Specification of osteoblast cell fate by canonical Wnt signaling requires Bmp2

Bone morphogenetic proteins (BMPs) and canonical Wnts (cWnts) play essential roles in the skeleton. BMPs, originally identified for their ability to induce de novo formation of bone and cartilage, are secreted ligands that act through cell surface complexes comprised of type I and type II BMP receptors to activate Smad1/5/8-dependent gene transcription (Salazar et al., 2016). cWnts are secreted proteins that signal through Lrp and Frizzled co-receptor complexes to stabilize intracellular pools of β-catenin and activate Tcf/Lef-dependent gene transcription (Baron and Kneissel, 2013). Genetic mutations in either of these pathways underlie a variety of severe skeletal pathologies including the high and low bone mass syndromes associated with mutations of the cWnt receptor Lrp5 (Boyden et al., 2002; Gong et al., 2001; Little et al., 2002) or fibrodysplasia ossificans progressiva, a disease characterized by inappropriate growth of endochondral bone at non-skeletal sites and caused by mutations of the type I BMP receptor ALK2 (Shore et al., 2006). For over a decade, recombinant BMPs, particularly BMP2 and BMP7, have been used in clinical orthopedic settings including non-union fractures, spinal fusions and oral surgery (Lo et al., 2012). More recently, neutralizing antibodies targeting antagonists of the cWnt pathway, such as SOST and DKK1, have entered clinical trials as systemic agents that enhance bone formation (Baron and Kneissel, 2013).

Given the abundance of information establishing clear roles for BMP and cWnt signaling in bone development and disease, and the fact that therapies based on each pathway are approved or under evaluation for use in humans, it is somewhat remarkable that the individual aspects of osteoblast specification and physiology controlled by either BMP or cWnt signaling remain largely undefined.

During development, bones of the appendicular skeleton are made by a Prx1 (Prrx1)⁺ subset of cells that arise in lateral plate mesoderm during limb bud outgrowth (Durland et al., 2008; Logan et al., 2002). Postnatally, undifferentiated cells of the Prx1 lineage persist in the marrow and periosteum of mature bone as distinct but complementary pools of Lepr⁺ or Grem1⁺ osteoprogenitors that can be activated by BMP2 for the purpose of bone growth or repair (Chan et al., 2015; Worthley et al., 2015; Zhou et al., 2014). We hypothesized that BMPs and cWnts play unique roles in specifying Prx1⁺ progenitors to the osteoblast cell fate. To test this, we utilized limb bud cells and bone marrow cells as models of Prx1⁺ cell populations to monitor osteogenic responses to cWnt activity in the presence or absence of Bmp2.

Immortalized cells from the E13.5 mouse embryonic limb bud (MLB13 cells) express endogenous Bmp2 and exhibit robust endochondral differentiation in response to recombinant BMPs (Rosen et al., 1993, 1994). We established a sub-clonal population of MLB13 (MLB13^WT) cells stably expressing siRNA against endogenous Bmp2 (MLB13^kdBmp2) (Fig. 1A). Osteoblast differentiation was induced for 48 h by supplementing the cell culture medium with ascorbic acid and β-glycerophosphate (osteogenic medium) with or without recombinant (r)BMP2 or Wnt3a. QPCR analysis revealed that levels of Prx1, Runx2 and Dlx5 mRNA were unaffected by knockdown of Bmp2, while Osx1 (Sp7) and the direct transcriptional target of Osx1, Alpl (Hojo et al., 2016), were dramatically diminished. Osx1 and Alpl mRNAs could be enhanced in MLB13^kdBmp2 cells by rBMP2 but not rWnt3. Wnt3a furthermore significantly diminished Dlx5 mRNA in MLB13^kdBmp2 but not MLB13^WT cells (Fig. 1B-F). MLB13^kdBmp2 cells treated with Wnt3a retained the ability to upregulate the TCF/Lef target gene Axin2 (Fig. 1G). MLB13^WT cells exhibited a dose-dependent increase in ALP activity in response to Wnt3a, whereas ALP activity in MLB13^kdBmp2 cells was ∼73% lower than MLB13^WT cells grown in osteogenic medium alone and remained unchanged by any concentration of Wnt3a (Fig. 1H).

We also examined Wnt-induced osteoblast differentiation of adult bone marrow stromal cells (BMSCs) lacking Bmp2 expression. BMSCs carrying a single copy of the TdTomato^{Flox-stop-Flox} reporter allele were used to optimize a protocol for in vitro recombination using Adeno-Cre. Fluorescence microcopy showed that nearly all TdTomato^{Flox-stop-Flox} cells transduced with Adeno-Cre but none of the cells transduced with Adeno-GFP were Tomato⁺ (Fig. 1I). BMSCs carrying two floxed alleles of Bmp2 (Bmp2^F/F) were transduced with Adeno-GFP (Bmp2^AdenoF/F) or Adeno-Cre (Bmp2^AdenoΔ/Δ), and osteoblast differentiation was induced with osteogenic medium with or without BMP2 or Wnt3a. After 10 days, BMSCs were stained with alkaline phosphatase (ALP) substrate to monitor induction of osteoblast differentiation and Alizarin Red dye to visualize calcification of the extracellular matrix. ALP activity and matrix calcification were comparable in Bmp2^AdenoF/F and Bmp2^AdenoΔ/Δ cells grown in control osteogenic conditions, enhanced in both genotypes of cells by rBMP2, and were greatest in both genotypes of cells treated with BMP2 and Wnt3a. By contrast, Wnt3a enhanced ALP activity and matrix calcification of Bmp2^AdenoF/F cells but not Bmp2^AdenoΔ/Δ cells, which exhibited even less osteoblast differentiation in response to Wnt3a than cells grown in osteogenic medium alone (Fig. 1J). BMSCs were analyzed on day 7 by QPCR to confirm recombination of Bmp2 (Fig. 1K) and assess osteoblast markers. Similar to results obtained in limb bud cells, mRNAs for Prx1 and Runx2 were not diminished by loss of Bmp2, while Osx1 and two Osx1 target genes, Alpl and Col1a1 (Hojo et al., 2016), were significantly reduced in Bmp2^AdenoΔ/Δ cells exposed to Wnt3a (Fig. 1L-P).

Prx1-Cre was used to conditionally delete Bmp2 in the limb bud mesenchyme at E9.5 when endochondral progenitors of the appendicular skeleton first appear during development (Durland et al., 2008; Logan et al., 2002). Bmp2 was efficiently ablated by endogenous expression of Prx1-Cre, was expressed in Bmp2^F/F BMSC and was upregulated by BMP2, Wnt3a and BMP2+Wnt3a together (Zhang et al., 2013) (Fig. 2A). Osteoblast lineage markers including Prx1, Runx2, Osx1, Alpl and Col1a1 were upregulated in Bmp2^F/F cells treated with BMP2 or Wnt3a, and most strongly expressed in Bmp2^F/F cells treated with BMP2 plus Wnt3a. These genes were also upregulated in Bmp2^Prx1Δ/Δ cells treated with BMP2 or BMP2 plus Wnt3a, to levels comparable to those in wild-type (WT) cells grown under corresponding conditions. However, with the exception of Prx1 and Runx2, Bmp2^Prx1Δ/Δ cells treated with Wnt3a did not upregulate the osteoblast differentiation markers Osx1, Alpl and Col1a1 to similar levels as observed in Bmp2^F/F cells (Fig. 2B-F). BMSCs from Bmp2^F/F or Bmp2^Prx1Δ/Δ mice produced comparable levels of calcified matrix (Fig. 2G,H) and immunoblot analysis showed they expressed similar amounts of Osx1 protein (Fig. 2I). BMP2 greatly enhanced matrix mineralization (Fig. 2G,H) and the abundance of Osx1 in Bmp2^F/F or Bmp2^Prx1Δ/Δ cells (Fig. 2I), confirming our in vivo observation that osteoprogenitors in Bmp2^Prx1Δ/Δ mice retain the ability to respond to exogenous BMP2 (Chappuis et al., 2012). In stark contrast, Wnt3a greatly enhanced the amount of mineralized matrix and Osx1 protein produced by WT cells but not by Bmp2^Prx1Δ/Δ cells (Fig. 2G-I), which made even less mineralized matrix and Osx1 protein than Bmp2^Prx1Δ/Δ cells grown in osteogenic medium alone (Fig. 2G, control). Concomitant treatment of cells with both BMP2 and Wnt3a also induced robust mineralization and Osx1 expression (Fig. 2G-I). Notably, this response was indistinguishable between WT and Bmp2^Prx1Δ/Δ cells, indicating that the ability of Bmp2^Prx1Δ/Δ cells to upregulate Osx1 and make calcified matrix in the presence of Wnt3a is fully rescued by exogenous BMP2. The Osx1 banding pattern in cells treated with BMP2 and/or Wnt3a was complex, with the appearance of several slower-migrating bands, suggesting that Osx1 undergoes post-translational modifications in differentiating osteoblasts, as seen with other Sp transcription factor family members (Waby et al., 2008). Regardless of culture supplements, Bmp2^Prx1Δ/Δ BMSCs had reduced levels of phosphorylated Smad1/5 but similar levels of phosphorylated Smad2/3 compared with Bmp2^F/F BMSCs, suggesting that total BMP signaling was reduced while TGF-β/activin signaling remained unchanged (Fig. 2I).

We further examined the possibility that diminished osteogenic response of Bmp2^Prx1Δ/Δ cells to Wnt3a can be explained by an inability of Bmp2-deficient cells to activate Tcf/Lef activity. Immunoblots revealed that β-catenin, an essential transcriptional mediator of cWnt signaling, was modestly reduced in Bmp2^Prx1Δ/Δ cells and markedly diminished in Bmp2^Prx1Δ/Δ cells treated with Wnt3a (Fig. S1A). QPCR showed that mRNAs encoding essential signaling molecules of the cWnt pathway are expressed in Bmp2^Prx1Δ/Δ cells grown in osteogenic medium with or without Wnt3a (Fig. S1B), as well as Bmp2^AdenoΔ/Δ cells grown in osteogenic medium with or without Wnt3a (Fig. S1C), suggesting that the diminished amounts of β-catenin protein detected in Bmp2^Prx1Δ/Δ cells grown with or without Wnt3a (Fig. S1A) occurs via a post-transcriptional mechanism. Despite diminished protein levels of β-catenin in Bmp2^Prx1Δ/Δ cells, the Tcf/Lef-target gene Axin2 (Jho et al., 2002) was strongly upregulated in Bmp2^Prx1Δ/Δ cells (Fig. S2B) or Bmp2^AdenoΔ/Δ cells following exposure to Wnt3a (Fig. S2C), similar to previous results obtained in MLB13 cells (Fig. 1G).

We next performed comparative transcriptome-level gene expression analysis. BMSCs from Bmp2^F/F or Bmp2^Prx1Δ/Δ mice were differentiated with osteogenic medium alone, BMP2, Wnt3a or BMP2+Wnt3a. RNA was harvested on day 7 to generate 24 samples. All 24 samples were used to prepare cDNA for traditional QPCR analysis. A subset of samples was also used to make cDNA libraries for microarray analysis: (1) one pooled sample of Bmp2^F/F extracts (n=3); (2) three unpooled samples of Bmp2^F/F+Wnt3a extracts; (3) three unpooled samples Bmp2^Prx1Δ/Δ+Wnt3a extracts; and (4) one pooled sample of Bmp2^Prx1Δ/Δ cells treated with BMP2+Wnt3a (n=3). All eight samples performed as a single tissue type (Fig. S2A), had similar gene-level mean signal intensities after normalization and summarization using Expression Console (Fig. S2B) and produced a >2-fold signal to background ratio (Fig. S2C). Pearson's correlation analysis demonstrated that the four sample types segregated into four distinct groups according to genotype and treatment (Fig. S2D).

Gene expression was calculated as linear fold change in Bmp2^F/F relative to Bmp2^Prx1Δ/Δ cells treated with Wnt3a. When the top 5000 most differentially regulated genes were used to perform hierarchical cluster analysis, samples from Bmp2^F/F or Bmp2^Prx1Δ/Δ cells treated with Wnt3a segregated into two distinct groups based on genotype (Fig. 3A). A total of 165 genes were downregulated >2 fold in Bmp2^F/F versus Bmp2^Prx1Δ/Δ (blue datapoints, Fig. 3B) and 114 genes were upregulated >2 fold in Bmp2^F/F versus Bmp2^Prx1Δ/Δ (red datapoints, Fig. 3B), where P<0.05. Differentially regulated genes were widely distributed throughout the genome and every chromosome (Fig. 3C).

We identified several pathways with ≥2 targets exhibiting ≥2-fold change in expression and a P-value ≤0.05 (Fig. 3D). Consistent with a study focused on cells lacking Bmp2, molecules in the BMP pathway, including Bmp2, and transcriptional gene targets of BMP signaling, including Smad6, Smad8, Id1 and Id3 (Table 1) were enriched. Intriguingly, the phenotype of Bmp2^Prx1Δ/Δ cells remains despite the fact that BMSC-derived osteoblasts expressed ≥17 other BMP/GDF ligands, and a variety of Type I and Type II BMP receptors that is sufficient to maintain some BMP signaling (Table 1 and Fig. 2I). Several modulators of Wnt signaling were altered in Bmp2^Prx1Δ/Δ cells treated with Wnt3a, including Lgr5, Lgr6, Wif1, Notum, and Sfrp2; however, many established Tcf/Lef target genes, including Axin2 and Lef1, were induced normally following exposure to Wnt3a, confirming that cWnt signaling was functional in the absence of Bmp2 (Table 2). Consistent with the mesenchymal and stem/progenitor cell nature of input samples, targets related to pluripotency, endochondral ossification and adipogenesis pathways were enriched (Fig. 3D) and included Lifr, FGFR2, Ihh, Ptch1, Mef2c and Sox9 (Table 3). Bmp2^Prx1Δ/Δ cells treated with Wnt3a expressed early markers of endochondral lineages, but expressed statistically low levels of Osx1, Dlx3 and Dlx5 (Table 3). A variety of G-protein-coupled receptors with established roles in osteoblasts were also enriched, including serotonin receptor type 2a (Htr2a) and Pthr1 (Pth1r). Bmp2^Prx1Δ/Δ cells treated with Wnt3a do not mineralize, and indeed these cells fail to induce Alpl and Phospho1, which are both poly-phosphate-metabolizing enzymes required during development for mineralization of the skeleton (Yadav et al., 2011), and Dmp1, a glycoprotein involved in bone/renal mineral homeostasis (Rowe, 2012) (Table 4). BMSC-derived osteoblasts expressed all 15 members of the carbonic anhydrase family – enzymes recently implicated in production of precursors to mature calcium hydroxyapatite crystals of vertebrate bone (Müller et al., 2013; Wang et al., 2014); Bmp2^Prx1Δ/Δ cells treated with Wnt3a expressed negligible levels of both Car3 and Car12 (Table 4).

QPCR was used to validate selected genes identified by microarray and furthermore evaluate their expression profiles in cells treated with BMP2 or BMP2 plus Wnt3a. As previously reported, Bmp2 and Bmp4 were upregulated by Wnt3a in Bmp2^F/F cells (Shu et al., 2005; Zhang et al., 2013) (Fig. 4A). Bmp2^F/F and Bmp2^Prx1Δ/Δ cells expressed similar levels of early endochondral lineage markers Prx1 and Runx2, but Bmp2^Prx1Δ/Δ cells treated with Wnt3a failed to acquire a committed osteoblast phenotype characterized by increased expression of Osx1, Dlx3 and Dlx5 (Fig. 4B). Bmp2^F/F and Bmp2^Prx1Δ/Δ cells expressed similar levels of Alpl and Phospho1, Car3, Car12 and Dmp1, but Bmp2^Prx1Δ/Δ cells treated with Wnt3a expressed only modest levels of these genes (Fig. 4B). Expression of osteoblast markers and mineral metabolism enzymes in Bmp2^Prx1Δ/Δ cells treated with Wnt3a could be rescued by complementation with exogenous BMP2 (Fig. 4B,C). In this experiment, the fundamental finding that Bmp2-deficient cells exhibit impaired progression to the Osx1/Dlx5⁺ cell fate phenotype in response to Wnt3a was clear; however, some responses to BMP2 were blunted, presumably as a result of a weak batch of BMP2. We therefore cultured primary BMSCs one more time, and used QPCR to verify that Bmp2^Prx1Δ/Δ cells are unable to induce Osx1 in response to Wnt3 without complementation by exogenous BMP2 (Fig. 4D). These samples were used to proceed with analysis of Grhl3, one of the most highly altered genes identified by comparative microarray analysis on Bmp2^Prx1Δ/Δ cells treated with Wnt3a.

Grhl3 encodes a transcription factor not previously recognized to play a role in osteoblast cell fate specification. In BMSCs, Grhl3 expression was induced by Wnt3a >50-fold in Bmp2^Prx1Δ/Δ cells and >117-fold in Bmp2^F/F cells. Grhl3 expression was not induced at the mRNA level by BMP2 alone, but was maximally induced (>300-fold) regardless of genotype in BMSCs co-stimulated with BMP2 and Wnt3a (Fig. 4E). In MLB13 cells, Grhl3 mRNA was diminished by knockdown of Bmp2, and enhanced by Wnt3a regardless of the Bmp2 genotype (Fig. 4F). The USCS Genome Browser (Kent et al., 2002) was used to examine the genetic locus of murine Grhl3 (Fig. S3A) and revealed two consensus Tcf/Lef-binding sites (Cadigan and Waterman, 2012; Frietze et al., 2012; Guenther et al., 2014) in regions of high to medium genomic conservation in vertebrates (Fig. S3B,C). Compared with MLB13 cells expressing plasmid-derived EGFP, Grhl3 mRNA was increased 12 h after transfection by a plasmid encoding Wnt1, but not by a plasmid encoding Wnt5. Grhl3 was also induced by 4 h of treatment with recombinant Wnt3a, and this induction was not sensitive to increasing titrations of cycloheximide (25-100 µg/ml), suggesting that protein translation is not required for Grhl3 induction by cWnt signaling (Fig. 4G) and that Grhl3 is a direct canonical Wnt target gene.

To examine the relevance of our microarray analysis, we first tested the model that Bmp2 is required at the Runx2/Osx1 transition during Wnt-activated differentiation of osteoblasts. Mice were bred to conditionally ablate Bmp2 in committed Osx1⁺ osteoblasts (Bmp2^F/F; Osx1-GFP::Cre or Bmp2^Osx1Δ/Δ) using the Osx1-GFP::Cre transgene (Rodda and McMahon, 2006). BMSCs were cultured in osteogenic medium with BMP2, Wnt3a or BMP2 plus Wnt3a. Alizarin Red staining revealed that Bmp2^F/F cells had partially mineralized their matrix; this was enhanced by treatment with either BMP2 or Wnt3a; and mineralization was maximized by dual exposure to BMP2 plus Wnt3a (Fig. 5A, left panels and B). Bmp2^Osx1Δ/Δ cells produced comparable levels of mineralized matrix under each corresponding culture condition, including after Wnt3a exposure (Fig. 5A, right panels and B). QPCR analysis on RNA from day 7 confirmed that Osx1-Cre had successfully recombined Bmp2 in Bmp2^Osx1Δ/Δ cells (Fig. 5C). Furthermore, Prx1, Runx2, Osx1, Dlx3 and Dlx5 levels were indistinguishable between Bmp2^F/F and Bmp2^Osx1Δ/Δ cells for each condition (Fig. 5D). And, consistent with results from Alizarin Red staining, we found that Bmp2^Osx1Δ/Δ cells are able to induce genes required for synthesis of mineralized bone matrix including Col1a1, Alpl, Phospho1, Car3 and Car12 (Fig. 5E).

We next investigated whether Grhl3 plays a role in osteoblast cell specification. To test if Grhl3 was sufficient to regulate osteoblast cell fate specification, we monitored specification of Osx1⁺ cells in MLB13^WT or MLB13^kdBmp2 cells transfected with human GRHL3. Immunoblot analysis showed that GRHL3 protein abundance is dramatically reduced by knockdown of Bmp2 and enhanced by rBMP2. Canonical Wnt3a or Wnt10b had no effect on GRHL3 protein expression, alone or in combination with BMP2 (Fig. 6A). Although a longer stimulation period was required (>3 days) than in experiments in BMSCs, Osx1 expression could be increased by treatment with Wnt3a or Wnt10b in MLB13^WT but not MLB13^kdBmp2 cells, which expressed 20-times lower levels of Osx1 than MLB13^WT (Fig. 6B). Strikingly, GRHL3 did not rescue the ability of Wnt3a or Wnt10b to induce Osx1 in MLB13^kdBmp2 cells (Fig. 6B), but induced Osx1 to levels indistinguishable from that of WT cells when expressed in the presence of rBMP2 (Fig. 6B). Ectopic expression of GRHL3 had minimal effects on Dlx3 and Dlx5 that did not correlate with specification of Osx1⁺ cell phenotype (not shown). Importantly, different classes of cWnt ligands and antagonists bind distinct extracellular domains of Lrp5/6 (Bourhis et al., 2011; Gong et al., 2010); however, Wnt3a and Wnt10b had similar inhibitory effects on Bmp2-deficient cells.

To test if Grhl3 was necessary for osteoblast differentiation, we monitored specification of Osx1⁺ cell fate in MLB13^WT cells transfected with siRNAs targeting Grhl3 (Fig. 6D). We also knocked down the highly conserved family member Grhl2 (Fig. 6C) since it is expressed in our cell models, can heterodimerize with Grhl3 (Ting et al., 2003b) and has been shown to compensate for Grhl3 during wound repair in skin (Boglev et al., 2011). Following knockdown, we stimulated cells for 48 h with BMP2, Wnt3a or BMP2+Wnt3a and then performed QPCR. Suppression of Grhl3 resulted in a statistically significant though modest decrease in the abundance of Prx1, Runx2, Dlx3 and Dlx5 mRNAs expressed in BMP2 treated cells (Fig. 6E-H, red and black bars). Modest upregulation of Prx1 by BMP2 was also blunted by suppression of Grhl2 (Fig. 6E). As seen previously, BMP2 treatment led to high magnitude upregulation of Osx1 and Alpl (45 and 20-fold, respectively), and this upregulation was blunted by 50-80% by introduction of Grhl2 and/or Grhl3 siRNAs (Fig. 6I,J). Grhl3 siRNA also completely abrogated the induction of Car12 by BMP2 (Fig. 6K). Expression of Col1a1 and Phospho1 were modestly reduced by Grhl3 siRNAs (Fig. 6L,M). By contrast, Car3 expression in response to Wnt3a greatly increased following suppression of Grhl2 (Fig. 6N), consistent with the fact that Grhl proteins are Janus factors and can therefore act as transcriptional activators or repressors at different genetic loci. Moreover, Grhl2 and Grhl3 siRNAs did not cause a non-specific decrease in global gene expression. These cumulative results strongly suggest that Grhl2 and Grhl3 are both necessary and sufficient for the mechanism by which BMP2 specifies progenitors to an Osx1⁺ osteoblast cell fate. Comprehensive statistical analysis is provided in Table S1.

Previous reports suggest Grhl3 is expressed in the condensing limb bud mesenchyme of E13.5 mouse embryos (Kudryavtseva et al., 2003), consistent with our findings that Grhl3 is expressed in E13.5 mouse limb bud cells. Consultation with the Eurexpress.org in situ hybridization database revealed that mRNAs for Grhl3 are enriched in the perichondrium of developing endochondral structures in the E14.5 hindlimb (Fig. 7A,B), where the expression domain of Grhl3 overlaps with Lrp5 and Bmp2 (Fig. 7C,D). In situ hybridization on E16.5 hindlimb and forelimb indicates Grhl3 expression is evident in the developing bone collar, hypertrophic chondrocytes, and cells within the newly forming subchondral bone (Fig. 7E). This Grhl3 expression domain corresponds to regions where newly forming osteoblasts appear during development (Salazar et al., 2016) and moreover, where Bmp2 is expressed when monitored using a Bmp2^ki(lacZ) gene replacement allele (Fig. 7F). Finally, we created standardized fractures in adult mouse femurs to monitor Grhl3 (and microarray hits) in an in vivo model of bone formation that relies on endogenous BMP2 and cWnt signaling mechanisms (Chen et al., 2007; Tsuji et al., 2006). QPCR analysis revealed that Bmp2, signaling molecules of the cWnt pathway, Grhl2 and Grhl3 continue to be expressed in marrow-free bone tissue in the adult skeleton (Fig. 7G,H). Bmp2 and cWnt signaling molecules, as well as surrogate markers of BMP signaling (Id3) and Wnt signaling (Axin2) were robustly induced 5 days following fracture (Fig. 7G). These endogenous signaling dynamics characteristic of early endochondral bone repair were accompanied by no changes in Grhl2, a modest increase in Runx2, and striking inductions of Prx1, Osx1 and Grhl3 (Fig. 7H). Car3, Car12, Alpl and Phospho1 were also induced at the fracture site (Fig. 7I), providing evidence that the molecular targets identified by microarray analysis in primary cells are biologically significant for osteoblast physiology during bone repair.

We used multiple mouse and primary cell models to test whether BMP and cWnt signaling play unique or redundant roles in the specification of new osteoblasts and subsequent production of bone matrix. Among the many BMPs expressed in bone, our data revealed a specific requirement for BMP2 in progression of Prx1⁺ progenitors through the Runx2/Osx1 transition. We identified Grhl3 as a novel transcription factor in the osteoblast lineage that contributes to specification of an Osx1⁺ cell fate in a BMP-dependent manner. These essential functions for BMP2 were enhanced by but not compensated for by cWnt activity in our primary cell model systems. We are currently completing a rigorous study of these findings in vivo.

The marrow and periosteum of adult bone provide essential niches for skeletal stem cells (Chan et al., 2015; Worthley et al., 2015) and other fate-restricted skeletal progenitors (Mizoguchi et al., 2014; Ono et al., 2014; Zhou et al., 2014) that give rise to new osteoblasts for bone growth and repair during postnatal life (Kassem and Bianco, 2015; Kfoury and Scadden, 2015). Skeletal stem cells (SSCs) first appear in the limb bud mesenchyme, where Prx1 is broadly expressed, and upon residency in adult bone, produce a variety of secreted niche factors including Bmp2, Wnt3a, TGFβ3 and Grem1 (Chan et al., 2015; Worthley et al., 2015; Yang et al., 2013). Accumulating data suggest that specific skeletal tissue types, such as bone or cartilage, can be induced from SSCs as well as ESCs (Craft et al., 2013) by controlling the timing and combination of growth factors presented to the cell. Understanding how each of these growth factors contributes, either alone or in combination, to a particular skeletogenic milieu has therefore emerged as a crucial step in understanding how to modulate bone mass or bone repair in humans with skeletal trauma or disease. The primary BMSC preparations used in this study expressed Prx1, Lepr, Nes, Pdgfra and Grem1, and therefore likely represent a heterogeneous population containing key skeletal progenitors identified in these recent studies. Of these markers, only Lepr was differentially expressed in Bmp2^Prx1Δ/Δ cells. Future studies will shed light on whether BMP2 regulates the expression of Lepr and/or abundance of Lepr+ progenitors.

Here, we focused on understanding how BMP2 accomplishes transcriptional control of osteoblast differentiation from endochondral progenitors. Bmp2^Prx1Δ/Δ cells successfully expressed transcriptional mediators of early endochondral bone formation including Sox2, Hand2, Msx2, Prx1, Dermo1 and Runx2. However, Bmp2^Prx1Δ/Δ cells expressed very low levels of late transcriptional mediators including Dlx3, Dlx5 and Osx1, a group of proteins that heterodimerize to induce genes essential for the physiology of mature osteoblasts (Hojo et al., 2016). These data revealed a block in the Runx2/Osx1 transition, consistent with the fact that Osx1 was originally cloned as a BMP2-responsive gene (Nakashima et al., 2002). Importantly, however, Osx1 expression is highly enriched in the skeleton and is therefore not indiscriminately expressed in all tissues undergoing active BMP signaling. We thus hypothesized that additional co-factors are required to target a BMP transcriptional response to the Osx1 promoter during osteoblast differentiation. We focused on Dlx3, Dlx5 and Grhl3, since microarray analysis indicated these were highly altered transcription factor genes in our samples. Since BMP2 is able to induce Osx1 in cells lacking Runx2 but not lacking Dlx5 (Lee et al., 2003; Ulsamer et al., 2008), we used Ensembl to search 13 kb of the murine Osx1 promoter for predicted Dlx binding motifs. Interestingly, we did not find Dlx3/5 binding sites, but did identify 12 predicted binding sites for Grhl3 intermingled with numerous Smad binding motifs (Gordon et al., 2014; Harrison et al., 2010).

We also found that Grhl3 is induced at the transcriptional level by cWnt signaling, and is regulated at the protein level by BMP2 signaling via as yet undefined mechanisms. Plasmid-based expression of GRHL3 specified Osx1⁺ cell fate of limb bud progenitors in a BMP2-dependent and Wnt3a-independent manner, while suppression of Grhl3 impaired the ability of BMP2 to induce Osx1. In vivo gene expression studies revealed that Grhl3 is co-expressed in regions where BMP and cWnt signaling contribute to bone development and bone repair. Grhl3 is therefore expressed at the appropriate time and place to be involved in developmental skeletogenesis and fracture repair. Our data strongly suggest that Grhl2 and Grhl3 play a necessary and sufficient role in the formation of Osx1⁺ cells. Importantly, these collective findings reconcile with a requirement for both BMP and cWnt/β-catenin signaling during specification of Osx1⁺ cells during skeletal development (Bandyopadhyay et al., 2006; Day et al., 2005; Hill et al., 2005). We are currently utilizing in vivo systems to evaluate Grhl3 as a potential nexus by which BMP2 and cWnt signaling cooperate to drive bone formation and bone repair in vivo. Further work is also needed to formally test the role of Grhl3 (and potential compensation by Grhl1 and Grhl2) in skeletal development.

A variety of human, mouse and fish studies suggest reduced expression of Grhl3, Dlx3, Dlx5 and Osx1 could explain the defects we found in calcium and phosphate metabolism (Dworkin et al., 2014; Han et al., 2011; Isaac et al., 2014; Lapunzina et al., 2010; Liu et al., 2015; Nguyen et al., 2013; Ting et al., 2003a). Mineralization of the endochondral skeleton is a complex biological process involving the formation of a calcium phosphate hydroxyapatite crystal that is deposited into an organic extracellular matrix composed primarily of type I collagen. Carbonic anhydrases (CAs) are zinc-dependent metalloenzymes that catalyze the hydration of carbon. Whereas in mammals, CAs are best known for their role in processing metabolic waste, regulating pH and fixing carbon, paleogenomic studies in sponges suggest that the ancestral CA was one of the first genes to confer the ability to synthesize skeletogenic structures through the formation of calcium carbonate (Jackson et al., 2007) and recent evidence suggests that calcium carbonate is an essential precursor crystal to the mature calcium phosphate hydroxyapatite crystals comprising endochondral bone (Müller et al., 2013). Although additional work clarifying the role of CA enzymes in bone metabolism is needed, we find that Car3 and Car12 are locally induced in bone following fracture.

In summary, we identify pre-Osx1⁺ progenitors as a crucial source and target of endogenous BMP2 required for the Runx2/Osx1 transition and show that BMP2 and cWnt play cooperative but non-redundant roles during osteoblast cell fate specification.

In vivo experiments were performed in compliance with the Guide for the Care and Use of Laboratory Animals and were approved by the Harvard Medical Area Institutional Animal Care and Use Committee (protocol #04043 to V.R.). Mice carrying floxed Bmp2 alleles (Bmp2^F/F) were bred to Prx1-cre mice (Logan et al., 2002) or Osx1-cre mice (Rodda and McMahon, 2006) to obtain Bmp2^F/F; Prx1-Cre mice (Bmp2^Prx1Δ/Δ) or Bmp2^F/F; Osx1-EGFP::Cre mice (Bmp2^Osx1Δ/Δ).

Standardized fractures with pin stabilization were made in femora of adult mice as previously described (Tsuji et al., 2006). Fractured and non-fractured contralateral controls femurs were collected 5 days post fracture and 6 mm of mid diaphysis bone were dissected, cleaned of bone marrow, snap frozen in liquid nitrogen, and processed for total RNA extraction according using Trizol and RNeasy Plus kit, as previously described (Salazar et al., 2013).

MLB13 cells were prepared from immortalized limb bud cells from E13.5 mouse embryos (MLB13 clone14) (Rosen et al., 1993, 1994) cultured and maintained in DMEM supplemented with 10% fetal bovine serum and 100 U/ml penicillin-G and 100 mg/ml streptomycin. Cells are contamination free and were authenticated by neomycin resistance, which was conferred during immortalization. MLB13 cells were transfected with Bmp2-shRNA plasmid [BLOCK-iT, Invitrogen (5′-TGCTGAGTTCAAGAA-GTCTCCAGCCAGTTTTG-GCCACTGACTGACTGG-CTGGACTTCTTGAACT-3′)] using Lipofectamine 2000 (Invitrogen), selected for 3 weeks in 20 mg/ml blasticidin S and cloned by limiting dilution. Where indicated, cells were seeded to confluence and pretreated with cycloheximide (Sigma Aldrich, diluted to 25-100 mg/ml in growth medium) for 1 h prior to 4 h stimulation with Wnt3a+cyclohexamide.

BMSCs were prepared from bone marrow from 4- to 6-month-old mice (male and female) collected as previously described (Salazar et al., 2013) and plated in BMSC medium [ascorbic acid-free α-MEM (Invitrogen) containing 20% FBS, 40 mM L-glutamine, 100 U/ml penicillin-G and 100 mg/ml streptomycin]. After 3 days, non-adherent cells were removed by vigorous washing. For osteoblast differentiation, BMSCs were seeded in BMSC medium at 35,000 cells/well in 96-well dishes (for Alizarin Red staining) or 300,000 cells per well in 24-well dishes (for RNA or protein). Confluent cultures were stimulated on day 1 with osteogenic medium (OM: BMSC medium plus 50 μg/ml ascorbic acid and 10 mM β-glycerophosphate) and recombinant proteins.

Recombinant proteins were added as indicated to culture medium: human BMP2 (200 ng/ml; Genetics Institute), and mouse Wnt3a and Wnt10 (40 ng/ml; R&D Systems).

Total RNA was harvested according to the manufacturer's instructions using the RNeasy Plus Kit (Qiagen). RNA was reverse transcribed with EcoDry Premix (Clontech). Data were normalized to β-actin, analyzed using the ΔΔCT method, and expressed as mean±s.d. (when biological replicates were established with pooled cells from multiple mice) or mean±s.e.m. (when biological replicates were established from unpooled cells from multiple mice) relative to Bmp2^F/F cells cultured in osteogenic medium. Student's t-test was utilized to calculate P-values.

Cells were plated at 2×10⁴ cells/well in 96-well dishes, stimulated with Wnt3a for 3 days, fixed in acetone/ethanol (50:50), and incubated with a substrate solution composed of 0.1 M diethanolamine, 1 mM MgCl₂ and 1 mg/ml p-nitrophenylphosphate. The reaction was quenched with 3 M NaOH and absorbance measured at 405 nm.

Human GRHL3 plasmid was provided by the Center for Cancer Systems Biology, PlasmID clone HsCD00376192. MLB13 cells were seeded to 90% confluence, transfected with a plasmid encoding human GRHL3, and then cultured for 3 days in osteogenic medium plus indicated growth factors.

Cells were seeded 200K cells/well in a 24-well dish and transfected according manufacturer's directions using RNAiMax (Invitrogen) and 10 pmol of indicated siRNA. Cells were stimulated 48 h with osteogenic medium with or without BMP2 and Wnt3a or with BMP2+Wnt3a.

Ad5-CMV-eGFP and Ad5-CMV-Cre-ires-eGFP viral particles were obtained from the University of Iowa Gene Transfer Vector Core. BMSCs were seeded at 40,000 cells/well in 96-well dishes for mineralization assays or 400,000 cells/well in 24-well dishes for QPCR, transduced for 6 h with 100 m.o.i of virus and 25 mM Hoechst in ascorbic acid-free α-MEM containing 2% FBS and L-glutamine, penicillin and streptomycin as for cell culture methods above. Cells recovered overnight in BMSC medium.

Cells were fixed 10 min in 10% neutral buffered formalin before staining with 1 mg Napthol AS-MX phosphate in 25 ml N,N-dimethylformamide plus 6 mg Fast Blue RR salt in 10 ml of buffer consisting of 0.1 M Tris-Base, 2 mM MgCl₂, pH 8.5.

For Alizarin Red staining, fixed cells were stained for 30 min in 0.4% aqueous solution of Alizarin Red S (Sigma). Alizarin Red S was eluted in 10% glacial acetic acid, pH was adjusted with 10% ammonium hydroxide, and absorbance measured at 405 nm.

Cells were scraped into RIPA buffer containing Halt protease and phosphatase inhibitors (Pierce) and homogenized with a QiaShredder column (Qiagen). Proteins were separated by SDS-page electrophoresis and immunoblotted with the following antibodies at 1:1000 dilution: p-Smad1/5 (Cell Signaling, 9511), p-Smad2/3 (Cell Signaling, 3101), β-catenin C-terminus (BD 61054), β-catenin N-terminus (Cell Signaling, 9562), osterix (Abcam, ab22552), V5 (Invitrogen, R961-25) and α-tubulin (Sigma, T6074).

Gene-level differential expression analysis was performed with Affymetrix Mouse ST 2.1 microarrays on BMSCs from Bmp2^F/F or Bmp2^Prx1Δ/Δ mice differentiated for 7 days with osteogenic medium alone or in medium plus BMP2 (100 ng/ml), Wnt3a (40 ng/ml) or both. Cultures were performed in biological triplicates using pooled cells from n≥3 mice/genotype. Eight cDNA libraries were made from 100 ng total RNA using WT Expression Kit (Ambion): one pooled sample from three independent cultures of Bmp2^F/F cells grown in osteogenic medium (33.33 ng each) or Bmp2^Prx1Δ/Δ cells with BMP2+Wnt3a (33.33 ng each); three samples from three independent cultures of Bmp2^F/F or Bmp2^Prx1Δ/Δ cells with Wnt3a. Data were analyzed with Expression Console and Transcriptome Analysis Console v.3.0 (Affymetrix). Gene expression was reported as linear fold change relative to Bmp2^Prx1Δ/Δ cells with Wnt3a, and P-values were calculated by one-way between-subject ANOVA for unpaired samples.

E16 mouse embryo hindlimbs were fixed, paraffin processed and sectioned in the sagittal plane using standard methods. In vitro transcription to generate riboprobes was performed using standard protocols and reagents (Promega). In situ hybridization with digoxigenin-labeled Grhl3 probe (kindly provided by Stephen Jane, Monash University, Australia) was carried out as described previously (Gamer et al., 2009). Hindlimbs from E16 mice were embedded in optimal cutting temperature compound (OCT) on dry ice and frozen sections were prepared for β-galactosidase staining as previously described (Kokabu et al., 2012).

We thank the University of Iowa Gene Transfer Vector Core (supported in part by the NIH and the Roy J. Career Foundation) for viral vectors; the Eurexpress.org ISH database for images of Bmp2, Lrp5 and Grhl3 expression; the Whitman lab for assembling V-5 tagged Grhl3 constructs; and Travis Burleson (Affymetrix) for microarray support.