Identification of a stem cell mediating osteoblast versus adipocyte lineage selection

Most skeletal fragility disorders are characterized by bone loss with a concurrent gain in marrow adipocytes 1–8. This suggests that a cell that forms adipocytes at the expense of osteoblasts is central to the pathogenesis of skeletal disorders. However, this cellular point of bifurcation between adipocyte and osteoblast differentiation pathways remains unknown. Here, we identify a new cell type defined by co-expression of skeletal stem cell and adipocyte precursor markers, 9–13 (CD24+CD29+ skeletal stem cells (SSCs)), that serves as a key cellular point of bifurcation between the osteoblast and adipocyte differentiation pathways, giving rise to closely related osteoblast and adipocyte lineage-restricted precursors. CD24+CD29+SSCs comprise a small fraction of SSCs, and only this fraction displays full stemness features, including the ability to undergo serial transplantation. In line with serving as the osteoblast/adipocyte bipotent cell, the “bone to fat” tissue remodeling occurring in models of postmenopausal osteoporosis or after high fat diet exposure occur in part by reprogramming these CD24+CD29+SSCs to change their output of lineage-restricted precursors. Lastly, as subcutaneous white adipose tissue displays a similar set of CD24+CD29+ stem cells and related lineage-restricted progenitors, these findings provide a new schema explaining the stem cell basis of bone versus adipose tissue production that unifies multiple mesenchymal tissues.

Developmental analysis corroborates this schema and indicates that it applies across multiple mesenchymal tissue types, including the marrow resident SSCs outside the periosteum (Fig 1d). In the early postnatal period, bone solely displays osteogenic output, only gaining the capacity to produce mature marrow adipocytes at approximately 4-6 weeks of age in mice. Thus, the rst skeletal adipocytes developmentally predate other populations previously considered as marrow adipocyte precursors, such as LEPR+ Cxcl12 abundant reticular (CAR) cells [18][19][20][21] . Similarly, cre lines labeling CAR cells only provide partial labeling of the skeletal adipocyte pool under many conditions. Thus, while CAR cells can contribute to marrow adipogenesis, such as with aging or irradiation, they cannot account for all marrow adipogenesis. In this respect, it is notable that CD24+CD29+SSCs do not overlap with LEPR+ cells (Extended Data Fig 1e). We observed that in the early postnatal period that bone solely contains CD24+CD29+ and CD29+ SSCs, whereas CD24+SSCs emerge at approximately 6 weeks of age with a timing matching adipogenesis (Fig 1d-e, Extended Data Fig 1b, 1f-g). Additionally, to see if this schema may apply to multiple mesenchymal tissues, subcutaneous white adipose tissue (sWAT) was similarly examined, nding that it also contained CD24+CD29+SSCs and abundant putatively adipogenic CD24+SSCs expressing the adipocyte lineage marker CD36 while lacking putatively osteogenic CD29+SSCs (Fig 1f, Extended Data Fig 1h-i) 22 . Similar ndings were observed in intrascapular and visceral adipose tissue (Extended Data Fig 1j-k).
In vivo tissue production assays were then used to directly establish the osteoblast/adipocyte bipotency of CD24+CD29+SSCs. Only CD24+CD29+SSCs and CD29+SSCs displayed intrinsic osteogenic capacity through the ability to form bone organoids in vivo after isolation and transplantation into the kidney capsule of secondary hosts (Fig 2a-b). Conversely, only CD24+CD29+SSCs and CD24+SSCs displayed adipogenic capacity in the same system, staining positive for LipidTOX, with CD24+CD29+SSCs generating bone organoids encapsulating a marrow-like space including Perilipin1+ adipocytes (Fig 2c-d).
Similarly, only CD24+SSCs and CD24+CD29+SSCs or their counterparts with identical surface immunophenotypes from sWAT were able to reconstitute adipogenesis when transplanted into the residual mammary fat pad of "fatless" A-Zip/F1 hosts 23 . Use of "fatless" A-Zip/F1 hosts allows for both assessment of adipogenesis and functional assessment of adipocyte reconstitution via rescue of the severe lipoatrophy of these hosts without the presence of confounding host adipocytes 11 . Only CD24+CD29+SSCs and CD24+SSCs and their sWAT counterparts were able to reconstitute adipogenesis after transplantation (Fig 2e-j, Extended Data Fig 2a-c). Not only did these two populations from either bone or fat sources physically reconstitute adipocytes at the gross, histologic and immunophenotypic levels, but they functionally reconstituted adipogenesis as seen through suppression of the diabetic phenotype associated with the severe lipoatrophy in A-Zip/F1 mice or through the reconstitution of systemic adipokine levels (Fig 2k i and iii, Extended Data Fig 2d i and iii) 11,23 . In keeping with the model that CD24+CD29+SSCs give rise to adipocyte committed and therefore non-stem CD24+SSCs, analysis of extended timepoints after transplantation showed that only the reconstitution by CD24+CD29+SSCs was durable, whereas the physical and functional reconstitution of adipogenesis by CD24+SSCs had largely abated by 20 weeks post transplantation (Fig 2k ii and  This nding that only a small CD24+CD29+ fraction within the current de nition of SSCs were capable of durable reconstitution of adipogenesis implied that CD24+CD29+ cells may be the only truly stem fraction of SSCs, thereby prompting a revision of the SSC de nition. To test this directly, each of the CD24/CD29-de ned fractions of SSCs was subjected to transplantation-based assessment of selfrenewal and differentiation hierarchy (Fig 3a). CD24+CD29+SSCs were capable of self-renewal, being able to maintain their immunophenotype through transplantation (Fig 3d-e). CD24+CD29+SSCs sat at the apex of their differentiation hierarchy, being able to generate CD29+SSCs and more mature CD105+, THY+ and 6C3+ derivatives when transplanted into wild type mammary fat pads (Fig 3d, Extended Data  Fig 3a). Interestingly, CD24+CD29+SSCs favored production of CD24+SSCs when transplanted into residual fat depots of A-Zip/F1 hosts, highlighting how differences in host environment regulate the lineage commitment of this bipotent population (Fig 3e). In contrast, CD29+SSCs did not give rise to the CD24+CD29+SSCs after transplantation (Fig 3f). CD24+SSCs showed only limited engraftment in wild type hosts (Fig 3b) but engrafted and rapidly differentiated into a 6C3+ population without generating other CD24/29-de ned SSC subsets when transplanted into A-Zip/F1 hosts (Fig 3c). CD24-CD29-SSCs failed to engraft after transplantation (Fig 3g). Therefore, CD24+CD29+SSCs are the only fully stem fraction of SSCs, as only CD24+CD29+SSCs self-renew and sit at the apex of their differentiation hierarchy.
Current methods for tracking the differentiation of speci c types of skeletal cells rely exclusively on cellular transplantation and subsequent re-isolation of derived populations, which, while a powerful approach, is limited by the tissue disruption associated with the transplantation procedure. To counter this limitation and determine lineage and differentiation relationships among skeletal cell types in their native environment, we modi ed recently described methods for tracking clonal and sub-clonal mitochondrial mutations across populations. Speci cally, we utilized PolG D257A/D257A mice with a mitochondrial DNA "mutator" phenotype that provide a rich substrate of mutations for analysis 24,25 . Each of the CD24/CD29-de ned SSC populations and other skeletal populations including CD105+, THY+, 6C3+ or THY+6C3+ cells were isolated by FACS from 3-month old PolG D257A/D257A femurs and were subjected to analysis of mitochondrial DNA mutations by ATAC-seq based mitoDNA capture 24 . A number of clonal variants were observed to be shared among CD24+CD29+, CD24+ and CD29+ SSCs, which, together with differentiation hierarchy studies, provides direct evidence that bipotency of CD24+CD29+SSCs is observed at the level of single clones (Extended Data Fig 3b-d). A cluster dendrogram was constructed in an unsupervised manner from these variants, placing CD24+CD29+SSCs at the rst branch of this dendrogram in line with our differentiation hierarchy model (Fig 3h). When considered alongside transplantation-based differentiation hierarchy studies, CD24+CD29+SSCs clonally transmit variants to each of the other populations isolated, and thus CD24+CD29+SSCs display multipotency for other skeletal cell types in the native bone environment (Fig 3i). Consistent with our model, analysis of the genomic DNA portion of the ATAC-seq data demonstrated broadly different epigenetic pro les for CD24+CD29+, CD24+ and CD29+ SSCs (Fig 3j, Extended Data Fig  3e). There was decreased accessibility of several genes associated with adipogenesis (Fasn, Cebpg, Stat5a, Stat6) accompanying the osteoblast commitment seen in CD29+SSCs. However, the reciprocal effect of silencing of osteoblast-lineage associated genes (Sp7, Spp1, Crtap) in CD24+SSCs was less pronounced (Fig 3k).
The physiologic bipotency of CD24+CD29+SSCs for osteoblast and adipocyte generation was con rmed in the native bone environment of irradiated A-Zip/F1 hosts using orthotopic transplantation directly into the femoral marrow cavity. Indeed, after orthotopic transplantation CD24+CD29+SSCs directly gave rise to both RUNX2+ osteoblasts in trabecular bone and to mature bone marrow adipocytes positive for FABP4 and LipidTOX staining (Fig 4a-d). In addition to these mature pools of cells, CD24+CD29+SSC graft-derived cells also incorporated into the growth plate (Fig 4e), in line with recent reports describing the growth plate region as a potential reservoir for skeletal stem cells 9, 26 . In keeping with this growth plate localization, a portion of the CD24+CD29+SSC derived cells are labeled by a PTHrP-mCherry reporter previously associated with a growth plate resident label retaining multipotent population 26 (Fig  4f). Similar to ndings in organoid systems, transplanted CD24+SSCs only gave rise to bone marrow resident LipidTOX+ adipocytes (Fig 4g-h), and CD29+SSCs only formed RUNX2+ osteoblasts (Fig 4i-j).
Both CD24+ and CD24+CD29+ SSCs provided functional reconstitution of adipogenesis in the orthotopic setting, as shown by signi cant suppression of the A-Zip/F1 diabetic phenotype and a signi cant increase in circulating adiponectin levels (Fig 4k).
Previously, the large number of markers needed to de ne SSCs and related progenitor populations has precluded determination of their location and, accordingly, the physical trajectory of differentiating skeletal cells. To address this, imaging mass cytometry was performed by applying a 13-color panel that included the ow cytometry markers used to de ne these populations (Fig 4l-n, Extended Data Fig 4a-r). Robust staining was observed in all regions except the growth plate, which was excluded from further analysis. Segmentation and gating on per cell marker expression were used to localize populations matching the ow cytometry de ned cell types studied above (Fig 4o-r, 4w). This revealed that CD24+CD29+SSCs were enriched in the secondary ossi cation center and the primary spongiosum near the growth plate and included many Nestin+ cells (Fig 4o, 4q Consistent with the physiologic absence of adipogenesis in the periosteum, we observed that periosteum lacked CD24+SSCs and only contained osteoblast committed CD29+ SSCs (Fig 4p). To elucidate whether there is a physical organization of differentiating cells derived from CD24+CD29+SSCs that matches their transplantation determined differentiation hierarchy, the cell types that serve as the nearest cellular neighbors of CD24+CD29+SSCs were calculated. In line with our differentiation hierarchy studies (Fig 3), the most likely nearest neighbors for CD24+CD29+SSCs, aside from other CD24+CD29+SSCs, are CD24+SSCs and CD29+SSCs or other CD29+ populations with a similar immunophenotype (Fig 4s-w, Extended Data Fig 4s-v). Thus, the physical proximity of these subsets supports the model of lineage bifurcation and differentiation hierarchy observed in transplantation systems and is consistent with a physical ow of differentiating cells stemming from CD24+CD29+SSCs. "Bone to fat" tissue remodeling is a signature of multiple disease states, including osteoporosis or obesity. Identifying CD24+CD29+SSCs as the point of adipocyte/osteoblast lineage bifurcation enables deconvolution of the relative contribution of alterations in lineage commitment versus changes in the differentiation/function of mature post-commitment cell types to these disorders. Indeed, in tandem with high fat diet (HFD)-associated positive energy balance (HFD-PEB) causing "bone to fat" remodeling of the skeleton (Extended data Fig 5a-d), the total numbers of adipogenic committed CD24+SSCs were signi cantly increased with a decrease in osteogenic committed CD29+SSCs (Fig 5a-b, Extended Data Fig  5e). To determine the cellular basis of how HFD-PEB acted on CD24/CD29-de ned SSC subsets to produce this "bone to fat" remodeling, each of these subsets were isolated and transplanted into hosts exposed to matching high fat or control low fat diets for 12 weeks (Fig 5c). HFD-PEB acted directly on transplanted CD24+CD29+SSCs to both decrease their osteogenic output and increase their adipogenic output (Fig 5d-f, Extended Data Fig 5f-g). Thus, HFD-PEB acts directly on CD24+CD29+SSCs to change their characteristic output of osteoblast versus adipocyte committed progenitors as a means to effect tissue remodeling. HFD-PEB also acted on transplanted lineage committed CD29+SSCs to decrease their bone formation capacity but did not introduce post-commitment lineage plasticity as CD29+SSCs did not display a gain in adipogenic capacity (Fig 5d-f). Consistent with HFD-PEB directly reprogramming CD24+CD29+SSCs, broad transcriptional changes in CD24+CD29+SSCs were observed after HFD, including upregulation of Bmp1, Bmp5 and adipogenesis associated genes (Adam8, Gas6, Foxc1) 5,28,29 while expression of genes associated with stemness or speci cally with SSCs was maintained (Bmp2, Klf4, Sox9, Hes1, Hey1) 9,10,30 (Fig 5g-h, Extended Data Fig 5h).
Next, we assessed whether direct reprogramming of CD24+CD29+SSCs is a convergent mechanism for tissue remodeling occurring in multiple skeletal disease states, including postmenopausal osteoporosis. Indeed, mice undergoing the ovariectomy model of post-menopausal osteoporosis displayed a signi cant expansion in adipogenic committed CD24+SSCs and contraction of osteogenic committed CD29+SSCs similar to that observed after HFD exposure (Fig 5i-j, Extended Data Fig 6a-g). This, likewise occurred due to ovariectomy directly reprogramming CD24+CD29+SSCs to skew these cells away from the production of osteoblast committed and towards adipocyte committed progenitors, as seen with CD24+CD29+SSCs isolated from unmanipulated donors and transplanted into overiectomized hosts (Fig 5k-n, Extended data  Fig 6h-k). This enhanced adipocyte and suppressed osteoblast lineage commitment by CD24+CD29+SSCs was durable and cell intrinsic, as seen through transplantation of CD24+CD29+SSCs from overiectomized donors into non-overiectomized hosts (Extended data Fig 6l-o). These ndings are consistent with the decreases in osteoblast numbers and bone formation and concurrent increases in marrow adipogenesis seen at late timepoints either post ovariectomy in mice or in human postmenopausal osteoporosis (Extended data Fig 6j-k) 4,8,31 . Thus, both ovariectomy and HFD-PEB act directly on CD24+CD29+SSCs to skew their production of lineage committed progenitors during the bifurcation step.
Taken together, we here identify that CD24+CD29+SSCs represent a key point of bifurcation between lineage committed CD24+SSC adipocyte precursors and CD29+SSC osteoblast precursors and the only fully stem fraction of SSCs. Skeletal disorders act directly on CD24+CD29+SSCs to increase their adipogenic commitment and decrease their osteogenic commitment, indicating a direct competition for lineage allocation in these cells in at least a subset of bone disorders. Insofar as these cells and their functions are preserved across both bone and sWAT tissue, this represents the rst example of a crosstissue schema identifying a discrete, physiologic stem cell identity shared across multiple mesenchymal tissue types.   bar 1mm (a) and 500µm (b). c, LipidTOX staining (green) con rms adipogenesis by the transplanted cells (tomato red). DAPI (blue) for nuclei. Scale bar 100µm. d, CD24+CD29+SSCs (tomato red) were allowed to form organoids after transplantation, and graft derived adipogenesis (Perilipin 1 staining, green) is shown. Scale bar 100µm. Enlarged view of dotted white boxes (i-ii). DAPI (blue) for nuclei. e, CD24+, CD24+CD29+, CD24-CD29-and CD29+ SSCs (top to bottom) from mouse femurs (left) or subcutaneous white adipose tissue (sWAT, right) were transplanted into A-Zip/MIP-GFP hosts. Reconstitution of adipose tissue by the graft compared to the contralateral residual adipose depots, representative gross images taken 12 weeks post-transplantation. f, H and E staining of the tissues in (e), derived from femoral skeletal cells. Scale bar 500µm. g, (i), Weight of reconstituted graft (transplants) adipose tissue derived from femoral skeletal cells in (e). Signi cant increase in adipose tissue produced by CD24+ (****p <0.0001) and CD24+CD29+ (****p <0.0001) SSCs. One-way ANOVA, Sidak's multiple comparison test; mean ± SEM; n=4 to 5, ns= non-signi cant. ii-iii, Transplanted CD24+ (ii) and  The location of CD24 and CD29 de ned SSC subsets in bone. a-e, Orthotopic transplantation of CD24+CD29+SSCs (green) into the femoral cavity of irradiated A-Zip/ MIP-GFP mice for 12 weeks. Images represent engrafted transplanted cells in host trabecular bone including osteoblast differentiation of graft cells as shown by RUNX2 (magenta) staining. Enlarged images for dotted yellow boxes (i-iii) (a). Orthotopically transplanted CD24+CD29+SSCs (green) engrafted in host bone marrow, differentiated into adipocytes shown by Fabp4 (red) (b) and LipidTox (red) staining (c), engrafted in the host epiphysis adjacent to Endomucin+ (red) endothelial cells (d), and localized into the host growth plate (white arrows) (e) . DAPI for nuclei. Scale bar 100µm (a, d), 50µm (b, c). f, FACS plots for Pthrp-mCherry+ cells isolated from the femurs of 15-day old PTHrP mcherry/+ mice. g-h, Transplanted CD24+SSCs (green) reconstituted into the bone marrow of secondary hosts (g), and differentiated into LipidTOX + (red) adipocytes (h). Scale bar 20µm. i-j, Transplanted CD29+SSCs (green) engrafted in host trabecular bone, differentiated into osteoblasts as shown by RUNX2 (magenta; enlarged view of dotted orange boxes) staining (i), and reconstituted in bone marrow staining negative for LipidTOX (red) (j). DAPI for nuclei. k, Intrafemoral  "Bone to fat" remodeling disorders reprogram lineage commitment by CD24+CD29+SSCs. a, Representative FACS plots showing the distribution of SSC subsets isolated from femurs of mice exposed to low fat diet (LFD, left) and high fat diet (HFD, right) for 4 (top) and 10 (bottom) weeks. b, Signi cant increases were seen in CD24+SSCs (*p = 0.0154 at 4 weeks; *p = 0.0476 at 10 weeks) (top), and CD24+CD29+SSCs (*p = 0.045 at 4 weeks; *p = 0.0257 at 10 weeks) (bottom) after HFD exposure.