Refining the identity of mesenchymal cell types associated with murine periosteal and endosteal bone

Single-cell RNA-seq has led to novel designations for mesenchymal cells associated with bone as well as multiple designations for what appear to be the same cell type. The main goals of this study were to increase the amount of single-cell RNA sequence data for osteoblasts and osteocytes, to compare cells from the periosteum to those inside bone, and to clarify the major categories of cell types associated with murine bone. We created an atlas of murine bone–associated cells by harmonizing published datasets with in-house data from cells targeted by Osx1-Cre and Dmp1-Cre driver strains. Cells from periosteal bone were analyzed separately from those isolated from the endosteum and trabecular bone. Over 100,000 mesenchymal cells were mapped to reveal 11 major clusters designated fibro-1, fibro-2, chondrocytes, articular chondrocytes, tenocytes, adipo-Cxcl12 abundant reticular (CAR), osteo-CAR, preosteoblasts, osteoblasts, osteocytes, and osteo-X, the latter defined in part by periostin expression. Osteo-X, osteo-CAR, and preosteoblasts were closely associated with osteoblasts at the trabecular bone surface. Wnt16 was expressed in multiple cell types from the periosteum but not in cells from endocortical or cancellous bone. Fibro-2 cells, which express markers of stem cells, localized to the periosteum but not trabecular bone in adult mice. Suppressing bone remodeling eliminated osteoblasts and altered gene expression in preosteoblasts but did not change the abundance or location of osteo-X or osteo-CAR cells. These results provide a framework for identifying bone cell types in murine single-cell RNA-seq datasets and suggest that osteoblast progenitors reside near the surface of remodeling bone.

Single-cell RNA-seq has led to novel designations for mesenchymal cells associated with bone as well as multiple designations for what appear to be the same cell type.The main goals of this study were to increase the amount of single-cell RNA sequence data for osteoblasts and osteocytes, to compare cells from the periosteum to those inside bone, and to clarify the major categories of cell types associated with murine bone.We created an atlas of murine bone-associated cells by harmonizing published datasets with in-house data from cells targeted by Osx1-Cre and Dmp1-Cre driver strains.Cells from periosteal bone were analyzed separately from those isolated from the endosteum and trabecular bone.Over 100,000 mesenchymal cells were mapped to reveal 11 major clusters designated fibro-1, fibro-2, chondrocytes, articular chondrocytes, tenocytes, adipo-Cxcl12 abundant reticular (CAR), osteo-CAR, preosteoblasts, osteoblasts, osteocytes, and osteo-X, the latter defined in part by periostin expression.Osteo-X, osteo-CAR, and preosteoblasts were closely associated with osteoblasts at the trabecular bone surface.Wnt16 was expressed in multiple cell types from the periosteum but not in cells from endocortical or cancellous bone.Fibro-2 cells, which express markers of stem cells, localized to the periosteum but not trabecular bone in adult mice.Suppressing bone remodeling eliminated osteoblasts and altered gene expression in preosteoblasts but did not change the abundance or location of osteo-X or osteo-CAR cells.These results provide a framework for identifying bone cell types in murine single-cell RNA-seq datasets and suggest that osteoblast progenitors reside near the surface of remodeling bone.
To develop a comprehensive understanding of how different cell types work together to create and maintain the vertebrate skeleton, it is essential to clearly identify the cell types involved.Initially, the cells that produce and maintain bone were defined by their morphology and location in histological sections (1-3), followed by studies using cultured cells, colonyforming assays, and transplant assays (4)(5)(6)(7)(8).Analysis of gene expression, either in isolated cells, primary cell cultures, or continuous cell lines, has also been used to define bone cell types (4).Moreover, detection of gene products in histological sections, via antibodies, in situ hybridization, or enzymatic activity, underpins much of our current knowledge of skeletal cell types.Conditional gene inactivation in cell types defined by expression of particular genes has also provided a wealth of information regarding the function of different genes and the cell types that express them (9).
Even with this abundance of knowledge, important questions remain regarding how different skeletal cell types are related to one another and regarding their full range of functions.For example, while it is clear that beta-catenin in a subset of cells targeted by Cre driver strains such as Osx1-Cre is essential for osteoblast specification and differentiation (10)(11)(12)(13), the location of these cells in the bone microenvironment and their relationship to other cells of the mesenchymal lineage remain unclear.Similarly, Dmp1-Cre driver strains are often used to study the role of particular genes in osteocytes (14,15).However, interpretation of results from such studies is complicated by the fact that the full range of cell types targeted by Dmp1-Cre strains is not known (15,16).These situations highlight the need for more precise definition of the cell types involved in skeletal formation and maintenance and which of these cells are targeted by various Cre driver strains.
Single-cell RNA-seq (scRNA-seq) provides a quantitative analysis of mRNA transcripts in hundreds to thousands of individual cells in a single experiment (17).Based on the transcriptomic profile of individual cells, they can be grouped or clustered so that cells with similar profiles are grouped together.Combining this information with the known specificity of gene expression obtained from other approaches can be used to identify or define clusters of cells in scRNA-seq datasets as a specific cell type.Several studies over the last 5 years have used scRNA-seq to define cells associated with murine bone, which include cells of the mesenchymal lineage, endothelial cells, and hematopoietic cells (18)(19)(20)(21)(22)(23)(24)(25)(26).Many of these studies have focused on defining types of mesenchymal stromal cells, which are thought to contain mesenchymal stem cells, as well as the progenitors of chondrocytes, osteoblasts, and adipocytes.
The first goal of this study was to increase the amount of sequencing data from osteoblasts and osteocytes because previous scRNA-seq studies included only small numbers of these cell types.A second goal was to better define the clusters, or cell types, associated with murine long bones.To accomplish these goals, we combined our results, which sequenced cells targeted by Osx1-Cre and Dmp1-Cre transgenes, with several published scRNA-seq datasets to develop a more comprehensive and uniform definition of mesenchymal cell types associated with murine long bones.Our results define 11 broad categories of cell types and identify which of these are targeted by Osx1-Cre and Dmp1-Cre in the periosteal and endosteal compartments of long bones.

Isolation of cells using Osx1-Cre and Dmp1-Cre
To increase the representation of osteoblasts and osteocytes in scRNA-seq datasets, we isolated cells targeted by Osx1-Cre or Dmp1-Cre transgenes by crossing these strains with Ai9 Cre reporter mice (27).We focused on cells associated with endosteal and trabecular bone in adult mice by removing the periosteum, epiphyses, and bone marrow from tibias and femurs, followed by sequential incubation of the remaining bone fragments with collagenase and EDTA.Cells expressing high levels of tdTomato were isolated by fluorescence-activated cell sorting (FACS) and subjected to scRNA-seq using the 10X Chromium platform (Fig. S1).For brevity, we will refer to these cells as endosteal but recognize that this preparation contains cells from cancellous bone as well as the endosteum.From two independent isolations from Dmp1-Cre;Ai9 mice, we obtained sequence from a total of 3845 cells (Fig. S1).From a single isolation from Osx1-Cre;Ai9 mice, we obtained sequence from 6071 cells.In a separate set of experiments, we also isolated targeted cells from the periosteum and cortical bone surface of long bones and obtained sequence from 4106 cells targeted by Dmp1-Cre and 15,113 cells targeted by Osx1-Cre (Fig. S1).Initial cluster analysis revealed that hematopoietic and endothelial cells made up only a small proportion of the sequenced cells from each isolation (Fig. S1).The presence of hematopoietic and endothelial cells in these preparations likely represents contamination or rare recombination of the Ai9 locus in these cells, with the latter idea supported by detectable levels of the tdTomato transcript in many cells in these clusters (Fig. S1).
The majority of the cells from each isolation appeared to represent different types of mesenchymal cells based on expression of known stromal cell and osteoblastic marker genes, such as type 1 collagen (Col1a1) and Pdgfra (Fig. S1).One approach to define cell types in our analysis would be to pool the results from the different experiments and perform a cluster analysis using only our own data.However, we and others have noted that scRNA-seq studies from independent laboratories often identify what appear to be the same or similar cell types (clusters) but use novel designations (23,28).In order to avoid propagating this growing complexity in cell designations, we sought to place our results into the context of published work by performing a pooled analysis including our data as well as multiple published datasets.To accomplish this, we reprocessed published datasets and performed canonical correlation analysis-based procedures to normalize our data with those from nine studies that focused on cells associated with murine bone and that used the 10X Genomics platform (18)(19)(20)(21)(22)(23)(24)(25)(26).By combing our new datasets with the published work, approximately 190,000 high-quality cells were included and projected on uniform manifold approximation and projection (UMAP) coordinates to uncover the structure of clusters representing different cell types (Fig. S2).
Even though our datasets included relatively small numbers of hematopoietic cells, the published datasets we used for harmonization contained sufficient numbers of these cell types so that definitive clusters representing all major hematopoietic cell types were apparent (Fig. S2).Similarly, clusters for endothelial cells, pericytes, and Schwann cells were also readily identified (Fig. S2).Because the goal of our study was to better define mesenchymal cell types, we focused the remainder of the study on this group of cell types.

Definition of mesenchymal clusters
Harmonization of the nine published datasets with our data produced a UMAP of 100,000 mesenchymal cells in 11 major clusters (Fig. 1, A-C).Because of the large number of cells included and because the cells were isolated using a variety of methods by independent laboratories (Fig. S3), this reference map may be useful as a guide for the analysis of future scRNAseq experiments using murine bone-associated cells.
The following is a brief description of each cluster.We endeavored to use cluster nomenclature that is consistent with the earliest published study describing that cell type (18,19).It is important to note that the number of cells in each cluster does not necessarily represent the relative abundance of that cell type in bone.The reason for this is that the procedures used to isolate the cells in each study selected for and against different cell populations (Fig. S3).In addition, some cell types, such as mature adipocytes, may not be represented at all due to their inability to survive the isolation processes that were used.

Chondrocytes
Cells in the chondrocyte cluster express high levels of Col2a1, Sox9, as well as the genes Snorc, Fxyd2, and Matn3 (Figs. 1, 2,  and S4).This cluster likely contains many stages of the chondrocytes present in long-bone growth plates as evidenced by subpopulations expressing Col10a1, Ihh, and Pthlh (Fig. S5).
with bone-associated fibroblasts, such as S100a4 and Dcn (19).We acknowledge that the term fibroblast is vague, but its use here reflects the uncertainty regarding the identity of the cell types in these clusters.In addition to S100a4, the fibro-1 cluster is defined by expression of Angptl7, Cilp2, and Anxa8 (Figs. 2 and S4).

Fibro-2
Many of the cells in the fibro-2 cluster express Ly6a and Cd34, which are often used as markers of progenitor cells (19,30).As noted by others, mesenchymal lineage cells that express these genes may represent mesenchymal stem cells or other mesenchymal progenitors (19).Cells in this cluster also express Clec3b and Mfap5, which, unlike Ly6a and Cd34, are not highly expressed by endothelial or hematopoietic clusters, suggesting that they may be more specific markers for skeletal mesenchymal progenitors (Figs. 2  and S4).

Tenocytes
Tenomodulin is one of the most tendon-specific gene products (31) and is a marker that defines a cluster that likely contains tenocytes (Figs. 2 and S4).Other genes enriched in this cluster include Mfap4, Eln, and Itgbl1.

Osteo-X
We have used the novel, and likely temporary, designation of osteo-X to define a cell population that highly expresses periostin (Postn), asporin (Aspn), and Tenascin-W (Figs. 2 and  S4).These genes are highly expressed in periosteal cells (32)(33)(34), suggesting that many of the cells in this cluster originated from the periosteum.Consistent with this idea, many more osteo-X cells were isolated from Osx1-Cre mice when periosteal tissue was included, compared to bones stripped of periosteum (Fig. 3A).

Adipo-CAR cells
High expression of the chemokine Cxcl12 has been used to define a population of stromal cells known as Cxcl12abundant reticular (CAR) cells (35).Baccin and co-workers classified CAR cells into two distinct subsets based on their gene expression profile, adipo-CAR and osteo-CAR (18).In our analysis, adipo-CAR cells express the adipogenic marker Adipoq in addition to high expression of the stromal marker Cxcl12 (Figs. 2 and S4).However, these cells express relatively low levels of the osteoblastic markers Col1a1 and Bglap (Figs. 2 and S4).Adipo-CAR cells also express high levels of Esm1 and Hp (Figs. 2 and S4).

Osteo-CAR cells
Cells in the lower region of the cluster designated osteo-CAR express stromal markers such as Cxcl12 and Adipoq, whereas cells in the upper region express osteoblastic markers such as Col1a1 and Bglap (Figs. 2 and S4).Transcripts that define this cluster include Limch1, Wif1, Angpt4, and Kcnk2 (Figs. 2 and S4).

Preosteoblasts
A cluster that expresses high levels of Spp1, also known as osteopontin, and Mmp13 was designated as preosteoblasts due to the continuity with the osteoblast cluster and expression of osteoblast marker genes in the upper portion of this cluster (Figs. 2 and S4).Other genes enriched in this cluster include Slit2 and Vdr.Additional support for the idea that these cells represent osteoblast progenitors will be presented in later sections of this study.

Osteoblasts
The osteoblast cluster was identified by high expression of the genes encoding osteocalcin (Bglap) and Col1a1 but lacking expression of known osteocyte-specific genes such as Mepe and Sost as well as lower levels of Dmp1 (Figs. 2 and S4).Cells in the osteoblast cluster also express high levels of Car3 and Entpd3, as noted by Tikonova and colleagues (20) (Figs. 2  and S4).

Osteocytes
This cluster expresses genes previously shown to be either produced specifically by osteocytes or highly expressed by osteocytes, including Mepe, Sost, Phex, and Dmp1 (Figs. 2 and   S4) (36).These cells continue to express high levels of gene products characteristic of osteoblasts including Col1a1 (Figs. 2  and S4).We compared the gene expression profile of our osteocyte dataset to previously published scRNA-seq (26,37) and bulk RNA-seq (38) osteocyte profiles (Fig. S6).Only three transcripts were present in all four profiles: Dkk1, Dmp1, and Irx5.Notably, two of the most osteocyte-specific transcripts, Sost and Mepe, were present in only two of the profiles, ours and the one from Youlten et al. (38).
Of the previously published scRNA-seq studies used for harmonization, only the study by Wang et al. reported a distinct cluster labeled as osteocytes (26).One reason for this may be the technical difficulty associated with isolating these cells from mineralized bone.Nonetheless, it is likely that such cells were present in most of the previous studies but were not uncovered as a separate cluster due to their small representation.Our integrated analysis uncovered 1840 osteocytes in total, including 1100 osteocytes from our experiments.
Osx1-Cre targets more osteo-X cells than Dmp1-Cre Having defined the clusters, we next compared the harmonized scRNA-seq results of endosteal cell preparations from Osx1-Cre and Dmp1-Cre and found that these driver strains labeled very similar cell populations, including osteo-X, osteo-CAR, adipo-CAR, preosteoblasts, osteoblasts, and osteocytes (Fig. 3, A and B).A small number of fibro-1, fibro-2, and chondrocytes were also labeled, but together these cell types represented less than 5% of the cells labeled by either transgene.A striking difference between the two transgenes was the percentage of cells in the osteoblast cluster, which represented approximately 30% of the Dmp1-Cre-targeted cells but approximately 60% of the cells targeted by Osx1-Cre (Fig. 3,  A-C).The significance of this observation is unclear but suggests that there may be a population of osteoblasts that are not targeted by Dmp1-Cre.
In the periosteal cell preparations, the greatest difference between Osx1-Cre and Dmp1-Cre was targeting of the osteo-X cluster (Fig. 3, A and B).While the percentages of osteo-X, preosteoblasts, and osteoblasts were similar to one another in Dmp1-Cre-targeted cells, the percentage of osteo-X cells targeted by Osx1-Cre was more than 3-fold greater than either preosteoblasts or osteoblasts (Fig. 3B).This preferential targeting was also evident in the expression pattern of the osteo-X marker transcript Postn (Fig. 3D).These results suggest that Osx1-Cre targets osteo-X cells more effectively than Dmp1-Cre does.
Because the difference in osteo-X targeting was not observed in the endosteal preparations, we explored whether there were significant differences in gene expression that might distinguish osteo-X cells in the periosteum from those at the endosteum.For this comparison, we also included published datasets from calvaria and calvarial sutures (24,25), both of which contain significant percentages of osteo-X cells (Fig. 4A).Comparison of osteo-X cells from the periosteum to osteo-X cells from the other sites revealed that a handful of transcripts, including the one encoding Wnt16, were greatly Cell types identified by scRNA-seq of murine bone enriched in periosteal cells (Fig. 4B).In situ hybridization confirmed expression of Wnt16 exclusively in periosteal cells (Fig. 4, C and D).However, expression of Wnt16 at the periosteum was not confined to osteo-X cells but also occurred in several other cell clusters (Fig. 4E).Thus, Wnt16 expression does not represent a unique property of periosteal osteo-X cells but rather a property of periosteal cells in general.The small level of differential gene expression between osteo-X cells from the periosteum versus the endosteum suggests that the preferential targeting of these cells by Osx1-Cre at one site but not the other is unrelated to differences in gene expression but instead may result from greater abundance of these cells at the periosteum.

Relationship of osteo-X to other clusters
To better characterize cells in the osteo-X cluster and to identify possible relationships to cells in other clusters, we sought to localize osteo-X cells in bone tissue sections using in situ hybridization with probes for transcripts highly expressed Cell types identified by scRNA-seq of murine bone by this cluster.To facilitate orientation, we first localized osteoblasts using a Bglap probe, which identified osteoblasts lining much of the cancellous and endosteal bone surface, with some staining at the periosteum in 4-month-old mice (Fig. 5A).No staining for Bglap was observed in the bone marrow and little or no staining was observed in osteocytes.Two of the defining transcripts for the osteo-X cluster are Aspn and Postn.Consistent with the idea that osteo-X cells are much more abundant in periosteal tissue than at the endosteum, both transcripts were highly expressed in cells at the periosteum but in few cells within the bone marrow cavity (Fig. 5, B and C).Strong staining was also observed in a structure with a location and morphology consistent with the groove of Ranvier (Fig. 5, B and C).Neither transcript was detected in hematopoietic cells in the bone marrow (Fig. 5, B  and C).Notably, transcripts for Postn but not Aspn were detected in cells associated with the cancellous bone surface (Fig. 5, B and C).Faint staining for both probes was detected in a subpopulation of blood vessels in the bone marrow and a few larger vessels in the diaphyseal region showed significant expression of Aspn (Fig. S7).
The presence of Postn-positive cells near the cancellous bone surface suggested that a subpopulation of cells in the osteo-X cluster may be in close association with either osteoblasts or their precursors.To explore this idea further, we used a probe for Cxcl12 to detect adipo-CAR cells and a probe for Limch1 to detect osteo-CAR cells since previous studies have shown that a population of CAR cells contributes to osteoblast formation, at least in cancellous bone (22).Consistent with earlier studies (35), Cxcl12-expressing cells were distributed throughout the bone marrow space, often associated with blood vessels (Fig. 5D).In contrast, osteo-CAR cells were much less abundant and primarily associated with the cancellous bone surface (Fig. 5E).Similar to osteo-CAR cells, cells expressing high levels of Spp1, a marker for cells we designated as preosteoblasts, were predominantly associated with the cancellous bone surface (Fig. 5F).
The close association of osteo-X (expressing Postn), osteo-CAR (expressing Limch1), preosteoblasts (expressing Spp1), and osteoblasts (expressing Bglap) suggests the possibility of a continuum of cells spanning from osteoblast progenitors to mature osteoblasts residing near the bone surface.If this is the case, one would anticipate that suppression of bone remodeling would eliminate mature osteoblasts but leave in place the precursors that are the target of coupling factors produced or controlled by osteoclasts (39).Therefore, to determine which, if any, of these cell populations are dependent on ongoing bone remodeling, we suppressed osteoclast formation in humanized Tnfsf11 mice using denosumab and examined the impact on different mesenchymal cell clusters using scRNAseq.Importantly, to ensure that our analysis of this new scRNA-seq experiment produced cluster annotations consistent with the cell clusters generated by the harmonized UMAP, we used the harmonized dataset and cluster annotation to construct a reference-based mapping model using the Azimuth pipeline (40).In essence, this approach allowed us to use the reference map to define the cell clusters in the new experiment.
We have previously shown that administration of denosumab to humanized RANKL mice once per 2 weeks for at least four doses potently suppresses bone remodeling (41).Therefore, we administered denosumab using this regime and then performed scRNA-seq analysis of cells associated with endosteal and cancellous bone but depleted of hematopoietic and endothelial cells to compare the mesenchymal cell distribution.
We found a dramatic reduction in osteoblast number in the denosumab-treated mice (Fig. 6, A and B), consistent with suppression of bone remodeling.The abundance of preosteoblasts was also slightly reduced as was expression of the preosteoblast marker Spp1 (Fig. 6, B and C).In contrast, the relative abundance of osteo-CAR and adipo-CAR cells was not dramatically altered (Fig. 6, A and B).Consistent with these results, in situ hybridization revealed an almost complete loss of Bglap-positive osteoblasts in vertebral cancellous bone in denosumab-treated mice (Fig. 6D).Preosteoblasts, as defined by Spp1 expression, showed robust staining on cancellous bone in the vehicle-treated mice, which was reduced but not abolished after denosumab (Fig. 6D).In contrast, the osteo-CAR marker Limch1 and the osteo-X marker Postn, displayed little change after denosumab administration (Fig. 6D).Together, these results demonstrate that osteoblasts, and to some extent preosteoblasts, are highly dependent on bone resorption and coupling factors for their existence in cancellous bone.However, the abundance of osteo-CAR and osteo-X cells and their close association with the cancellous bone surface do not require on-going bone remodeling.
While osteo-X and osteo-CAR exhibit characteristics consistent with a role as osteoblast progenitors, they do not possess commonly accepted characteristics of mesenchymal stem cells or skeletal stem cells, such as expression of Ly6a (Sca1) and Cd34 (Fig. S4).Cells in the cluster that we have designated fibro-2 do express these genes and have been proposed by others to represent early mesenchymal progenitors analogous to PαS cells (21).To determine the location of such cells in adult remodeling bone, we performed in situ hybridization with probes for Clec3b and Mfap5, two transcripts highly expressed in fibro-2 cells along with Ly6a and Cd34.We selected Clec3b and Mfap5 rather than Ly6a and Cd34 since the latter two genes are also highly expressed by endothelial progenitor cells (42).Both transcripts were easily detected in cells in the outer layers of the periosteum, consistent with the presence of early mesenchymal progenitors at this site observed by others (Fig. S8) (43).In contrast, we did not detect cells expressing either transcript within the bone marrow cavity or associated with cancellous bone or blood vessels (Fig. S8).These results suggest that the cells represented by the fibro-2 cluster reside predominantly, if not exclusively, at the periosteum.

Discussion
With the advent of scRNA-seq, the characteristics that define a cell type or the number of different cell types associated with a particular tissue are becoming increasingly difficult to agree upon (44).One benefit of our harmonization approach is that it allowed us to identify what may be a minimum number of general mesenchymal cell types associated with murine bone.An important caveat to this view is that our approach only identified cell types that can be sequenced using the 10X Genomics platform.For example, cells with a profile consistent with mature bone marrow adipocytes were not present in any of the datasets that we used Cell types identified by scRNA-seq of murine bone and this is likely due to an inability of such cells to survive the isolation and sequencing methods.Even with this limitation, our approach may facilitate future sequencing projects by providing marker genes that can be used to guide consistent cluster identification.
The heterogeneity within many of the clusters identified here indicates that they contain multiple different cell types.The chondrocyte cluster, for example, clearly contains resting, proliferating, and hypertrophic, or perhaps prehypertrophic, growth plate chondrocytes.Similarly, close examination of the heatmaps for each cluster reveals multiple distinct patterns of gene expression within each cluster.Whether most of these differences represent distinct differentiation states or whether they represent a given cell type in different modes of function will require significant additional study and agreement within the field.
The significance of the restriction of Wnt16 expression to the periosteum is unclear.Periosteal-specific expression of

Cell types identified by scRNA-seq of murine bone
Wnt16 is consistent with its demonstrated role in periosteal but not cancellous bone formation (45,46).It is noteworthy that many of the cell types expressing Wnt16 at the periosteum are also present within the bone marrow cavity.Expression specifically at the periosteum suggests that a stimulating Wnt16 production exists exclusively in the periosteal environment.Accumulating evidence suggests that skeletal muscle directly promotes bone formation (47,48), and it is possible that muscle cells provide a signal that stimulates Wnt16 expression in the periosteal environment.Regardless of the mechanism underlying the restricted expression of Wnt16, its production specifically in periosteal cells provides a useful genetic marker for the presence of periosteal cells within scRNAseq datasets.We were unable to detect Wnt16-expressing cells by in situ hybridization at the endosteum, on cancellous bone, or within the bone marrow cavity in either growing or adult mice.Thus, any bone-associated cells expressing this gene in scRNA-seq datasets are likely derived from periosteal tissue.
The identity and location of skeletal stem cells or mesenchymal stem cells remain active areas of debate and investigation.In our analysis, the only mesenchymal cell cluster to express accepted markers of mesenchymal stem cells, such as Ly6a (Sca1) and Cd34, is fibro-2.Fibro-2 cells, as defined by expression of Clec3b and Mfap5, were easily detected at the periosteum and surrounding tissue but not within the bone marrow cavity.This finding suggests that osteoblast progenitors within the bone marrow cavity of adult mice may not be derived from cells with classical stem cell markers.This scenario is consistent with previous reports suggesting that the majority of osteoblasts produced during physiological bone remodeling are derived from committed replicating progenitors rather than stem cells (49,50).
What is the relationship, if any, of either adipo-CAR or osteo-CAR cells to osteoblasts?Lineage tracing studies in adult mice suggest that at least a portion of osteoblasts on cancellous bone are derived from some type of CAR cell.Specifically, transient activation of a Cxcl12-CreERT2 transgene at 2 months of age led to labeling of more than 30% of trabecular osteoblasts and this labeling remained constant until at least 18 months of age (22).Notably, cortical osteoblasts were not labeled at any time in these same mice.Similarly, Adipoq-Cre targets almost all CAR cells, as defined by Cxcl12-GFP transgene expression, and while few osteoblasts were labeled at 4 weeks of age, about 20% were labeled at 6 months of age (51).Whether the osteoblasts labeled in these studies were derived from adipo-CAR cells, osteo-CAR cells, or both is unclear.Nonetheless, the long-term labeling of osteoblasts by transient activation of the Cxcl12-CreERT2 transgene suggests that some of the CAR cells that it targets represent selfrenewing osteoblast progenitors in adult remodeling bone.
In the present study, we localized osteo-CAR cells, as defined by Limch1 expression, to the endosteal and trabecular bone surface of adult mice.Further, we showed that in actively remodeling bone, these cells are associated with osteoblasts on the bone surface and that they overlap with or are associated with preosteoblasts, as defined by Spp1 expression.Strikingly, suppression of bone remodeling using denosumab led to the elimination of osteoblasts and suppression of Spp1 expression in preosteoblasts but no change in the abundance of osteo-CAR cells, which remained in close association with the bone surface.Based on their location, their independence from bone remodeling and the lineage-tracing studies noted earlier, we suggest that osteo-CAR cells or a subpopulation of osteo-CAR cells represent a continuous source of osteoblast progenitors in adult remodeling bone.
The study that initially proposed the term "osteo-CAR" to refer to cells expressing both Cxcl12 and osteoblast-lineage genes reported that they were localized to arteriolar and nonvascular niches (18).In contrast, our study localized these cells primarily to regions near endosteal and trabecular bone surfaces.As noted by the authors of this earlier work, the extreme signal intensity of the alkaline phosphatase staining that they used to identify osteoblastic cells may have prevented detection of CAR cells near the bone surface (18).
Our rationale for proposing that osteo-CAR, rather than adipo-CAR cells, serve as osteoblast progenitors is based largely on the close proximity of osteo-CAR cells to the bone surface and to existing teams of osteoblasts.Transient activation of an Osx1-CreERT2 transgene at E14.5 labeled a population of osteoblast progenitors that remained spatially restricted to the region of the femur that existed at the time of labeling even up to 10 months of age (52).This later finding reveals that during normal bone remodeling, osteoblasts are derived from progenitors that do not migrate large distances, suggesting that osteo-CAR cells are better positioned to serve as osteoblast progenitors than the majority of adipo-CAR cells.
If osteo-CAR cells do represent an important source of osteoblasts in remodeling bone, it might be expected that deletion of beta-catenin using Dmp1-Cre would inhibit osteoblast formation due to the fact that Dmp1-Cre targets CAR cells.However, it is important to note that Dmp1-Cre targets only about 30% of CAR cells, as defined by Cxcl12-GFP transgene expression (53).Therefore, loss of beta-catenin in only 30% of osteoblast progenitors would not be expected to completely inhibit osteoblast formation, which is consistent with the phenotype of mice lacking beta-catenin in Dmp1-Cre-targeted cells (54).
As opposed to life-long production of osteoblasts by a single type of resident mesenchymal or skeletal stem cell, evidence is accumulating that osteoblasts are derived from a variety of different progenitors depending on the age of the animal, the location of the osteoblasts in the skeleton, and the need for repair (28).Results presented herein suggest that osteo-X and osteo-CAR cells represent potential sources of osteoblast progenitors located in close proximity to the endosteal and cancellous bone surfaces in adult mice.The developmental origin of osteo-X and osteo-CAR cells will require additional study but a likely source is the cell type that is marked by an Osx1-CreERT2 transgene and that gains access to the bone interior at an early stage of endochondral bone formation (55).While initially associated with invading blood vessels, such cells may give rise, over time, to progenitors residing near the bone surface.The presence of osteoblast progenitors near the bone surface is consistent with the ability of osteocyte-derived Cell types identified by scRNA-seq of murine bone sclerostin to inhibit osteoblast formation as sclerostin function depends on local membrane retention by Lrp4 (56).It will be important in future studies to determine whether osteo-X and osteo-CAR cells do in fact serve as osteoblast progenitors and whether their ability to do so is altered in that lead to low bone formation.

Experimental procedures
Animal housing and care Dmp1-Cre, Osx1-Cre, Ai9, and humanized Tnfsf11 mice have been described previously (12,27,41,57).WT C57BL/6J mice were obtained from the Jackson Laboratory.Mice were socially housed at 2 to 5 animals per cage using a blend of onequarter inch corncob bedding and white enrichment paper both produced by Andersons Incorporated.The animal colony was specific pathogen-free based on the Division of Laboratory Animal Medicine's exclusion list.Mice were provided ad libitum water and an irradiated Purina diet of either 5V5M (for breeders) or 5V5R (for maintenance).The temperature range in the room was 68 to 79 F with a set point of 71 ± 2; additionally the humidity range was 30% to 70%.The room was on a 12:12 h light cycle, and the illumination was 364 lux measured 1 m from the floor.

Study approval
All animal procedures were reviewed and approved by Institutional Animal Care and Use Committee of the University of Arkansas for Medical Sciences.

Cell isolation
For all experiments using mice harboring the Cre driver transgenes and the tdTomato reporter allele (Ai9), mice between the ages of 8 and 14 weeks were used.To obtain cells associated with endosteal and trabecular bone, femurs and tibias from a single mouse were cleaned of soft tissues and the periosteum was removed by scraping with a scalpel.After removing the epiphyses, bone shafts were cut open lengthwise using a scalpel and bone marrow cells were removed by flushing with PBS containing 1% bovine serum albumin (BSA).The remaining bone fragments were then cut into smaller pieces (approximately 1 mm in length) and placed into a single well of a 6-well culture dish containing Hank's Balanced Salt Solution and two Wunsch units of LiberaseTM (Milli-poreSigma).The cells were incubated for 20 min at 37 C with shaking after which the released cells were collected by pipetting and incubated on ice.The bone fragments were digested a second time for 20 min with LiberaseTM and the cells saved as above.The bone fragments were then incubated in PBS containing 5 mM EDTA and 0.1% BSA for 20 min at 37 C with shaking.The released cells were collected and saved as above.The bone fragments were subsequently exposed to alternating exposures of either LiberaseTM or EDTA to yield a total of four LiberaseTM digestions and three EDTA incubations.All fractions were pelleted at 300g for 10 min and resuspended in 0.5 ml PBS containing 1% BSA.Fractions were then pooled, filtered through a 70-μm filter, pelleted at 300g for 10 min, and resuspended in less than 1 ml sorting buffer (PBS containing 0.5% BSA and 2 mM EDTA).
To obtain cells from the periosteum, intact femurs were dissected from two mice.Muscles were carefully removed using a scissors without damaging the periosteum.The femurs, including articular cartilage, were placed into a 50 ml conical centrifuge tube containing Hank's Balanced Salt Solution with calcium and two Wunsch units of LiberaseTM (Milli-poreSigma) and incubated for 20 min at 37 C with shaking.The released cells were collected by pipetting and stored in PBS containing 1% BSA on ice.Bones were then washed twice with 50 ml of PBS and incubated in calcium-free PBS with 5 mM EDTA and 0.1% BSA and incubated for 20 min at 37 C with shaking.The released cells were collected and stored on ice.The femurs were digested one more time for 20 min with LiberaseTM and the cells were collected.All cells collected were then pooled, filtered through a 70-μm filter, and pelleted at 300g for 10 min.Cells were then resuspended in 200 μl sorting buffer for sorting.
Cells from mice harboring the Ai9 allele were sorted using a BD FACS Aria III with a 100 μm nozzle to collect tdTomatopositive cells.For mice not harboring a reporter allele, cells were obtained by LiberaseTM and EDTA incubation as above but instead of isolation by FACS, cells were depleted of hematopoietic cells and endothelial cells using a lineage depletion kit (Miltenyi Biotec, cat.no.130-090-858), followed by CD45, CD117, and CD31 microbeads (Miltenyi Biotec, cat.nos.130-0520301, 130-091-224, and 130-097-418) according to the manufacturer's instructions.Cells in the flow-through were pelleted at 300g for 10 min and resuspended in 20 μl of sorting buffer.

10X Genomics sequencing
Single cells isolated from digested bones were stained with ReadyProbes cell viability imaging kit, blue/green (Thermo Fisher Scientific, catalog # R37609) and manually counted using a hemocytometer under an EVOS M7000 microscope (Thermo Fisher Scientific).Immediately following cell counting, samples were processed using Chromium Next GEM Single Cell 3 0 Reagent Kits v3.1 (Dual Index) as described in the manufacturer's instructions (10X Genomics).In brief, aiming for 8000 cells per library, single-cell suspensions with more than 70% live cells were loaded onto Chromium Controller (10X Genomics) to generate gel beads-inemulsions.Then, copartitioned cells were lysed, primers were released from gel beads, and barcoded full-length cDNA was produced and amplified.3 0 gene expression libraries were generated from cDNA by fragmentation, end repair, A-tailing, adaptor ligation, and index PCR amplification.The concentration and size distribution of final libraries were assessed by Qubit 1X dsDNA high sensitivity assay (Thermo Fisher Scientific, catalog # Q33231) and the fragment analyzer system (Agilent).Libraries were sequenced either on a NextSeq500 or NovaSeq 6000 (Illumina) with paired-end mode (read1: 28 cycles, read 2: 90 cycles, i7: 10 cycles, i5: 10 cycles) to generate a minimum of 20,000 read pairs per cell.

Bioinformatic analysis of scRNA-seq
The fastq files were preprocessed using Cell Ranger software version 6 (https://www.10xgenomics.com/support/software/cell-ranger/, 10X Genomics) to produce feature-barcode matrixes.The alignments were performed using mouse reference genome mm10.The feature-barcode matrixes were imported for further analysis in R suite software using Surat version 4.2.0, https://www.r-project.org/(40).Cells with between 500 and 3000 transcripts were included for further analysis.The harmonization between samples was performed using a canonical correspondence analysis method based on the top 50 principal components and 6000 most variable features to minimize batch effect.The harmonized results were used for clustering using Louvain algorithm with multilevel refinement and UMAP for dimension reduction.The gene specific markers of individual clusters and differential expressed genes were identified using MAST algorithm for cell-type identification (58).The normalized expression kernel-weighted density values were calculated for selected marker genes using Nebulosa (59) then visualized on either UMAPs or violin-box plots.The expression level of the selected genes was plotted based on normalized expression value of relative count per 10,000 (CP10K).The integrated datasets were used as the input to Azimuth workflow (40) to construct a reference map and the annotated cell type models for reference-based mapping analysis of denosumab administration experiments.The results of the integrated datasets of this study are provided at https://uamscmdr.shinyapps.io/bone_cells_explorer_all, enabling query and visualizing of annotated cell types, datasets, and genes.

Denosumab administration
Denosumab (Prolia, 60 mg/ml) was purchased from the UAMS pharmacy, aliquoted into 0.5 ml sterile microtubes, and stored at 4 C protected from direct light.Thirteen-month-old humanized Tnfsf11 male mice were injected subcutaneously with either vehicle (saline) or denosumab at a dose of 10 mg/ kg body weight, in a total volume of 100 μl, once every 2 weeks (Q2W) for either 4 or 5 doses.In both cases, bones were harvested 2 weeks after the final dose.

Figure 1 .
Figure 1.Harmonization of new and existing bone mesenchymal cell datasets.Results shown in this figure were derived from single-cell RNA-seq analysis of 100,000 mesenchymal cells of which 26,535 were isolated and sequenced in the present study.A, heat map of transcript abundance for the major genes driving definition of 11 mesenchymal cell clusters.Cell names and color codes are shown at the top.B, UMAP representation of the 11 major clusters of mesenchymal cells.Names and color codes same as in A. C, UMAP feature plots showing expression of representative cell type-specific transcripts.Red denotes high expression.UMAP, uniform manifold approximation and projection.

Figure 2 .
Figure 2. Major transcripts defining mesenchymal cell clusters.violin plots showing expression levels of four major transcripts defining each of the 11 major mesenchymal cells clusters using the same dataset used in Figure 1.The box plots show the maximum value, first quartile value, median value, third quartile value, and the minimum value.Cell names and color codes are the same as in Figure 1.

Figure 3 .
Figure 3. Osx1-Cre targets periosteal osteo-X cells better than Dmp1-Cre.A, UMAPs of the cells targeted by Dmp1-Cre or Osx1-Cre in endosteal and trabecular bone (endo) or in periosteal bone (peri).Cell names and color code are the same as in Figures 1 and 2. B, pie charts depicting the relative number of cells in each cluster targeted by Dmp1-Cre or Osx1-Cre at each site.Please note that cells from only 2 of 4 mice were included for the Osx1-Cre periosteoal analysis.C and D, UMAP feature plots showing the abundance of Bglap and Postn transcripts in the same cell preparations shown in panel A. Postn, periostin.

Figure 4 .
Figure 4. Periosteal-specific expression of Wnt16.A, UMAP representations of all the periosteal cells isolated in the present study (periosteum), all of the endosteal and trabecular bone-associated cells isolated in the present study (endosteum), cells isolated by Ayturk et al. (24) (calvaria), or cells isolated by Holmes et al. (25) (suture).B, differential gene expression plots comparing expression of genes in the osteo-X cluster from the indicated cell preparations.Circle size corresponds to the ratio of Wnt16 expression between the two indicated bone sites.C, in situ hybridization using a probe for Wnt16 and a femoral bone section from a 7-day-old male C57BL/6 mouse.Red denotes Wnt16 expression.Arrowheads indicate location of periosteal bone surface.Dapb is a bacterial transcript used as a negative control.D, in situ hybridization was performed on a femoral bone section from a 4-month-old female C57BL/6 mouse.Images at the right are higher magnifications of the boxed areas in the left panel.E, UMAP feature plots showing expression of Wnt16 in cells from the indicated preparations.Red denotes high expression.

Figure 5 .
Figure 5. Localization of osteo-X and osteo-CAR cells.RNAScope-based in situ hybridization of femoral bone sections from a 4-month-old female C57BL/ 6 mouse using the probes for transcripts expressed by the following cell types: osteoblasts, defined by Bglap (A); osteo-X, defined by Aspn (B); osteo-X, defined by Postn (C), adipo-CAR, defined by Cxcl12 (D); osteo-CAR, defined by Limch1 (E), and preosteoblasts, defined by Spp1 (F).Higher magnification images for each section are shown below in each panel.Red denotes transcript expression.Arrowheads in panel B and C indicate location of groove of Ranvier.Aspn, asporin; CAR, Cxcl12-abundant reticular; Postn, periostin.

Figure 6 .
Figure 6.Suppression of remodeling does not alter location of osteo-X and osteo-CAR cells.A, UMAP representations of mesenchymal cells isolated from endosteal and trabecular bone by negative selection from mice treated with vehicle (Veh) or denosumab (Dmab).Cell names and color codes are the same as in earlier figures.B, bar graphs indicating the relative abundance of the cell clusters shown in panel A. C, violin plot of Spp1 expression in cells of the preosteoblast clusters shown in panel A. D, two-color RNAScope in situ hybridization using the indicated probes and vertebral bone sections from mice treated with vehicle or denosumab.CAR, Cxcl12-abundant reticular; uniform manifold approximation and projection.