AP-1 family members act with Sox9 to promote chondrocyte hypertrophy

The SRY-box (Sox) transcriptional regulator Sox9 is a central regulator of endochondral skeletal development (reviewed by Akiyama and Lefebvre, 2011; Lefebvre and Dvir-Ginzberg, 2016). Initial Sox9 expression in condensing mesenchymal cells precedes active chondrogenesis (Bi et al., 1999). Subsequently, Sox9 expression is maintained in proliferating immature chondrocytes and post-mitotic prehypertrophic chondrocytes (Akiyama et al., 2005; Dy et al., 2012). Sox9 transcription terminates in early hypertrophic chondrocytes (Dy et al., 2012). However, Sox9 protein persists through hypertrophic development: only terminal hypertrophic chondrocytes at the hypertrophic-osteoblast interface of the bone shaft lack Sox9 (Dy et al., 2012).

Genetic analysis in the mouse has shown that Sox9 is necessary for establishing chondrocytes in the cranial, axial and appendicular skeleton (Akiyama et al., 2002; Bi et al., 2001; Mori-Akiyama et al., 2003). Moreover, Sox9 is sufficient to initiate chondrogenic programs when activated in mesenchymal stem cells (Ikeda et al., 2004), embryonic stem cells (Ikeda et al., 2004) and human dermal fibroblasts (Ikeda et al., 2004; Ohba et al., 2015; Outani et al., 2013). Inactivating mutations in human SOX9 result in campomelic dysplasia, an autosomal dominant disorder characterized by hypoplasia of the endochondral skeleton and bowing of skeletal elements (Foster, 1996; Giordano et al., 2001; Kwok et al., 1995). Molecular studies indicate that Sox9 promotes expression of a broad spectrum of cartilage matrix components, promotes cell division and cell survival, and inhibits chondrocytes from adopting an alternative pathway of osteoblast development (reviewed by Lefebvre and Dvir-Ginzberg, 2016).

Sox9 directs chondrogenic programs in concert with other transcriptional components, notably the related Sox members Sox5 and Sox6. The combinatorial interaction of Sox5, Sox6 and Sox9 at Acan and Col2a1 enhancers promotes the expression of these key genes within mitotic chondrocytes (reviewed by Akiyama and Lefebvre, 2011; Lefebvre, 2002). Although genome-wide binding analysis of Sox9 and Sox6 indicates that they co-regulate genomic targets within proliferating chondrocytes (Liu and Lefebvre, 2015), Sox action in hypertrophic chondrocyte development is less clear.

Genetic removal of Sox9 broadly within both proliferating and prehypertrophic chondrocytes prevents hypertrophic progression. In the absence of Sox9, chondrocytes switch fate generating ectopic osteoblasts (Dy et al., 2012). Runx2, a key transcriptional regulator promoting hypertrophic chondrocyte development, is expressed in mitotic and postmitotic chondrocytes of the growth plate where Runx2 has been proposed to counter Sox9's transactivation of early chondrocyte targets such as Col2a1 (Cheng and Genever, 2010; Zhou et al., 2006). Sox9 has also been reported to interact with Mef2 to activate Col10a1, a matrix-encoding gene specifically expressed by hypertrophic chondrocytes (Dy et al., 2012). Furthermore, other reports suggest that a Gli-Sox9 interaction at a distal enhancer of Col10a1 suppresses Col10a1 expression in proliferating chondrocytes (Leung et al., 2011). Together, these studies suggest that Sox9 actions on chondrocyte programs are temporally and spatially regulated by interplay with a variety of additional transcriptional regulators.

Direct analysis of DNA binding has been particularly insightful in distinguishing conflicting regulatory models. Recently, we performed a genome-scale analysis comparing Sox9-DNA interactions in somite-derived and neural crest-derived chondrocytes (Ohba et al., 2015). Analysis of motif recovery identified a highly significant enrichment of an activator protein-1 (AP-1) motif, suggesting co-integration of AP-1 factor engagement into a Sox9-driven chondrocyte regulatory network. AP-1 family members have been linked to osteoblast and osteoclast regulation (reviewed by Wagner, 2002; Wagner and Eferl, 2005) and joint formation (Kan and Tabin, 2013) where Sox9 is either absent or present only at low levels. However, Fosl2 mutants show a reduction in the hypertrophic chondrocyte zone and in mineralized hypertrophic cartilage matrix (Karreth et al., 2004) whereas Jun removal delays chondrocyte hypertrophy specifically in the baso-occipital bone (Behrens et al., 2003).

Here, we explored Sox9, Jun and Fosl2 interactions in the developing mouse skeleton. Jun and Fosl2 are specifically expressed in prehypertrophic and early hypertrophic chondrocytes where endogenous Sox9 levels are maximal. ChIP-seq analysis shows extensive overlap between Jun- and Sox9-bound genomic target regions in mouse primary rib chondrocytes. DNA association likely reflects both direct Jun DNA binding through AP-1 motifs, and indirect protein-protein association with Sox9 mediated by the DNA-binding region of Jun. In vivo and in vitro functional assays suggest that AP-1 and Sox9 co-regulatory pathways promote the chondrocyte transition to post-mitotic hypertrophic chondrocytes and activation of the hypertrophic chondrocyte marker gene Col10a1. These findings highlight the cooperative interplay of transcriptional programs underpinning development of the mammalian skeleton.

Our recent ChIP-seq analysis of Sox9 action in mammalian chondrocytes identified a striking enrichment of an AP-1 factor binding motif (TGAG/CTCA), in addition to the expected Sox9 dimer and monomer motifs (Ohba et al., 2015), suggesting that AP-1 family members might act in conjunction with Sox9 to regulate the chondrogenic program.

To address potential AP-1 action, we performed RNA-seq on neonatal rib cartilage to examine the expression of all AP-1 family members that are known to interact with the recovered AP-1 motif: Jun (previously known as c-Jun), Junb, Jund, Fos (cFos), Fosb, Fosl1 (Fra1) and Fosl2 (Fra2). Several members of the AP-1 family showed significant rib chondrocyte expression through this analysis (Table S1). Given that AP-1 factors interchangeably form functional homodimer and heterodimer protein complexes (reviewed by Eferl and Wagner, 2003), the data raised the possibility of considerable heterogeneity in endogenous AP-1 complexes. Among the expressed family members, Jun and Fosl2 show strong expression within hypertrophic chondrocytes (see below) and both are linked to chondrocyte development (Behrens et al., 2003; Hess et al., 2003; Kan and Tabin, 2013; Karreth et al., 2004); these two AP-1 family members were selected for further study.

In situ hybridization to limb and rib skeletal elements with Jun and Fosl2 probes showed strong specific expression of both factors within prehypertrophic and early stage hypertrophic chondrocytes where Col10a1 is first activated (Fig. 1A-D′; Fig. S1A). Immunostaining showed high levels of Jun and Sox9 within Ki67⁻Sp7⁺ prehypertrophic chondrocytes extending into the initial domain of Col10a1⁺ hypertrophic chondrocytes (Fig. 1E-I′; Fig. S1B). These data suggest a potential interplay between Sox9 and AP-1 members at this key transition point in cartilage development.

To address directly AP-1 action within this chondrocyte population, we performed ChIP-seq with anti-Jun antibody on rib chromatin preparations previously used to define Sox9 interactions in vivo (Ohba et al., 2015). The populations used in both this study and the previous one (Ohba et al., 2015) include proliferating and prehypertrophic chondrocytes, but not mature hypertrophic chondrocytes. A Cisgenome analysis (Ji et al., 2008; Jiang et al., 2010) of the ChIP-seq data predicted a large number of Jun interaction sites: 99,162 peaks were significantly enriched above the input control. Motif recovery by MEME-ChIP (Bailey et al., 2009) and Cisgenome predicted a significantly enriched AP-1 consensus binding motif within the dataset (Fig. 2A; Table S2). An AP-1 motif was centered within Jun ChIP-seq peaks (Fig. S2A) and was present within a large percentage of recovered peaks (67.6%: 66,986 of 99,162) consistent with direct engagement of AP-1 complexes containing a Jun component. In summary, the general features of the dataset point to a comprehensive, highly confident set of genomic interactions for Jun within developing chondrocytes.

Next, we investigated the overlap of Jun ChIP-seq peaks with Sox9 Class II ChIP-seq peaks identified in our previous study (Ohba et al., 2015). The Class II peaks represent direct Sox9-bound cis-regulatory modules closely linked to the activation of chondrocyte-specific gene regulatory programs (Ohba et al., 2015). Surprisingly, the majority of Sox9 Class II peaks (81%: 16,895 of 20,862) overlapped with Jun peaks (Fig. 2B); multiple interaction sites were prominent around genes encoding matrix proteins that are expressed at high levels in chondrocytes (Fig. 2C; Fig. S3). Furthermore, we observed a striking correlation in the relative levels of engagement between Jun and Sox9 at Jun-Sox9 co-bound regions (referred to as ‘Jun-Sox9 shared peak regions’ hereafter) as was reflected by the number of DNA sequence reads for bound regions (Fig. 2D,E). Importantly, Sox9 and Jun both associate strongly around Col10a1, a direct target of Sox9 regulation expressed in hypertrophic chondrocytes (Fig. 2F). Binding around Col10a1 highlighted a previously identified hypertrophic DNA enhancer module 4 kb upstream of Col10a1 that is sufficient to direct hypertrophic chondrocyte-restricted reporter gene expression in vivo (Fig. 2F) (Chambers et al., 2002; Gebhard et al., 2004; Leung et al., 2011; Li et al., 2011; Riemer et al., 2002; Zheng et al., 2009).

Interestingly, a comparison of DNA regions associated with Jun alone (referred to as ‘Jun-only peak regions’ hereafter) with those co-associated with Jun and Sox9 (Jun-Sox9 shared peaks) revealed several distinguishing features that suggest the latter are the most relevant to chondrocyte development. First, Jun-Sox9 shared peak regions showed a higher number of Jun sequence reads per peak and higher peak ranking consistent with stronger and/or more frequent engagement of Jun at sites co-engaged with Sox9 than Jun alone (Fig. 2G; Fig. S2B). Second, although both of these groups show strong enrichment of an AP-1 motif centered in peak regions indicative of bona fide Jun interactions with DNA targets (Fig. S2A), a Sox9 motif is only enriched in Jun-Sox9 shared peak regions (Fig. S2A; Table S3). Furthermore, the strong centering of Jun and Sox9 motifs within ChIP-seq peaks suggests direct DNA engagement of both factors within some shared cis-regulatory elements but also the potential for indirect association through protein-protein interactions (see later). Third, comparison of Jun-binding data with our published analysis of active chondrocyte enhancer signatures (p300 binding, H3K4Me2 and H3K27 acetylation enrichment and RNA polymerase II engagement; Ohba et al., 2015) showed that active enhancer signatures predominated only within Jun-Sox9 shared peak regions (Fig. 2H). Together, these data lend additional weight to the biological relevance of the Jun and Sox9 component of the dataset in active transcriptional processes within chondrocytes. Fourth, the biological relevance of Jun-Sox9 shared peaks was underscored by GREAT gene ontology (GO) analysis associating bound regions with neighboring potential target genes (McLean et al., 2010). The Jun-Sox9 shared peak regions showed a highly significant association with skeletal GO terms; the top GO Biological Process term was ‘cartilage development’ and analysis of MGI expression terms predicted skeletal structures (Fig. 2I). Jun-only peaks also showed strong matrix organization GO term enrichment (Fig. S2C), suggesting that many of these identified weaker target sites around the same set of genes engaged by Sox9, a conclusion underscored by the observation that the predicted Sox9 target gene set is essentially entirely encompassed by the predicted Jun-target gene set (Fig. 2J). Given all the evidence here that Sox9 co-association identifies a chondrocyte-related Jun/AP-1 signature with the strongest feature of active enhancers, we focused our analysis on this component of the Jun dataset.

To understand the molecular mechanisms underlying the interplay of Jun and Sox9 during chondrogenesis, we analyzed the distribution of AP-1 and Sox9 motifs in Jun-Sox9 shared peak regions. Compared with matched controls (position-matched random control regions bioinformatically calculated from the genome), we observed a strong, significant co-occurrence of Sox9 and AP-1 motifs: 39% in the whole dataset (Fig. 3A; P-value=1.8e-14) and 65% in the subset of peaks with a Sox9 dimer-binding motif (Fig. S4A; P-value<2.2e-16). Only 16% of predicted peaks failed to display a Sox9 or AP-1 binding site suggesting that direct DNA binding of Sox9, AP-1, or both factors, was the primary mode of Sox9 and Jun interaction at Jun-Sox9 shared peak regions (Fig. 3A). The Jun-Sox9 shared peak regions containing both Sox9 and AP-1 motifs showed a higher Sox9 ChIP-seq signal than those with only Sox9 motifs though each showed a comparable active enhancer signature (Fig. 3B; Fig. S4B). These data suggest an enhanced engagement of Sox9 when AP-1 was bound at adjacent DNA sites.

Although these results are consistent with cooperative binding interactions between Sox9 and AP-1 complexes at respective target sites, a significant number of the shared regions have only an AP-1 or Sox9 site suggesting protein-protein association between Sox9 and AP-1 complexes (Fig. 3A). Consistent with this view, Jun and Fosl2 co-immunoprecipitated with Sox9 when DNA-templated interactions were removed by digestion with the DNA nuclease benzonase (Fig. S4C). To further identify potential protein-protein interaction domains between Sox9 and Jun, we performed co-immunoprecipitation using a series of deletions directed by the domain architecture of each factor (Fig. 3C). These studies indicated that the N-terminal region of Sox9 interacts with a region of Jun encompassing the DNA-binding domain and adjacent proline/glutamine-enriched stretch (Fig. 3D).

We confirmed the protein-DNA-mediated and protein-protein-mediated interaction between Sox9 and AP-1 on the chondrocyte genome by ChIP-reChIP assay using the same chromatin preparations as those used in Sox9 and Jun ChIP-seq (Fig. 3E,F); enrichment was detected in (1) Sox9-AP-1-shared peaks with Sox9 and AP-1 motifs and (2) Sox9-AP-1-shared peaks with only Sox9 motif, but not in a Sox9-only peak with Sox9 motif, with more enrichment detected in (1) than in (2). These data, when considered with those in Fig. 3B and Fig. S4B, suggest that co-occupation of Sox9 and Jun within a cis-regulatory module may be promoted by the co-occurrence of each factor's recognition motif.

To understand the interplay of direct (protein-DNA mediated) and indirect (protein-protein mediated) Sox9 and AP-1 engagement on transcription of target genes, we generated a series of luciferase reporter constructs combining different enhancer-motif configurations with a basal promoter element: 4x48-Col2a1 (Fig. 4A; four copies of a 48 bp enhancer from Col2a1 intron 1 that contains multiple Sox9 binding motifs but no AP-1 motifs; Kan et al., 2009); 3xSox9BS(Col9a1) (Fig. 4B; three copies of a 17-nucleotide Sox9 dimer-binding motif from the Col9a1 enhancer; Ohba et al., 2015); 7xAP-1BS (Fig. 4C; seven copies of the optimal AP-1 motif); and 3xSox9BS(Col9a1)+7xAP-1BS (Fig. 4D; an artificial enhancer element combining Sox9 and AP-1 regulatory modules above).

As expected, Sox9 activated the 4x48-Col2a1 and 3xSox9BS(Col9a1) reporters (Fig. 4A,B). However, Sox9-mediated activation was reduced on co-transfection with Jun and Fosl2 (Fig. 4A,B). By contrast, TAM67, a dominant-negative variant of Jun that retains DNA binding but has no transcriptional activation capability, specifically suppressed endogenous AP-1 function (Brown et al., 1993), but enhanced Sox9-mediated activation of the reporter, presumably reflecting TAM67 inhibition of endogenous AP-1 complexes (Fig. 4A,B; Fig. S4D). As expected, Jun alone, or together with Fosl2, promoted activity of the 7xAP-1BS reporter; however, their activity was suppressed by Sox9 (Fig. 4C). By contrast, maximal activation of the 3xSox9BS(Col9a1)+7xAP-1BS reporter required both Sox9 and AP-1 family members and reporter activity was reduced by co-transfection with TAM67 (Fig. 4D). Together, these data suggest that protein-protein association between Sox9 and AP-1 family members reduced the transactivation activity of either complex evident in the response of target enhancers where direct binding sites are only present for one Sox9 or AP-1 factor. However, an additive transcriptional activity occurs when the DNA target contains binding sites for both Sox9 and AP-1. Consequently, the overall effect of AP-1 members on Sox9 regulation of chondrocyte targets is likely to be complex, reflecting the distribution of each factor's direct binding sites within the regulatory genome and non-DNA-dependent co-interaction.

To understand the functional significance of the interplay between Sox9 and AP-1 in chondrocyte differentiation, we utilized a drug, SR11302, that inhibits AP-1-mediated transcriptional activation (Fanjul et al., 1994). We confirmed first that SR11302 inhibited Jun-mediated activation of its target in a dose-dependent manner (Fig. S4E). Next, we analyzed the effects of SR11302 on the expression of a Col10a1 enhancer-driven mCherry reporter (Col10a1:mCherry) (Maye et al., 2011) in primary mouse transgenic chondrocytes. Col10a1 is expressed in conjunction with Jun and Fosl2 activation at the hypertrophic chondrocyte transition (Fig. 1). SR11302 downregulated Col10a1:mCherry reporter activity (Fig. 5A,B) and the endogenous levels of Col10a1 mRNA (Fig. 5C). In addition, SR11302 inhibited normal activation of an endogenous Col10a1:mCherry transgenic reporter in intact femur explants, reducing the resulting hypertrophic zone and restricting ossification (Chung et al., 2001) (Fig. 5D). Whereas the levels of Col2a1 (predominantly but not exclusively expressed in proliferating immature chondrocytes), Sox9, and a number of genes displaying similar expression levels within distinct chondrocyte cell types (Actb, Epyc and Ndufv2) were not significantly changed on SR11302 treatment, expression of prehypertrophic chondrocyte-enriched genes (Sp7, Mef2c and Ihh) was elevated (Fig. 5C; Fig. S4F). Collectively, these data support a model wherein AP-1 family engagement promotes a progressive hypertrophic chondrocyte program.

To test this model further, we co-expressed Fosl2 and Jun, or expressed TAM67 or a GFP control, in neonatal mouse primary chondrocytes. Ectopic expression of Fosl2 and Jun downregulated expression of markers of proliferative and prehypertrophic chondrocytes (Col2a1, Sox9, Sp7, Mef2c and Ihh), but upregulated Col10a1, an early indicator of hypertrophic development (Fig. 5E); expression of Mmp13, a marker of the most mature mineralizing hypertrophic chondrocytes that do not show either Jun or Fosl2 activity (Fig. 1A,A′,B,B′,H), was unaltered on the ectopic expression of Fosl2 and Jun (Fig. 5E). Ectopic expression of TAM67 exerted opposite effects on expression of these genes to those invoked by Fosl2 and Jun (Fig. 5E). Next, we examined the effect of ectopic expression of Fosl2 and Jun on the developing zebrafish skeleton using Tol2 transposase for transposon-mediated transgenesis in the zebrafish embryo (Fig. 5F). Ectopic expression resulted in premature mineralization of endochondral cartilage elements (Fig. 5G-I). By contrast, mineralization was delayed on ectopic expression of TAM67 (Fig. 5G-I). Together, these data support findings from earlier gain-of-function and dominant-negative approaches in vitro and in vivo.

To examine Sox9 function in the hypertrophic program directly, we used an Sp7 (also known as osterix)-driven CRE transgenic mouse strain (Rodda and McMahon, 2006) to remove Sox9 activity specifically at the prehypertrophic transition, contrasting with other studies that have removed Sox9 from the columnar chondrocytes (Dy et al., 2012; Ikegami et al., 2011). Prehypertrophic chondrocyte-specific removal of Sox9 led to an expansion of the prehypertrophic chondrocyte zone (Fig. S5A,B) and pronounced loss of Col10a1-expressing hypertrophic chondrocytes (Fig. S5C,D; see also S5C′,D′ for higher magnification), providing evidence in support of Sox9 action in promoting progressive development of hypertrophic chondrocytes. In this, Col10a1 appears to be one likely shared target of Sox9 and Fosl2/Jun regulation from our primary ChIP-seq data (Fig. 2F) with strong binding to an in vivo-verified hypertrophic enhancer (Fig. 6A), transfection studies examining transcriptional responses in transfected primary rib chondrocytes (Fig. 6B), and data reviewed earlier (Gebhard et al., 2004; Leung et al., 2011; Li et al., 2011; Riemer et al., 2002).

To examine this possibility further, we tested the effects of Sox9 or Jun on reporter gene expression driven from the Sox9/Jun-bound Col10a1 enhancer (Fig. S6). Individually, each factor upregulated reporter activity, and reporter activity was further enhanced when both Sox9 and Jun were transfected together (Fig. 6C, pGL-Col10WT-enh). Consistent with the direct DNA-dependent interaction of each factor, Sox9 and AP-1 were no longer able to activate the reporter when their predicted binding sites were mutated (Fig. 6C, pGL-Col10SM-enh and pGL-Col10AM-enh). Collectively, these data suggest that Col10a1 is one key, direct target upregulated by Sox9 and AP-1 in hypertrophic chondrocyte development.

Our data argue for a complex interplay between Sox9 and at least two AP-1 family members, Jun and Fosl2, in chondrocyte development. Whereas Jun binds a large array of DNA regions independently of Sox9, the strongest Jun-bound regions closely overlap with the large majority of Sox9 DNA targets in developing rib chondrocytes. Two distinct modes of Sox9 and AP-1 factor co-engagement were observed with distinct transcriptional consequences. Sox9 can form protein-protein complexes with Jun and Fosl2. At enhancers where direct binding of only Sox9 or AP-1 complexes is possible, protein-protein interactions attenuate the direct binding factor's transcriptional activity. By contrast, when Sox9 and AP-1 engage together through their cognate DNA-target sequence, their combined actions elevate expression above levels induced by either factor alone. Given our recent findings (Ohba et al., 2015) and those of other groups (Bhandari et al., 2012; Liu and Lefebvre, 2015; Oh et al., 2014, 2010) that the transcription of Sox9-specific chondrocyte targets is governed by integrating information from multiple independent enhancer elements, the overall impact of AP-1 factor activation on any given Sox9/AP-1 target gene is not easy to predict as it will reflect the combined action of multiple binding activities within any given enhancer and the integration of information from multiple enhancers; the inhibitory action of protein-protein interactions could function to fine-tune expression of target genes. However, the balance of the evidence herein suggests that the combined action of AP-1 factor input promotes a progressive hypertrophic chondrocyte program with Sox9. In addition, AP-1 family members other that Jun and Fosl2 may be expressed in proliferating chondrocytes and participate in Sox9-mediated gene regulation at the mitotic phase, although this study has not addressed this possibility. Extensive studies have linked AP-1 family action to bone formation, bone modeling, joint formation and intervertebral disc development (see Introduction). However, there is only limited prior evidence suggesting AP-1 action in chondrogenesis. Ectopic expression of Jun and JunD in vitro in chicken sternum chondrocytes has been reported to inhibit hypertrophic gene expression (Kameda et al., 1997). By contrast, in vivo Fosl2 removal from mammalian chondrocytes is reported to reduce the hypertrophic chondrocyte zone and associated calcified matrix, consistent with a positive role in hypertrophic chondrocyte development (Karreth et al., 2004).

Though Jun mutants do display joint phenotypes we saw no obvious phenotype in the remainder of the neonatal skeleton (Kan and Tabin, 2013). By contrast, small molecule-mediated AP-1 factor inhibition provided evidence in favor of an AP-1 input enhancing hypertrophic development. However, the precise action of this small molecule inhibitor is unclear, as is the extent of AP-1 family member inhibition. RNA-seq analysis indicates multiple AP-1 family members are expressed in hypertrophic chondrocytes in addition to Jun and Fosl2 as studied here. The potential for significant redundancy among family members precludes establishing a clear genetic resolution to the full potential of AP-1 family regulatory action in chondrocytes.

In contrast to AP-1 factors, Sox9 has been extensively studied in early chondrogenenesis. Sox9 is absolutely required for specification of mitotic chondrocytes (reviewed by Lefebvre, 2002). However, genetic studies have been less clear on the role of Sox9 in later, post-mitotic, hypertrophic chondrocyte development. Early deletion of Sox9 from Prrx1- or Col2a1-expressing cells blocks initiation or ongoing chondrogenesis, respectively (Akiyama et al., 2002). Further, spatial and temporal refinement of Sox9 removal from chondrocytes has supported a continuing role in promoting the hypertrophic chondrocyte transition (Dy et al., 2012; Ikegami et al., 2011).

Interestingly, our data show that Sox9 nuclear levels are highest at the prehypertrophic boundary precisely where Jun and Fosl2 expression initiate. Further, Sox9 and AP-1 family members co-bind each other's cis-regulatory regions, suggesting that cross- and auto-regulatory interactions might contribute to their observed patterns of gene activity. Removing Sox9 activity at the point of the prehypertrophic transition in vivo demonstrated that Sox9 is essential for the cellular progression to hypertrophic chondrocytes, consistent with the published studies discussed above. Further, drug-mediated attenuation of AP-1 factor activity blocked hypertrophic chondrocyte development, suggesting that both Sox9 and AP-1 have positive roles in the hypertrophic component of the endochondral skeletal program. One prominent co-regulated target in our study is Col10a1: Col10a1 activity in primary chondrocytes in vitro depends directly on Sox9 and AP-1 factor engagement in agreement with transfection studies of AP-1 factors in cell culture (Gebhard et al., 2004) but contrasting with transgenic studies that show that loss of a single Sox9-binding site within the context of a Col10a1 enhancer element results in premature activation of reporter in proliferating chondrocytes (Leung et al., 2011). The differences between these studies most likely reflect the limitation of analysis of single binding events in single enhancer elements where the in vivo picture is multiple binding events in multiple enhancer elements.

Previous studies have argued that Gli factors facilitate Sox9 inhibition of Col10a1 expression in proliferating chondrocytes (Leung et al., 2011). Interestingly, hedgehog responsiveness, as indicated by upregulation of Ptch1 and Gli1 targets, is lost at the prehypertrophic chondrocyte boundary (St-Jacques et al., 1999). Thus, the hypertrophic transition may act as a regulatory switch point for Sox9 control of hypertrophic target gene activity removing negative interactions (Gli), and activating de novo AP-1 factors to promote hypertrophic gene activity with Sox9.

Parathyroid hormone-like hormone (Pthlh; also known as Pthrp) signaling through the Pthlh receptor is a key inhibitor of the hypertrophic chondrocyte program (reviewed by Kronenberg, 2003). Pthlh is reported to induce AP-1 factors and enhance their DNA-binding activity in vitro, and dominant-negative forms of Fos have been reported to abolish Pthlh-mediated suppression of chondrocyte maturation in culture (Ionescu et al., 2001). In addition, Pthlh specifically activates Sox9 by phosphorylating Sox9 at serine 181 through protein kinase A exclusively in prehypertrophic chondrocytes, although the role of phosphorylated Sox9 has not been well studied (Huang et al., 2001).

Collectively, our in vivo and in vitro studies provide several lines of evidence supporting a direct ongoing role for Sox9 in the early stages of the hypertrophic program. Further, the data suggest that the actions of Sox9 in promoting the hypertrophic phase are augmented by the co-regulatory input of AP-1 family members at enhancer modules, a regulatory mode initiated on chondrocyte commitment to the hypertrophic program.

These studies suggest that the cell type-specific actions of Sox9 and AP-1 family members are crucially linked to the timing of chondrocyte development. Our Sox9 ChIP-seq analysis of primary chondrocytes reflects a mixed population of chondrocytes at different stages of chondrogenesis. Consequently, the exact timing of Sox9 action on a given enhancer is unclear and will require a significant technological improvement to be able to address in vivo given the small number of cells present within any skeletal element. However, the close concordance of Jun engagement at DNA targets with Sox9 and prehypertrophic-specific expression of Jun suggest that Jun interactions follow those of Sox9. Interestingly, Jun binds to many more genomic regions than does Sox9, but these ‘Jun-only’-associated regions do not display an active enhancer signature.

In this view, Sox9 likely provides the chondrocyte-context to a Jun regulatory input in chondrogenesis. Jun binds to a broad array of enhancers, many of which, such as those regulating Col2a1, were active under Sox9 control at an earlier stage (Ohba et al., 2015), whereas others, such as Col10a1, switch on in hypertrophic cells. As most of the early-activated enhancers remain active in the early phase of hypertrophic chondrocyte development, Jun and Fosl2 complexes formed at the hypertrophic transition could provide an additional layer of regulatory input into these early-initiating enhancers while providing de novo activating input into Col10a1 enhancers.

Although most of the Jun-Sox9 shared peak regions can be accounted for by a direct interaction of at least one of these factors, a small number (16%) of them show no readily identifiable Sox9 or AP-1 motif. This may reflect degeneracy of target sites. Alternatively, these data may point to additional interactions with other transcriptional regulators. In line with this, we recovered additional motifs within the Jun ChIP-seq dataset that might reflect an indirect association of Jun to the chondrocyte genome through engagement with other transcriptional regulatory factors. Of note, our analysis of Sox9-binding sites recovered significant enrichment for Runx-, Nfat-, Ets- and AT-rich homeodomain protein-binding motifs (Ohba et al., 2015). Of these possible co-operating factors, Runx2 is known to play an essential role in hypertrophic chondrocyte development and is normally targeted by Jun (Ding et al., 2012; Li et al., 2011; Papachristou et al., 2005; Yoshida et al., 2004) and Runx2 is reported to upregulate both Col10a1 and Sox9 in the Col10a1 expression domain (Ding et al., 2012), suggesting an interacting, positive regulatory loop. However, counter evidence suggests Sox9 upregulates Bapx1 (Nkx3-2), a negative regulator of Runx2 (Yamashita et al., 2009). Further, protein-protein interactions have been documented between Sox9 and Runx2 that are reported to attenuate Sox9-mediated transcriptional activation of target genes, suggesting a complex regulatory interplay (Zhou et al., 2006). Future studies developing strategies to incorporate Runx2, Bapx1, Mef2c and other transcriptional regulators of the hypertrophic program will help to decipher the regulatory logic of this key developmental event in mammalian skeletogenesis.

All experiments in this study were carried out in strict accordance with the recommendation in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health, USA. The procedures were approved by the Institutional Animal Care and Use Committee of The University of Southern California (IACUC # 11892).

Cryosections of postnatal day 1 (P1) Swiss Webster mouse skeletal samples were subjected to in situ hybridization with digoxigenin-labeled riboprobes. See supplementary Materials and Methods for further details of sample preparation and riboprobes.

ChIP was performed on P1 mouse rib chondrocytes as previously described (Ohba et al., 2015), using specific antibodies against Jun. ChIP-seq libraries were constructed using ThruPLEX DNA-seq Kit (Rubicon Genomics, R400429). See supplementary Materials and Methods for further details of chromatin preparation, antibodies, and library construction.

ReChIP was performed by sequential precipitations of chromatin using specific antibodies against Sox9 and Jun. DNA was analyzed by conventional polymerase chain reaction (PCR) and quantitative PCR (qPCR). See supplementary Materials and Methods for further details of the reChIP procedure, antibodies, and PCR primers.

Sequencing was performed in the Epigenome Center of the University of Southern California. Data were analyzed as previously described (Ohba et al., 2015); alignment of sequence reads, peak calling, de novo motif analysis, and gene ontology analysis of genes associated with obtained peaks were performed using ELAND, CisGenome software (Ji et al., 2008; Jiang et al., 2010), MEME-ChIP (Bailey et al., 2009), and GREAT (McLean et al., 2010) and DAVID (Huang da et al., 2009), respectively. See supplementary Materials and Methods for further details of these analyses.

Ribs were digested with collagenase D from P1 Col10a1:mCherry transgenic mice. Fluorescence-activated cell sorting was performed to isolate RNA from 100,000 mCherry-positive chondrocytes. Sequencing was performed as mentioned above. Sequence reads were mapped with TopHat2 and quantified with Partek software. See supplementary Materials and Methods for further details of RNA isolation and sequencing library construction.

Plasmids expressing genes of interest were constructed by cloning of cDNA into mammalian expression vectors; cDNA was obtained by a PCR-based strategy. Luciferase reporter constructs were generated by cloning of annealed oligonucleotides into luciferase reporter vectors. Retroviral vectors were based on the pMX vector platform, and viral particles were produced in plat-E cells. See supplementary Materials and Methods for further details of cloning procedures.

Nuclear extracts of HEK293T cells transfected with plasmids were subjected to co-IP, followed by SDS-PAGE of precipitated materials and western blotting. See supplementary Materials and Methods for further details of nuclear extract preparation, co-IP reaction, immunostaining of blotted membranes, and image acquisition.

Chondrocytes were obtained from ribs of P1 Swiss Webster mice or Col10a1:mCherry transgenic mice by enzymatic digestion and subjected to cultures. mRNA expression was analyzed by reverse transcription (RT)-qPCR. See supplementary Materials and Methods for further details of cell isolation, cultures and mRNA expression analysis.

Femurs were obtained from embryonic day (E) 12.5 Col10a1:mCherry transgenic mice and cultured with 1 μM of SR11302 (R&D Systems, 2476) or in DMSO in α-MEM/10% fetal bovine serum with antibiotics (penicillin-streptomycin; Sigma-Aldrich, P4333). Culture medium was changed daily with fresh SR11302 or DMSO.

For zebrafish overexpression, Fosl2-P2A-Jun or TAM67 cDNAs, and E1 Col2a1 enhancer-reporter constructs were cloned into a Tol2 construct containing a zebrafish gata2 basal promoter (Kwan et al., 2007). Forty nanograms of each overexpression construct was injected together with 50 ng of Tol2 transposase into 1-cell-stage fertilized zebrafish embryos. The skeletal pattern was visualized by Alcian Blue and Alizarin Red staining of developing zebrafish (Walker and Kimmel, 2007). The extent of mineralization was quantified by measuring the red (Alizarin Red) pixels using ImageJ analysis of skeletal images.

Luciferase and effector constructs were transfected into HEK293T cells with FuGENE-6 transfection reagent (Promega, E2691). After 48 h, luciferase activities were measured with the Dual-Glo Luciferase Reporter Assay System (Promega, E2920). For transfection of cells cultured on 96-well culture plates, 100 ng of total DNA was mixed with 0.3 μl FuGENE-6 reagent in DMEM per well. In the DNA mix, 5 ng of a thymidine kinase promoter driving a Renilla control construct was mixed with 10 ng of the experimental firefly luciferase construct. Up to 5 ng of Sox9-expressing construct and up to equal amounts of Fosl2 construct and Jun construct were used as effectors (up to 70 ng combined). A GFP construct was added to make up each DNA transfection to 100 ng DNA per well.

Cryosections of E16.5 Sp7Cre;Sox9^+/+ or Sp7Cre;Sox9^flox/flox limbs were subjected to immunostaining using specific antibodies, in combination with secondary antibodies. See supplementary Materials and Methods for further details of sample preparation, antibodies and staining procedures.

We thank Jill McMahon, Gohar Saribekyan, Lora Barsky, Bernadette Masinsin, Seth Ruffins, Riana Parvez and Charles Meyer Nicolet from the USC Histology Core, the USC Flow Cytometry Core at the USC Imaging Center, and the USC Epigenome Center for providing technical assistance. We thank Dr Anton Valouev for his bioinformatics analysis assistance and Drs Toshiyuki Ikeda and Taku Saito for experimental materials. We are also grateful to Dr Gage Crump and Crump laboratory members for help with zebrafish studies, and Drs Henry M. Kronenberg, Clifford J. Tabin and Ung-il Chung for their helpful inputs.