Common cellular origin and diverging developmental programs for different sesamoid bones

Sesamoid bones constitute a special group of bones that are integral to the skeletons of many vertebrate species (Abdala et al., 2017; Bizarro, 1921; Corina Vera et al., 2015; Jerez et al., 2009; Maisano, 2002; Ponssa et al., 2010; Sarin et al., 1999). Estimated to have evolved 200 million years ago (Carter, 1998), sesamoid bones are usually small and flat with morphological resemblance to the sesame seed, hence their name (Vickaryous and Olson, 2007). In addition, sesamoid bones are typically embedded within tendons, notably in proximity to joints. However, despite these common characteristics, sesamoids vary tremendously in aspects such as location and penetrance both within and across species, and the type of joints they are associated with. This high degree of variability imposes a challenge on providing a comprehensive model for sesamoid bone development.

The patella, also known as the kneecap, is the largest sesamoid bone in mammals and is part of the patellofemoral joint, one of the two joints that compose the knee (Samuels et al., 2017). The patella is positioned within the patellar groove, which is located at the anterodistal part of the femur and is separated from it by the synovial patellofemoral joint. The patellar groove enables the patella to have a controlled slide during flexion and extension of the lower leg segment, whereas the synovial joint permits this high degree of flexibility while maintaining the integrity of the articular cartilage of the long bones that flank the synovial capsule (Schindler and Scott, 2011). The patella has a crucial effect on the stability and mechanics of the knee. It absorbs disruptive forces that are applied to the quadriceps tendon, within which it is embedded, and enhances the moment arm and resulting extension force of the quadriceps muscle (Bizarro, 1921; Lennox et al., 1994; Mottershead, 1988; Sarin et al., 1999; Sutton et al., 1976).

The smaller fabella is located opposite the patella at the back of the mammalian knee. Although it is similarly embedded within the lateral head tendon of the gastrocnemius muscle in both human and mouse, the attachment site of this tendon varies between the two species. In humans, the fabella is located behind the lateral condyle of the femur, whereas in mice it is found on its lateral side (Jin et al., 2017; Phukubye and Oyedele, 2011; Sarin et al., 1999). The fabella lacks articular cartilage and is, in addition, connected by the fabellofibular ligament to the fibular tip and to the proximolateral side of the meniscus by two separate bundles of fibers (Jin et al., 2017). These anatomical differences imply that, unlike patella, the function of the fabella is not to facilitate and enhance movement but rather restrict it by anchoring and stabilizing the knee, thus preventing possible external rotation of the tibia or hyperextension of the knee (Hauser et al., 2015).

In mice, digit sesamoids can be found in pairs at any metacarpo/metatarsophalangeal and proximal interphalangeal joints of any digit, at which they are dorsally embedded within the flexor digitorum tendon and anteriorly share articulation with the metacarpal/metatarsal bones. In contrast to the patella, they are unable to freely glide over these articulated joints, as they are physically connected to the proximal tip of the proximal phalanx by a fibrocartilaginous joint (Doherty et al., 2010). Thus, digit sesamoids serve as a guide for the long flexor tendons in directing and controlling the course of the strings during flexion of the digits (Wirtschafter and Tsujimura, 1961).

To date, the prevailing model for sesamoid development suggests that these bones are induced within the tendons that wrap around joints by mechanical forces that are generated by embryonic movement (Hall, 2005; Parsons, 1904). However, we recently reported that the patella forms by a different developmental mechanism (Eyal et al., 2015). Briefly, we demonstrated that the patella originates from chondroprogenitors that co-express both Sox9 and scleraxis (Scx) and are distinct from the Sox9⁺/Scx⁻ chondroprogenitors that give rise to the long bone cartilaginous anlagen (Blitz et al., 2013; Eyal et al., 2015; Eyal et al., 2018 preprint). These progenitors form in juxtaposition to the femur and are later separated from it by the application of a joint formation program. TGFβ and BMP4 signaling pathways were shown to be required for patella precursor specification and differentiation, respectively. Finally, we found that mechanical load is not necessary for patella formation, but is needed for its separation from the femur by regulating the formation of the patellofemoral joint. Although our findings pertain specifically to patella formation, they raise the question of whether our model for patella development could be applied to other sesamoid bones.

In this study, we test our model of patella development on the development of the fabella and digit sesamoids, which, as mentioned, are different from the patella in their anatomical, histological and functional properties. We show that all examined sesamoid bones originated from Sox9⁺/Scx⁺ progenitors that were induced independently of mechanical stimuli. We further show that these progenitors require TGFβ signaling for their specification, BMP2 signaling for their growth, and redundant interaction between BMP4 and BMP2 signaling for their differentiation. We provide evidence that suggests the sesamoids may develop away from the skeleton, or separate from the bone shaft through mechanical load-dependent formation of a synovial joint, or through mechanical load-independent formation of a fibrocartilaginous joint.

Previously, we described a mechanism for patella development (Eyal et al., 2015). To investigate the generality of this mechanism, we analyzed comparatively development of the patella with that of two other subgroups of sesamoid bones, namely the lateral fabella and the metacarpophalangeal sesamoids. The two are similar to the patella in their full penetrance in both human and mice, but are dissimilar from it in their anatomy, histology and functions. We first compared the anatomy of the three sesamoid bones in mice using both micro-CT scans and histological sections (Fig. 1). Our results showed that, by E18.5, all sesamoid bones have formed and were embedded within tendons. The patella, which was separated from the femur by the patellofemoral synovial joint on its dorsal side, was ventrally embedded within the quadriceps tendon (Fig. 1A,Ci,Cii). The lateral fabella was embedded within the tendon of the lateral head of the gastrocnemius muscle, which attaches the dorsolateral condyle of the femur, and connected at its base to the lateral meniscus by the fabellofibular ligament (Fig. 1A,Di,Dii). The third metacarpophalangeal sesamoid bone (III-MCPS) was separated from the metacarpal bone by synovial space on its ventral aspect and embedded within the flexor digitorum tendon on its proximodorsal aspect (Fig. 1A,Ei,Eii). Micro-CT imaging showed that the III-MCPS was distinct from the proximal phalanx; yet, histologically we observed an unfamiliar tissue that connected the III-MCPS and the posterior tip of the proximal phalanx (Fig. 1Eii; enlarged boxed area), most likely the primordium of the fibrocartilage tissue that connects the metacarpophalangeal sesamoid bone to the proximal phalanx in adult mice (Doherty et al., 2010).

Patella precursors, which co-express Sox9 and Scx, are added secondarily to the Sox9⁺/Col2a1⁺ chondrocytes of the femur anlage (Eyal et al., 2015). To examine whether the lateral fabella and III-MCPS originated from similar progenitors, we first sought to find the developmental stage at which these sesamoids start to form. For that, we harvested and sectioned embryonic wild-type (WT) knees or carpal digits at sequential developmental stages from E13.5 to E17.5 and immunostained them using antibodies for SOX9, a marker for chondroprogenitors, and COL2A1, a marker for chondrocytes. At E14.5, Sox9⁺/Col2a1⁻ cells marked the undifferentiated lateral fabella precursors, whereas Sox9⁺/Col2a1⁺ differentiated chondrocytes marked the femur and tibia (Fig. 1Diii). Notably, whereas the Sox9⁺/Col2a1⁻ patella precursors developed in juxtaposition to the femur, lateral fabella precursors were not connected to either femur or tibia (Fig. 1Ciii,Diii).

The Sox9⁺/Col2a1⁻ III-MCPS precursors were first observed at E16.5, subsequent to the differentiation of both metacarpal and proximal phalanx anlagen. Interestingly, at that stage, these precursors extended continuously from the proximodorsal aspect of the proximal phalanx (Fig. 1Eiii). This implies that later in development, they will have to detach from the proximal phalanx, similar to the process seen in the patella (Eyal et al., 2015).

Having established the initiation stage of these sesamoids, we next examined their cellular origin. For that, we performed a lineage tracing experiment using the Scx-Cre driver mouse crossed with tdTomato reporter mouse. We observed in sagittal sections of E18.5 knees and forelimb digits that patella, lateral fabella and III-MCPS cells were of Scx⁺ lineage (Fig. 1Civ,Div,Eiv). Next, we confirmed that, as seen in patella progenitors, both lateral fabella and III-MCPS progenitor cells co-expressed Sox9 and Scx. For that, sectioned knees and carpal digits of transgenic ScxGFP embryos were immunostained for SOX9 to highlight chondrocytes and, specifically, Sox9⁺/Scx⁺ chondroprogenitors. We observed large populations of Sox9⁺/Scx⁺ chondroprogenitors at the presumptive sites of patella, lateral fabella and III-MCPS formation (Fig. 1Cv,Dv,Ev). Collectively, these results confirmed that both lateral fabella and III-MCPS precursors formed in accordance to the patella model, namely from Sox9⁺/Scx⁺ chondroprogenitors that formed following differentiation of the long bone primordium cells.

Sesamoid bone development has been assumed to be induced by mechanical signals that are produced by embryonic movement (Parsons, 1904; Parsons, 1908). Yet, we have reported that patella induction was mechanical load-independent (Eyal et al., 2015). We therefore re-examined the paradigm of mechanical load-dependent induction of sesamoid bones by studying the formation of lateral fabella and III-MCPS in paralyzed homozygous mdg mutant embryos; heterozygous mdg embryos were used as controls (Pai, 1965). Examination of histological sections through knees and carpal bones of E17.5 mutants showed, as was previously described, that the patella was developing (Fig. 2A,A′). Similarly, we observed that both lateral fabella and III-MCPS were formed in the mutant as well (Fig. 2B-C′). These results provide conclusive support for our model by indicating that the induction of sesamoid bone formation is independent from muscle-induced mechanical load.

Another feature of patella development is the regulation of its Sox9⁺/Scx⁺ precursors by both TGFβ and BMP4 signaling pathways (Eyal et al., 2015). Having identified a similar cellular origin, we proceeded to examine the involvement of these pathways in lateral fabella and III-MCPS development. In addition, we examined the involvement of Bmp2, a paralog of Bmp4 (Gamer et al., 2018; Pignatti et al., 2014). For that, Tgfbr2, Bmp4 or Bmp2 were specifically knocked out (cKO) from early limb mesenchyme using Prx1-Cre as a deleter mouse (Prx1-TgfβRII^floxed; Prx1-Bmp4^floxed; Prx1-Bmp2^floxed) (Chytil et al., 2002; Liu et al., 2004; Logan et al., 2002; Ma and Martin, 2005; Selever et al., 2004); Prx1-Cre-negative embryos were used as controls. Then, sections through the knees and digit bones of E17.5 mutant and control embryos were stained using Safranin-O (Fig. 3A-F″). Our examination revealed that similar to patella, both lateral fabella and III-MCPS were completely absent in Prx1-TgfβRII^floxed mutants (Fig. 3A-B″). Focusing on the BMP pathways, we found that in contrast to our previous findings for patella formation (Eyal et al., 2015), loss of Bmp4 in limb mesenchyme had no effect on the formation of lateral fabella and III-MCPS (Fig. 3C-D″). Similarly, examination of the Prx1-Bmp2^floxed mutants revealed that all sesamoids were formed; however, all displayed growth retardation (Fig. 3E-F″).

Our finding that the exclusive loss of either Bmp2 or Bmp4 did not abolish the formation of lateral fabella and III-MCPS raised the possibility that these BMPs have a redundant function (Bandyopadhyay et al., 2006). To study this hypothesis, we crossed Prx1-Bmp2^floxed; Bmp4^floxed/+ mice with Prx1-Bmp2^floxed; Bmp4^floxed mice to generate mutant embryos that lack both Bmp2 and Bmp4 in limb mesenchyme; Prx1-Cre-negative embryos were used as controls. As seen in Fig. 3G-H″, the loss of a single copy of Bmp4 allele on the background of a homozygous Prx1-Bmp2^floxed mutation resulted in fabella aplasia, whereas Prx1-Bmp2^floxed; Bmp4^floxed double cKO resulted in aplasia of all three sesamoid bones that were examined (Fig. 3I-J″). Collectively, these results suggest that sesamoid bones are regulated by both TGFβ and BMP signaling.

Although digit sesamoids are distinct from the proximal phalanx in neonatal mice, at E16.5 the III-MCPS precursors appeared to be continuous with the cartilage anlage of the proximal phalanx (Fig. 1Eiii). By E18.5, these precursors were detached from the proximal phalanx, although they remained connected to it by a fibrocartilage-like tissue (Fig. 1Eii). To date, little is known about the mechanisms that underlie the genesis of the fibrocartilaginous joint. To study the origin of the fibrocartilage joint cells, we performed lineage tracing experiments using either Scx-Cre or Sox9-CreER^T2 driver mice crossed with tdTomato reporter mice. As seen in sagittal sections of E18.5 forelimb digits of Scx-Cre: tdTomato embryos, the fibrocartilage joint cells were all Scx⁺ (Fig. 4A). In parallel, sagittal sections of E18.5 forelimb digits of Sox9-CreER^T2: tdTomato embryos, to which tamoxifen was administered at E13.5, showed positive labeling of the fibrocartilage joint cells (Fig. 4B). Labeling of the fibrocartilage joint cells by both Scx-Cre: tdTomato and Sox9-CreER^T2: tdTomato suggests that these joint cells are descendant of Sox9⁺/Scx⁺ progenitors.

To further characterize molecularly the process of fibrocartilage joint formation, we performed double-fluorescent in situ hybridization using probes for cartilage marker Sox9, tendon cell markers Scx and Tnmd, and joint-forming cell markers Gdf5 and Tppp3. As has been previously reported, Scx and Tnmd expressions were observed within the tendon that envelops the patella on its ventral aspect, whereas Gdf5 and Tppp3 were expressed within the forming patellofemoral joint (Fig. 4C-F). However, within the tissue that connects the metacarpophalangeal sesamoid bone and the proximal phalanx, Scx expression was observed (Fig. 4C′), whereas expression of Tnmd, Gdf5 or Tppp3 could not be detected (Fig. 4D′-F′).

Recently, Iroquois genes irx5a and irx7 were shown to mark the forming cartilaginous hyoid joint in the zebrafish (Askary et al., 2015), whereas Irx1 was reported to be highly expressed within the digit joints in mice (McDonald et al., 2010). We therefore examined the involvement of Irx genes in sesamoid bone development. For that, we performed double-fluorescent in situ hybridization using complementary RNA probes against Sox9 and Irx1. As seen in Fig. 4G′, Irx1 was specifically expressed within the connecting tissue between the metacarpophalangeal sesamoid bone and the proximal phalanx, but was excluded from other joint compartments in the digits and from the patellofemoral synovial joint (Fig. 4G,G′). The combined expression of Irx1 and Scx and the absence of Tnmd in the connective tissue between the metacarpophalangeal sesamoid bone and the proximal phalanx suggest that Irx1 and Scx can serve as markers for fibrocartilaginous joint.

Sesamoid bones are small auxiliary bones that are highly variant in positioning, size and number (Bizarro, 1921; Jerez et al., 2009; Pearson and Davin, 1921b; Ponssa et al., 2010; Sarin et al., 1999). To date, the mechanism that regulates the development of these bones and allows the variations among them is unknown. Here, by comparing the development of three different sesamoid bones, namely patella, lateral fabella and III-MCPS, we identify common molecular and cellular mechanisms that underlie these processes. Interestingly, we also identify points of diversion in their developmental program, which could provide a mechanistic explanation to the variability within this family of bones (Fig. 5).

Our finding that all three studied sesamoid bones originated in Sox9⁺ and Scx⁺ precursors is interesting, because this lineage also contributes to the formation of bone superstructures, protrusions that project from the long bone surface and serve as attachment sites for tendons (Blitz et al., 2009; Blitz et al., 2013). These observations led us to suggest that bone superstructures and sesamoid bones can be placed under one category. Nevertheless, whereas bone superstructures protrude from the long bone shaft, sesamoid bones are auxiliary to it. Our finding that both patella and III-MCPS are separated from their bone of origin by a joint formation process offers a strategy for transforming a bone superstructure into a sesamoid bone, namely by detachment. Another way to determine between the development of a bone superstructure or a sesamoid bone is by regulating the spatial distribution of Sox9⁺/Scx⁺ precursors. If the condensation of the Sox9⁺/Scx⁺ precursors is adjacent to the target bone, then a bone superstructure will develop, as in the case of the deltoid tuberosity (Blitz et al., 2013; Sugimoto et al., 2013). Conversely, if these cells condense away from the target bone, a sesamoid bone will be formed, as in the case of the lateral fabella (Fig. 5).

An example for such transitional plasticity between sesamoid bones and superstructures is given by olecranon development. In mammals, the olecranon forms as a bone superstructure at the proximal tip of the ulna, whereas in some anurans (frogs) it develops as a detached patella-like sesamoid bone (Ponssa et al., 2010), accordingly dubbed patella ulnaris. Presumably, formation of the olecranon as a sesamoid bone in earlier vertebrates preceded its formation as a fused superstructure in mammals. Interestingly, studies in several mutated mouse lines have also demonstrated this developmental plasticity. Compound Hox11^aadd mutant embryos were shown to exhibit a detached olecranon, similar to the patella ulnaris of anurans (Koyama et al., 2010). Another example was given by homozygous Pbx1 mutant embryos, which developed a detached sesamoid-like deltoid tuberosity (Selleri et al., 2001).

The detachment mechanism was hypothesized a century ago by Pearson and Davin (1921a, b), who suggested that sesamoid bones originate from existing bone eminences that detached from long bone shafts. F. G. Parsons proposed a contrary hypothesis, suggesting that sesamoid bones develop independently yet could, in some cases, attach to neighboring long bones and become a bony eminence (Parsons, 1904; Parsons, 1908). Our data reconcile between these two opposing models by placing both bone eminences and sesamoid bones under one category of cartilage elements that develop from secondary Sox9⁺/Scx⁺ chondroprogenitors. Moreover, they share a molecular regulation mechanism, as both require TGFβ/BMP signaling for their development (Blitz et al., 2013; Eyal et al., 2015).

Interestingly, our findings also suggest that both genetic and mechanical regulation are involved in facilitating developmental diversity, as indicated by both global and specific genetic regulation of the TGFβ/BMP signaling pathways and by the requirement of mechanical stimuli for patellofemoral joint formation. Our results show that TGFβ signaling is necessary for the formation of all three sesamoids. Moreover, we show that BMP2 and BMP4 are redundantly necessary for sesamoid bone formation. Interestingly, however, whereas BMP2 signaling regulated the growth of all three bones, BMP4 specifically regulated differentiation of patella progenitors.

Another example for the diversity in the regulation of sesamoid bone development is the involvement of mechanical signals. First, we found no effect for lack of these signals on the formation of the lateral fabella. Similarly, the III-MCPS was formed and separated by the formation of a fibrocartilaginous joint in the absence of mechanical signaling. Yet, as we have previously shown, patella separation through the formation of a synovial joint is mechanical load-dependent (Eyal et al., 2015). In addition to the difference in mechanosensitivity, cells at the presumptive site of the developing fibrocartilaginous joint displayed a distinct expression profile compared with that of developing synovial joints. They expressed the tendon marker Scx and joint marker Irx1, but failed to express other markers such as Tnmd, Gdf5 or Tppp3. These findings shed light on the little known process of fibrocartilaginous joint formation.

Altogether, these variations in the developmental programs of the different sesamoid bones may provide an initial explanation for the diversity in anatomy and function of this unique group of bones.

To conclude, our results provide further insight into the origin, formation and development of sesamoid bones. We suggest that these bony elements originate from pools of Sox9⁺/Scx⁺ chondroprogenitors. We further show that some sesamoid bones form initially in juxtaposition to the long bone anlagen and are later separated by different programs of joint development. In contrast, we show that other sesamoids, such as the lateral fabella, can form as a separate entity. Finally, we demonstrate that a combination of genetic and mechanical regulatory programs is applied to achieve anatomical and ultimately functional variation during sesamoid bone development.

All experiments involving mice were approved by the Institutional Animal Care and Use Committee (IACUC) of the Weizmann Institute. For all timed pregnancies, plug date was defined as E0.5. For harvesting of embryos, timed-pregnant females were euthanized by cervical dislocation. The gravid uterus was dissected out and suspended in a bath of ice-cold phosphate-buffered saline (PBS) and the embryos were harvested after removal of the placenta. Tail genomic DNA was used for genotyping by PCR. All experiments were performed in at least three biological repeats (i.e. three embryos from three separate litters).

Mice that were heterozygous for the mutation muscular dysgenesis (mdg) (Pai, 1965) were kindly provided by G. Kern (Innsbruck Medical University, Austria). Scx-Cre and ScxGFP transgenic mice were obtained from R. Schweitzer (Shriners Hospital for Children Research Division, Portland, OR, USA) and R. L. Johnson (The University of Texas MD Anderson Cancer Center, Houston, TX, USA). Sox9-CreER^T2 mice (Soeda et al., 2010) were kindly provided by the laboratory of Haruhiko Akiyama (Kyoto University, Japan). C57/Bl6 mice (The Jackson Laboratory) were used for micro-CT analyses.

The generation of Prx1-Cre (Logan et al., 2002), floxed-Tgfbr2 (Chytil et al., 2002), floxed-Bmp4 (Liu et al., 2004; Selever et al., 2004), floxed-Bmp2 (Ma and Martin, 2005), ScxGFP (Pryce et al., 2007), Sox9-CreER^T2 (Soeda et al., 2010) and Rosa26-tdTomato (Madisen et al., 2010) mice have been described previously. To create mdg mutant mice, animals heterozygous for the mutation were crossed; heterozygous mdg embryos were used as a control. To create Prx1-Tgfbr2, Prx1-Bmp4 or Prx1-Bmp2 mutant mice, floxed-Tgfbr2, floxed-Bmp4 or floxed-Bmp2 mice were mated with Prx1-Cre-Tgfbr2, Prx1-Cre-Bmp4 or Prx1-Cre-Bmp2 mice, respectively. To create compound Prx1-Bmp2;Bmp4 mutant mice, floxed-Bmp2;Bmp4 mice were mated with Prx1-Cre-Bmp2;Bmp4 mice. As a control, Prx1-Cre-negative embryos were used. For genetic lineage tracing analysis, Scx-Cre or Sox9-CreER^T2 mice were crossed with homozygous Rosa26-tdTomato mice. Induction of Cre recombinase was performed at various pregnancy stages by administration of 0.03 mg/g tamoxifen/body weight in corn oil by oral gavage (stock concentration was 5 mg/ml).

Before scanning, limbs of postnatal day (P) 14 C57/Bl6 mice (Jackson) were dissected, fixed overnight in 4% paraformaldehyde (PFA)/PBS and dehydrated in an ethanol series to 100%. Samples were then scanned ex vivo in 100% ethanol using the Xradia MicroXCT-400 (Zeiss) at 30 kV and 4.5 W or 40 kV and 8 W.

For preparation of paraffin sections, embryos were harvested at various ages, dissected and fixed in 4% PFA/PBS at 4°C overnight. After fixation, tissues were dehydrated in an ethanol series to 100% and embedded in paraffin. The embedded tissues were sectioned at 7 µm and mounted onto slides.

For preparation of OCT-embedded sections, embryos were harvested at various ages, dissected and fixed in 1% PFA/PBS at 4°C overnight. Fixed embryos were then dehydrated gradually, first in 15% sucrose for 4-6 h at room temperature and then in 30% sucrose overnight at 4°C. Next, samples were dissected and soaked in 15% sucrose/50% OCT for 30-60 min and then embedded in OCT. Frozen samples were immediately sectioned at 10 µm and mounted onto slides.

Safranin-O/Fast Green staining was performed following standard protocols. Double fluorescent in situ hybridizations (dbFISH) on paraffin sections were performed using digoxigenin (DIG)- and fluorescein (FITC)-labeled probes (Shwartz and Zelzer, 2014). See Table S1 for probes. Probes were detected using anti-DIG-POD (Roche, 11207733910, 1:300) and anti-FITC-POD (Roche, 11207733910, 1:300), followed by Cy3- or Cy2-tyramide labeled fluorescent dyes, according to the instructions of the TSA Plus Fluorescent Systems Kit (PerkinElmer).

For immunofluorescence staining for SOX9 and COL2A1, 7 μm thick paraffin sections of embryo limbs were deparaffinized and rehydrated in water. Antigen was then retrieved in 10 mM citrate buffer (pH 6.0), boiled and cooked for 10 min in a microwave oven. In order to block non-specific binding of immunoglobulin, sections were incubated with 7% goat serum, 1% bovine serum albumin dissolved in PBST (PBS+0.1% Triton X-100+0.01% sodium azide). Following blockage, sections were incubated overnight at 4°C with primary anti-SOX9 antibody (1:200, AB5535, Millipore). Then, sections were washed in PBST and incubated with Cy3-conjugated secondary fluorescent antibodies (1:100, 611-165-215, Jackson ImmunoResearch). After staining for SOX9, slides were washed in PBST and fixed in 4% PFA at room temperature for 10 min. Then, slides were incubated with proteinase K (Sigma-Aldrich, P9290), washed and post-fixed again in 4% PFA. Next, sections were washed and incubated overnight at 4°C with primary anti-COL2A1 antibody (1:50, II-II6B3, the Developmental Studies Hybridoma Bank). The next day, sections were washed in PBST and incubated with Cy2-conjugated secondary fluorescent antibodies (1:200, 615-222-214, Jackson ImmunoResearch). Occasionally, slides were counterstained using DAPI. Slides were mounted with Immu-mount aqueous-based mounting medium (Thermo Fisher Scientific).

For immunofluorescence staining for SOX9 and ScxGFP, 10 μm cryostat sections of embryo limbs endogenously labeled for ScxGFP were used. SOX9 immunofluorescence staining was performed as described above, using primary SOX9 antibody and secondary Cy3-fluorescent antibodies.

We thank N. Konstantin for expert editorial assistance and all other members of the Zelzer laboratory for their advice and suggestions. We thank R. Schweitzer for providing the ScxGFP mice, and H. Akiyama for providing the Sox9-CreER^T2 mice. We thank H. Vega from the Weizmann Institute Department of Design, Photography and Printing for designing the model.