A dynamic history of gene duplications and losses characterizes the evolution of the SPARC family in eumetazoans

Abstract

The vertebrates share the ability to produce a skeleton made of mineralized extracellular matrix. However, our understanding of the molecular changes that accompanied their emergence remains scarce. Here, we describe the evolutionary history of the SPARC (secreted protein acidic and rich in cysteine) family, because its vertebrate orthologues are expressed in cartilage, bones and teeth where they have been proposed to bind calcium and act as extracellular collagen chaperones, and because further duplications of specific SPARC members produced the small calcium-binding phosphoproteins (SCPP) family that is crucial for skeletal mineralization to occur. Both phylogeny and synteny conservation analyses reveal that, in the eumetazoan ancestor, a unique ancestral gene duplicated to give rise to SPARC and SPARCB described here for the first time. Independent losses have eliminated one of the two paralogues in cnidarians, protostomes and tetrapods. Hence, only non-tetrapod deuterostomes have conserved both genes. Remarkably, SPARC and SPARCB paralogues are still linked in the amphioxus genome. To shed light on the evolution of the SPARC family members in chordates, we performed a comprehensive analysis of their embryonic expression patterns in amphioxus, tunicates, teleosts, amphibians and mammals. Our results show that in the chordate lineage SPARC and SPARCB family members were recurrently recruited in a variety of unrelated tissues expressing collagen genes. We propose that one of the earliest steps of skeletal evolution involved the co-expression of SPARC paralogues with collagenous proteins.

1. Introduction

Vertebrates appeared more than 500 Ma, and have acquired specialized cell types able to produce mineralized extracellular matrix found in cartilage, bone and teeth [1]. Skeletal development relies on a myriad of extracellular proteins that create biochemical conditions compatible with biomineralization [2,3]. Hence, a comprehensive analysis of the evolutionary history of extracellular proteins expressed by skeletal cells is essential to improve our understanding of the molecular changes that contributed to the emergence of vertebrates. For instance, the fibrillar collagen proteins are the principal component of the vertebrate mineralized matrix [4] and their evolution has been studied in great details. Fibrillar collagen orthologues are present in sponges, cnidarians and bilaterians [5–8], and have been recruited into the vertebrate mineralizing skeletal matrix to perform a structural role by facilitating the nucleation of hydroxyapatite crystals [1,9,10]. The SPARC homologues represent another gene family that is intimately linked to skeletal evolution for a variety of reasons. First, SPARC proteins bind calcium and act as extracellular collagen chaperones [11,12]. Second, the SPARC family predates the emergence of vertebrates [13–17]. Third, among osteichthyans (bony fishes and tetrapods), most species exhibit two paralogues called SPARC and SPARCL1 that are expressed in skeletal cells [12,18–26]. Finally, the SPARCL1 paralogue underwent a series of tandem duplications, thereby giving rise to the small calcium-binding phosphoproteins (SCPP) gene members [20,21]. The SCPP members are expressed in cartilage, bone and teeth and encode highly acidic proteins regulating extracellular matrix biomineralization [18,27–34]. Therefore, while SPARC and SPARCL1 are neither sufficient in vitro nor required in vivo for biomineralization to occur [35–37], the SPARC family has been instrumental in the emergence of vertebrate genes involved in skeletal tissue mineralization.

Here, we present new data supporting a novel scenario for the evolutionary history of the SPARC family. Most importantly, both phylogeny and synteny conservation analyses reveal that an ancient gene duplication in the last common ancestor of eumetazoans produced SPARC and SPARCB, a gene first described here. We further show that three independent losses eliminated one of the two paralogues in the lineage of the cnidarians, protostomes and tetrapods. Consequently, cnidarians exhibit a single gene which is actually orthologous to SPARCB and not SPARC. In order to shed light on the ancestral expression pattern of SPARC family genes, we analysed and compared the embryonic expression pattern of homologues from all major vertebrate and invertebrate chordate groups. We show that SPARC family members have independently been recruited to embryonic tissues expressing collagenous proteins, providing novel insights regarding the earliest molecular changes involved in skeletal matrix evolution.

2. Material and methods

(a). Phylogenetic analysis

Amino acid sequences of the SPARC and SPOCK families were obtained from the NCBI, the JGI and the Ensembl databases (see the electronic supplementary material, table S1). For Oncorhynchus mykiss and Pimephales promelas, EST sequences corresponding to putative SPARCB orthologues were assembled using CAP3 [38]. The resulting nucleotide sequences were translated in the adequate frame. The amino acid sequences were all aligned using Clustal Omega [39] and the alignment was manually corrected using SeaView [40]. Only amino acids located between the Kazal and the SPARC domains were kept for the phylogenetic reconstruction. Bayesian inference (BI) tree was inferred using MrBayes v. 3.1.2 [41,42], with the model recommended by ProtTest [43] under the Akaike information criterion (we used WAG+Γ + I model because LG is not implemented in MrBayes), using the CIPRES Science Gateway [44]. Two independent runs were performed, each with 4 chains. A burn-in of 25 per cent was used and the consensus tree was calculated for the remaining trees. Maximum likelihood (ML) analysis was performed using PhyML [45] with the model recommended by ProtTest (LG + Γ + I) and aLRT support for branches was computed. The phylogenetic tree obtained using ML had a topology consistent with the one obtained by BI (see the electronic supplementary material, figure S1).

(b). Synteny conservation analysis

We searched human orthologues for all the predicted amino acid sequences found in amphioxus scaffolds 131, 329 and 562 using BLASTP with default parameters on the Ensembl database. For all the sequences that had hits localized in human chromosomes 4, 5 or/and 10, orthology was assessed by phylogenetic reconstruction. The aligned amino acid sequences of the corresponding gene families were retrieved from Ensembl. Other sequences were obtained from Genbank and the amphioxus sequences were thereafter added using clustalW [46]. Finally, alignments were manually corrected in SeaView [40] and ML trees were constructed with PhyML [45] implemented in SeaView with the WAG + Γ + I model of protein evolution, and aLRT was computed for branch support.

(c). Cloning and in situ hybridization

Partial cDNAs from Mus musculus SPARC and SPARCL1, Xenopus tropicalis SPARC, Danio rerio SPARC, SPARCL1 and SPARCB, Branchiostoma lanceolatum SPARC/SPARCL1 and SPARCB were cloned by RT-PCR in pGEM-T Easy vector (Promega). Primers used for each gene are listed in the electronic supplementary material, table S2. The Ciona intestinalis SPARC/SPARCL1 (cilv13k23) and SPARCB (ciad43l22) clones were obtained from Nori Satoh's Gene Collection Plates. Dioxigenin-labelled RNA probes were synthesized using appropriate enzymes according to the manufacturer's instructions (Roche). Ripe animals of the Mediterranean amphioxus (B. lanceolatum) were collected in Argelès-sur-Mer (France), and gametes were obtained by heat stimulation [47,48]. Fixation and whole-mount in situ hybridization were performed as described in [49], except the chromogenic reaction which was performed using BM Purple [50]. Mouse embryos were dissected in PBS, fixed in 4 per cent paraformaldehyde and whole-mount in situ hybridization was performed according to Wilkinson [51]. In situ hybridization on frog embryos was performed according to Harland [52] and Maldonado-Agurto et al. [53]. In situ hybridization on zebrafish embryos was performed according to [54]. In situ hybridization on ascidian embryos was performed according to Hudson & Yasuo [55] and Wada et al. [56].

3. Results

(a). A new SPARC gene family member

In our search for SPARC and SPARCL1 orthologues in metazoan genomes, we encountered a third gene that was not previously described and that we named SPARCB. Vertebrate SPARCB orthologues are absent from tetrapod genomes, but were identified in zebrafish (D. rerio) and coelacanth (Latimeria chalumnea) genomes. We also found EST sequences corresponding to SPARCB genes in two other teleost species, O. mykiss and P. promelas. SPARCB genes code for proteins that have the same general domain organization as SPARC, with a signal peptide followed by a Kazal domain and a SPARC domain. However, the zebrafish SPARCB does not possess a large region (typical of vertebrate SPARCL1) separating the signal peptide from the Kazal domain, nor does it display the acidic region previously described for bilaterian SPARC and SPARCL1 (not shown and electronic supplementary material, figure S2). In fact, this atypical organization has previously been described for the Nematostella vectensis SPARC1–4 proteins [13], which, as we show here, belong to the SPARCB subgroup. In non-vertebrates, we found SPARCB orthologues in amphioxus (Branchiostoma floridae), tunicates (C. intestinalis), ambulacrarians (Strongylocentrotus purpuratus and Saccoglossus kowalevskii) and cnidarians (N. vectensis and Hydra magnipapillata), but not in protostomes (annelids, molluscs, nematodes and arthropods). In order to unambiguously assess orthology relationships between SPARC, SPARCL1 and SPARCB, we performed phylogenetic reconstruction, using SPOCK family sequences as an outgroup. The phylogenetic tree shows that teleost, coelacanth and non-vertebrate deuterostome SPARCB proteins are orthologous and group with the cnidarian sequences (figure 1a and electronic supplementary material, figure S1). In agreement with published literature, the vertebrate SPARC and SPARCL1 paralogues cluster as a monophyletic clade (figure 1a and [22]). Our results also reveal that the genomes of tunicates, amphioxus, ambulacrarians and protostomes harbour a unique gene (referred to as SPARC/SPARCL1 hereafter), which is co-orthologue to the vertebrate SPARC and SPARCL1 genes (figure 1a and electronic supplementary material, figure S1). Remarkably, no SPARC/SPARCL1 orthologues are present in cnidarians (figure 1a and electronic supplementary material, figure S1).

Figure 1.

Paralogy and orthology relationships of the SPARC gene family members in eumetazoans. (a) Bayesian inference phylogenetic tree of the SPARC family. SPOCK family was used as outgroup. Posterior probabilities are shown on each node. Scale bar indicates the number of substitutions per site. See the electronic supplementary material, table S1 for accession numbers and name abbreviations. (b) Synteny conservation among vertebrate and amphioxus SPARC chromosomal regions. The amphioxus SPARCB and SPARC/SPARCL1 genes are located on three overlapping genomic contigs numbered 329, 562 and 131 (i). This amphioxus region is co-orthologous to regions of the human (ii) chromosomes 4 (containing SPARCL1), 5 (containing SPARC) and 10, as well as regions of the zebrafish (iii) chromosomes 1 (containing SPARCL1), 7, 14 (containing SPARC), 17 (containing SPARCB) and 12.

(b). Synteny conservation analysis

To decipher the evolutionary history of the chordate SPARC/SPARCL1/SPARCB family, we looked at the genomic position of SPARC/SPARCL1 and SPARCB genes of species that possess both genes (i.e. B. floridae, C. intestinalis, S. kowalevskii, S. purpuratus, L. chalumnea and D. rerio). We found that they are located on separate contigs, scaffolds or chromosomes, except in amphioxus. We then analysed the synteny conservation between amphioxus genomic regions around SPARC/SPARCL1 and SPARCB, and the human and zebrafish genomes. As shown in figure 1b, the amphioxus homologues are located on the scaffold 329 (containing SPARCB) and on the scaffolds 131 and 562 (overlapping at the level of SPARC/SPARCL1). All three scaffolds can be assembled into a single and continuous 2,8 Mb genomic fragment, showing that the amphioxus SPARC/SPARCL1 and SPARCB are syntenic (figure 1b). We then systematically compared all the predicted coding sequences from this amphioxus region with the human genome at the Ensembl database website (http://www.ensembl.org/) using BLASTP. We focused on the best BLAST hits that aligned to three well-defined paralogous regions (paralogons) of the human genome located on chromosomes 4 (containing SPARCL1), 5 (containing SPARC) and 10 (see figure 1b and [57]). This approach led us to identify 25 amphioxus genes (see the electronic supplementary material, table S3), for which orthology with vertebrate genes was unambiguously confirmed (see the electronic supplementary material, figure S3 and S4). Therefore, we show that the amphioxus genomic region containing SPARC/SPARCL1 and SPARCB harbours many genes whose human orthologues are found around the human SPARC or SPARCL1 genes or in the chromosome 10 paralogon. For most of these amphioxus genes, we were also able to detect conservation of synteny in zebrafish (figure 1b; electronic supplementary material, figure S4 and table S3). We found that the zebrafish SPARCB is located in a region of chromosome 17 which is orthologous to the aforementioned human chromosome 10 paralogon (figure 1b). Combined to our phylogenetic analyses, these findings strongly suggest that the SPARC/SPARCL1 and SPARCB genes were linked in the chordate ancestor, and further confirm the fact that SPARC and SPARCL1 result from two whole-genome duplication events that occurred in the vertebrate lineage [22,58].

(c). Developmental expression of the chordate SPARC family genes

In order to explore the developmental evolution of the SPARC family members in the chordate lineage, we undertook the analysis of their expression during embryogenesis using whole-mount in situ hybridization in species belonging to the three chordate phyla. Below, we describe the expression of the SPARC/SPARCL1 gene in amphioxus and tunicates, as well as its two vertebrate orthologues SPARC and SPARCL1. We subsequently describe the expression of the SPARCB members found in tunicates, amphioxus and zebrafish.

In amphioxus, SPARC/SPARCL1 expression is first detected at the early neurula stage in the paraxial mesoderm (figure 2a; electronic supplementary material, figure S5). At the mid-neurula stage, the expression increases while remaining restricted to the forming somites (figure 2b). In the larva, SPARC/SPARCL1 is expressed in the ventral and paraxial mesoderm and in the notochord (figure 2c). In ascidians, ubiquitous transcripts of SPARC/SPARCL1 were detected during early cleavage stages (not shown). Specific expression begins at late neurula/early tailbud stage in the notochord (figure 2d). At early tailbud stage, lower levels of transcripts were also detected in the central nervous system and anterior-most epidermis, as well as a strong expression in the notochord (figure 2e). In later tailbud stages, expression was detected specifically in the notochord (figure 2f).

Figure 2.

Comparative analysis of gene expression patterns of chordate SPARC homologues. (a–f) SPARC/SPARCL1 in situ hybridizations, amphioxus embryos are at (a) the early neurula, (b) mid–late neurula and (c) larvae stages. (d) Ascidian embryos are at the late neurula, (e) tadpole and (f) larvae stages. (g–r) SPARC in situ hybridizations, zebrafish embryos are at stages (g) 18 hpf, (h) 24 hpf, and (i,j) 72 hpf. Amphibian embryos are at stages (k) 21, (l) 27 and (m) 31. Mouse embryos are at stages (n) E7.5, (o) E8.5, (p) E11.5, (q,r) E10.5. (s–z) SPARCL1 in situ hybridizations, zebrafish embryos are at stages (s) 24 hpf, (t) 48 hpf and (u) 72 hpf. Mouse embryos are at stages (v) E7.5, (w) E8.5, (x) E9.5, (y) E10.5 and (z) E11.5. (a′–h′) SPARCB in situ hybridizations, amphioxus embryos are at the (a′) mid–late neurula, late (b′) neurula and (c′) larvae stages. Ascidian embryos are at the early (d′) gastrula, (e′) neurula, (f′) early tailbud and (g′,h′) late tailbud stages. Embryos are oriented with their anterior and dorsal sides leftward and upward, respectively, except for (a,d,k,e′) (dorsal view); (n–r,v–z) (anterior upward, dorsal towards the left); and (d′) (vegetal pole view), (h′) (ventral views). For mouse embryos, the images were inverted and the right side of the embryos is shown. A summary of the phylogenetic relationships of these genes is indicated on the right. Description of the different expression patterns can be found in the main text.

We next examined the expression patterns of the osteichthyan SPARC and SPARCL1 paralogues. In zebrafish, SPARC is first expressed in the otic placode at the beginning of somitogenesis (not shown). The expression in the otic vesicle persists until at least 72 hpf (figure 2g–j). At 18 hpf, expression is also detected in the notochord and in the somites (figure 2g). At 24 hpf, SPARC is expressed in the same territories as well as in the epidermis (figure 2h). The epidermal expression persists until 72 hpf (figure 2j). In 48 hpf zebrafish embryos, SPARC expression is detected in the epidermis, the posterior notochord, the somites, the otic vesicle and the apical ectodermal ridge of the pectoral fin (figure 2i). The 72 hpf embryos display a similar SPARC expression, except for the notochord where it is no longer detected (figure 2j). This expression pattern is in agreement with previously published data [59]. Regarding amphibians, our results in X. tropicalis are consistent with the previously published expression patterns of SPARC in Xenopus laevis embryos [60]. The X. tropicalis SPARC gene is specifically expressed in the notochord and neural plate of stage 21 neurula (figure 2k). In stage 26 embryos, SPARC is detected in the notochord, the presomitic mesoderm and the anterior region (figure 2l). In stage 31 tadpoles, transcripts are fading in the presomitic mesoderm and are detected at the level of the notochord, otic vesicle, lens, branchial arches and midbrain (figure 2m). In mouse, SPARC is strongly expressed in extraembryonic tissues at stages E7.5–E8.5 (figure 2n,o). At E8.5, expression is also detected in the brain (figure 2o). From E9.5 onward, SPARC is mainly expressed in the epidermis as well as in the vascular system (figure 2p,q). At E11.5, the expression is also detected in the heart of the embryo (figure 2p,r).

As SPARCL1 has been lost in amphibians [20,21], we focused on the D. rerio and M. musculus orthologues. In zebrafish, restricted SPARCL1 expression is first detected in 24 hpf stage embryos in the lens, the brain and at a lower level in the notochord (figure 2s). The lens and notochord expression persists until at least 72 hpf (figure 2t,u). Additional expression in the brain floorplate is observed at 48 and 72 hpf (figure 2t,u). At this later stage, SPARCL1 is also expressed in the oral cavity (figure 2u). In E7.5 mouse embryos, SPARCL1 is specifically expressed in the node (figure 2v). At E8.5, it is detected in the brain, the tailbud and the somites (figure 2w). At E9.5, the expression is strong in the dorsal part of the somites and in the tailbud (figure 2x). SPARCL1 is also expressed in the brain, mainly in the hindbrain, and in the branchial arches at this stage (figure 2x). At E10.5, expression is still observed in somites, branchial arches and brain, as well as in the vascular system and in the otic placode (figure 2y). The same holds true for embryos at E11.5, with an additional expression in the forelimb buds (figure 2z).

The onset of the amphioxus SPARCB expression occurs at the mid–late neurula stage, specifically in the differentiating notochord (figure 2a′; electronic supplementary material, figure S5). At the late neurula stage, before mouth opening, the expression persists in the notochord and appears in the pharyngeal endoderm (figure 2b′). In amphioxus larvae, SPARCB is expressed in the anterior-most part of the pharyngeal endoderm (oral cavity) and in the notochord (figure 2c′). In ascidians, SPARCB expression is first detected at the early gastrula stage in the endoderm precursors (figure 2d′). During neural plate stages, transcripts can be detected in the a-line-derived part of the neural plate (figure 2e′). At early tailbud stages, expression is detected in the posterior epidermis at the tip of the tail (figure 2f′). Finally, during later tailbud stages expression is seen in the three palp precursors (figure 2g′). Expression is stronger, and begins earlier, in the ventral most palp precursor (figure 2h′). In zebrafish, only a weak SPARCB expression could be detected in the lens between 24 and 48 hpf (see the electronic supplementary material, figure S5).

4. Discussion

(a). A new evolutionary scenario for the SPARC gene family

Both phylogenetic reconstructions and synteny conservation data reveal a surprisingly rich history of gene duplications and losses for the eumetazoan SPARC family, and the unsuspected existence of the SPARCB paralogy group. Based on our results, we propose the following scenario for the evolution of the SPARC gene family in eumetazoans (figure 3). A SPARC gene was present in the ancestor of eumetazoans and was duplicated before the divergence between cnidarians and bilaterians, giving rise to the SPARC/SPARCL1 and SPARCB paralogues. The SPARC/SPARCL1 gene was thereafter lost in cnidarians, whereas the SPARCB gene was lost in the ancestor of all protostomes. The fact that SPARC/SPARCL1 and SPARCB orthologues are present in the genomes of zebrafish, ambulacrarians, amphioxus and ascidians shows that both genes were retained in the deuterostome ancestor. In the vertebrate lineage, two whole-genome duplication events [58] gave rise to SPARC and SPARCL1, and to two SPARCB paralogues. In tetrapods, however, both SPARCB paralogues were lost. Moreover, only SPARC was retained in amphibians [20–22]. In teleosts, an additional genome duplication occurred was followed by extensive gene losses [61], resulting in the retention of one SPARC, one SPARCL1 and one SPARCB gene. Finally, the paucity of SPARCB sequences in teleost databases suggests that this gene was recurrently eliminated in this lineage. Because the zebrafish SPARCB gene is only weakly expressed in the lens, a tissue in which SPARC and SPARCL1 transcripts are abundant (figure 2i,j,t,u; electronic supplementary material, figure S5), it is tempting to propose that functional redundancy might have facilitated the loss of SPARCB in most other teleosts. Our synteny conservation analysis between amphioxus and osteichthyans shows that SPARC/SPARCL1 and SPARCB were linked in the genome of the ancestral chordate. No synteny conservation could be found in the other non-chordate deuterostomes that have both SPARC/SPARCL1 and SPARCB genes, which could be due to genomic rearrangements or to incomplete genome assembly. We, therefore, propose that SPARC/SPARCL1 and SPARCB arose from a tandem duplication that occurred in the last common ancestor of eumetazoans.

Figure 3.

Evolutionary scenario of the SPARC gene family in eumetazoans. The diagram depicts a phylogenetic tree of metazoan species whose SPARC homologues were analysed in this study. Whole-genome duplications are indicated at the base of each branch in which they were inferred to have occurred. We have considered here a scenario with no gene loss before these duplications and we have indicated the gene losses that probably occurred between these duplications and the ancestor of osteichtyans or teleosts. On the right the presence of SPARC family members with their syntenic relationships in different eumetazoan species is shown.

(b). Shared and derived expression patterns of SPARC homologues

Several conserved gene expression domains are evident in chordates for SPARC/SPARCL1 orthologues. First, a notochordal expression of SPARC/SPARCL1 orthologues is evident in amphioxus, tunicates, teleosts and amphibians (figure 2c,d–f, g,k–m,s) suggesting that this feature was already present in the last common ancestor of chordates. Second, the amphioxus SPARC/SPARCL1, the zebrafish and amphibian SPARC, and the mouse SPARCL1 orthologues are expressed in the paraxial mesoderm (figure 2a–c,g–h,k–m,w–z). Hence, the paraxial mesoderm-specific expression also represents a synapomorphy of chordates, but has been lost in ascidians concomitantly with the acquisition of a highly derived mode of embryogenesis. The fact that different vertebrate paralogues are expressed in this region (i.e. SPARCL1 in mouse and SPARC in zebrafish and amphibian) suggests independent losses of regulatory enhancers, as proposed by the duplication–degeneration–complementation model [62]. Third, we find that several tissues display a conserved expression of SPARC/SPARCL1 homologues, such as the brain (figure 2e,k–m,o), the otic vesicle (figure 2g–j,m,y) and the lens (figure 2i–j,m,s–t). Finally, some expression territories of SPARC or SPARCL1 are lineage-specific and include the teleost apical ectodermal ridge (figure 2i,j), the mouse extraembryonic tissue, vasculature, epidermis and heart (figure 2n–r) and the mouse node (figure 2v).

The expression patterns of the SPARCB orthologues in amphioxus, tunicates and zebrafish greatly differ from each other (figure 2a′–h′; electronic supplementary material, figure S5), suggesting that SPARCB function evolved faster than its SPARC/SPARCL1 paralogue. Below, we will discuss how these rapid changes of SPARCB expression patterns, and particularly its recruitment in the amphioxus notochord and the tunicate palps, might be informative regarding the evolution of the vertebrate skeleton.

(c). Is the origin of the vertebrate skeleton linked to the co-option of genes involved in collagenous matrix formation?

Our results strengthen the idea that the convergent co-expression of SPARC family members with collagen genes is a recurrent phenomenon in bilaterian evolution [22,63]. In the fruitfly, the SPARC/SPARCL1 orthologue functionally interacts with collagen-IV to stabilize the basal lamina of embryonic epithelia [11,15,16]. The ascidian and amphioxus SPARC/SPARCL1 orthologues and the osteichthyan SPARC genes are expressed in the notochord, a structure expressing a variety of fibrilar and non-fibrilar collagens [7,8,64]. It has been shown that fibrilar collagens from clade A, B and C were independently recruited to the notochord in the chordate ancestor [7]. Remarkably, such a convergent phenomenon also characterizes the SPARC family members, as demonstrated by the notochordal expression shared by the amphioxus SPARCB and its chordate SPARC/SPARCL1 paralogues (figure 2c,f,k,s,b′). Finally, in developing ascidian palps, the SPARCB gene is co-expressed with the fibrilar collagen gene CiFCol4 (this study and [7]). Altogether, these data suggest a close functional relationship between SPARC family members and collagenous proteins, which is supported by the presence of a well-conserved collagen-binding site in SPARC, SPARCL1 and SPARCB proteins (see the electronic supplementary material, figure S2). We, therefore, propose that one of the earliest steps of skeletal evolution involved the co-expression of SPARC/SPARCL1/SPARCB and collagenous proteins in an available pool of embryonic, multipotent, mesenchymal cells. Clearly, future directions in the field of skeletal evolution will involve the comparison of regulatory network architectures from unrelated chordate tissues that nevertheless have experienced a selective pressure to convergently acquire a SPARC-dependent collagenous extracellular matrix.