Molecular mechanism of synovial joint site specification and induction in developing vertebrate limbs Free

During vertebrate embryogenesis, the limb is initially perceptible as a small bud containing a single cartilaginous anlage that grows in length and undergoes repeated branching and segmentation, to eventually give rise to almost all of the mature skeletal elements. Segmentation of the limb skeletal anlage gives rise to three distinct segments: the upper limb (stylopod), the lower limb (zeugopod) and the hand/foot (autopod) (Oster et al., 1988). Across vertebrates, the length ratios of these segments considerably vary to accommodate functional locomotor and positional behavioral differences between species (e.g. human versus bat). The length ratios of each of these three distinct segments depend on the correct specification of a segmentation site or future joint within the cartilaginous skeletal template. It is well known that the proximal joint (i.e. elbow or knee) is determined first with more distal joints (i.e. wrist or ankle) forming later. This progressive specification is linked to the proximo-distal (P-D) patterning of skeletal elements, as proposed by the two-signal model (Tabin and Wolpert, 2007), albeit the precise mechanism is unclear.

According to the two-signal model, limb P-D patterning is determined by reciprocal antagonism between two diffusible signals: retinoic acid (RA), which is secreted from the limb flank, and fibroblast growth factors (FGFs), which are secreted from the distal epithelial structure called the apical ectoderm ridge (AER). Mutual antagonism, coupled with growth-induced separation of these two signaling centers, leads to dynamic changes in gene expression profile, eventually resulting in nested expression of a set of transcription factors along the proximo-distal axis of the limb – Meis1/2 (proximal), Hoxa11 (intermediate) and Hoxa13 (distal) (Mercader et al., 1999; Nelson et al., 1996). It was speculated that a joint is induced at the border of the expression domain of two adjacent segment markers (Galloway et al., 2009). Lineage tracing experiments in chick and mouse have failed to demonstrate a strict correlation between the expression of a proposed segment marker and the founder cell population for the segments (Delgado et al., 2020; Sato et al., 2007; Scotti et al., 2015). Such non-congruity has also been reported in invertebrate limb development (Milán and Cohen, 2000). Furthermore, Meis1/2 compound loss of function abolishes condensation of the stylopod, whereas zeugopod development is largely impaired (Delgado et al., 2020). On the other hand, the loss of Hox11 activity leads to the formation of dramatically short zeugopodial elements, due to a failure in the formation of the normal growth plate at the two ends of zeugopod bones. However, the zeugopod primordia are specified normally (Boulet and Capecchi, 2004; Fromental-Ramain et al., 1996). Taken together, it appears that none of these genes is individually important for specifying segment identities. Furthermore, the recent data from our group and earlier studies show that the expression of Meis1/2 precedes the expression of Hoxa11 and Hoxa13 in the developing limb bud, and double conditional knockout of Meis1/2 alters the expression domains of Hoxa11 and Hoxa13, suggesting that Meis1/2 dictates the establishment of the proximo-distal pattern of the limb bud (Delgado et al., 2020; Mercader et al., 1999; Nelson et al., 1996).

Expression of several known joint markers, such as Gdf5, Atx and Jun, has been reported at the segmentation sites, but how the segment-specific homeobox genes induce joint structures at their boundaries, if at all, remains unknown (Kan and Tabin, 2013; Storm and Kingsley, 1999). Finally, to date, no gene has been identified to induce the characteristic features of a joint, including the flattening of cartilage cells and eventual segmentation.

To investigate how segmentation sites (future joints) in the limbs are specified, we modelled our work on findings from vertebrate somitogenesis (Dubrulle and Pourquié, 2004) due to the remarkable similarity in segmentation and the signaling gradients across a tissue. During somitogenesis, single contiguous pre-somitic mesoderm (PSM) undergoes repetitive segmentation events to form discrete somites along the anterior-posterior (A-P) axis. This process is controlled by reciprocal gradients of RA and FGF signaling, which dictate the sites of PSM segmentation in the AP axis wherein RA signaling induces differentiation and FGF signaling keeps the PSM cells in a proliferative state (Dubrulle and Pourquié, 2004). Therefore, we hypothesized that, akin to the segmentation of presomitic mesoderm, the reciprocal RA and FGF signaling gradients in the developing limb bud regulate the position of the putative joint site (Fig. S1).

In this article, we identify the homeodomain-containing transcription factor Barx1 as a molecule that is expressed in all the presumptive interzone sites in a developing limb, and its misexpression is sufficient to induce an ectopic interzone. Furthermore, we show that RA and FGF signaling gradients, together, specify the expression domain of Barx1 in the first presumptive interzone site, and thereby specify the knee joint site. We demonstrate that the Barx1 expression domain is not induced at the border of Meis1 and Hoxa11 expression domains at the early stages of limb development. However, subsequent to interzone formation, the Barx1 expression domain is coincident with the border of Meis1 and Hoxa11 expression domains. Our data suggest that RA-FGF signaling gradients independently control the expression domains of the putative segment markers and Barx1; only after interzone induction are these domains are further refined to demarcate specific segments and their borders.

We perturbed the RA-FGF signaling gradient using the classical bead implantation technique. We first implanted individual beads soaked in varying concentrations of RA or DEAB (reversible RA signaling inhibitor) in the proximal end, or FGF8 or SU5402 (FGFR1 inhibitor) in the distal end in HH20-21 chick hind limb buds (Hamburger and Hamilton, 1951; Mohammadi et al., 1997; Sachidanandan et al., 2008). We investigated the lengths of the femur (stylopod) and tibia (zeugopod), and their length ratios after HH30, i.e. once all the skeletal elements are defined. Implantation of individual beads gave normal limbs with no significant change in femur-tibia length ratios (Fig. S2A-C,A′-D′). Higher doses of SU5402 (250 µM or above) resulted in limb truncations (Fig. S2D).

To further explore the roles of these factors, we next attempted to perturb both signaling gradients such that the activity of one signaling pathway was raised at one end while simultaneously reducing the activity of the other signaling pathway at the other end. Through the use of progressively lower concentrations of individual chemicals, we eventually optimized pairs of concentrations such that only length ratios are altered without causing perceptible skeletal abnormalities. Simultaneous implantation of a RA-soaked bead (10 µM) in the proximal end and a SU5402-soaked bead (50 µM) in the distal end resulted in the treated limbs having similar overall lengths to the contralateral control limbs. However, the treated limbs exhibited a higher femur-to-tibia length ratio compared with the contralateral control limbs (Fig. 1A,A′). The change in the femur-tibia length ratios was observed due to the increase in the length of the stylopod and simultaneous decrease in the length of the zeugopod (Table S1). In contrast, implantation of a DEAB-soaked bead (500 µM) in the proximal limb and an FGF8-soaked bead (0.1 µg/µl) in the distal limb resulted in lower femur-to-tibia length ratios compared with the untreated contralateral limb, without any gross malformation or overall size differences compared with the treated limbs (Fig. 1B,B′). This change in the femur-tibia length ratios was observed due to the decrease in the length of the stylopod and simultaneous increase in the length of the zeugopod (Table S2). Thus, manipulation of RA-FGF signaling gradients alters the relative location of the proximal-most joint site in the limb skeletal primordia.

Changes in the ratio of the femur to tibia length may be caused by either a shift in the segmentation site of the presumptive knee joint in the early developing limb or by differential growth rates of the skeletal elements during late limb development, or a combination of both (Cooper et al., 2013). To distinguish between these possibilities, the identification of an early molecular marker of segmentation was required. The segmentation site in the cartilage anlagen is a morphologically conspicuous band of dense flat cells, referred to as the interzone. Removal of the interzone leads to the failure of segmentation and subsequent loss of joint formation (Holder, 1977). Molecularly, the interzone is characterized by the expression of genes such as Gdf5 and Atx (Enpp2) and by the loss of Col2a1. However, the most frequently used markers of interzone, such as Gdf5 and Atx, are typically expressed in broader domains, which become restricted only once the interzone forms (Ray et al., 2015; Storm and Kingsley, 1999). Furthermore, none of the reported joint cartilage markers is known to induce the interzone in the developing cartilage anlage. In the context of a screen to identify genes involved in early joint development, we identified Barx1 as an early interzone marker, the expression of which is sustained in late joints (Singh et al., 2016). This homeobox-containing gene was earlier identified to be expressed in the developing facial primordia, stomach and developing joints (Barlow et al., 1999). We found that Barx1 expression can be detected as early as HH21-22, before segmentation occurs, in a region described as the ‘opaque patch’ or putative interzone (Fell and Canti, 1934) (Fig. 2). In corroboration, multiple Barx-binding sites were identified in an upstream enhancer region for mouse and human Gdf5, the mutation of which resulted in the loss of joint-specific expression of LacZ in developing joints in mouse (Chen et al., 2016). Using the chick system, we found a similar effect (Fig. S3). Because, in the developing cartilage, Gdf5 is exclusively expressed in the interzone, our data( taken together with existing literature) suggest a potential conserved role for Barx1 in interzone induction. However, the role of Barx1 in the segmentation of limb cartilage or joint induction remains unexplored.

To test the role of Barx1 in interzone induction, we first cloned the full-length Barx1 gene and its constitutively active version, Barx1-VP16, into avian retroviral vector RCAS and then electroporated the right hind limb bud of HH14 chicken embryos with either RCAS-Barx1 or RCAS-Barx1-VP16 (Logan and Tabin, 1998). Barx1 gain-of-function constructs produced grossly malformed limbs (Fig. 3A,F). Because of the severe shortening of the electroporated limbs, variations in the region and depth of infection in the developing shaft of the skeletal elements, and variations in depth of tissue under observation, it is nearly impossible to accurately identify the correct comparable regions of control limb for scrutiny. Therefore, experiments were designed to efficiently use internal controls that would eliminate contralateral and test limb mismatch problems. Internal controls have been an established method for analysis in similar studies (Hartmann and Tabin, 2001; Kumar et al., 2018). Infected cells were detected by 3C2 immunoreactivity and expression of marker genes was compared between the 3C2-positive infected cells (Fig. 3B,E,F′,I) and immediately neighboring 3C2-negative uninfected cells. Chondrocytes infected with RCAS-Barx1 or RCAS-Barx1-VP16 presented with a dense, flattened interzone-like morphology (Fig. 3C,E′); they expressed Atx (Fig. 3D,G) and expression of Col2a1 was downregulated (Fig. 3B,E″,F″). Moreover, right hind limb buds co-electroporated with both RCAS-Barx1-VP16 and the aforementioned Gdf5-LacZ reporter vector, showed increased LacZ activity at these sites (Fig. S4), suggesting a role for Barx1 in Gdf5 expression as well as in joint induction. Phenotypes observed upon RCAS-Barx1-VP16 electroporation (Fig. 3E-F″) were more pronounced than with RCAS-Barx1 electroporation (Fig. 3A-D). Although RCAS-Barx1-VP16-infected cells (Fig. 3I) expressed Gdf5 (Fig. 3H), we could not detect the same activity in RCAS-Barx1-infected cells. In the contralateral limbs, Atx and Gdf5 expression was observed in most of the interzones (Fig. 3J,K), and the length of the contralateral limb skeleton was substantially greater than the limb skeletons electroporated with Barx1 gain-of-function constructs (compare the scale bars in Fig. 3J,K with this in Fig. 3F′,F′′). In addition, although in the contralateral limb skeleton, these markers were expressed in domains parallel to each other, in the RCAS-Barx1 infected skeletons, the expression pattern was irregular and confined to the infected domains. In order to test the necessity of Barx1 in interzone induction, we attempted to knockdown Barx1 levels using RNA interference, following methods described by Singh et al. (2018), and using CRISPR- mediated knockout in chick (Cong et al., 2013) and mouse (Roh et al., 2018). However, none of these approaches worked. The Barx1 ORF has 69.1% GC content, with some long stretches having up to 80% GC content. For this work, we could not RT-PCR amplify the Barx1 ORF. We had to synthesize the gene chemically. It is possible that RNA interference- or CRISPR-mediated gene knockout failed due to the high GC content in the ORF (Reynolds et al., 2004; Shojaei Baghini et al., 2021). Therefore, to investigate the effect of loss of Barx1 function, we cloned a dominant-negative version of Barx1 (Barx1-ENGR, created by in-frame fusion of Barx1 with Engrailed transcriptional repressor domain) in the RCAS backbone and performed electroporation as described previously (Kamei et al., 2011). Barx1-ENGR mis-expression in the interzone region exhibits mutually exclusive domains of Barx1-ENGR expression and interzone markers, which indicates that Barx1-ENGR mis-expression leads to the loss of expression of Gdf5 and Atx in the infected cells of the interzone. However, Barx1-ENGR failed to upregulate Col2a1 expression in the infected cells and did not convert flattened interzone cells to a rounded morphology (Fig. S5). Therefore, it appears that Barx1 is crucial for interzone formation. However, further experiments are required to assess its exact role.

We next investigated whether perturbation of RA-FGF signaling gradients resulted in the shift of the Barx1 expression domain around the time of interzone formation (HH24-25). Following the protocol described above, simultaneous implantation of RA- (10 µM) and SU5402-soaked (50 µM) beads shifted the Barx1 expression domain distally, i.e. away from the body flank, whereas implanting DEAB- (500 µM) and FGF8-soaked (0.1 µg/µl) beads shifted the Barx1 expression domain proximally, i.e. closer to the body flank, in HH24-25 treated limbs when compared with their respective contralateral control limb (Fig. 4A-B″). The change in the ratios was observed due to reciprocal changes in the lengths, both proximal and distal to the Barx1 expression domain (Tables S3 and S4). The observed shifts in the Barx1 expression domain are in line with changes in skeletal element length ratios observed at later developmental stages, and suggest that shifts in the segmentation site upon perturbing RA-FGF signaling gradients at the early stage contribute to altered length ratios observed at later developmental stages.

The question that remains unanswered is how is the joint site specified? Two possibilities exist, which are not necessarily mutually exclusive: (1) segments are determined first and joints are induced where the segments meet; or (2) joints are determined first and segments are later specified on either side of the joints. In order to investigate these possibilities, we used the third generation in situ hybridization chain reaction (Choi et al., 2018), and examined Barx1 expression with reference to the domains of expression of known putative segment markers Meis1, Hoxa11 and Hoxa13 (Fig. 4C-C″). At HH24 or earlier, there is no sharp boundary between Meis1 and Hoxa11 expression domains. We next investigated the relative expression domains of the putative segment markers and Barx1 in untreated limbs and in limbs where RA-FGF signaling was perturbed. In unperturbed HH22 and HH24 embryonic chick limb bud contexts, the Barx1 expression domain normally resides immediately adjacent to the distal boundary of Meis1 expression and within the Hoxa11 expression domain, while at HH28, the Barx1 expression domain is at the border of Meis1 and Hoxa11 expression domains. Simultaneous implantation of RA- and SU5402-soaked beads at HH20-21 led to the distal expansion of the Meis1 expression domain, resulting in partial overlap of the Barx1 and Meis1 expression domains. In contrast, implantation of DEAB- and FGF8-soaked beads at HH20-21 led to the retraction of Meis1 expression proximally, resulting in an increase in the gap between Barx1 and Meis1 expression domains in HH24 chick limbs (Fig. 4D-E′ and Table S5). The uncoordinated change in relative expression domains of Meis1 and Barx1 suggests that the expression of the putative segment markers and the location of interzone induction are not tightly coupled. It appears that initially the expression of putative segment markers and Barx1 are independently regulated; once the interzone is formed, the Barx1 expression is confined to the border of Meis1 and Hoxa11 expression.

We observed that the domain of Barx1 expression relative to the limb field alters depending on the manipulation of RA and FGF signaling gradients. One of the ways that it can happen is if the manipulation results in differential cell proliferation in the proximal and distal regions of the limb bud. To investigate this, RA-FGF signaling gradients were perturbed, as described above, and embryos were harvested 24 h after bead implantation. The number of proliferating cells in the proximal and the distal regions of the limb bud, as judged by phosphohistone 3 immunoreactivity, was analyzed. We observed that perturbing RA-FGF signaling gradients leads to generic increase or decrease in cellular proliferation that is consistent along the proximo-distal axis (Fig. S6 and Table S6).

Finally, we sought to address the question of how RA-FGF reciprocal gradients specify the location of the Barx1 expression domain. In maxillary primordia, FGF signaling positively regulates Barx1 expression (Barlow et al., 1999; Tucker et al., 1998). However, in the limb bud, Barx1 is expressed at a distance from both FGF or RA signaling sources, suggesting that, in this context, neither FGF nor RA signaling alone can induce Barx1 expression (Fig. 2A). To investigate the regulation of Barx1 expression by RA or FGF, we implanted RA- and FGF8-soaked beads individually close to the putative Barx1 expression site in HH22-23 limbs. We observed that within 6 h of bead implantations, the endogenous domain of Barx1 expression was abolished, suggesting high activity of either RA or FGF signaling inhibits Barx1 expression (Fig. 5A,B and Table S7). We produced similar results upon upregulating RA signaling by in ovo electroporation of constitutively active RAR or by upregulating FGF signaling by in ovo electroporation of RCAS-FGF8 (Fig. S7A,B and Table S7). We next downregulated RA signaling by in ovo electroporation of dominant-negative RAR and FGF signaling by in ovo electroporation of dominant-negative MAPKK or diffusible FGFR1 (Delfini et al., 2005; Novitch et al., 2003; Sen et al., 2005). We observed that each of these manipulations downregulated Barx1 expression (Fig. 5C,D and Fig. S7C). We further corroborated this finding, in part, using the mouse model system, where we found that the expression of a Barx family member was abolished in Meis1/Meis2 double conditional knockout mice (Fig. S7F,F′) (Delgado et al., 2020). Because a high dose of RA or FGF signaling abolishes Barx1 expression and loss of RA or FGF signaling or a RA downstream transcription factor (Meis1/2) also abolish Barx1 expression, it appears that the induction of Barx1 expression requires a specific low threshold level of both RA and FGF signaling. To test this, we attempted to manipulate RA and FGF signal gradients in such a manner that no overlapping domain of RA and FGF signals existed in the developing limb bud. We therefore soaked beads in varying concentrations of DEAB and SU5402, simultaneously implanted them at the sites of their action in HH22-23 limb buds, and then monitored Barx1 expression after 12 h of treatment. Contrary to our expectation that the absence of signals would abolish Barx1 expression, we observed that implantation of DEAB- (500 µM) and SU5402-soaked (50 µM) beads not only kept endogenous Barx1 expression intact, albeit at a lower intensity, but also led to ectopic expression of Barx1 near the SU5402 source (Fig. 5E,G,H). Furthermore, we observed precocious Barx1 expression after 6 h of implantation of DEAB- (500 µM) and SU5402-soaked (50 µM) beads in HH20 limb buds (Fig. 5F). For our hypothesis to be true, i.e. that low levels of RA and FGF signaling are required for Barx1 expression, there should be a low level of RA signaling, far from the source of RA, in the distal limb where SU5402-soaked beads are implanted. To investigate this, in ovo electroporation of the RA-signaling reporter RARE-AP was performed. We observed that the RA signaling domain extends beyond the first segmentation site and/or the expression domain of Meis1, a known downstream target of RA signaling in developing chick limb bud. Inhibiting the activity of the Raldh2 enzyme by implanting a DEAB-soaked bead reduces the level of RA signaling activity in the distal limb bud (Fig. S8). Taken together, we propose that a narrow range of low concentrations of both RA and FGF signaling is required for Barx1 expression.

When embryos implanted with DEAB- (500 µM) and SU5402-soaked (50 µM) beads at HH22-23 are allowed to develop to later developmental stages, we observed a kink and reduced ossification in the zeugopod at the SU5402-soaked bead implantation site. Such limbs also displayed features of ectopic/expanded interzone formation, as seen by lowered levels of cartilage-specific Safranin-O dye staining, cell-flattening and ectopic expression of Atx and Gdf5 around the SU5402 source (Fig. 6). Even around the endogenous interzone, a kink in the skeletal anlage forms and reduced ossification are observed. These data provide further support that Barx1 expression is sufficient for the induction of interzone-like features.

Neither the factor(s) that induces cartilage segmentation in a developing limb-bud nor the molecular mechanisms that dictate the sites of expression of such a factor have been identified to date. Here, we began by considering somitogenesis as a paradigm, and initially arrived at the well-known RA and FGF signaling pairs as candidate factors to investigate further. However, limb segmentation is distinct from somitogenesis. For example, as segmentation of pre-somitic mesoderm occurs in a periodic manner from anterior to posterior, with periodicity being maintained by a known molecular clock involving an intricate network of factors (Jouve et al., 2000; Palmeirim et al., 1997), no detailed clock has been reported in the formation of limb joints (Pascoal et al., 2007). Furthermore, despite several attempts, we could not detect the presence of active Notch signaling during limb cartilage segmentation, which is known to be responsible for pre-somitic mesoderm segmentation (Sato et al., 2002).

Nevertheless, taking hints from molecular networks/gradients involved in the segmentation of pre-somitic mesoderm, we separately individually or simultaneously perturbed the RA or/and FGF signaling gradients across the early limb. Only when we perturbed both signaling pathways simultaneously, e.g. when one is upregulated while the other is downregulated, did we observe significant changes in femur-to-tibia length ratios while keeping the overall limb lengths similar. The absence and presence of changes in skeletal element length ratios upon individual and dual bead implantations, respectively, is possibly due to the action of a known compensatory, mutually antagonistic relationship between these pathways (del Corral et al., 2003; Moreno and Kintner, 2004; Vermot et al., 2005; Yashiro et al., 2004). In previous studies in which only RA or FGF signaling was perturbed, limbs exhibited abnormalities (Mariani et al., 2008; Yashiro et al., 2004). However, perturbing the same signaling pathways by bead implantation, as carried out by us, did not induce limb abnormalities. These differences in phenotypes can be attributed to the duration over which the signaling pathways were perturbed, i.e. permanent for genetic knockout versus transient and short-lived for bead implantation. However, at a very high concentration of RA (i.e. specifically when the beads were placed in the distal end of the limb bud) or SU5402, we did observe limb deformities. We experimented with numerous combinations of RA/FGF agonists/antagonists at different concentrations to arrive at the specific combinations where the manipulations resulted only in shifting of the joints site and no other gross developmental abnormalities.

In this study, we relied on the gene Barx1 as an early interzone marker. Previously, it has also been reported that in mouse mandible cultures, FGF signaling positively regulates Barx1 expression, and that ectopic activity of Barx1 leads to the transformation of incisors to appear more molar-like (Tucker et al., 1998). Interestingly, in the zebrafish model, genetic manipulation experiments of barx1 demonstrate that Barx1 represses the formation of the jaw joint and promotes cartilage development in the craniofacial skeleton (Nichols et al., 2013). This observation is further supported by Barx1 knockdown experiments that lead to the upregulations of expression of joint markers such as Gdf5 and Chrd (chordin), and the downregulations of osteochondrogenic markers, such as Col2a1 and Runx2a, and the odontogenic marker Dlx2b (Sperber and Dawid, 2008).

Therefore, to investigate the role of Barx1 in joint induction in chick embryonic appendicular skeleton, we took advantage of gain- and loss-of-function studies. Through ectopic upregulation of Barx1 activity, we found ectopic interzone-like structures and ectopic expression of interzone markers. Thus, our data strongly suggest that at least in the developing chick appendicular skeleton, Barx1 is sufficient to induce the interzone. This observation is in stark contrast with observations in the zebrafish craniofacial skeleton. Although the reason for this difference is not immediately understood, it may relate to the different tissue lineages underlying the development of each structure, with the craniofacial skeleton arising from neural crest cells, while the limb skeleton arises from lateral plate mesoderm cells.

Before Barx1, several other molecules, e.g. Gdf5, Wnt9a and Jun, were implicated in interzone induction during limb development. The mice bearing mutation in the Gdf5-coding region exhibited features of brachypodism, i.e. loss of digital joints (Storm and Kingsley, 1999). However, gain-of-function studies for Gdf5, in both chick and mice, did not induce ectopic interzone or joints, but suppressed endogenous joints (Storm and Kingsley, 1999; Tsumaki et al., 2002). Misexpression of Wnt9a, previously known as Wnt14, could induce expression of several interzone markers, such as autotaxin and Gdf5, and downregulate the expression of Col2a1 in the cartilage anlagen (Hartmann and Tabin, 2001). However, the chondrocytes affected by Wnt9a misexpression did not exhibit the characteristic flattened morphology of the interzone cells. A subsequent study where Wnt9a activity was depleted demonstrated no major effect on interzone initiation and/or joint morphogenesis (Später et al., 2006). Jun, a transcription factor belonging to the AP1 family, is expressed in an interzone-restricted manner. The loss of limb-specific Jun activity led to upregulated Col2a1 expression and downregulation of the interzone marker Gdf5. However, the interzone was specified normally in the Jun mutant mice. Joint development was largely normal, albeit with some defects (Kan and Tabin, 2013). Therefore, none of the genes described in the literature so far to have interzone-specific expression seem to be important for interzone induction per se. Barx1 is the first molecule identified that induces all the features of an interzone. However, misexpression of Barx1 does not lead to ectopic joint formation. Joint morphogenesis requires movement of the adjacent skeletal elements around each other, which in turn requires the involvement of muscles and appropriate attachment of tendons/ligaments, etc. (Rolfe et al., 2014; Singh et al., 2018). It is unlikely that Barx1 can induce appropriate patterning of the muscles and attachment of tendons/ligaments at the site of an ectopic interzone, leading to the morphogenesis of a fully functional joint.

Next, although we could not knock down or knock out Barx1 in chick or mouse, due to technical limitations, we were able to inhibit Barx1 function in the limb using a dominant-negative Barx1-EnR construct. Misexpression of this construct blocked the expression of joint markers such as Atx and Gdf5, but could not induce a reversal of cell shape (from flattened to round) or upregulate Col2a1. It is possible that by the time Barx1-EnR infection was established in limb mesenchymal cells, interzone specification had already occurred and the Barx1-EnR fusion protein could only block further joint differentiation.

There is a second possibility. As demonstrated, full-length Barx1 does not elicit the full range of phenotypes of interzone induction, but Barx1-VP16 does, suggesting that Barx1 needs other collaborating partners. On the other hand, Barx1-Enr blocks some of the features of interzone but not all, suggesting that there is functional redundancy with other proteins, possibly Barx2. As joint morphogenesis is a key feature for vertebrate development, it is likely that there are several layers of redundancy built in the program. Barx1 is certainly a key player but it appears that there are other molecules that play a key role in induction of joint site. Unfortunately, existing literature on Barx1 or Barx2 loss of function in mice does not shed any light on the necessity of Barx proteins in joint induction: Homozygous Barx1-null embryos survive only until E13 but are largely normal compared with wild-type littermates of the same stage (Kim et al., 2005); Barx2-null mice, which survive until adulthood, also lack any obvious joint phenotype (Olson et al., 2005). It is possible that a conditional mutant of Barx1 or a Barx1/Barx2 double mutant, driven by an early limb mesenchyme-specific Cre such as Prx1-Cre or Dermo1-Cre, could result in establishing the necessity of Barx proteins in interzone/joint induction.

In chick, whereas Barx1 is expressed in a joint-specific manner, Barx2 is expressed in a graded manner near the FGF signaling source, during early developmental stages. Conversely, in mice, whereas Barx2 is expressed in a joint-specific manner, Barx1 is expressed in a graded manner near the FGF signaling source, during early developmental stages (Fig. S7). Therefore, it is possible that there is role-reversal in the functions of Barx1 and Barx2 in chick and mouse, as has been seen for other gene orthologs between chick and mouse (Nieto, 2018).

In the chick system, the expression domain of Barx1 is farthest from both RA and FGF signaling centers, yet both gain-of and loss-of-function of either RA or FGF signaling abolish Barx1 expression. Therefore, it appears that a critical, yet low, threshold of both these signals is required for Barx1 expression. In this context, it should be noted that implantation of a DEAB-soaked bead in the proximal end of the limb, along with implantation of a SU5402-soaked bead in the distal end of the limb, resulted in the induction of ectopic Barx1 expression around the SU5402-soaked bead site, and the consequent differentiation of an ectopic interzone. Co-electroporation of a RARE-AP reporter construct along with pCAG-mCherry demonstrated that RA signaling extends close to the distal end. It is likely that, due to DEAB-soaked bead implantation, the RA signaling front receded proximally; with the simultaneous downregulation of FGF signaling, another zone was created that had optimum concentrations of both RA and FGF signaling, which led to ectopic Barx1 expression.

Analysis of putative enhancers (5 kb upstream) of chick Barx1 and Barx2, as well as mouse Barx1 and Barx2 revealed a closely spaced cluster of Meis1- and Fos-Jun (AP1)-binding sites (Fig. S9). Meis1 is a transcription factor acting downstream of the RA signaling, and AP1 is downstream of FGF signaling (Kim et al., 1998; Mercader et al., 2000). This supports our hypothesis that both RA and FGF signaling are required for Barx gene expression in chick and mouse. However, what remains unclear is why and how chick Barx1 and mouse Barx2 are expressed in joint-specific manners while chick Barx2 and mouse Barx1 are not. Expression of Barx2 in mouse is abolished in the absence of Meis1 and/or Meis2, suggesting that Meis genes are involved in the induction of Barx genes. However, in chick, at an early stage, HH22-24, Meis1 expression does not overlap the Barx1 domain (Fig. 4C,C′). ChIP-Seq analysis with Meis1, Meis2 and AP1, in both species, will be needed to gain further clarity.

Another molecular framework that we could have investigated would have been to test a hypothesis that the border between putative segment markers acts as a molecular organizer for the induction of joint sites. To this end, our observations suggest that the initial domains of expression of the putative segment markers do not precisely mark the future segments. This observation is in keeping with lineage tracing data reported earlier (Delgado et al., 2020; Sato et al., 2007). Yet we note that, at later embryonic stages, the expression of Barx1 does reside at the border of Meis1 and Hoxa11. Our data, taken together with existing literature, suggest that RA-FGF reciprocal gradients are involved in initiating the expression of the putative segment markers, as well as specifying the location of expression of the segmentation gene Barx1. Subsequently, once Barx1 induces the interzone, the expression domains of putative segment markers become restricted to the segments without any overlap and Barx1 continues to be expressed at the border.

Our data suggest that the expression of Barx1 and/or Barx2 and the putative segment markers is not tightly coupled. There may be several reasons for this. One possibility is that the threshold requirement of RA and/or FGF signaling for Meis1, Meis2, Hoxa11, Hoxa13 and Barx1/2 is different from each other. There is some evidence in the literature to support such a speculation. For example, RA can induce expression of Meis1 over a broad range of concentrations. On the other hand, FGF signaling inhibits Meis1 expression only when the level of RA signaling is low (Capdevila et al., 1999; Mercader et al., 2000). The regulation of Hoxa11 expression by RA signaling appears to be position dependent, i.e. RA inhibits Hoxa11 expression at its proximal border, whereas it promotes Hoxa11 expression at its distal border (Tabin and Wolpert, 2007). Furthermore, it has been speculated that the expression of Hoxa11 is independent of FGF signaling (Hashimoto et al., 1999; Vargesson et al., 2001). RA signaling inhibits Hoxa13 expression, whereas FGF signaling stimulates its expression (Hashimoto et al., 1999; Vargesson et al., 2001). Our experimental studies suggest that the expression of Barx1 requires a narrow range of low signaling levels of both RA and FGF. Additionally, Meis1 and/or Meis2 regulates the domain of Hoxa11 and Hoxa13. However, this regulation is not direct (Delgado et al., 2020). Therefore, the possibility exists that a distinct set of transcriptional regulators, in conjunction with downstream effectors of RA and/or FGF signaling are needed for the expression of Meis1, Meis2, Hoxa11, Hoxa13, Barx1 and Barx2, which in turn leads to de-coupling of expression of these genes.

Although the present study is limited to the specification of only the first joint, it is possible that the positioning of subsequently more distal joints of the limb might also be dictated by the interplay of RA and FGF signaling pathways. Newly formed joints express Raldh2 and may in turn act as new RA sources, similar to newly formed somites during somitogenesis (Dubrulle and Pourquié, 2004) (Fig. S10). As skeletal element lengths and their ratios are crucial parameters in affording appropriate mechanical advantages to species to meet locomotion and feeding needs across their life span (Schultz, 1944; Smith and Savage, 1956), our studies additionally posit that the doses of RA and FGF signaling must be subjected to strong evolutionary pressure in and across species. In future studies, this hypothesis should be tested.

Fertilized White Leghorn Chicken eggs were obtained from the Central Avian Research Institute of India (Bareilly, UP, India), from Chandra Shekhar Azad Agricultural University (Kanpur, UP, India) and from Ganesh Enterprises (Nankari, Kanpur, UP, India). Eggs were incubated at 38°C in a humidified chamber to be treated and/or harvested at specific stages of development, as assessed by Hamburger and Hamilton staging criteria (Hamburger and Hamilton, 1951).

Meis1/Meis2 double conditional knockout mice were generated by recombining Meis1 and Meis2 floxed alleles with Dll1^Cre, as previously described (Delgado et al., 2020). Mice were handled in accordance with Centro Nacional de Investigaciones Cardiovasculares (CNIC) Ethics Committee, Spanish laws and the European Union (EU) Directive 2010/63/EU for the use of animals in research. All mouse experiments were approved by the CNIC and Universidad Autónoma de Madrid Committees for ‘Ética y Biene-star Animal’ and the area of ‘Protección Animal’ of the Community of Madrid with reference PROEX 220/15.

Embryos were harvested in phosphate-buffered saline (PBS), hindlimbs were dissected and fixed overnight at 4°C in 4% paraformaldehyde (PFA) in PBS. Post-fixation the tissues were washed in PBS twice for 5 min each and then dehydrated to 100% ethanol via an ethanol gradient from 25% ethanol in PBS to 50% ethanol in PBS to 75% ethanol in water with 5 min incubation in each. The tissues were then dehydrated in 100% ethanol twice for 10 min each and then cleared in xylene. Next, the tissues were treated with 1:1 solution of xylene:paraffin wax for 20 min at 65°C and then kept in 100% paraffin wax at 65°C for 1 h, followed by treatment in 100% paraffin wax at 65°C for 8-9 h. The tissues were then embedded in fresh paraffin wax and kept at 4°C. Paraffin wax-embedded limbs were sectioned along the para-sagittal plane using microtome to obtain 5 μm sections on poly-L-lysine coated glass slides.

Paraffin wax-embedded tissue sections were initially baked on hot plate at 60°C for 1 h, followed by dewaxing in xylene three times for 5 min each. The sections were next treated with 100% ethanol twice for 5 min each and then rehydrated to PBS via an ethanol gradient from 75% ethanol to 50% ethanol to 25% ethanol for 5 min each.

The embryos were harvested and eviscerated in PBS and fixed in 95% ethanol for at least 2-3 days followed by an overnight fixation in 100% acetone. Next, the tissues were stained for 2-3 days with a mixture of one volume of 0.3% Alcian Blue solution in 95% ethanol:one volume of 0.1% Alizarin Red solution in 70% ethanol:one volume of glacial acetic acid:17 volumes of 70% ethanol. Post-staining, the tissues were cleared in 1% potassium hydroxide and photographed under a dissection microscope. Limbs were dissected out from the skeleton preparations and were carefully imaged from the same angle to avoid any foreshortening. The skeletal element lengths were measured by ImageJ software using Straight line tool. The skeletal element lengths were measured along their two ends. Femur-to-tibia length ratios were determined for both treated and contralateral control limb, and statistical significance of variation in length ratios was determined using a paired Student's t-test (two-tailed) method.

cDNA clones used to make digoxigenin-labeled antisense riboprobes generated by in vitro transcription are: Atx (ChEST 520a5), Barx1 (ChEST 52p17), Col2a1 (ChEST 197l15) and Raldh2 (564a12). The cDNA clones for FGF8 and Cyp26b1 were generated by cloning in pBS and TA backbones, respectively. RNA in situ hybridization was performed as described previously for chick (Singh et al., 2017).

cDNA clones used to make digoxigenin-labelled antisense riboprobes generated by in vitro transcription are same as discussed above. Whole-mount in situ hybridization experiments were performed as per the protocol of Wilkinson and Nieto (1993) . Limbs were dissected out and were carefully imaged from the same angle to avoid any foreshortening. Approximate center of Barx1 expression was determined as the point of highest Barx1 signal intensity. A line joining the point of the distal-most extremity of limb bud and approximate center of Barx1 expression was drawn and extended to the body flank. The lengths proximal and distal to Barx1 expression were measured using ImageJ software using Straight line tool. Ratio of lengths proximal and distal to Barx1 expression were determined for both treated and contralateral control limb, and the statistical significance of variation in length ratios was determined using a paired Student's t-test (two-tailed).

HCR probe sets and HCR amplifiers tagged with Alexa Fluor 488 against Meis1 and Hoxa13, with Alexa Fluor 546 against Hoxa11, with Alexa Fluor 594 against Barx1 and Dusp6, and with Alexa Fluor 647 against Gdf5 were commercially synthesized by Molecular Instruments. The chick tissues were cleared using the CUBIC method (Gómez-Gaviro et al., 2017) and HCR in situ hybridization was performed as described previously for chick (Choi et al., 2018). Images were acquired using Leica Stellaris 5 (Fig. 4C-C′′) or Leica MZ10F (Fig. 4D-E′).

Fig. 4D-E′ images had background fluorescence. Therefore, an intensity threshold of 30 for Fig. 4D,D′ and 50 for Fig. 4E,E′ for the signal in green (i.e. for Meis1 expression) was set using ImageJ software. A yellow line was drawn approximately tracing the domain with signal intensity above the set threshold. Points of maximum Barx1 expression intensity (using ImageJ software) and the proximal-most tip of the limb bud were marked (as white spots). A line passing through the point of maximum Barx1 expression intensity and approximately parallel to the tangent to proximal limb curvature at the proximal-most point was drawn (marked as red lines). A line perpendicular to the red line and joining the farthest point of the Meis1 expression domain was drawn (marked as white dotted line) to determine the expansion or regression of the Meis1 expression domain. The length of this line was determined using ImageJ software. Distances proximal and distal with respect to the Barx1 expression domain were represented with negative and positive values, respectively.

Safranin-O staining was carried out on de-paraffinized rehydrated sections that were counterstained in Hematoxylin and Fast Green, rinsed in acetic acid and stained with Safranin-O (0.1% in water). Sections were subsequently dehydrated and mounted in DPX Mountant (ThermoFisher Scientific, 18404).

The bright-field images were converted to black and white images, followed by inversion to greyscale images using ImageJ software. An oval-shaped selection of the same-sized area was used for both the control and treated limbs to measure Barx1 mean signal intensity. The statistical significance was determined using a paired Student's t-test (two-tailed).

The constructs, RCAS-Barx1 (1 µg/µl), RCAS-Barx1-VP16 (1 µg/µl), RCAS-Barx1-ENR (0.9 µg/µl), pCAG-dnRAR (1.5 µg/µl), pCAG-caRAR (1.5 µg/µl), pCIG-MKKdn (1.5 µg/µl), pSV1-dFGFR1 (1.5 µg/µl), RCAS-FGF8 (1.5 µg/µl), pGL3-RARE-AP (2 µg/µl), Hsp68-R37-LacZ (2 µg/µl) and Hsp68-R37 Mut4-LacZ (2 µg/µl) were combined with 0.5 µg/µl pCAGGS-mCherry and 1% Fast Green injected between the somatic LPM and splanchnic LPM using a microinjector. The embryos were initially lowered by removing 2-3 ml of albumin and a window was made to visualize the embryo under the vitelline membrane. This vitelline membrane was shorn near the hindlimb field and bathed in 100 μl of a sterile PBS+pen strep solution (ThermoFisher Scienific, 10378016). At stage HH14, the DNA and Fast Green dye mix was injected into the embryonic space between the somatic LPM and splanchnic LPM at a concentration of 2 μg/μl using a microinjector. As soon as the DNA was injected, a platinum cathode was placed within the albumin beneath the yolk sac while an L-shaped anode was placed in parallel to the embryo over the hindlimb field before electric pulses (10 V, 50 ms pulse-on, 950 ms pulse-off, five repetitions) were applied. Care was taken that the electrodes did not touch the embryo when the electric field was applied.

Simultaneous implantation of a RA-soaked bead (10 µM) in the proximal end and a SU5402-soaked bead (50 µM) in the distal end, or a DEAB-soaked bead (500 µM) in the proximal end and FGF8-soaked bead (0.1 µg/µl) in the distal end of HH20-21 chick hind limb buds was performed and the embryos were harvested 24 h after bead implantation i.e. at HH24-25. The limbs were embedded in paraffin wax and sectioned at 12 µm. Subsequently, immunohistochemistry for phosphohistone H3 (pH3) (Sigma, H0412), a mitotic marker, was carried out (as described by Singh et al., 2017). The numbers of pH3-positive cells were counted for the contralateral control limbs as well as the treated limbs. Equal number of sections (six for HH24 limbs, four for HH24-25 or HH25 limbs) from contralateral and treated limbs were analyzed to count pH3-positive cells. A line joining the distal-most tip of the limb bud and an approximate mid-point of proximal flank was drawn. The limb-field was split into proximal and distal halves by drawing another line parallel to the proximal flank and passing through the mid-point of the previously drawn line.

The avian retroviral vector RCASBP(A) was used to deliver constitutively active (cBarx1-VP16), full-length (cBarx1) and dominant-negative (cBarx1-ENGR) versions of chicken Barx1 (cBarx1), and were named as RCASBP(A)-cBarx1-VP16, RCASBP(A)-cBarx1 and RCASBP(A)-cBarx1-ENGR, respectively. Chicken Barx1 ORF was chemically synthesized and procured from GenScript cloned as pUC57-cBarx1 and thereafter subcloned into a RCAS vector.

AG1-X2 ion exchange resin beads (150-200 µm diameter; BioRad) were soaked for 45 min in retinoic acid (Sigma, R2625)/Diethylaminobenzaldehyde (DEAB, Sigma, D86256)/SU5402 (Tocris, 3300) dissolved in DMSO, achieving mentioned dilutions. Beads were briefly stained in Fast Green and rinsed three times in saline solution. Bead insertion was performed close to the body flank or most distal border of the AER in chick limbs at stages indicated. Control beads were incubated in DMSO alone and treated as described above with RA, DEAB or SU5402. Affi-gel blue beads (BioRad) were rinsed in phosphate-buffered saline three times for 10 min and incubated (room temperature, 1 h) with recombinant FGF8b (R&D Systems) at the indicated dilutions on a culture dish. Control beads were incubated in PBS alone and treated as described above for FGF8b. For bead experiments involving simultaneous perturbation of RA and FGF signaling gradients, the concentrations of RA, SU5402, DEAB and FGF8 were serially titrated down such that resulting limbs form without any gross morphological defects.

The sections were deparaffinized, rehydrated in PBS via an ethanol gradient as described above, followed by post-fixation in 4% PFA for 5 min. Sections were then washed in PBT for 5 min three times. For detecting RCAS-infected cells, sections were preblocked with MST for 30 min before incubating with anti-GAG 3C2 antibody (AMV-3C2 from DHSB) at 1:5 dilution in MST for overnight at 4°C. Immunofluorescence was detected using the secondary antibody Alexa Fluor 488 conjugated anti-mouse IgG (1:200; Jackson Immunoresearch Laboratories, 115-545-003). The tissues were counterstained with DAPI and mounted in Vectashield antifade reagent (Vector Laboratories; H-1000).

Embryos with the Gdf5-LacZ reporter (Chen et al., 2016) electroporated were fixed for 45 min in fresh 4% PFA in PBS at room temperature, followed by washing three times in wash buffer, containing 2 ml 1 M MgCl₂,10 ml 1% deoxycholate, 2 ml 10% NP40 and 988 ml 0.1 M sodium phosphate (pH 7.3). Embryos were stained in X-Gal staining buffer (1 mg/ml) in the dark at room temperature until signal developed. After staining, embryos were briefly washed in wash buffer, post-fixed in 4% PFA for 2 h and imaged under stereomicroscope.

In silico transcription factor-binding site analysis for the transcription factors AP1 and MEIS1 has been performed using ConTra v3 in the 5 kb promoter region from the transcription start site of the BARX1 and BARX2 genes with Mouse (Mus musculus) and chicken (Gallus gallus) as reference organisms. For this analysis, a core stringency of 0.95 and a similarity stringency of 0.85 were used. For MEIS1 transcription factor, the position weight matrices V$MEIS1_02, M01419, V$MEIS1_01, M00419, MA0498.2 and MA0498.2 were considered; for the AP1 transcription factor, position weight matrices MA0940.1, MA0940.1, V$AP1_01, M00517, V$AP1_Q2_01, M00924, V$AP1_Q6_01, M00925, V$AP1_Q4_01,M00926, V$AP1_C,M00199, V$AP1_Q4,M00188, V$AP1FJ_Q2,M00172, V$AP1_Q2,M00173 and V$AP1_Q6,M00174 were considered. We used stringencies of 0.95 for both the core and similarity matrix matches (matrix score) for schematic representation of transcription binding sites.

We thank Olivier Pourquié for generously providing pCIG-MKK1dn and pSV1-diffusible FGFR constructs, Cliff Tabin for providing the pGEMT-Gdf5, pBS-Meis1 construct, Vikas Trivedi for guidance in designing HCR probes, and Pradip Sinha for the Leica Stellaris 5 confocal microscope facility. We thank Debojyoti Chakraborty, Riya Rauthan, Jyoti Tripathi and Mahima Bharti for providing assistance in imaging HCR probe-stained whole embryos. We also thank Cliff Tabin and Jonaki Sen for helpful discussions.

The peer review history is available online at https://journals.biologists.com/dev/lookup/doi/10.1242/dev.201335.reviewer-comments.pdf.