Skeletal stem cell fate defects caused by Pdgfrb activating mutation

Two platelet-derived growth factor receptors (PDGFRs) have been identified in mammals, PDGFRα and PDGFRβ, which bind to five PDGF ligands. PDGFRs play crucial and distinct roles in embryo development by regulating the proliferation, migration, survival and differentiation of many mesenchymal-derived cells that populate all tissues and organs (Andrae et al., 2008; Hoch and Soriano, 2003; Klinkhammer et al., 2018). It has recently been discovered that humans with gain-of-function mutations in PDGFRB exhibit a spectrum of phenotypes affecting the skeleton and other connective tissues in an autosomal-dominant fashion (Guérit et al., 2021). These mutations result in constitutive PDGFRβ signaling, causing Penttinen syndrome (MIM 601812) or Kosaki overgrowth syndrome (MIM 616592). Both disorders progressively affect the skeleton, beginning in childhood. Other activating variants of PDGFRB are associated with a milder phenotype, infantile myofibromas (MIM 228550), which does not affect the skeleton. Penttinen syndrome, with PDGFRB mutations V665A or N666S (mutated in the first kinase domain), is characterized as a premature aging condition with osteoporosis, scoliosis, lipoatrophy, dermal atrophy, aneurysms and acro-osteolysis (Bredrup et al., 2019; Johnston et al., 2015). Kosaki overgrowth syndrome, with PDGFRB mutations P584R or W566R (mutated in the juxtamembrane domain), is characterized by tall stature, elongated long bones, enlarged hands and feet, distinctive facial features, scoliosis, hyperelastic skin, aneurysms, myofibromas and neurodegeneration (Minatogawa et al., 2017; Takenouchi et al., 2015). The pathological mechanisms of the human disorders are still unknown.

We have previously demonstrated that mice with a gain-of-function D849V mutation in Pdgfrb (mutated in the second kinase domain, corresponding to the human D850 residue) died between 2-3 weeks of age after developing a postnatal wasting phenotype (He et al., 2017; Olson and Soriano, 2011). Surprisingly, these phenotypes were modulated by signal transducer and activator of transcription 1 (STAT1). Mice with Pdgfrb^D849V mutation but lacking Stat1 (Pdgfrb^+/D849VStat1^−/− mice) survived until 8-9 weeks, becoming overweight and with widespread connective tissue overgrowth (not obesity), thick calvarias, abnormally curved spine and enlarged rib cage (He et al., 2017). Although the murine model clearly identified Stat1 as a gene modulating the phenotype spectrum driven by Pdgfrb^D849V, many questions remain about the target cell types and signaling pathways underlying such striking phenotypes. Moreover, the role of PDGFRβ in the skeleton is not well understood, as PDGFRβ appears to be redundant for skeletal development based on the normal skeletal phenotypes of global and osteoblast-specific knockouts in mice (Bohm et al., 2019; Brun et al., 2020; Soriano, 1994).

Ligand binding to wild-type PDGFRβ induces receptor dimerization, which activates the kinase activity of the receptor and results in autophosphorylation of intracellular tyrosine residues that activate downstream signaling pathways (Lemmon and Schlessinger, 2010). Gain-of-function PDGFRβ mutations disrupt the inactive conformation of the receptor, leading to constitutive kinase activity and autophosphorylation. PDGFRs use a variety of signaling pathways to mediate their effects on cell behavior, including PI3K, MAPK, PLCγ and STAT1/3/5 (Demoulin and Essaghir, 2014; Heldin and Westermark, 1999; Tallquist and Kazlauskas, 2004). PDGFRβ is particularly important for pericyte development and function, but it is also expressed on fibroblasts, osteoblasts and stem/progenitor cells (Andrae et al., 2008).

Skeletal stem cells (SSCs) residing in bone and bone marrow (BM) are responsible for postnatal bone development, tissue homeostasis and repair (Bianco and Robey, 2015). A single SSC at the apex of skeletal lineages can give rise to chondrocytes, osteoblasts, adipocytes and fibroblasts. Recent findings with in vivo lineage tracing and single-cell transplantation have increased the rigor of SSC biology and improved our understanding of SSC heterogeneity (Ambrosi et al., 2019; Serowoky et al., 2020). Perisinusoidal vasculature in BM is surrounded by SSCs expressing PDGFRβ, PDGFRα, CD146 [melanoma cell adhesion molecule (MCAM)], nestin (Nes), LepR and Cxcl12, whereas arterial vasculature is associated with SSCs expressing PDGFRβ, PDGFRα, stem cell antigen 1 (Sca1; also known as Ly6a) and LepR (Méndez-Ferrer et al., 2010; Morikawa et al., 2009; Sacchetti et al., 2007; Zhou et al., 2014). The resting zone of the growth plate harbors SSCs expressing Grem1, Col2a1, parathyroid hormone-related peptide (PTHrP; Pthlh) and Itgav (Chan et al., 2015; Mizuhashi et al., 2018; Newton et al., 2019; Worthley et al., 2015). The periosteum and cranial sutures contain SSCs expressing PDGFRβ, Gli1, Axin2, cathepsin K (Ctsk) and αSMA (Debnath et al., 2018; Ortinau et al., 2019; Shi et al., 2017; Yang et al., 2013; Zhao et al., 2015). PDGFRβ is broadly expressed in SSCs, but its role has not been examined. We speculated that PDGFRβ signaling could be functional in SSCs, and hypothesized that elevated Pdgfrb^D849V mutant signaling in SSCs could alter stem cell functions to generate skeletal disorders.

To explore the PDGFRβ^D849V skeleton, we generated a cohort of mice and measured their weights and bone lengths. As Stat1 is an important modifier of PDGFRβ^D849V phenotypes (He et al., 2017), four offspring genotypes were established with the Sox2-Cre driver for germline activation of Pdgfrb^D849V and/or deletion of Stat1-floxed alleles by crossing Pdgfrb^+/[s]D849VStat1^flox/flox and Stat1^+/−Sox2-Cre^+/− mice. The resulting four genotypes are: Stat1^+/− (hereafter designated as S^+/−), Stat1^−/− (S^−/−), Pdgfrb^+/D849VStat1^+/− (KS^+/−), and Pdgfrb^+/D849VStat1^−/− (KS^−/−) (Fig. 1A). The Pdgfrb allele expressing D849V is designated K because the mutation is in the kinase domain. We used two genetic controls, S^+/− and S^−/−, that do not show growth or survival defects (He et al., 2017; Meraz et al., 1996). KS^+/− mice died by 3 weeks of age with features of autoinflammation and wasting. KS^−/− mice, however, were rescued in survival at 3 weeks and died at around 8-9 weeks, consistent with previous findings (He et al., 2017). Thus, body weights were collected from live mice at 2-9 weeks of age, and bone length data were collected from tibias at 3 or 6-8 weeks of age. In the resulting growth curve, KS^+/− and KS^−/− mice were clearly distinguishable by their obvious wasting and overgrowth phenotypes, respectively (Fig. 1B). Whole-body X-ray showed smaller bones in KS^+/− at 2 weeks of age (Fig. 1C) and larger bones in KS^−/− at 5-7 weeks of age (Fig. 1D) with an abnormally curved spine, as previously reported (He et al., 2017). Dissected 3-week-old KS^+/− tibias were shorter than KS^−/− and control tibias (Fig. 1E), and by 6-8 weeks the KS^−/− tibias were significantly longer than controls (Fig. 1F).

To determine whether SSCs were altered in PDGFRβ^D849V-expressing mice, we performed SSC quantification and colony formation assays. SSCs were immunophenotyped from limb bones with enzymatic digestion and flow cytometry using two well-established SSC markers, PDGFRα and Sca1 (PαS) (Houlihan et al., 2012; Morikawa et al., 2009) (Fig. S1A). The percentage of PαS SSCs in the stromal fraction (excluding hematopoietic and endothelial cells) was calculated at 3 and 6 weeks of age. The PαS percentage was similar between all genotypes at 3 weeks old (Fig. 2A), but it was significantly increased in cells isolated from KS^−/− bones at 6 weeks old (Fig. 2B). To characterize stem cell function, we sorted PαS SSCs from 3-week-old and 6- to 9-week-old bones and performed colony formation assays. PαS SSCs showed an approximately ninefold increase in colony formation compared with freshly isolated whole bone marrow (WBM) (Fig. 2C). At 3 weeks there was a decrease in the number of colonies generated by KS^+/− PαS SSCs compared with S^+/− or S^−/− cells, whereas the number of colonies was increased in KS^−/− cells (Fig. 2D,E). The colonies were classified into groups based on size (small, <5 mm²; medium, 5 mm²<x<36 mm²; large, >36 mm²) to examine the expansion capacity of control and mutant SSCs. KS^+/− SSCs generated decreased colony numbers of all three sizes, and KS^−/− increased medium size colonies (Fig. 2E). Also, 6- to 9-week-old KS^−/− SSCs increased the number of colonies formed compared with controls (Fig. 2F,G). Western blotting confirmed constitutive activation of PDGFRβ and knockout of STAT1 in SSC-derived cells (Fig. 2H). These results show that mice with PDGFRβ^D849V exhibit changes in the number of PαS SSCs and colony forming unit activity in parallel to the wasting and overgrowth phenotypes displayed in vivo, consistent with the idea of intrinsic defects in mutant SSCs mediating skeletal phenotypes.

PαS-derived cells can differentiate into osteoblasts or adipocytes when treated with differentiation cocktails (Morikawa et al., 2009). However, we noted that our colonies were only partially differentiated (Fig. S1B,C), suggesting cellular heterogeneity within each colony. As an approach to evaluate cellular heterogeneity in SSC-derived polyclonal colonies, we performed single-cell RNA-sequencing (scRNA-seq). In principle, freshly isolated PαS SSCs would be attractive for scRNA-seq, but the KS^+/− mutant mice suffer from systemic autoinflammation that inhibits colony formation (Fig. 2D,E) and would strongly influence gene expression. To escape this condition, we used SSC-derived colonies that had been cultured for 14 days in maintenance medium instead of freshly isolated cells. Single-cell suspensions of colony-forming units-fibroblasts (CFU-Fs) from the four genotypes were subjected to scRNA-seq using 10x Genomics Chromium platform. We integrated scRNA-seq data from eight samples, representing the four genotypes in duplicate, to generate clusters using the Seurat package (Butler et al., 2018; Stuart et al., 2019). Data integration with canonical correlation analysis (CCA) removed batch effects (Fig. S2A). Each cluster was grouped based on signature genes that were differentially expressed between clusters (Fig. 3A). A heatmap of signature gene expression is shown in Fig. 3B (full list in Table S1). These markers, combined with current literature and gene ontology, were used to suggest cell types or cell fates represented by each cluster (Fig. 3C,D; Table S2). Clusters of biological duplicates were distributed similarly within each genotype (Fig. S2B). As shown by uniform manifold approximation and projection (UMAP) plot (Fig. 3A), we found ten clusters tentatively containing SSCs, intermediate skeletal stem and progenitor cells (SSPCs), chondrocyte precursors, osteoblast precursors or adipocyte precursors. All ten clusters were conserved across the four genotypes (Fig. S2C). These cell-type designations may represent potential cell fates rather than fully differentiated cell types. Our interpretations are based on current biological and gene ontology knowledge, but have not been functionally validated to determine their characteristics in vivo.

As a percentage of all cells, SSCs and SSPCs were more abundant than precursors (Fig. 3E). The most abundant was cluster 0 (26.6%, SSPCs with Pdgfra^high), followed by cluster 1 (21.9%, SSPCs with Sca1^high) and cluster 2 (11.1%, SSCs with Pdgfra^high and Sca1^high). Given the initial seeding of PαS SSCs, cluster 2 should be related to the origin of other clusters owing to high expression of Pdgfra and Sca1. The mesoderm transcription factor Prrx1 was most highly expressed in cluster 2 and was broadly expressed in other clusters, as expected for SSCs and their progeny (Fig. S2D). Pdgfrb was moderately expressed in most clusters including SSCs (cluster 2), but was downregulated in chondrocyte precursors (clusters 6, 8 and 9) (Fig. 3C). The SSC and SSPC clusters broadly expressed several previously identified stem cell markers including Cxcl12, Ctsk, Cd164, Itgav, podoplanin (Pdpn) and Grem1, but others were barely detected including Nes, Cd146 and Lepr (Fig. S2D). Previously, Nes-CreER⁺ BM cells were not enriched in CFU-Fs, suggesting that Nes expression should not be detectable in CFU-F (Zhou et al., 2014). Similarly, CD146 was found not to be an SSC marker on murine BM stromal cells (Chou et al., 2012). As mouse BM stromal cells begin to express Lepr after 3 weeks of age (Zhou et al., 2014), the absence of Lepr in our data is probably due to collecting PαS SSCs from 2- to 3-week-old mice.

The remaining clusters, 3 through 9, could be precursors rather than differentiated cells, because the cells were cultured under conditions to support clonal expansion without differentiation. Representative markers for each precursor cluster are summarized in Fig. S2E. Provisionally, cluster 3 represents osteogenic precursors expressing metalloproteinase 13 (Mmp13), alkaline phosphatase (Alpl) and osterix (Sp7). Cluster 4 represents adipocyte precursors expressing lipoprotein lipase (Lpl), fatty acid-binding protein 4 (Fabp4), haptoglobin (Hp) and adiponectin (Adipoq). Cluster 5 represents osteoblastic precursors highly expressing collagens Col1a1 and Col1a2 and moderately expressing actin alpha 2 (Acta2), transgelin (Tagln) and myosin light chain 9 (Myl9). Clusters 6 and 8 represent actively proliferating chondrocyte precursors highly expressing cell cycle genes (Cdk1, Mki67 and Pcna) and early chondrogenic markers (Col2a1 and Grem1). As we regressed out cell cycle genes (S and G2/M phase genes) during dimensional reduction of scRNA data, many S phase cells were distributed throughout all the clusters (Fig. S2F). However, clusters 6 and 8 remained prominent for cell cycle, chromosome and mitosis gene signatures (Table S2). Cluster 7 represents chondrogenic precursors with expression of semaphorin 3c (Sema3c) and proline arginine-rich end leucine-rich repeat protein (Prelp). Cluster 9 represents chondrocyte precursors with expression of aggrecan (Acan), unique cartilage matrix-associated protein (Ucma), collagens Col9a1 and Col2a1, and the crucial chondrogenic transcription factor Sox9 (Akiyama et al., 2002). Taken together, scRNA transcriptomics reveals the heterogeneity and potential cell fate transition of PαS SSC-derived cells.

To identify changes in cell clusters and gene expression due to PDGFRβ^D849V, we compared scRNA data of four genotypes representing S^+/−, S^−/−, KS^+/− and KS^−/−. Pdgfrb mRNA was moderately decreased in KS^+/− and KS^−/− genotypes, as shown previously in dermal fibroblasts (He et al., 2017), and Stat1 mRNA was absent from S^−/− and KS^−/− (Fig. 4A). We compared the proportion of each cluster within the total population of each genotype and found subtle shifts in osteogenic and chondrogenic precursors. The proportional ratio of osteogenic clusters (clusters 3 and 5) was decreased in both mutants KS^+/− and KS^−/− compared with S^+/− and S^−/− controls (Fig. 4B,C; Table S6). This suggests that PDGFRβ^D849V signaling could suppress osteogenesis independently of Stat1 status. Meanwhile, proliferating chondrocytes (clusters 6 and 8) and Acan^high-chondrogenic precursors (cluster 9) were increased in KS^+/− compared with all other genotypes (Fig. 4B,C). KS^−/− only slightly increased the proliferating chondrogenic precursors (cluster 8), but less than KS^+/−. This suggests PDGFRβ^D849V-mediated chondrogenic defects, potentially modulated by Stat1 status.

To identify gene expression changes in osteogenic clusters, we analyzed scRNA data with the Database for Annotation, Visualization and Integrated Discovery (DAVID) analysis (Huang et al., 2009) (Tables S3,S4). We found that KS^+/− and KS^−/− cells downregulated osteogenesis-related genes including Mmp13, Col1a1, Col1a2, pentraxin 3 (Ptx3), plasminogen activator inhibitor-1 (Serpine1) and pigment epithelium-derived factor (Serpinf1) (Fig. 4D). Their expression levels were decreased in almost all KS^+/− and KS^−/− clusters, suggesting a broad impact across skeletal lineages (Fig. 4E). To examine osteogenic differentiation in vitro, we isolated heterogeneous skeletal cells from 3-week-old compact bone (see Materials and Methods). These cells include SSCs, SSPCs and progenitors with osteoblast potential. After culturing them with a standard cocktail of osteoblast differentiation inducers, KS^+/− and KS^−/− cells showed significantly reduced alkaline phosphatase (ALP) staining, consistent with defective osteoblast differentiation (Fig. 4F), and significantly reduced Alizarin Red (AR) staining, suggesting defective mineralization (Fig. 4G).

Next, to examine bone mass and mineralization in mice, we used micro computed tomography (microCT) to examine 3-week-old tibias from the four genotypes. KS^+/− tibias decreased bone volume to total volume (BV/TV) and decreased bone mineral density (mg HA/ccm) in the cortical bone of the diaphysis (Fig. 5A) and the proximal tibial metaphysis (Fig. 5B) compared with S^+/− and S^−/− tibias. KS^+/− microCT scans also displayed osteoporosis-like pores in the cortical bone (Fig. 5A). KS^+/− tibias decreased trabecular thickness, but not trabecular number and separation (Fig. 5B). In contrast, KS^−/− tibias at 3 weeks had no pores and partially normalized BV/TV and bone mineral density in both cortical and trabecular regions (Fig. 5A,B). There were no changes in trabecular number, thickness and separation in KS^−/− compared with S^+/− and S^−/−. However, 5- to 8-week-old KS^−/− tibias displayed reduced BV/TV and bone mineral density in the cortical bone of the diaphysis (Fig. 5C) and the proximal tibial metaphysis (Fig. 5D) compared with age-matched S^+/− and S^−/− tibias. Trabecular numbers and thickness were decreased, but separation was increased in KS^−/−. We performed Calcein double staining to quantify mineral appositional growth rate (MAR). Histomorphometric analysis of cortical bone at 3 weeks showed decreased bone growth in KS^+/−, which was normalized in KS^−/− (Fig. 5E). Interestingly, 6-week-old KS^−/− cortical bone showed increased MAR compared with controls (Fig. 5F), but the Calcein labeling was thicker and more diffuse, which is consistent with incomplete or delayed mineralization. In summary, as suggested by transcriptomics, we found that PDGFRβ^D849V impairs osteogenesis and leads to osteopenia that becomes severe early in KS^+/− mice (by 3 weeks) and later in KS^−/− mice (by 6 weeks).

The growth and survival phenotypes (Fig. 1B), and the time required to develop osteopenic phenotypes (Fig. 5), are very different between KS^+/− and KS^−/− mice. Furthermore, in SSC-derived polyclonal colonies, KS^+/− prominently increased chondrogenic clusters (clusters 1, 6, 8 and 9) (Fig. 4B,C) including Sox9^high chondrogenic precursors (cluster 9) compared with others (Fig. 6A). We further analyzed differentially expressed genes (DEGs) specific to the KS^+/− genotype (Tables S3,S4). Among 34 genes specifically upregulated in KS^+/− versus the other three genotypes, cartilage development genes were highly enriched, including Sox9, Col2a1, H19, imprinted maternally expressed transcript (H19) and Acan (Fig. 6B). Increased Sox9, Col2a1 and H19 expression was not limited to cluster 9, but also showed increased expression in other clusters (Fig. 6C; Fig. S3A). Col2a1 and H19 are known downstream targets of Sox9, which directly regulates collagen 2 production (Bell et al., 1997). Sox9 also indirectly promotes collagen 2 expression via a long non-coding RNA H19 and its micro-RNA, miR675 (Dudek et al., 2010). This suggests that increased Sox9 in KS^+/− SSCs promotes chondrogenic proliferation and commitment. To evaluate chondrogenesis in vivo, we analyzed Safranin-O-stained tibias at 3 weeks old. KS^+/− exhibited greater cartilage area in the tibia head, although KS^−/− also showed mildly increased cartilage area (Fig. 6D). Note that KS^+/− tibias were significantly smaller in size (Fig. 1C), and therefore the proportional area of KS^+/− cartilage is much larger than the three other genotypes. We evaluated Sox9 expression in vivo in femurs from 3-week-old mice. Sox9⁺ cell numbers were increased in both KS^+/− and KS^−/− genotypes, each with a distinct distribution in the distal femur (Fig. 6E). KS^+/− showed ectopic expansion of Sox9⁺ cells above the growth plate, whereas KS^−/− showed increased Sox9⁺ chondrocytes in the proliferating and prehypertrophic zones within the growth plate (Fig. 6E). To examine chondrogenic differentiation in vitro, we isolated heterogeneous skeletal cells from 3-week-old compact bone, which includes progenitors with chondrogenic potential. We found that KS^+/− cells dramatically increased pellet size compared with the others after 21 days in culture (Fig. 6F). Interestingly, although many KS^+/− cells appeared to have matured into chondrocytes in matrices that were strongly positive for Toluidine Blue and Safranin-O, cells in the pellet core appeared to be undifferentiated (Fig. 6F). KS^+/− cells often generated irregular donut-shaped or rod-shaped pellets, unlike spherical pellets formed by other genotypes. KS^−/− cells generated chondrocyte pellets similar to controls. We further investigated transcriptional changes in KS^+/− pellets using in situ hybridization. Upregulated H19 and Acan mRNA expression was evident in KS^+/− pellets (Fig. S3B), consistent with scRNA data (Fig. 6B,C). Col10a1 (collagen, type X, alpha 1) expression was also increased in KS^+/− pellets (Fig. S3B). TUNEL-positive area was not increased in the majority of KS^+/− pellets (Fig. S3C). Therefore, increased KS^+/− chondrogenesis in vitro was not linked to apoptosis. Together, these findings suggest that PDGFRβ^D849V and STAT1 upregulate Sox9 and enhance chondrogenesis in vitro and in vivo.

As KS^+/− mice showed wasting due to PDGFRβ-STAT1-mediated auto-inflammation, and KS^+/− dermal fibroblasts intrinsically overexpress interferon-stimulated genes (ISGs) (He et al., 2017), we examined whether KS^+/− cells from bone also exhibit an ISG signature. In KS^+/− heterogeneous skeletal cells, a panel of ISGs was highly overexpressed (Fig. S4A). In KS^+/− SSC-derived cells, the ISGs Ifi27, Bst2, Isg15, Ifit1, Psmb8, Mx1 and Stat1 were upregulated throughout all clusters (Fig. S4B-D). However, in SSC-derived cells the expression of ISGs was very low and was detected in few cells regardless of genotype. It has been shown that other types of stem cells are intrinsically protected from interferon responses (Burke et al., 1978; Eggenberger et al., 2019), and this may be true for SSCs.

PDGF signaling in osteolineage cells promotes osteoclast activity through the expression of colony stimulating factor 1 (Csf1) and receptor activator of nuclear factor kappa-B ligand (Rankl; Tnfsf11) (Brun et al., 2020; Chen et al., 2015). From our mutant scRNA-seq data (Table S3), we found changes in several genes that are linked to osteoclastogenesis (Table S10), although we did not observe increased Csf1 or Rankl. Instead, positive regulators of osteoclasts such as Igfbp5 (in KS^+/−) and Igf1 (in both KS^+/− and KS^−/−) were upregulated. Negative regulators were downregulated including clusterin (Clu), cystatin C (Cst3), Cxcl15 and fibromodulin (Fmod) in KS^+/−, and Serpinf1 in both KS^+/− and KS^−/−. These changes suggest enhanced osteoclastogenesis in the context of PDGFRβ^D849V. To evaluate osteoclasts in vivo, we performed tartrate-resistant acid phosphate (TRAP) staining. The TRAP-positive area covering trabecular bone surface approximately doubled in both KS^+/− and KS^−/− mutant femurs at 3 weeks of age (Fig. S5A). These findings suggest that PDGFRβ^D849V promotes osteoclast activity through multiple mechanisms. PDGFRβ^D849V-mediated osteoclastogenesis may be also dependent on RANKL, as many of the differentially expressed osteoclast regulators work through RANKL (Kram et al., 2017; Strålberg et al., 2013; Wang et al., 2006) (Table S10).

Although both KS^+/− and KS^−/− mutants showed defective mineralization, only KS^−/− showed overgrowth in vivo. From scRNA-seq, we found that Igf1 and Socs2 were highly upregulated in KS^−/− SSC-derived cells and were upregulated to a lesser amount in KS^+/− cells, compared with controls (Fig. 7A). Increased Igf1 and Socs2 mRNAs were detected in most KS^+/− and KS^−/− clusters, from SSCs to precursors (Fig. 7B). Igf1 and Socs2 are direct transcriptional targets of STAT5. The STAT5-IGF1 axis is crucial for the biological effects of growth hormone receptor signaling (Chia et al., 2006; Udy et al., 1997), and SOCS2 is involved in negative feedback on STAT5 (Greenhalgh et al., 2002). It is known that PDGFRβ can phosphorylate STAT5 (Valgeirsdóttir et al., 1998), although the biological significance is unclear. Interestingly, Stat5a mRNA was overexpressed by KS^+/− SSC-derived cells and Stat5b was overexpressed by KS^−/− cells (Fig. 7C). However, the role of STAT5 in signaling and gene activation is mainly regulated by phosphorylation. Both KS^+/− and KS^−/− SSC-derived cells exhibited constitutive STAT5 phosphorylation (Fig. 7D), and STAT5 was also constitutively phosphorylated in heterogeneous skeletal cells from both mutants (Fig. 7E). The antibodies used for these western blots detect both STAT5 proteins and therefore cannot discern whether there is differential activation of the two isoforms in KS^+/− and KS^−/− cells. Based on these results, we suggest that the STAT5-IGF1 axis could be a key signaling pathway for PDGFRβ-mediated overgrowth. Serum IGF1 was unchanged between 3-week-old controls and KS^−/− mice, which were not yet overgrown (Fig. 7F), but was very low in 3-week-old KS^+/− mice, which were moribund (Fig. 7F). IGF1 levels were, however, elevated in culture medium of KS^+/− and KS^−/− skeletal cells (Fig. 7G). Therefore, PDGFRβ mutant skeletal lineages may promote skeletal overgrowth by increasing IGF1 levels locally, rather than systemically. We suggest that both KS^+/− and KS^−/− mice have potential for overgrowth due to increased signaling through the STAT5-IGF1 axis. But KS^+/− mice do not exhibit the overgrowth phenotype because of growth-suppressive (and lethal) effects of Stat1-mediated autoinflammation.

Recent characterization of SSC heterogeneity and elucidation of SSC niches has greatly expanded our understanding of skeletal biology. Depletion of SSCs by conditional diphtheria toxin (DT) expression in mice has been considered one of the essential in vivo approaches to demonstrate the importance of SSCs in skeletal growth. For example, depletion of PTHrP⁺ cells, Cxcl12-abundant reticular (CAR) cells, Grem1⁺ cells, or LepR⁺ cells using DT has successfully demonstrated the importance of these diverse stem cell types in chondrocyte, osteoblast and adipocyte maintenance (Mizuhashi et al., 2018; Omatsu et al., 2010; Worthley et al., 2015; Zhou et al., 2014). However, there is still much to learn in terms of signaling pathways and genes regulating SSC functions. In this study, we found that increased PDGFRβ signaling alters PαS SSC abundance and the fate of SSC-derived cells in parallel to phenotypes affecting the skeleton. Our results suggest that constitutive PDGFRβ signaling in SSCs precedes alterations in osteogenic and chondrogenic differentiation. These cell-autonomous and skeletal lineage-autonomous defects, combined with systemic effects of constitutive PDGFRβ signaling (including interferonopathy-like autoinflammation), lead to osteopenia and skeletal phenotypes that are reminiscent of humans with PDGFRB mutations.

Colony formation assays have historically been used to assay for the enrichment of cells with stem cell properties, but it has been challenging to identify molecular targets or cellular pathways owing to the rarity of SSCs. In our case, it was important to isolate KS^+/− SSCs and culture them to isolate the intrinsic gene expression changes from the effects of severe autoinflammation in vivo. However, it is known that marker expression and biosynthetic outputs change with time in culture. Furthermore, as illustrated by our data, sorted population of stem cells becomes heterogeneous in a short time. Nevertheless, with scRNA transcriptomics of cultured SSC-derived polyclonal colonies, we were able to tentatively designate SSCs and progenitor cell types and then identify molecular alterations that were linked to phenotypes in KS^+/− and KS^−/− mutant mice. Thus, our work suggests the utility of using scRNA-seq even with heterogeneous cultured cells to provide new molecular understanding of diseases. Our results could potentially be refined even further by characterizing individual SSC clones through in vivo transplantation (Chan et al., 2015).

We found that wasting KS^+/− mutants and overgrown KS^−/− mutants both exhibit impaired osteogenesis in vivo. SSC-derived cells from both mutants decreased osteogenesis-related genes such as Col1a1, Col1a2 and Serpinf1, and skeletal cells from both mutants showed reduced osteogenic differentiation and mineralization capacities. Human loss-of-function mutations in COL1A1 or COL1A2 (MIM 166200) or SERPINF1 (MIM 172860) result in osteogenesis imperfecta, with decreased mineralization and brittle bones (Byers and Pyott, 2012). On the other hand, DEGs between KS^+/− and KS^−/− mutants suggest that Stat1-dependent mechanisms are also involved. We highlighted increased Sox9 expression in KS^+/− mice and SSC-derived cells as an intrinsic source of osteogenesis defects because Sox9 overexpression in mice leads to impaired osteogenesis and dwarfism (Akiyama et al., 2004; Zhou et al., 2006). KS^+/− osteogenesis defects in vivo are likely compounded by autoinflammation (Redlich and Smolen, 2012). The kidneys are crucial for maintaining healthy bones by regulating phosphorous and calcium levels in the blood. A PDGFRβ gain-of-function mutation controlled by Foxd1Cre^tg has been shown to cause mesangioproliferative glomerulonephritis and renal fibrosis with progressive anemia in mice (Buhl et al., 2020). However, as renal phenotypes only began to appear at 6 weeks of age, renal dysfunction is unlikely to initiate skeletal phenotypes in our mice, but it may contribute to later progression in KS^−/−.

Our study suggests that Pdgfrb^D849V signaling through STAT1 promotes Sox9 expression and chondrocyte proliferation. We found that KS^+/− skeletal cells formed irregularly shaped chondrocyte pellets. The non-spherical morphology may occur because of excessive cartilage matrix. For example, BM stromal cells obtained from familial osteochondritis dissecans (FOCD) patients (ACAN loss-of-function mutant) produce enlarged and irregularly shaped chondrocyte pellets with highly upregulated cartilage matrix proteins including collagen 11 and proteoglycans (Xu et al., 2016). We found that Acan, Col2a1 and Col11a1 were all upregulated in KS^+/− colonies. We do not know how Pdgfrb^D849V-STAT1 regulates Sox9, but STAT3 has been shown to directly regulate Sox9 in chondrogenesis (Hall et al., 2017) and STAT1 can heterodimerize with STAT3.

Pdgfrb^D849V increases functional SSC numbers and SSC-derived cells overexpress growth signaling genes (i.e. Igf1 and Socs2). STAT5 may be an important mediator of overgrowth because Igf1 is a direct transcriptional target of STAT5, which is directly activated by PDGFRβ (Chia et al., 2006; Valgeirsdóttir et al., 1998). STAT5 is constitutively phosphorylated in KS^+/− and KS^−/− cells, but only KS^−/− induces overgrowth. As discussed above, there are additional effectors in KS^+/− to counterbalance potential overgrowth, such as STAT1-dependent autoinflammation and Sox9 expression. Furthermore, KS^+/− cells upregulated the expression of IGF binding proteins (IGFBP3, 6 and 7) (Tables S3,S4), potentially blocking IGF1 signals. Increased IGF1 in KS^−/− mice appears to occur locally in skeletal cells rather than systemically, because increased IGF1 was detected in conditioned medium from mutant cells but not in mutant mouse circulation. We do not know whether IGF1 causes overgrowth through autocrine signaling or by signaling between cell types. Tissue-specific genetic studies will be needed to investigate the involvement of a PDGFRβ-STAT5-IGF1 axis and to identify IGF1-responsive cells.

Although the D849V mutation in our study is not the same as mutations seen in Penttinen syndrome and Kosaki overgrowth syndrome, we believe our findings here are closely related to the human conditions. Osteopenia, often associated with bone fractures, is a common feature in both human diseases. For example, two Kosaki overgrowth syndrome patients developed fractured tibias and compression fractures in the spine, causing deformation of their bones (Foster et al., 2020; Minatogawa et al., 2017). Further, a Penttinen Syndrome patient exhibited osteoporosis with multiple fractures (Johnston et al., 2015). Similarly, we found osteopenia in Pdgfrb^D849V mice regardless of phenotype on the wasting-overgrowth spectrum. A better understanding of the pathogenic mechanisms of different activating PDGFRβ mutations will be obtained through future genetic models that reproduce the specific human PDGFRB mutations seen in Penttinen and Kosaki syndromes. The current work puts forth SSCs as a conceptual tool for considering pathological changes in the PDGFRβ mutant skeleton as resulting from altered stem cell functions.

Mouse strains Pdgfrb^D849V (#018435), Sox2-Cre^tg (#008454) and Stat1^flox (#012901) are available at the Jackson Laboratory. All procedures performed on mice were approved by the Institutional Animal Care and Use Committee of the Oklahoma Medical Research Foundation, USA. Mice were maintained on a mixed C57BL/6;129 genetic background with a standard mouse chow diet (5053, Purina). Mutant KS^+/− and KS^−/− mice were compared with age and sex-matched littermate control S^+/− and S^−/− mice.

PαS cells were isolated by the method of Houlihan et al. (2012) with some modifications. FACS buffer was made of 1× HBSS (Gibco), 2% fetal bovine serum (FBS), 1× penicillin/streptomycin (Gibco), 1 mM ethylenediaminetetraacetic acid (EDTA, VWR) in autoclaved H₂O. Limbs of control and mutant mice at 3, 6 and 9 weeks old were dissected and soaked in 70% ethanol for 2 min, followed by careful removal of adherent muscles. BM was flushed 2-3 times with sterile PBS using a 26 G×1/2 needle and 1 ml syringe (BD Biosciences). BM-free bones were cut into small pieces with sterile scissors until bones became paste (usually 3 min for 3-week-old and 5 min for 6- to 9-week-old mice). The bone paste from each mouse was transferred to a 15 ml conical tube and enzymatically digested with 15 ml 0.2% type II collagenase (Worthington) in DMEM (Corning) for 1 h with agitation at 37°C. Digested bone paste was filtered through a 70 µm cell strainer and collected in a 50 ml conical tube on ice. Bone paste remaining on the filter was collected with 2.5 ml FACS buffer, gently tapped in a mortar with a pestle and filtered into the same 50 ml conical tube on ice. This was repeated until the total volume of filtrate reached 50 ml. After centrifugation at 280 g for 10 min at 4°C, 1 ml ice-cold sterile H₂O was used to remove red blood cells for 6 s. Then 1 ml 4% FBS in PBS and 13 ml FACS buffer were added, followed by filtering with 70 µm cell strainer into a new 50 ml tube. After centrifugation at 280 g for 5 min, re-suspended cells were incubated with fluorophore-conjugated antibodies (Table S9) on ice for 20 min in the dark. PαS cells were sorted using MoFlo XDP Cell Sorter (Beckman Coulter) or FACSAria III (BD Biosciences) by negative gating with Zombie Green™ (live/dead cell dye), CD31 (PECAM1), CD45 (PTPRC), TER119 (Ly76) and positive gating with PDGFRα and Sca1 (Fig. S1A). For colony formation assays, 2000 PαS cells or freshly isolated WBM cells were cultured in a 6-well plate with 20% FBS (mesenchymal stem cell-qualified, Gibco), 25 units/ml penicillin/streptomycin and 2 mM L-glutamine in alpha MEM (Gibco) for 2 weeks under hypoxia (5% oxygen) at 37°C. Colonies were fixed with 4% paraformaldehyde (PFA) for 15 min and stained with 0.5% Crystal Violet for quantification.

Single-cell suspensions of PαS SSC-derived colonies were harvested from CFU-F assays after culturing for 2 weeks as explained above. Trypsinized cells were filtered through a 40 µm cell strainer. A total of eight samples as four batches were generated in duplicate with one male and one female of each genotype. Each time, two samples (maximum 20,000 cells per each sample) were loaded for the 10x Genomics Chromium platform, barcoded and sequenced on the Illumina NovaSeq SP flowcell platform at a depth of 400 million reads per sample. See Table S5 for batch information and actual cell numbers sequenced and analyzed. Using the 10x Genomics Cellranger software (version 3), sequences were demultiplexed to extract cellular barcodes, cDNA inserts and unique molecular identifiers (UMIs) and the cDNA sequences were aligned to the murine mm10 genome reference to count UMIs. Quality control, data integration, clustering and gene expression analysis were performed with Seurat package V3 for R (Butler et al., 2018; Stuart et al., 2019). Cells with >8000 genes, <1500 genes or >8% of genes mapping to mitochondrial genes were removed as poor quality or doublets. Eight samples were individually normalized by using the log transformation method in Seurat. Data integration of eight samples to remove batch effect was carried out with CCA, with the top 2000 variable genes identified by FindVariableFeatures function in Seurat (Butler et al., 2018). Cells were categorized into S, G1 or G2/M phase by scoring cell cycle-associated gene expression (Kowalczyk et al., 2015). Principal component analysis (PCA) was performed by regressing out cell cycle, mitochondrial and ribosomal gene expression. The top 100 PCs were selected to perform dimensional reduction using UMAP. We determined a cluster resolution of 0.4 (total 12 clusters) using Clustree R package (graphic-based cluster resolution analyzer) (Zappia and Oshlack, 2018). Among the 12 clusters, ten were selected for analysis, because two of the clusters only included about ten cells each. A heatmap was generated with the top ten genes (Table S1) in each cluster determined by FindAllMarkers function in Seurat. For gene signature analysis, DEGs of each cluster were determined by FindMarkers function in Seurat with 0.25 log-scale fold-change threshold between two groups (i.e. genotypes). DEGs were further analyzed using DAVID to determine gene ontology and pathway identification (Huang et al., 2009). Cell types of clusters were tentatively defined with a signature gene list, gene ontology and literature as SSCs, SSPCs, chondrocyte precursors, osteoblast precursors and adipocyte precursors. To identify genotype-specific genes and clusters, we split the integrated data into four groups (individually S^+/−, S^−/−, KS^+/− and KS^−/−). DEGs between four groups were identified using the FindMarkers function in Seurat. The scRNA data were deposited as GSE183404 in the Gene Expression Omnibus of the National Center for Biotechnology Information.

Primary skeletal cells were isolated by enzymatic digestion of compact bone (Zhu et al., 2010). BM-free bones were collected as described above and were cut into small pieces (1-3 mm³) with sterile scissors. The bone chips were transferred into a 1.6 ml microcentrifuge tube and enzymatically digested with 1.5 ml of 1 mg/ml type II collagenase (Worthington) in alpha MEM (Corning) plus 10% FBS for 1.5 h with agitation in a 37°C incubator. The bone chips were washed with 1 ml of alpha MEM three times, seeded into a 6-well plate and maintained with alpha MEM plus 10% FBS, 25 units/ml penicillin/streptomycin and 2 mM L-glutamine at 37°C. Passages 4 to 6 were used for tri-lineage differentiation assays. For osteoblast differentiation, 2×10⁵ cells were seeded in a 24-well plate. On the following day, they were treated with alpha MEM supplemented with 10% FBS, 10⁻⁷ M dexamethasone, 10 mM β-glycerol-phosphate and 50 mM ascorbate-2-phosphate. Medium was changed three times per week for 2 weeks (early osteoblast differentiation). The samples were subjected to ALP staining with ALP buffer [100 mM Tris-HCl, 100 mM NaCl, 5 mM MgCl₂, 0.05% Tween-20 in deionized water (adjusted at pH 9.5)] plus 0.02% 5-bromo-4-chloro-3-indolyl phosphate (BCIP) and 0.03% nitro blue tetrazolium (NBT). For mineralization, 1×10⁵ cells were seeded in a 96-well plate and treated with the osteoblast differentiation medium mentioned above. After 4 weeks, samples were stained with 40 mM Alizarin Red. For chondrogenesis, 1×10⁶ cells were pelleted in a 15 ml conical tube at 280 g for 6 min at 4°C and cultured for 3 weeks with the MesenCult™-ACF chondrogenic differentiation kit (Stemcell Technologies). Pellets were fixed with 4% PFA for 1 h, embedded in paraffin, sectioned at 8 µm and stained with 0.05% Toluidine Blue or 1% Safranin-O.

For osteogenesis using PαS-derived colonies, colonies cultured for 2 weeks under hypoxia were further treated with the osteogenesis inducers as described above for an additional 2 weeks under normoxia and stained with ALP (Fast Green as a counterstain). For adipogenesis, 2-week-cultured colonies were treated with alpha MEM supplemented with 10% FBS, 10⁻⁶ M dexamethasone, 250 µM IBMX, 10 µg/ml insulin and 5 nM rosiglitazone for 2 days. Colonies were maintained with alpha MEM plus 10% FBS and 10 ng/ml insulin for an additional 8 days by changing three times in a week. Adipocytes were stained with Oil-O-Red stain (Fast Green as a counterstain).

IGF1 levels were measured in serum and skeletal cell culture supernatant using ELISA (Sigma-Aldrich, RAB-0229) according to the manufacturer's protocol. Blood was collected from four genotypes at 3 weeks old, incubated at room temperature for 30 min, and centrifuged at 1000 rpm (94 g) for 10 min at 4°C. Clear supernatant was stored at −20°C until used. Primary skeletal cells (passages 4-5) isolated from compact bone of four genotypes at 3 weeks were cultured in a 24-well plate until confluent. After washing three times with sterile PBS, 500 µl serum-free fresh alpha MEM was replaced and cultured for 72 h. Cell culture supernatant was filtered through a 20 µm cell-strainer and stored at −20°C until used. Both serum and cell culture supernatant were diluted to 1:100 for ELISA.

For PαS SSC-derived cells, plates of colonies were starved in medium containing 0.2% FBS for 24 h. For skeletal cells, cells were starved in medium containing 0.1% FBS for 24 h and treated with 10 ng/ml PDGF-BB (R&D Systems) for 10-15 min. Cells were then lysed in RIPA buffer [50 mM Tris (pH 7.4), 1% NP-40, 0.25% sodium deoxycholate, 150 mM NaCl, 0.1% sodium dodecyl sulfate] with 1 mM NaF, Na₃VO₄, PMSF and 1× protease inhibitor cocktail (Complete, Roche). Pierce BCA assay was used to determine protein concentration. Aliquots of 5-10 µg of protein were separated by SDS-PAGE (using parallel gels for proteins of similar size), transferred to nitrocellulose membranes, blocked with 5% BSA and incubated with primary antibodies (Table S9) overnight at 4°C. Membranes were probed with horseradish peroxidase (HRP)-conjugated secondary antibodies at 1:5000 (Jackson ImmunoResearch) in 5% milk. Pierce ECL Western blotting substrate (Thermo Fisher Scientific) and autoradiography film (Santa Cruz Biotechnology) were used to develop blots.

Total RNA was isolated from cultured primary skeletal cells enriched from compact bones using Trizol (Thermo Fisher Scientific). cDNA reverse transcription was performed with Superscript III RT (Invitrogen) and random primers. Quantitative PCR was performed using a Bio-Rad iCycler with iQ TM SYBR green master mix and designated primers (Table S8).

Tissue was fixed overnight in 10% neutral buffered formalin (NBF; Sigma-Aldrich) or 4% PFA (Electron Microscopy). Bones were decalcified in 0.5 M EDTA solution (pH 7.4) for 2 weeks at room temperature. For cartilage staining, fixed tissues were embedded in paraffin and sectioned at 6 µm thickness followed by Hematoxylin and Eosin (H&E) or Safranin-O staining (Fast Green and Hematoxylin as counterstain). For TRAP staining, fixed tissues were decalcified in 0.5 M EDTA solution at 4°C, dehydrated in 20% sucrose in PBS and cryosectioned at 8-10 µm thickness followed by Naphthol AS-BI phosphoric acid and diazotized Fast Garnet GBC stains (Fast Green as counterstain) (Sigma-Aldrich, 387A). For immunostaining, PFA-fixed tissues were cryosectioned at 8-12 µm. Slides were stained with primary antibody (Table S9) and conjugated secondary antibody (Jackson ImmunoResearch), with DAPI (Sigma-Aldrich) as a nuclear stain. For TUNEL staining, pellets were cryosectioned at 6 µm, permeabilized with 0.1% Triton X-100 and 0.1% sodium citrate for 2 min on ice, incubated with 50 µl TUNEL reaction mixture (5 µl enzyme solution and 45 µl label solution) (Roche) for 1 h at 37°C, and stained with DAPI. Imaging was performed using a Nikon Eclipse 80i microscope with a digital camera.

Probes were generated from template sequences (Table S7). Chondrocyte pellets were fixed with 4% PFA for 1-3 h, dehydrated in 20% sucrose in PBS and cryosectioned at 6 µm thickness. Antisense digoxigenin (Dig)-labeled RNA probes were synthesized using Dig RNA labeling mix (Roche) from cDNA template sequences. Hybridized probes were detected with anti-Dig Fab fragments (Roche) and developed with NBT/BCIP.

To measure the MAR of bone formation, double Calcein stain was performed. Calcein (10 mg/kg, Sigma-Aldrich) in 0.9% NaCl and 2% NaHCO₃ was intraperitoneally injected in 3- or 6-week-old mice with a 4-day interval. Mice were harvested 2 days after the second injection. Paraffin sectioning was performed following the method of Porter et al. (2017): femurs were harvested with adherent muscles and fixed with formalin for 2 days, then cleared with 10% KOH for 96 h. The tissues were embedded in paraffin and sectioned at 6 µm thickness, pre-soaking the paraffin blocks in 1% KOH immediately before sectioning. Imaging was performed using a FITC filter with a Nikon Eclipse 80i microscope with a digital camera. To get clear fluorochrome images with minimal background autofluorescence, we adjusted ‘Shadows’ in the levels histogram in the software (NIS Elements D 3.2) using the same setting across samples. For microCT, tibias were harvested and stored in 70% ethanol at 4°C, then analyzed to measure bone volume (mm³), total volume (mm³), mean/density (mg HA/ccm), trabecular numbers (1/mm), trabecular thickness (mm) and trabecular separation (mm) with a VivaCT 40 microCT (Scanco Medical) at an X-ray tube voltage of 70 kVp and an X-ray current of 114 µA for 2- to 3-week-old bones, or 55 kVp and 85 µA for older bones.

The size of CFU-F and chondrocyte pellets, the distance between double Calcein stain, and coverage area for TRAP, ALP and AR were quantified with ImageJ software (National Institutes of Health). The Set Scale function was used to convert units from pixels to mm or µm with known scale of images. Regions of interest (ROIs) were selected using the oval or polygon selection tool to measure the area of CFU-F and pellet images. The straight tool was used to measure the distance between double Calcein stain. Coverage area were calculated by 8-bit gray scale conversion and thresholding to measure area fraction within ROIs. Statistical calculations were performed using GraphPad Prism 9 with one-way or two-way ANOVA or unpaired, two-tailed t-test. Data are represented as mean±s.e.m.

We thank Jacquelyn C. Herron, Shouan Zhu, Mary Beth Humphrey, Timothy M. Griffin and members of the Olson lab for their assistance and helpful discussion. We also thank the Microscopy Core and Mouse Phenotyping Core Facilities (associated with P30-GM114731) of the Oklahoma Medical Research Foundation Centers of Biomedical Research Excellence.

The peer review history is available online at https://journals.biologists.com/dev/article-lookup/doi/10.1242/dev.199607.