The transcription factor scleraxis (Scx) is required for tendon development; however, the function of Scx is not fully understood. Although Scx is expressed by all tendon progenitors and cells, only long tendons are disrupted in the Scx−/− mutant; short tendons appear normal and the ability of muscle to attach to skeleton is not affected. We recently demonstrated that long tendons are formed in two stages: first, by muscle anchoring to skeleton via a short tendon anlage; and second, by rapid elongation of the tendon in parallel with skeletal growth. Through lineage tracing, we extend these observations to all long tendons and show that tendon elongation is fueled by recruitment of new mesenchymal progenitors. Conditional loss of Scx in mesenchymal progenitors did not affect the first stage of anchoring; however, new cells were not recruited during elongation and long tendon formation was impaired. Interestingly, for tenocyte recruitment, Scx expression was required only in the recruited cells and not in the recruiting tendon. The phenotype of Scx mutants can thus be understood as a failure of tendon cell recruitment during tendon elongation.
Development of the limb musculoskeletal system requires the coordinated differentiation and growth of the skeleton, muscles and tendons, however, the patterning cues and mechanisms that allow for integrated development are largely unknown. Although it is well established that the primary skeletal structures (i.e. the long bones) and muscles can form independently of one another, a growing body of evidence suggests that the formation of tendons requires signals from the adjacent skeletal and muscle tissues (Arvind and Huang, 2017; Blitz et al., 2009; Chen and Galloway, 2014; Edom-Vovard and Duprez, 2004; Huang, 2017; Kardon, 1998). Tendon function in integrating muscle and skeleton into a unified system suggests that the regulation of tendon growth may be a key mechanism by which the musculoskeletal system maintains coordinated growth to establish proportions optimal for mechanical tension and force transfer. Despite the functional importance of tendons, however, there is still very little known about the cell and molecular events that regulate their growth and development.
In the mouse limb, the distal skeletal segments are controlled by long tendons that extend from muscles in the arm (zeugopod) and insert into skeletal sites within the wrist or paw (autopod) (Watson et al., 2009). Autopod and zeugopod tendons are formed as separate modular units that are independently regulated, such that tendons in one segment can develop in the absence of tendons in the other segment (Huang et al., 2015; Kardon, 1998). While autopod tendon induction depends on cartilage, the zeugopod tendon components are first induced at E12.5 as short-range progenitors that anchor muscle to the skeleton, followed by tendon differentiation and elongation in parallel with skeletal growth. In the absence of muscle, zeugopod tendon induction is initiated, but subsequent differentiation and growth fails at E13.5 and the tendon anlagen are lost (Huang et al., 2015). This biphasic mode of long tendon development (anchoring and elongation) provides a useful framework at the tissue level, but the cellular dynamics and molecular regulators of this process have not been established.
One intriguing marker for tendon is the transcription factor scleraxis (Scx), which is expressed by all tendon progenitors and cells, yet its functional role in tendon development remains poorly understood (Murchison et al., 2007; Schweitzer et al., 2001; Yoshimoto et al., 2017). The Scx−/− mutant displays a strong tendon phenotype; however, not all tendons are equally affected (Murchison et al., 2007). The long-range tendons of the trunk and limbs are either missing or severely deficient, but short-range anchoring tendons appear normal. Importantly, the ability of tendon to connect muscle to skeleton is not impaired in the absence of Scx, as the Scx−/− mutant is viable after birth, despite limited mobility due to disruption of the long tendons. This suggests that Scx may be functionally important for tendon growth and elongation.
One paradigm underlying tissue and organ development follows a basic assumption that progenitor cells are specified at an early stage and that these cells then differentiate and proliferate to form the final tissue/organ (Gilbert, 2010). There is emerging evidence, however, suggesting that this simple paradigm is not applicable for all tissues/organs. In the joint, it has been shown that joint progenitor cells may be continuously induced and allocated (Shwartz et al., 2016). In this study, we show that tendon elongation in the mouse limb and tail is fueled by recruitment of new mesenchymal progenitors. We further find that Scx is not required for tendon anchoring, but Scx expression in mesenchymal progenitors is required for their recruitment during tendon elongation, underlying the disruption of long tendons in the Scx mutant.
Long tail tendons develop via establishment of a short-range tendon anlage followed by tendon elongation
Integration of the musculoskeletal tissues is first established at E12.5, when short-range anchoring tendons attach muscles to the skeleton. Subsequent elongation of the short-range tendon anlage into the zeugopod allows for coordinated development of all musculoskeletal elements during growth (Huang et al., 2015). Although this model was initially conceptualized based on observations in limb tendon development, the limb as a model system poses additional challenges for mechanistic studies due to the complex patterning of several tendons (which are initially formed as fused units that subsequently separate) as well as the modular nature of the autopod and zeugopod segments during tendon development (Huang et al., 2015; Watson et al., 2009). To bypass these complexities, we turned to the tail tendons as a simplified model system to identify the cell and molecular mechanisms underlying anchoring and elongation. Tail tendons are also advantageous because they are the longest tendons in the body and undergo the most dramatic growth during development.
The fully formed mouse tail consists of long tendons that extend from muscles that reside in the base of the tail to each vertebral bone within the tail (Fig. 1A) (Shinohara, 1999). Movement of each skeletal element is governed by dorsal and ventral tendons organized in quadrants (Fig. 1A,B). As all tendons extend from muscles at the base of the tail, the tendons that insert closest to the tip of the tail are much longer than those that insert closer to the base. To determine whether tail tendon development is governed by similar principles of anchoring and elongation, we examined the early stages of tail development. Whole-mount confocal imaging of E13.5 ScxGFP embryos showed repeating short-range ScxGFP+ tendon segments along the tail, attached to their respective vertebral structures (Fig. 1C). The shortest tendon segments were found closest to the tip of the tail (white arrow). Whole-mount immunostaining for myosin heavy chain for muscle showed that each of these short-range tendon segments is directly attached to muscle (Fig. 1C′,C″). Between E13.0 and E15.5, there is dramatic elongation of these short-range tendon segments to form the long-range tendons of the tail, concurrent with skeletal growth (Fig. 1D). These results show that long tail tendons also undergo a developmental process of tendon anchoring followed by tendon elongation, similar to the long tendons of the limb.
Sox9 lineage tracing shows tail tendon elongation is driven by recruitment of new tendon progenitors
Having established that long tail tendons undergo elongation, we next examined the cellular basis for tendon elongation and asked whether the cells driving tendon elongation are completely derived from the early population of anlage progenitors or whether there is subsequent recruitment of new cells. To address this issue, we used Sox9Cre to distinguish early tendon progenitors because it has previously been demonstrated that the cells forming the early skeletal insertion site are derived from bipotent Scx+/Sox9+ progenitors, established at E12.5 (Blitz et al., 2013; Sugimoto et al., 2013). The transient expression of Sox9 in early tendon progenitors and the consistent localization of Sox9 lineage (Sox9lin) tendon cells to regions near the skeletal insertion suggested that Sox9 in tendons may be a distinctive marker for the anchoring anlagen progenitors. We therefore examined the tails of Sox9Cre embryos combined with the Rosa26-TdTomato (RosaT) Cre reporter at E12.5. Imaging by whole-mount confocal microscopy and in transverse sections showed that the ScxGFP+ tail tendon anlagen are completely derived from RosaT+, Sox9lin cells, demonstrating that Sox9 is indeed a useful marker for the anchoring tendon in the tail (Fig. 2A,B-B″,C-C″).
As the early tail tendon anlagen is composed of Sox9lin progenitors, we used the inducible Sox9CreERT2 mutant to label these cells at E12.5 via tamoxifen injection and determined whether new tendon progenitors are recruited during tendon elongation. To evaluate the distribution of labeled cells along the tendon, we focused our analyses to two distinct regions in the tail in order to (1) follow individual tendons as they insert into the skeleton as well as (2) capture tendons well away from their skeletal insertions. As tail tendons are extremely long and numerous (Fig. 1), we determined that the best region to follow individual tendons as they insert was close to the tail tip (because most of the other tendons would have already inserted proximally). A series of consecutive transverse sections taken across the length of the tail tip at E16.5 showed that tendon cells further from the insertion (tendon 5, L1) are devoid of Sox9lin cells, while tendon cells near/at the insertion (tendon 1, L1 and L2; tendon 2, L3 and L4) are strongly composed of Sox9lin cells (Fig. 2D,E). To confirm that tendon cells away from the skeletal insertions are non-Sox9lin, we also collected sections closer to the tail base where the tendons attach to muscle. Analysis of these sections revealed that the bulk of the tendon midsubstance length was almost completely derived from non-Sox9lin cells (ScxGFP+, RosaT−), indicating that tendon elongation is likely fueled by recruitment of new progenitors and not by expansion of the original anchoring population (Fig. 2F,G).
As previous studies showed dramatic variability in the contribution of Sox9lin cells in tendons (Soeda et al., 2010), we also examined Sox9Cre-labeled tail tendons at E16.5 in transverse sections. To our surprise, Sox9Cre lineage tracing showed that the full length of dorsal tail tendons (from skeletal insertion to muscle origin) is almost entirely composed of Sox9lin cells (Fig. S1). Notably, ventral tail tendons were composed of largely non-Sox9lin cells, which suggested that Sox9 itself is likely not playing a functional role in tendon development (Fig. S1). This is also supported by other studies showing that deletion of Sox9 by ScxCre does not result in loss of tendon (Blitz et al., 2013). Collectively, the analyses of Sox9CreERT2 and Sox9Cre tail tendons indicate that the anchoring tendon is initially composed of Sox9lin cells, while new tendon progenitors are subsequently recruited during tendon elongation. In addition, these progenitors may be either Sox9lin or non-Sox9lin.
Conditional deletion of Scx shows that Scx is required in mesenchymal cells to enable their recruitment during tail tendon elongation
As both the anchoring tendon progenitors and the recruited mesenchymal cells in dorsal tendons are derived from Sox9lin progenitors, we next tested Scx function using Sox9Cre to target both cell populations in dorsal tendons. In Scx−/− mutants, tail tendons are severely disrupted. If Scx is required in anchoring progenitors, we expected ScxSox9Cre mutants to phenocopy Scx−/− mutants in that new cells will not be recruited, resulting in a failure of tendon elongation. Conversely, if Scx is required only in the recruited mesenchymal cells, then mutant cells will be excluded from the elongating tendon. When we examined transverse sections of tails from ScxSox9Cre; RosaT embryos at E18.5, we found that dorsal tail tendons were still present; however, the mutant tendons were much smaller than wild-type tendons (Fig. 3A-D). Analysis of RosaT expression showed that mutant dorsal tendons at the midsubstance were almost completely composed of non-Sox9lin cells (Fig. 3C,D). Remarkably, these data showed that out of a large population of mutant cells (RosaT+), only wild-type cells were capable of recruitment, suggesting that Scx is required in mesenchymal cells for their recruitment into tendon. Reduced dorsal tendon size in the mutant was likely a result of the restricted availability of wild-type cells in the surrounding milieu, because in wild-type embryos these tendons are fully composed of Sox9lin cells. Moreover, as the anchoring tendon is targeted by Sox9Cre, this result also indicates that Scx function is required exclusively in the recruited mesenchymal cells and not in the recruiting tendon, because loss of Scx in the anchoring tendons in the dorsal tail tendons of these mutants did not preclude cell recruitment. Interestingly, the size of ventral tendons (which normally recruit a mix of Sox9lin and non-Sox9lin cells) appeared relatively normal in ScxSox9Cre mutants, which may be due to the larger availability of wild-type cells in the vicinity of these tendons (Fig. S1).
Conditional deletion and embryonic cell injections show that Scx is required for tendon cell recruitment in the limb
Our experiments in the mouse tail showed that tendon elongation is fueled by recruitment of new cells and that Scx is required for their recruitment. To determine whether this is a general mechanism for long tendon growth, we evaluated the complex tendons within the limb by whole-mount confocal imaging. Initial analyses of Sox9Cre; RosaT; ScxGFP limbs showed broad Cre activity in a variety of tissues, including skin and nerves; as a result, RosaT expression was overwhelming in maximum intensity projections and did not reflect true Sox9lin cell contribution to tendon (Fig. S2). We therefore developed a non-biased approach using a customized algorithm in Matlab software to eliminate the RosaT signal in ScxGFP-negative cells, thereby restricting RosaT signal to tendon and related connective tissues (Fig. 4A,B). Analysis of Sox9lin cells in limb tendons of Sox9Cre; RosaT embryos at E12.5 revealed that the short-range tendon anlage that mediate connections between skeleton and muscle are mostly composed of Sox9lin cells, similar to our findings for the E12.5 tail. Interestingly, the short-range tendons that integrate the muscles in the zeugopod with their respective autopod tendons are also mostly composed of Sox9lin cells (Fig. 4A, Fig. S3). The Sox9lin population in tendons was largely restricted to the wrist where the muscles attached, and the autopod components of these tendons were composed of mixed or non-Sox9lin cells. When we evaluated Sox9Cre; RosaT embryos at a later stage (E16.5), we found a mixed composition of Sox9lin and non-Sox9lin cells in most tendons within the zeugopod, although the tendon regions near the wrist and elbow remained strongly RosaT+ for most tendons (Fig. 4B,C). Tamoxifen labeling of Sox9lin cells at E12.5 showed that most of the tendon at E16.5 was derived from recruited cells, similar to the distribution in the tail tendons (Fig. 4D). Deletion of Scx in Sox9lin cells using Sox9Cre resulted in a dramatic exclusion of mutant cells from the zeugopod tendons, similar to the phenotype observed in the tail (Fig. 5). Limb tendon diameters were noticeably smaller, although not to the same extent as the mutant dorsal tail tendons.
Because of the mixed composition of Sox9lin and non-Sox9lin cells in limb tendons, we tested the role of cell recruitment and Scx function via an additional experimental approach of injecting labeled mesenchymal cells into developing limbs in utero and testing the recruitment of such cells into developing tendons. To label limb mesenchyme, we used the Prx1Cre driver that targets all cells in the early limb bud mesenchyme. We isolated RosaT+ mesenchymal cells from Prx1Cre; RosaT; ScxGFP limb buds at E12.5, and directly injected these cells into the limb buds of stage-matched ScxGFP embryos via transuterine microinjection (Fig. 6A). As only donor cells are labeled, the presence of RosaT+ cells incorporated within a tendon would indicate cell recruitment. However, location alone was not sufficient for a RosaT+ cell to be considered recruited, unless the labeled cell also adopted a tenogenic fate by expressing ScxGFP. Systematic analysis of consecutive transverse sections taken from digit to elbow showed numerous examples of non-recruited donor cells embedded within the mesenchyme that were not associated with any tendon and did not express ScxGFP (white arrows, Fig. 6B,B′,C,C′). Occasionally, however, donor cells expressing ScxGFP were found within tendon, matching our criteria for recruitment (yellow arrows, Fig. 6B,B′). Out of 26 wild-type injected embryos, seven embryos had at least one positive recruitment event. Positive recruitment involved only one or two cells, and was not limited to any particular tendon or region within a tendon.
Having demonstrated cell recruitment into tendons using transuterine cell injection, we next wanted to test the role of Scx in this experimental paradigm. Injecting Scx−/− mutant cells into wild-type limbs would closely parallel the genetic experiments described above. However, as we found that injection of wild-type cells into wild-type limbs resulted in modest recruitment efficiency (WT-WT, ∼27%), a very large number of injections would be required to determine a definitive failure of the recruitment of Scx−/− cells. We therefore opted to perform the complementary experiment by injecting labeled wild-type cells into Scx−/− mutant limbs (WT-Mutant) (Fig. 6D). When wild-type cells were injected into mutant limbs, we found that all limbs (4/4) showed positive recruitment events. Moreover, although positive recruitment in the WT-WT experiments produced typically 1-2 cells within a tendon, positive recruitment events in WT-Mutant consisted of massive incorporation of wild-type cells (Fig. 6E′,E″,F,F′,G,G′). Whole-mount imaging showed extensive RosaT+ signal within tendon, which was confirmed via transverse sections showing large numbers of recruited cells for several consecutive sections, highlighting preferential recruitment of injected wild-type cells in an environment where most mesenchymal cells are Scx−/− and therefore cannot be recruited to the tendon. Collectively, these results indicate that cell recruitment is a driving force in tendon elongation and that Scx function is required in the mesenchymal cells that are recruited. The ability of Scx−/− tendons to recruit wild-type cells further shows that for cell recruitment, Scx activity is required exclusively in the recruited cells and there is no role for Scx activity in the recruiting tendon.
In this study, we show that the long tendons in the body are typically formed by anchoring and elongation, and further identify a fundamental mechanism underlying elongation. We show that additional mesenchymal cells are recruited after induction of the anchoring tendon, which fuels the dramatic extension of long tendons during growth. In addition, Scx expression in mesenchymal cells is required for their recruitment, thus establishing a new and crucial function for this essential tendon transcription factor (Fig. 7).
Scx was the first transcription factor identified for tendons, and to date remains the earliest known marker for distinguishing tendon and connective tissue progenitors during embryonic development (Chen and Galloway, 2014; Schweitzer et al., 2001). The early and distinctive expression of Scx initially suggested that Scx may be a master regulator of the tendon cell fate; however, analyses of the Scx−/− mutant subsequently revealed that tendons can still develop in the absence of Scx, although many of the tendons are grossly impaired (Murchison et al., 2007; Schweitzer et al., 2001). One confounding aspect of the Scx−/− mutant was the divergent phenotypes of two types of tendons, which can be generally classified as short-range and long-range tendons. Although short-range tendons (such as the intrinsic lumbrical tendons of the autopod or the flattened tendon sheets connecting the intercostal muscles to the ribs) are completely unaffected in the Scx−/− mutant, long-range tendons (such as the long tendons of the limbs and tail) are severely reduced or missing. In the context of long tendon formation, we now demonstrate that the Scx−/− phenotype is a failure of tendon elongation, and, more specifically, a failure of tendon cell recruitment during elongation. This is consistent with our previous results, which showed normal tendon induction at E12.5 in the Scx−/− mutant (during the stage of tendon anchoring), with the manifestation of the tendon phenotype at E13.5 (during the stage of tendon cell recruitment and elongation) (Murchison et al., 2007). In the current study, we also found that deletion of Scx in the early population of Sox9lin anchoring tendon progenitors at E12.5 did not negate the ability of these cells to subsequently recruit wild-type cells, further supporting the specific activity of Scx in the newly recruited mesenchymal cells.
Although we have identified this new and exciting function for Scx, it also plays other roles in tendon biology. For example, it has been shown that Scx drives expression of Col1a1, and collagen matrix content is greatly reduced in Scx−/− tendons, indicating that Scx may also have a later role in tendon matrix synthesis (Léjard et al., 2007; Murchison et al., 2007). The loss of the flexor digitorum profundus tendons in the metacarpal regions of the autopod also suggests a failure of progenitor cell condensation, independent of tendon elongation. Finally, it has previously been shown that Scx drives Bmp4 expression in bone eminence progenitor cells, which is required for cartilage differentiation to form the secondary skeletal structures, such as bone ridges and tuberosities (Blitz et al., 2013, 2009).
Although in this study we used the expression of Sox9 as a fortuitous marker of early tendon progenitor cells, we found no evidence for a functional role for Sox9 in either tendon cell anchoring or recruitment during elongation. This is consistent with previous reports, which showed that long tendons are not noticeably disrupted in Sox9ScxCre mutants, although the skeletal enthesis and tuberosities were lost. In both our tail and limb models, Sox9lin and non-Sox9lin cells can be recruited into tendons. However, although the composition of Sox9lin cells varied dramatically between tendons, the relative composition of Sox9lin cells was remarkably constant for each tendon. For example, the dorsal tendons of the tail are always highly composed of Sox9lin cells, whereas the ventral tendons are of mixed composition, suggesting that cell recruitment is likely driven by proximity of mesenchymal cells to the growing tendon. In cases where most surrounding cells are from Sox9lin, the tendon would be heavily composed of such cells. Consistent with this hypothesis, in the ScxSox9Cre mouse, where Scx mutant cells are no longer recruited, tendon size is reduced only in tendons that normally incorporate a large number of Sox9lin cells. Consequently, the tendon phenotype of ScxSox9Cre is significantly milder than that of Scx−/− tendons in which several long tendons are lost (including all of the long tendons of the tail), likely due to the ability of ScxSox9Cre tendons to still recruit wild-type cells and maintain structural integrity.
The long tendons of the limb and body are a common anatomic feature in mammalian species, but are not universal across tetrapod clads. In more primitive vertebrates such as amphibians, the zeugopod and tail are highly muscular and short tendons mediate the attachment of these muscles to the skeleton (Diogo et al., 2009) (Ashley-Ross, 1992; Diogo et al., 2009; Walthall and Ashley-Ross, 2006). While long tendons are observed in the amphibian autopod, the developmental program governing the autopod tendons is distinct from that of the long zeugopod tendons, and elongation of autopod tendons may involve divergent mechanisms (Huang et al., 2015). While cell recruitment appears to be a fundamental mechanism driving elongation, it is not clear whether recruitment is part of the original mechanism underlying all tendon formation or whether it is a later evolutionary addition specific to the new developmental program regulating long tendon elongation. From a functional perspective, the shift from short to long tendons may be advantageous. It has been shown that increasing the length of tendons relative to the length of muscle fascicles increases the energy storage capacity of tendons, enhancing muscle function while reducing the metabolic cost of muscle contraction (Alexander, 2002). Interestingly, cell recruitment does not occur as a substitute for tenocyte proliferation (which is sustained throughout the stages of tendon elongation), but is an additional mechanism for generating new cells. This indicates that proliferation alone may be insufficient for the tendon to keep pace with the growth of the underlying skeletal structures. The dramatic extension of the tendons during evolution may have necessitated a new, additional cellular mechanism for growth, to enable the remarkable increase (and variation) in limb length that occurred through mammalian evolution.
MATERIALS AND METHODS
Existing mouse lines used in these studies have been previously described: ScxGFP tendon reporter (Pryce et al., 2007), Sox9Cre (Akiyama et al., 2005), Sox9CreERT2 (Soeda et al., 2010), Scx−/− (Murchison et al., 2007), Scxf/f (Murchison et al., 2007), Prx1Cre (Logan et al., 2002) and Ai14 Rosa26-TdTomato reporter (RosaT) (Madisen et al., 2010). All mice were crossed with ScxGFP to enable visualization of tendon cells.
Limb bud cell isolation and transuterine microinjection into E12.5 limbs
Stage-matched timed matings were set for donor embryos (Prx1Cre; RosaT; ScxGFP) and host embryos (ScxGFP or Scx−/−; ScxGFP). On the day of injection, RosaT+; ScxGFP+ forelimb and hindlimb buds were isolated from E12.5 donor embryos and rinsed in sterile PBS. Limb bud cells were isolated by incubation in digestion solution (0.1% collagenase, 0.1% trypsin in PBS) at 37°C for 10 min. Cells were then resuspended in DMEM at a concentration of 1.5×106 cells/ml and kept on ice until injection. For injections, the uterine horns of pregnant dams were externalized by ventral laparotomy and transilluminated to visualize the host E12.5 embryo (Gubbels et al., 2008; Huang et al., 2013; Wang et al., 2012). Cell solution (10 nl) was microinjected through the uterus into the presumptive wrist region of the nascent limb. Host embryos with injected limbs were harvested at E14.5 and whole-mount limbs were visually scanned for the presence of RosaT+ signal, which is indicative of successful injection. Limbs showing any positive expression were then processed for histology and cryosectioned from digit tips to elbow. Every section was collected and inspected to identify recruited cells into tendon.
Standard protocols for whole-mount and section immunostaining were performed as previously described (Murchison et al., 2007). Patterning of limb and tail tendons, and muscles was acquired in serial transverse sections (Watson et al., 2009). A monoclonal antibody for My32 (Sigma-Aldrich M4276, 1:750) was used to detect muscle-specific type II myosin heavy chain (MHC).
Whole-mount confocal microscopy and image processing
Mouse forelimbs or tails from E12.5 to E15.5 were fixed in 4% paraformaldehyde overnight at 4°C. Following fixation, whole-mount immunostaining for MHC was carried out for E12.5 and E13.5 limbs and tails as previously described (DeLaurier et al., 2006). Tissues were then incubated in Sca/e2 solution at 4°C until cleared (Hama et al., 2011). A Zeiss LSM780 laser scanning confocal microscope was used to acquire 10 tiled z-stack images. Image processing was carried out using Zeiss Zen software to stitch tiles and obtain maximum intensity projection images. Transverse sections were imaged using the Zeiss AxioImager with Apotome for optical sectioning. As Sox9Cre labeling is broad in the limb, whole-mount confocal Sox9Cre; RosaT; ScxGFP images were processed using custom MATLAB software to eliminate non-ScxGFP contributions to better enable visualization of Sox9lin cells in tendon.
We thank Brian Pryce for helpful discussions and technical assistance.
Conceptualization: A.H.H., R.S.; Methodology: A.H.H., L.W., J.V.B., R.S.; Software: B.M.B.; Validation: A.H.H., B.M.B., J.V.B., R.S.; Formal analysis: A.H.H., J.V.B.; Investigation: A.H.H., S.S.W., L.W., J.V.B., R.S.; Resources: H.A., J.V.B., R.S.; Data curation: A.H.H.; Writing - original draft: A.H.H., R.S.; Writing - review & editing: A.H.H., R.S.; Visualization: A.H.H., R.S.; Supervision: R.S.; Project administration: R.S.; Funding acquisition: A.H.H., R.S.
This work was supported by the National Institutes of Health/National Institute of Arthritis and Musculoskeletal and Skin Diseases (R01AR055640 and R01AR055973 to R.S.; R01DC008595 to J.V.B.), by Shriners Hospitals for Children (85410-POR-14 to R.S.) and by the Arthritis Foundation (to A.H.H.). Deposited in PMC for release after 12 months.
The authors declare no competing or financial interests.