The evolution of a mineralized skeleton occurred in early vertebrates, in a variety of tissues including superficial dermal scales and teeth, together with internal cartilages, and perichondral bones (Ørvig, 1951; Donoghue and Sansom, 2002). In the internal skeleton, several cell types are associated with biomineralization, and the most studied cell model in mammalian organisms is the osteoblast active in the endochondral ossification process (Long and Ornitz, 2013). These osteoblasts are derived from periosteal tissues or from hypertrophic transdifferentiated chondrocytes (Tsang et al., 2015). The process of endochondral ossification, or replacement of cartilage matrix by bone marrow and bone trabeculae, is absent from chondrichthyans and has long been thought to be a derived feature specific to osteichthyans (reviewed in Donoghue and Sansom, 2002; Hirasawa and Kuratani, 2015), although recent paleontological data has challenged this view (Brazeau et al., 2020). Also known to mineralize their matrix are the chondrocytes, not only at the ossification front of endochondral bone growth (in the case of hyaline cartilage), but also in stable forms of mineralized cartilage such as fibrocartilages and other forms of cartilage displaying striking similarities to bony tissues (Beresford, 1981; Dyment et al., 2015; Paul et al., 2016; Pears et al., 2020; Berio et al., 2021). Even though both perichondral bones and cartilaginous tissues displayed mineralization in the earliest forms of mineralized internal skeletons (Ørvig, 1951; Min and Janvier, 1998; Donoghue et al., 2006; Johanson et al., 2010, 2012; Pears et al., 2020), mineralizing cartilages have been understudied from a genetic and evolutionary perspective in extant vertebrates. A better understanding of the genetic underpinning of the mineralizing chondrocytes is therefore necessary to understand the early steps of the evolution of endoskeletal mineralization in vertebrates.

The evolution of vertebrate endoskeletal mineralization has been discussed in the light of the two rounds of whole-genome duplication (2Rs). These duplications occurred before the diversification of extant jawed vertebrates (Nakatani et al., 2021) and generated gene families with diverging gene functions which may have produced the genetic toolkit required for the cellular ability to mineralize an extracellular matrix (Zhang and Cohn, 2008). The evolution of the SPARC/SPARC-L/SCPP gene family has been of great interest in this perspective (Kawasaki and Weiss, 2003; Kawasaki et al., 2005; Kawasaki, 2009; Bertrand et al., 2013; Enault et al., 2018), and is summarized in Figure 1. SPARC-L and SPARC are two paralogues having originated from the 2Rs (Kawasaki and Weiss, 2003; Kawasaki et al., 2005; Kawasaki, 2009; Bertrand et al., 2013; Enault et al., 2018). In bony fishes, independent local duplications at the SPARC-L locus generated SPARC-L1 and SPARC-L2 and a variable number of tandemly located genes coding for Secretory Calcium-binding PhosphoProteins (SCPPs) that have evolved rapidly since their origin (see Supplementary Figure S1 and Kawasaki and Weiss, 2003; Kawasaki et al., 2005; Enault et al., 2018). Hence outside of amniotes, homology relationships between SCPP duplicates are obscured by independent gene gains and losses together with a high rate of sequence divergence (Kawasaki, 2009). No SCPP genes have been identified in cartilaginous fish genomes, making the chondrichthyan SPARC-L gene the single orthologue to all SCPP genes of bony vertebrates (see Figure 1, Supplementary Figure S1 and Ryll et al., 2014; Venkatesh et al., 2014; Enault et al., 2018).

The SPARC gene (Secreted Protein Acidic and Rich in Cysteine, formerly coined Osteonectin) encodes a matricellular protein which is one of the most abundant non-collagenous matrix proteins in mammalian and teleost bone (Schreiweis et al., 2007; Kessels et al., 2014). Secreted by osteoblasts, the SPARC protein functions in mineralized tissues by binding both collagen fibrils and calcium, but also by signaling to bone cells (reviewed by Rosset and Bradshaw, 2016). In osteichthyans, the expression of SPARC is evolutionary conserved in osteoblasts as well as in odontoblasts (Holland et al., 1987; Li et al., 2009; Espinoza et al., 2010; Enault et al., 2018). In chondrichthyans having secondarily lost the bone tissue (and the osteoblast cell type), SPARC is highly expressed in odontoblasts (Enault et al., 2018). The single SPARC-L gene in cartilaginous fishes is expressed in enameloid secreting cells in teeth and scales of the catshark Scyliorhinus canicula (Enault et al., 2018). In osteichthyans it seems that SPARC-L1 and SPARC-L2 are not specifically expressed nor functionally required in the skeleton (McKinnon et al., 2000; Bertrand et al., 2013). In addition, SPARC-L2 was independently lost in tetrapods and teleosts, and SPARC-L1 was also lost in amphibians (see Figure 1 and Kawasaki et al., 2007; Bertrand et al., 2013; Enault et al., 2018), suggesting that these two genes are functionally dispensable. Rather, in osteichthyans, SCPP family members are key players of skeletal mineralization. Within amniote SCPP genes, Bone sialoprotein (BSP), Osteopontin (OPN or SPP1) and Dentin matrix protein 1 (DMP1) are strongly expressed by osteoblasts and their protein products are stored in the mineral phase of bone tissue (Ustriyana et al., 2021). Most members of this family are also expressed and functional during tooth development in mammals (either in the production of enamel or/and dentin, reviewed by Nikoloudaki, 2021). In the clawed frog Xenopus tropicalis and the zebrafish Danio rerio the expression of distinct SCPP members has been reported in ameloblasts, odontoblasts, and osteoblasts (Kawasaki et al., 2005; Kawasaki, 2009; Espinoza et al., 2010; Enault et al., 2018). Overall, our knowledge of the evolution of the expression of SPARC, SPARC-L and the SCPP members during cartilage mineralization remains limited, and, in this study, we examined the expression of these genes during endoskeletal development in Xenopus tropicalis and Scyliorhinus canicula.

Our findings in Xt reveal an evolutionary conservation of the cartilaginous expression of the SCPP genes in tetrapods. Indeed, similarly to the situation in Xt, SPARC is not expressed in mouse chondrocytes (Holland et al., 1987). Rather, SCPP genes such as DMP1 and BSP are expressed and required for mouse cartilage development (Chen et al., 1991; Ye et al., 2005; Bouleftour et al., 2014; Fujikawa et al., 2015). As indicated by other studies (Yagami et al., 1999; Bandyopadhyay et al., 2008), gene expression in cartilaginous elements can be subdivided in two distinct domains which we will use to discuss our results. First, SCPP genes become activated at late stages of hypertrophy, when the cartilage matrix becomes replaced by bone marrow at the mammalian diaphysis (Chen et al., 1991; Fujikawa et al., 2015). Likewise, in Xt, DMP1 is exclusively expressed at the diaphysis (Figure 3F), and SCPPA2 exhibits a much stronger expression at the diaphysis than the epiphysis region (Figures 3G,G′). A similar situation is observed at the level of the Xt vertebrae, where the expression of SCPP genes tightly correlates with cartilage maturation and mineralization in the neural arch (for BSP, DMP1, and SCPPA2) as well as in the ventral vertebral region (for BSP and SCPPA2). Second, the SCPP genes harbor a stronger expression in the non-mineralized peripheral cartilage, as observed in mouse for osteopontin (Heilig et al., 2016) and DMP1 (Fujikawa et al., 2015). This situation is similar to the expression of Xt SSCP genes in chondrocytes located in the vicinity of the bone matrix in long bones and vertebrae (Figures 3, 4). Such a peripheral cartilage domain expresses specific genes, as reported in chick (Bandyopadhyay et al., 2008), and undergoes ectopic mineralization in mutant mouse animals for the Mgp (Marulanda et al., 2017) and Trps1 (Napierala et al., 2008) genes. In summary, SCPP genes from frog and mouse are expressed in the mature cartilage of the diaphysis and neural arches, as well as in peripherally located chondrocytes.

Available expression analyses did not report any cartilaginous expression for SCPP genes in teleosts (Kawasaki et al., 2005; Kawasaki, 2009; Weigele et al., 2015). Rather, the expression of the SPARC gene has been associated to cartilage development in zebrafish, gilthead seabream and the cichlid mouth breeder (Estevao et al., 2005; Redruello et al., 2005; Rotllant et al., 2008; Estevao et al., 2011; Weigele et al., 2015), albeit not in medaka, at least at the examined stages (Renn et al., 2006). Hence the reported cartilaginous expression patterns in teleosts (SPARC positive and SCPP negative) are opposite to the tetrapod situation (SPARC negative and SCPP positive), which might be related to drastic difference in the mode of endochondral ossification between these two groups (Cervantes-Diaz et al., 2017). In this respect, our data in the chondrichthyan representative Sc is instrumental to understand the evolution of the expression of these genes in the jawed vertebrate endoskeleton.

As no SCPP genes have been reported in chondrichthyan genomes to date, we focused on the evolutionarily related gene SPARC-L (Kawasaki and Weiss, 2003; Bertrand et al., 2013; Venkatesh et al., 2014; Enault et al., 2018). Our finding that Sc SPARC-L is not expressed in the vertebral cartilage is further confirmed by the robust and expected Sc SPARC-L expression in the odontodes present on the same sections and serving as convenient internal positive controls (Enault et al., 2018). By contrast, SPARC expression clearly co-localizes with sites of vertebral mineralization. In the neural arches of early stage 32 embryos, SPARC is restricted to the mineralizing peripheral cartilage matrix, thereby paralleling the expression of SCPP genes in frog (Figures 2, 4) and mouse (Fujikawa et al., 2015; Heilig et al., 2016). Hence, we uncover a novel evolutionarily conserved cartilage domain, as defined by peripherally located chondrocytes expressing SPARC in chondrichthyans and the SCPP genes in tetrapods. One difference is that this domain mineralizes in chondricthyans, but not in tetrapods. We propose that dosage variations between pro- and anti-mineralizing proteins might explain evolutionary differences between vertebrate lineages, as might be the case for instance for MGP which is a well-conserved cartilage mineralization inhibitor (Yagami et al., 1999; Espinoza et al., 2010; Viegas et al., 2013; Marulanda et al., 2017; Leurs et al., 2021). By examining the centrum of Sc specimens from different developmental stages we show that a ring of SPARC expression is already present in Alizarin red-negative early stage 32 embryos, suggesting that the presence of the SPARC protein plays a crucial role in the initiation of mineralization. The functional interaction between the mammalian SPARC and collagen 1 proteins is important for mineralization (Termine et al., 1981). However, as the shark chondrocytes embedded within a mineralizing cartilage matrix neither expresses collagen 1a1 nor collagen 1a2 (Enault et al., 2015), Sc SPARC function might be related to other aspects of matrix mineralization, such as its interaction with calcium and hydroxyapatite crystals (Engel et al., 1987; Maurer et al., 1995; Fujisawa et al., 1996).

Our data suggest that chondrichthyans are more similar to teleosts than to tetrapods because the Sc mineralizing cartilage is SPARC positive and SPARC-L negative. Here, we propose two evolutionary models to explain these divergent expression patterns (see Figure 5 and Supplementary Figure S1). Both models are based on the assumption that SPARC, SPARC-L, and SCPP share some level of functional redundancy, at least during chondrogenesis, as suggested by the fact that both SPARC and SCPP proteins are extracellular transglutaminase substrates (Aeschlimann et al., 1995; Forsprecher et al., 2011) and bind calcium ions (Engel et al., 1987; Chen et al., 1992; Maurer et al., 1995; Klaning et al., 2014). The “multiple losses” model is reminiscent of the Duplication Degeneration Complementation (DDC) phenomenon (Force et al., 1999) and involves at least three independent changes which abrogated the cartilaginous expression of SPARC, of SPARC-L or of the SCPP genes in distinct lineages (Figure 5, left panel). The “expression swap” model involves two changes and implies that, in the tetrapod lineage, cartilaginous expression was gained for the SCPP genes and lost for SPARC (Figure 5, right panel). While the SPARC/SPARC-L/SCPP family exhibit a unique evolutionary trajectory (Bertrand et al., 2013; Enault et al., 2018), the “expression swap” model would be similar to the dynamic expression turnover observed between members of the Keratin, Vitellogenin, and Globin vertebrate families (Finn et al., 2009; Vandebergh and Bossuyt, 2012; Opazo et al., 2015). Both scenarios imply regulatory changes that switched off (“multiple losses model”) or on (“expression swap model”) several SCPP genes (Figure 5). From a regulatory perspective, the idea of coordinated change in the expression of tandemly located SCPP genes is consistent with the fact that BSP and DMP1 are included within the same topological association domain in human chromatin (Krietenstein et al., 2020), and that multiple genes can be co-regulated by the same enhancer (reviewed in Zheng and Xie, 2019). To the best of our knowledge, current experimental evidence is not sufficient to discriminate between the two models shown in Figure 5. Hence, a detailed analysis of the expression and regulation of SPARC/SPARC-L/SCPP genes in chondrocyte from a broader array of chondrichthyan and osteichthyan species will be required to shed light on the genetic mechanisms involved in the evolution of cartilage mineralization that originated deep in the vertebrate lineage (Min and Janvier, 1998; Donoghue et al., 2006; Johanson et al., 2010; 2012).