) and CD200⁺,ITGAV⁺,CD105^- cells represent the most primitive self-renewing and multipotent skeletal stem cell (SSC) population, while lineage negative and ITGAV⁺,CD105⁺ cells represent a more committed SSPC population (Gulati et al., 2018). Our gene expression analyses observed broad expression of Cd200 across all clusters including the osteoblasts; however, Eng (encoding CD105) was more restricted to SSPC clusters 3, 4, 5 and enriched specifically within cluster 4 (Figure 5h–i and k). When assessing the precise cellular overlap of these factors, we observed 128 of 953 SSPCs to be Itgav+, Cd200+, Cd105-. While this most primitive SSC population exists among our SSPCs analyzed via scRNA-seq, they were relatively evenly distributed among cells within clusters 3, 4, and 5 (Figure 5g–I and k). These data suggest that hypertrophic chondrocytes likely undergo a process of dedifferentiation to generate even the most primitive SSCs; however, these cells themselves may be heterogeneous due to the differential expression of genes that allow for their segregation into multiple distinct clusters within our Seurat analysis.

In an attempt to further delineate each of the SSPC clusters in our Seurat analysis, we analyzed the data to identify genes restricted in their expression to a single SSPC cluster. A limited number of genes were ultimately found to be uniquely expressed within solitary SSPC clusters; however, much more commonly we observed genes to be enriched within specific clusters with limited expression within the other one or two SSPC clusters (Figure 4—figure supplement 2).

SSPCs are known to reside within specific niches often associated with blood vessels (Zhou et al., 2014a). Previously, we noted that some of the hypertrophic chondrocyte derived tdTOMATO reporter cells appeared to associate with vascular structures. To assess this directly, we performed immunofluorescent staining for ENDOMUCIN (EMCN) on bone sections of Col10a1Cre; Rosa26^fs-tdTomato mice at 2 months of age and identified a number of tdTOMATO+ reporter cells directly adjacent to EMCN expressing endothelial cells (Figure 4—figure supplement 3a). To ensure that these tdTOMATO+ cells were not pericytes or vascular smooth muscle cells (vSMCs), we assessed our differential gene expression data for genes that serve as common markers for these specific cell types, including Acta2, Nes, Kcnj8, Rgs5, Cspg4, and Mcam (Chasseigneaux et al., 2018; Yamazaki and Mukouyama, 2018). We observed limited or no gene expression for these markers in SSPCs of clusters 3, 4, or 5 (Figure 4—figure supplement 3b-g).

To finally parse the relatedness or relationship between SSPC and osteoblast cell clusters, we performed Monocle three pseudotime cell trajectory analysis of the Seurat data. These data indicate a clear hierarchy of cellular differentiation and relatedness between all cell clusters in an unbranched or continuous trajectory. Clusters 4 and 5 shared a common trajectory, with cluster 4 being slightly earlier in pseudotime. These clusters then give rise to cluster 3, which further differentiates into the osteoblast clusters 2 and then 1 (Figure 5j). Taken together these data indicate that cluster 3, 4, and 5 are hypertrophic chondrocyte-derived SSPCs that are capable of becoming osteoblasts and have unique niche localizations with at least some of these SSPCs being vessel associated.

Since most bone-marrow-associated SSPCs have the potential to become either osteoblasts or adipocytes during development, aging, regeneration, and/or disease, we further examined our differential gene expression data for indicators of this competency. While we observed limited expression of the osteoprogenitor transcription factor Sp7 in the SSPC clusters 3, 4, and 5, we noticed much broader and robust expression of Runx2, the earliest master transcriptional regulator of osteoblast specification, throughout all clusters and especially concentrated in the SSPC clusters 3, 4, and 5 (Figure 6a). RUNX2 has been observed in LEPR⁺ cells and appears to prime cells for osteoblastic specification and differentiation (Yang et al., 2017). We have already demonstrated in Figure 1 the efficiency by which hypertrophic chondrocytes can contribute to the osteoblast lineage, and that these osteoblasts are part of a differentiation trajectory that can originate from hypertrophic chondrocyte-derived SSPCs.

Recent data has shown that some LEPR⁺ SSPCs also express PPARγ (encoded by Pparg) and ADIPONECTIN (encoded by Adipoq) (Zhong et al., 2020). Similar to this report we observed Pparg and Adipoq expression within Lepr-expressing SSPCs of clusters 3, 4, and 5 (Figure 6b–c). While PPAR-gamma is necessary and sufficient to induce adipogenesis of SSPCs, ADIPONECTIN enhances osteogenesis but is primarily secreted by adipocytes (Tencerova and Kassem, 2016). Given the expression of these adipogenic regulators and marker genes, we asked if hypertrophic chondrocyte-derived cells in clusters 3, 4, and 5 represent mature adipocytes or whether these cells could ultimately give rise to mature adipocytes in vivo. Our differential gene expression analysis did not indicate the presence of Perilipin (Plin1), a mature lipid laden adipocyte marker, in SSPC clusters 3, 4, or 5, suggesting that none of these cells isolated utilizing our procedures were representative of mature adipocytes (Figure 6d). Providing evidence for an ultimate adipogenic fate for some of these SSPCs, Col10a1Cre; Rosa26^fs-tdTomato mice at 2months and 13months of age displayed co-localization of the tdTOMATO reporter and PERILIPIN antibody staining (Figure 6e). By 13 M of age, approximately 25% of adipocytes labeled with PERILIPIN were also tdTOMATO reporter positive (Figure 6f). None of these double-labeled cells are either known to express Col10a1, nor could we detect CRE expression within any of these cells. Therefore, our results suggest that hypertrophic chondrocyte-derived SSPCs are also a significant contributor to the adipogenic fate; however, these mature cells likely eluded our scRNA-seq analyses due to the technical nature for isolating and sorting lipid laden adipocytes and their limited availability in the bone marrow at 2 months of age.

To assess SSPC contribution to the osteoblast and adipocyte lineages, we also performed reporter studies utilizing LeprCre; Rosa26^fs-tdTomato and AdipoqCre; Rosa26^fs-tdTomato mice. Utilizing bone sections of P0 LeprCre; Rosa26^fs-tdTomato reporter mice, we observed reporter positive signals in both-marrow-associated cells and some bone lining cells; however, OCN⁺, tdTOMATO⁺ osteoblasts were not apparent (Figure 4—figure supplement 4a). By 2 M of age we observed bone lining cells, OCN-expressing osteoblasts, and osteocytes that were tdTOMATO reporter positive (Figure 4—figure supplement 4a – white arrows), similar to data previously reported (Zhou et al., 2014a). As expected, a limited number of PERILIPIN⁺, tdTOMATO⁺ adipocytes were also observed within the bone marrow of 2 M old LeprCre; Rosa26^fs-tdTomato mice (Figure 4—figure supplement 4a). Similarly, P0 AdipoqCre; Rosa26^fs-tdTomato bone sections exhibited marrow-associated and bone lining reporter-positive cells that not yet contributed to OCN+ osteoblasts; however, by 2 M of age tdTOMATO⁺ and OCN⁺ osteoblasts were observed, as well as tdTOMATO⁺ osteocytes and adipocytes (PLIN1+) were apparent in Adipoq Cre; Rosa26^fs-tdTomato bone sections (Figure 4—figure supplement 4b – white arrows). In contrast with these studies, Col10a1Cre; Rosa26^fs-tdTomato descendants demonstrate a relatively quick commitment to the osteoblast lineage that is observed at P0 and earlier, while Col10a1CreERT2; Rosa26^fs-tdTomato mice document that bone lining cells and mature OCN-expressing osteoblasts can be generated from hypertrophic chondrocytes within 1 day to 2 weeks (Figure 1a–g). This highlights the potential for functional differences between sub-populations of LEPR and ADIPOQ expressing SSPCs based on their particular origin and/or niche.

Col10a1Cre-derived marrow-associated SSPCs constitute only a portion of LEPR⁺ SSPCs and LeprCre-derived marrow-associated SSPCs contribute to nearly 100% of all CFU-F colonies, therefore we reasoned that Col10a1Cre-derived marrow-associated SSPCs would contribute to only a fraction of total CFU-Fs and that this may provide a method for separating hypertrophic chondrocyte-derived progenitors from those derived from alternative sources such as the perichondrium/periosteum (Zhou et al., 2014a). Following plating at clonal density and 10 days of growth, 50–60% of CFU-F colonies derived from Col10a1Cre; Rosa26^fs-tdTomato mice exhibited relatively homogeneous tdTOMATO reporter expression while the other colonies were largely devoid of tdTOMATO positivity at both 2 months and 8 months of age (Figure 7a–c). To achieve a better understanding of the similarities and differences between progenitors derived from these different sources at the transcriptomic level, we isolated three tdTOMATO⁺ and three tdTOMATO^- CFU-F colonies and performed individual RNA-isolations and submitted for bulk RNA-sequencing. Of the differentially expressed genes reaching significance (p < 0.05, LogFC >1), the three tdTOMATO⁺ CFU-Fs clustered separately from three tdTOMATO^- CFU-Fs on the basis of only 363 differentially expressed genes (227 enriched in tdTOMATO⁺ CFU-Fs and 136 enriched in tdTOMATO^- CFU-Fs) out of approximately 12,000 genes with an average expression level above biological thresholds (Figure 7d and Methods for threshold calculations). We observed Lepr, Grem1, Pdgfra, Adipoq, Pparg, Itgav, Cd200, and Cxcl12, all highly expressed above biological thresholds in all CFU-Fs, indicating a general maintenance of SSPC phenotype while in culture and a striking similarity between progenitors from disparate sources.

To more readily identify differences between tdTOMATO⁺ and tdTOMATO^- CFU-Fs, we performed gene ontology analysis for biological processes and found two interesting terms arise due to enriched expression of genes within the tdTOMATO⁺ or hypertrophic chondrocyte derived CFU-Fs; cell population proliferation and regulation of vascular development (Figure 7e; Mi et al., 2019; Gene Ontology Consortium, 2021; Ashburner et al., 2000). Regulation of vascular development was interesting, although not a surprising finding, since we know that terminal hypertrophic chondrocytes secrete factors to initiate vessel invasion from the periosteum and we had identified a number of marrow-associated hypertrophic chondrocyte derived SSPCs localized near blood vessels. The fact that the tdTOMATO⁺ or hypertrophic chondrocyte derived CFU-Fs were enriched in genes associated with cellular proliferation was unanticipated. To this end, we investigated EdU incorporation into all CFU-Fs from Col10a1Cre; Rosa26^fs-tdTomato mice at day 5 in culture that consisted of at least 5 cells. We observed a greater percentage of cells in tdTOMATO⁺ CFU-Fs that had incorporated EdU as compared to tdTOMATO^- CFU-Fs (Figure 7f - top). By day 10 in culture, tdTOMATO⁺ CFU-Fs on average were larger and displayed more colonies with 200+ cells as compared to tdTOMATO^- CFU-Fs (Figure 7f - bottom). To determine whether this proliferative advantage could be observed in SSPCs immediately harvested from Col10a1Cre; Rosa26^fs-tdTomato bone marrow, we performed cell cycle analysis on our scRNA-seq dataset and found that as a percentage of the population there was an equal or greater percentage of tdTOMATO⁺ SSPCs (cluster 3, 4, and 5) in a proliferative state (S, G2, or M phase) as compared to osteoblasts (clusters 1 and 2) (Figure 7g). This unexpectedly high proliferative capacity for hypertrophic chondrocyte-derived SSPCs may be reflective of the fact that hypertrophic chondrocytes eventually re-enter the cell cycle as they transition fates within the marrow space (Park et al., 2015; Yang et al., 2014a).

To determine whether tdTOMATO⁺ and tdTOMATO^- CFUs contained inherent differences in their functional capacities to undergo osteogenic and adipogenic differentiation in vitro, we performed CFU-osteoblastic (CFU-Ob) and CFU-adipogenic (CFU-Ad) assays. Colonies composed of cells from either hypertrophic or non-hypertrophic origins were able to differentiate at similar capacities in vitro when measured by Alkaline Phosphatase, Von Kossa, and Oil Red O staining (Figure 8—figure supplement 1). However, when we performed kidney capsule transplantations to generate bone ossicles in vivo, there was a much more apparent functional difference between tdTOMATO⁺ and tdTOMATO^- cells. The tdTOMATO⁺ cells were able to generate a full ossicle with bone marrow, vessels, and adipocytes (Figure 8a -top and Figure 8b). Conversely, the tdTOMATO^- cells could only generate bone with no vascular invasion or adipocyte generation (Figure 8 -bottom and Figure 8b). These apparent differences when transplanted suggest that hypertrophic chondrocyte-derived SSPCs are much more supportive of vascular invasion and marrow formation than SSPCs derived from other cell sources.

The Col10a1Cre mouse line was a generous gift from Dr. Klaus von der Mark and has been previously described (Park et al., 2015). The Col10a1CreERT2 mouse line was a generous gift from Dr. Kathryn Cheah (University of Hong Kong). The Rosa26^fs-tdTomato, Pdgfra^H2B-GFP, LeprCre, and AdipoqCre mouse lines were obtained from Jackson Laboratories (007909, 007669, 008320, and 028020, respectively). All animal work was approved by Duke University Institutional Animal Care and Use Committees (IACUC). Both sexes of mice were utilized for experiments.

Harvest and sectioning: Animals were harvested at embryonic day 16.5 and postnatal day 0, 2–4 months, and 13 months of age. Tissues were washed in 1 x PBS and then fixed overnight with 4% paraformaldehyde (PFA - Sigma). P0 animals were treated overnight with 14% EDTA (Sigma), while 2 M old animals were treated for 10 days. Animals used for PERILIPIN labeling were not decalcified. Tissues were then transferred to 30% sucrose (Sigma) for 2 days. Samples were imbedded in Shandon Cryomatrix (ThermoFisher Scientific) and cryosectioned at 10 um, and slides were stored at –20 °C until staining.

For ENDOMUCIN and LEPR staining: Samples were prepared for staining as published (Kusumbe et al., 2015) 2 M animals were harvested and fixed overnight in 4% PFA. Samples were decalcified overnight in 14% EDTA. Tissues were transferred to 20% sucrose and 2% polyvinylpyrrolidone (PVP - Sigma) overnight. Samples were embedded in 8% gelatin (Sigma), 2% PVP, and 20% sucrose and cryosectioned at 30–200 um with slides stored at –20°C until staining.

Staining: Slides were brought to room temperature and rehydrated in PBS. Antigen retrieval was performed as shown in Appendix 1. Blocking was performed in 1% goat serum or 3% BSA. Slides were incubated with antibodies at concentrations in Appendix 1 overnight at 4 °C in a humidified dark box. Slides were washed and incubated with secondary antibodies (Invitrogen catalog number in Appendix 1 1:200) for 30–60 min at room temperature in the dark. Slides were imaged on a Leica DMI3000B microscope using a Leica DFC3000G camera with the latest Leica imaging suite, a Zeiss 710 or Zeiss 780 inverted confocal microscope running the latest ZEN software. Images were reconstructed and pseudocolored on Fiji and quantified using built-in quantification measures.

Cell isolation E16.5: Single cells were isolated by removing soft tissue on mechanical dissociation using scissors and then placed in 15 mL conical tubes with 2 mg/mL collagenase type II (Gibco 17101–015) in aMEM +20% FBS rotating at 37 °C. A pipet was used for further mechanical dissociation at 1- and 2 hr. Cells were removed at 2 hrs and subjected to red blood cell lysis for 30 s. Cells were stained with DAPI for 30 min on ice. Cells were washed with PBS and placed in PBS + 2% FBS for fluorescent activated cell sorting (FACS) on a B-C Astrios cell sorter. Gating was performed for single cells, DAPI, and tdTOMATO as shown in Figure 2—figure supplement 1a (Figure generated using FlowJo version 10). Cells were sorted into PBS + 10% FBS.

Cell isolation 2M-old: Single cells were isolated from bisected tibia and femur of 2 M Col10a1Cre; Rosa26^fs-tdTomato mice centrifuged at 4 °C twice pulsed up to 5000 x g. A 27 G needle attached to a syringe was then used to flush through the centrifuged bones. Bones and marrow were placed in 3 mg/mL of Collagenase D (Roche 11088866001) for 30 min in hypoxia. Red blood cells were lysed and strained through a 70 µm filter. Cells were stained with DAPI for 30 min on ice. Cells were washed with PBS and placed in PBS + 2% FBS for fFACS on a B-C Astrios cell sorter. Gating was performed for single cells, DAPI, and tdTOMATO as shown in Figure 2—figure supplement 1b (Figure generated using FlowJo version 10). Cells were sorted into PBS + 10% FBS.

10 x Transcriptome library prep: Cells were transferred to the Duke Molecular Physiology Institute Molecular Genomics core where they were combined with a master mix that contained reverse transcription reagents. The gel beads carrying the Illumina TruSeq Read one sequencing primer, a 16 bp 10 x barcode, a 12 bp unique molecular identifier (UMI) and a poly-dT primer were loaded onto the chip, together with oil for the emulsion reaction. The Chromium Controller partitions the cells into nanoliter-scale gel beads in emulsion (GEMS) within which reverse-transcription occurs. All cDNAs within a GEM, that is from one cell, share a common barcode. After the RT reaction, the GEMs were broken and the full length cDNAs cleaned with both Silane Dynabeads and SPRI beads. After purification, the cDNAs were assayed on an Agilent 4,200 TapeStation High Sensitivity D5000 ScreenTape for qualitative and quantitative analysis.

Enzymatic fragmentation and size selection were used to optimize the cDNA amplicon size. Illumina P5 and P7 sequences, a sample index, and TruSeq read two primer sequence were added via End Repair, A-tailing, Adaptor Ligation, and PCR. The final libraries contained P5 and P7 primers used in Illumina bridge amplification. Sequence was generated using paired end sequencing (one end to generate cell specific, barcoded sequence and the other to generate sequence of the expressed poly-A tailed mRNA) on an Illumina NextSeq500 with 400 M read per run.

Analysis: The primary analytical pipeline for the SC analysis followed the recommended protocols from 10 X Genomics. Briefly, we demultiplexed raw base call (BCL) files generated by Illumina sequencers into FASTQ files, upon which alignment to the mouse reference transcriptome, filtering, barcode counting, and UMI counting were performed using the most current version of 10 X’s Cell Ranger software. We used the Chromium cell barcode to generate feature-barcode matrices encompassing all cells captured in each library.

E16.5 timepoint: We sequenced a median of 4,868 genes per cell, a mean of 107,918 reads per cell, and reached a sequencing saturation of 44.6%. The secondary statistical analysis was performed using the latest R package of Seurat 4 (Hao et al., 2020). In Seurat, data was first normalized and scaled after basic filtering for minimum gene and cell observance frequency cut-offs ( > 200 genes, 15% mitochondrial genes, genes expressed in >5 cells). We thresholded tdTomato expression at two or higher to exclude a minimal number of cells that were separated via FACS but were tdTomato negative by gene expression. We observed some separation of clusters based on cell cycle and regressed out the cell cycle score during default SCTransform (Hafemeister and Satija, 2019).

After quality control procedures were complete, we performed linear dimensional reduction calculating principal components using the most variably expressed genes in our dataset (2000 variable genes, npcs = 35). Significant principal components for downstream analyses are determined through methods mirroring those implemented by Macosko et al, and these principal components were carried forward for two main purposes: to perform cell clustering and to enhance visualization (Macosko et al., 2015). Cells were grouped into an optimal number of clusters for de novo cell type discovery using Seurat’s FindNeighbors and FindClusters functions (resolution = 0.5) graph-based clustering approach with visualization of cells being achieved through the use of UMAP, which reduces the information captured in the selected significant principal components to two dimensions. Differential expression of relevant genes was visualized on UMAP plots to reveal specific individual cell types. Additional downstream analyses included examining the cellular distribution and differential gene expression of a priori genes of interest and closer examination of genes associated within the identified 8 cell clusters. Monocle analysis was performed using the Monocle3 package after cds was generated from Seurat using SeuratWrappers (Cao et al., 2019; Qiu et al., 2017a; Qiu et al., 2017b; Satija et al., 2020; Trapnell et al., 2014). GEO accession number GSE190616.

Two-month timepoint: We sequenced a median of 2,820 genes per cell, a mean of 261,899 reads per cell, and a sequencing saturation of 92.3%. The secondary statistical analysis was performed using the last R package of Seurat 3 (Butler et al., 2018; Konopka, 2019; R Development Core Team, 2013; Stuart et al., 2019; Wickham et al., 2020). In Seurat, data was first normalized and scaled after basic filtering for minimum gene and cell observance frequency cut-offs (200–4000 genes, 7% mitochondrial genes, genes expressed in >5 cells). We then closely examined the data and performed further filtering based on a range of metrics in an attempt to identify and exclude possible multiplets (i.e. instances where more than one cell was present and sequenced in a single emulsified gel bead). We thresholded tdTomato expression at two or higher to exclude a minimal number of cells that were separated via FACS but were tdTomato negative by gene expression.

After quality control procedures were complete, we performed linear dimensional reduction calculating principal components using the most variably expressed genes in our dataset (2000 variable genes, npcs = 25). The genes underlying the resulting principal components are examined in order to confirm they are not enriched in genes involved in cell division or other standard cellular processes (subsetting out percent.mito). Significant principal components for downstream analyses are determined through methods mirroring those implemented by Macosko et al, and these principal components were carried forward for two main purposes: to perform cell clustering and to enhance visualization (Macosko et al., 2015). Cells were grouped into an optimal number of clusters for de novo cell type discovery using Seurat’s FindNeighbors and FindClusters functions (resolution = 0.4) graph-based clustering approach with visualization of cells being achieved through the use of UMAP, which reduces the information captured in the selected significant principal components to two dimensions. Differential expression of relevant genes was visualized on UMAP plots to reveal specific individual cell types. Additional downstream analyses included examining the cellular distribution and differential gene expression of a priori genes of interest and closer examination of genes associated within the identified 5 cell clusters. Cell cycle analysis of the scRNA-seq dataset was preformed using a gene list and commands packaged in Seurat (Tirosh et al., 2016; Wickham, 2019). Monocle analysis was performed using the Monocle3 package (num_dim = 25) (Cao et al., 2019; Qiu et al., 2017a; Qiu et al., 2017b; Trapnell et al., 2014). GEO accession number GSE179148.

The Colony Forming Unit – Fibroblastic (CFU-F) assay was performed as published, with the exception of placing cells in hypoxia based on a recent publication demonstrating reduction of contaminating cells (Guo et al., 2020; Robey et al., 2014). In short, cells were isolated from bisected tibia and femur of 2-month-old Col10a1Cre; R26-tdTomato mice centrifuged at 4°C twice pulsed up to 5 K xG. Following red blood cell lysis (Roche), cells were plated at 750 K cells per 15 cm tissue culture plate in complete media (a-MEM, 20% FBS, 1% penicillin-streptomycin) and glutamax (Gibco) after straining through a 70 µm filter. After incubation in hypoxia (2% oxygen, 5% CO2, 37 C) for 3 hr, plates were washed three times with Hanks Balanced Salt Solution (Gibco) and replaced with complete media and glutamax. Media was changed on days 3 and 6. Colonies were utilized for analysis using a Leica DMI3000B microscope using a Leica DFC3000G camera with the latest Leica imaging suite, bulk RNA sequencing, or EdU labeling. TOMATO fluorescence was used to identify positive and negative colonies.

For CFU-OB, osteogenic differentiation was initiated after day 10 colony growth by replacing media with aMEM supplemented 10% FBS, 1% penicillin-streptomycin, 50 µg/mL L-ascorbic acid (Sigma-A4544), and 10 mM B-glycerophosphate (Sigma-G9422). Alkaline phosphatase and Von Kossa (Ricca) staining were performed per general histological procedures. For CFU-AD, adipogenic diffeentiation was performed as published; replacing media with aMEM supplemented with 10% FBS, 1% penicillin-streptomycin, 1.7 uM Insulin (Sigma), 1 µM Dexamethasone (Sigma), and 0.5 mM IBMX (Sigma) at day 10 colony growth (Yu et al., 2019). Oil red O (Sigma) staining was performed per general histological procedures.

Colonies were imaged using a Leica DMI3000B microscope using a Leica DFC3000G camera with the latest Leica imaging suite to identify TOMATO positive and negative colonies grown as in the CFU-F assay section to day 10. Colonies were prepared using Corning 10 mm cloning cylinders and vacuum grease to provide single colony isolation. RLT buffer was pipetted directly into the cloning cylinder and then transferred to ethanol for processing using the RNeasy Micro kit (QIAGEN). RNA was provided to the Duke Center for Genomic and Computational Biology core for quality assessment and sequenced 50 bp single-end read on an Illumina HiSeq4000.

RNA-seq data was processed using the TrimGalore (https://github.com/FelixKrueger/TrimGalore) toolkit which employs Cutadapt to trim low-quality bases and Illumina sequencing adapters from the 3’ end of the reads (Martin, 2011). Only reads that were 20nt or longer after trimming were kept for further analysis. Reads were mapped to the GRCm38v73 version of the mouse genome and transcriptome using the STAR RNA-seq alignment tool (Dobin et al., 2013; Kersey et al., 2012). Reads were kept for subsequent analysis if they mapped to a single genomic location. Gene counts were compiled using the HTSeq tool (Anders et al., 2015). Only genes that had at least 10 reads in any given library were used in subsequent analysis. Normalization and differential gene expression was carried out using the DESeq2 Bioconductor package with the R statistical programming environment (Huber et al., 2015; Love et al., 2014; R Development Core Team, 2013). The false discovery rate was calculated to control for multiple hypothesis testing. Biological thresholds were determined by taking the log2 expression value plotted on a histogram. A bimodal distribution is observed for low expressed genes followed by a larger Guassian distribution. The split point between these was used as a value cutoff for biological relevance. GEO accession number GSE179174.

Gene ontology analysis was performed using geneontology.org Analysis Type: PANTHER Overrepresentation test released 02-24-2021. Annotation version and release date: release date 01-01-2021 and version 10.5281/zenodo.4437524 for biological processes using Mus musculus as the reference. Bulk RNA sequencing data was organized into log fold changes > 1 and p-value < 0.05 for input into the enrichment analysis. A negative log fold change in the dataset represents higher expression in TOMATO positive colonies while positive log fold change represents higher expression in TOMATO-negative colonies. Test type: Fisher’s exact. Correction: Calculate False Discovery Rate (Mi et al., 2019; Gene Ontology Consortium, 2021; Ashburner et al., 2000).

Colonies were grown as in CFU-F assays (in a six-well plate at 500 K cells per well instead of 15 cm plate) until day 5. EdU staining was performed following the protocol of Click-iT Plus EdU Cell Proliferation Kit (Invitrogen – C10637). In short, EdU was added to growth media to allow a half media change of the cells. The cells were incubated at in hypoxia at 37 °C for 4 hours. Cells were fixed, permeabilized, and treated with staining solution. Cells were imaged using a Zeiss Axio Observer Z1 microscope at ×5 magnification tiled across 70% of each well and stitched together using the latest Zen Blue software. Number of cells per colony calculations were performed at day 10 of plates grown in parallel to the EdU day 5 plate.

Cells from Col10a1Cre; Rosa26^fs-tdTomato mice at age 2 months were isolated as in Colony-Forming Unit Assays above. Cells were plated at high density on 10 cm plates for 6 days at 37 °C in hypoxia. Cells were trypsinized from the plates and were stained with DAPI for 30 min on ice. Cells were washed with PBS and placed in PBS + 2% FBS for FACS. Gating was performed for single cells, DAPI, and tdTOMATO similar to Figure 2—figure supplement 1. Cells that were tdTOMATO⁺ and tdTOMATO^- were isolated from each of 5 plates (each from an individual mouse) and transplanted into the kidney capsule (one transplant per mouse – two male/3 female) after being resuspended and absorbed into a Gelfoam matrix at 200–300 k cells per transplant. Transplants were isolated at 8 weeks post transplantation, washed in 1 x PBS, and then fixed overnight with 4% paraformaldehyde (PFA - Sigma). The samples were then decalcified overnight with 14% EDTA (Sigma). Tissues were then transferred to 30% sucrose (Sigma) for 2 days. Samples were imbedded in Shandon Cryomatrix (ThermoFisher Scientific) and cryosectioned at 20 µm, and slides were stored at –20 °C until staining. Staining was performed as above in Tissue Analysis.

Following isolation techniques noted in the Colony Forming Unit Assay methods above, cells were resuspended in complete media and stained for 30–45 min in conjugated antibodies (noted in above table). Cells were spun down following staining and resuspended in PBS + 2% FBS. Flow cytometry was performed on a BD FACSCanto II machine.

Statistical analyses were performed using two-tailed, unpaired Student’s t-test. A p-value < 0.05 was considered significant.