Transcriptome analysis

Total RNA of various embryonic and larval developmental stages (eggs, blastulae, gastrulae, swimming larvae, and metamorphosing larvae) and adult A. digitifera was isolated [34]. Total RNA was then fragmented into about 200 bp lengths and an RNA-seq library was prepared using a TruSeq RNA Sample Prep Kit v2 (Illumina). Libraries were sequenced (100 bp paired-end reads) using an Illumina GAIIx platform. Sequence data were mapped to the A. digitifera genome using TopHat ver. 2.0.8b [35]. Differential gene expression analysis and calculations of FPKM (Fragments Per Kilobase of exon per Million mapped fragments) for each A. digitifera gene model were performed using Cufflinks ver. 2.1.1 [36] with default parameters.

Proteomic analysis

The skeleton used was from the same A. digitifera colony from which the genome was decoded [34]. Specimens were stirred in 12.5% NaClO and washed with H₂O several times to remove soft tissue. The skeleton was mechanically crushed and ground into fine powder, and treated with NaClO again to remove remaining contaminants. The sample was washed in MilliQ water and then air-dried. The powder (10g) was decalcified with 1M acetic acid overnight and the de-calcified solution was centrifuged to separate insoluble material and solution. The insoluble pellet or acid insoluble matrix (AIM) was washed with MilliQ water several times. Then, the pellet was suspended in solubilization buffer (1% SDS, 10mM DTT, 50mM Tris-HCl (pH 8.0)). The acid soluble matrix (ASM) was precipitated from the soluble fraction using chloroform/methanol precipitation. Briefly, four volumes of methanol, one volume of chloroform, and three volumes of MilliQ water were added to one volume of soluble fraction. After mixing by vortexing, the sample was centrifuged at 14,000g, at 4°C for 5 min. The top, aqueous layer was removed. Then four volumes of methanol were added and centrifuged at 14,000g, 4°C for 5 min. The methanol was removed by pipetting and then the sample was air-dried. Finally, the sample was suspended in solubilization buffer.

ASM and AIM samples were dissolved in sample loading buffer for SDS-PAGE. Proteins were separated in a 10%-20% gradient gel, visualized with silver staining and CBB staining, and size-fractionated by dividing the gel horizontally into 16 pieces. Each piece of the gel was digested with trypsin for LC/MS/MS analysis as described previously [37]. Briefly, peptide digests were analyzed using a capillary liquid chromatography system (Ultima3000; DIONEX) connected online to a mass spectrometer (LTQ-XL, Thermo Scientific). Raw spectra were processed using SEQUEST software to extract peak lists [38]. Resulting peak lists were analyzed using an in-house MASCOT (ver. 2.3.2) server against A. digitifera sequences included in gene models and also against RNA-seq sequences, which were retrieved from MarinegenomicsDB (http://marinegenomics.oist.jp/genomes/gallery). The false discovery rate was set to 0.05. Proteins with a spectral count of more than 2 in each sample were treated as identified in this study.

Characterization and annotation of SOMP sequences

Protein sequences identified through proteomic analysis were analyzed using the InterProScan 5.3 platform [39] in order to find conserved functional domains. Signal peptide prediction was conducted with SignalP 4.0 [40], and non-classical protein secretion information was retrieved with SecretomeP 2.0 [41]. Proteins with high NN-scores (0.5 or higher) were regarded as possible extracellular proteins. Transmembrane domains were assessed with TMHMM 2.0 software [42]. Phosphorylation sites of amino acid sequences were predicted using NetPhos 2.0 Server [43]. All of these programs were run using default settings and thresholds.

In order to search for proteins with domain architectures similar to those of other metazoans, functional domain searching, described above, was also applied to a wide range of animals for which whole genome sequences are available. Genomic data used are listed in S1 Table.

For phylogenetic analysis, sequences of diverse metazoan animals were retrieved from public databases as described in S2 Table. Multiple sequence alignments of amino acid sequences were made with MUSCLE 3.8 [44] with default parameters, and visualized with Jalview version 2.8.2 [45]. The alignment was prepared with Gblocks version 0.91b [46] under default settings in order to remove poorly aligned positions if needed. Then, phylogenetic analysis was conducted using the maximum likelihood (ML) method in RAxML 8.1.3 [47]. The best-fit model of amino acid evolution for the tree was selected using ProteinModelSelection.pl script provided in the RAxML package. Neighbor-joining (NJ) trees were also obtained using ClustalW version 2.1 [48]. Reliability of the topology was checked by bootstrap analysis on the basis of 1,000 replicates for the ML and NJ methods. This dataset was also analyzed using PhyloBayes 3.3b [49] with the site heterogeneous mixture CAT model and two independent Markov chains. Convergence between the two chains was confirmed by comparing the frequency of their bipartitions.

Identification of coral skeletal matrix proteins

In order to identify coral skeleton matrix proteins, A. digitifera gene models version 2 and a transcriptome database [34] were searched. Fig 1 and S3 Table list A. digitifera SOMPs. Proteins from ASM and AIM were separated by gel electrophoresis (S1 Fig) and their analysis after tryptic digestion allowed us to identify 30 proteins (Fig 1, S3 Table). The presence of all these SOMPs in the skeletal proteome was supported by more than one unique peptide identified by LC/MS/MS analysis (S3 Table). All proteins found in ASM were also detected in AIM. Comparisons among three coral skeleton proteomes (A. digitifera, A. millepora, and S. pistillata) showed that 27 A. digitifera proteins had significant sequence similarity to skeletal matrix proteins of at least one of the other two coral species, and the remaining three proteins (USOMP -9 to -11) were newly identified in the A. digitifera skeletal proteome (S2 Fig). Twelve SOMPs were conserved among three species, while fourteen proteins were shared between Acropora species but not found in S. pistillata. One protein (vitellogenin) was detected from the A. digitifera and S. pistillata proteomes, but not from A. millepora. Four SOMPs of A. digitifera were not described in the proteome of the closely related A. millepora, while 12 SOMPs in the A. millepora proteome have not been detected in the A. digitifera skeleton.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. List of skeletal organic matrix proteins in A. digitifera.

See also S3 Table. a: The presence of a signal peptide at the N-terminus predicted by SignalP [40]. b: Possible extracellular localization predicted by SecretomeP [41]. c: Presence of a transmembrane domain predicted by TMHMM [42]. d: Phosphorylation detected by MS/MS. e: Orthologous protein in the A. millepora proteome [29]. f: Orthologous protein in the S. pistillata proteome [26].

https://doi.org/10.1371/journal.pone.0156424.g001

Protein localization and classification

The in vivo localization of proteins predicted using SignalP, SecretomeP, and TMHMM software showed that almost all of the identified proteins are either secretory or transmembrane proteins (Fig 1, S3 Table). Sixteen SOMPs have been categorized as “Transmembrane” or “Extracellular matrix-like” based on their predicted localization. Both the “Cysteine-rich” proteins and “Acidic” proteins are secretory. The former are rich in cysteine residues while the latter have high percentages of aspartic acid and a signal peptide; two of them also contain a transmembrane domain near the N-terminus. Finally, six secretory SOMPs are called “Uncharacterized” after Ramos-Silva et al. [29], since they do not have any conventional conserved domains.

Gene expression profile

RNA-seq data obtained from embryonic and larval stages (e.g., egg, blastula, gastrula, planula, and polyp) and from adults included 29,446,547 read pairs for egg, 27,117,706 for blastula, 29,692,832 for gastrula, 29,400,663 for planula, 29,385,101 for polyp and 26,597,707 for adult. Photographs of Acropora digitifera developmental stages are shown in Fig 1 of Shinzato et al. [34]. Polyp and adult RNA libraries correspond to stages in which this species forms a calcareous skeleton (Fig 2). Gene expression analysis using RNA-seq showed three expression patterns. First, eight genes encoding cadherin, neurexin, EGF and laminin G domain-containing proteins (dcps), REJ dcp, zona pellucida dcp, multi-copper oxidase (MCO), and laminin G dcp, were expressed in earlier developmental stages before skeleton formation, as well as during polyp and adult stages (Pattern 1). Second, 18 genes that encode MAM and LDLr dcp, TSP and VWA dcp, CUB dcps, vitellogenin-like protein, EP-like proteins, galaxins, Cys-rich protein, SAP2, SAARPs, USOMP-1, and 9 began to be expressed in the planula or polyp stage (Pattern 2). Third, four genes (mucin4-like, SAP-1, USOMP-10, and -11) were expressed only during the adult stage (Pattern 3).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Gene expression patterns of SOMPs and SEM images of coral skeletons of polyp and adult stages.

(a) Heatmap of SOMP gene expression at different developmental stages. Three expression patterns are indicated at the left: continuous expression before and during calcification (Pattern 1), strong expression commencing during the planula or polyp stages (Pattern 2), and exclusive expression in adult stage (Pattern 3). The color key represents FPKM of normalized log₂ transformed counts. Orange to red color intensity indicates higher gene expression. Polyp and adult stages, in which corals generate calcium carbonate skeletons, are boxed in black. (b, c) SEM images of polyp (b) and adult (c) skeletons. The individual polyp used here was fixed seven days after settlement. Soft tissue of each sample was removed with 10% NaOH. After settlement, the disk-like structure was deposited onto the substrate. Then, a network structure similar to that of an adult skeleton began to form.

https://doi.org/10.1371/journal.pone.0156424.g002

Annotation of Acropora digitifera SOMPs

We searched functional domains in SOMP sequences and compared them to proteins having the same or similar domain architecture encoded in other metazoan animal genomes (Fig 3). Domain architectures of transmembrane and extracellular matrix-like SOMPs are shown in Fig 4. Other cnidarian (Nematostella and Aiptasia) proteins with similar domain architecture are also depicted for comparison. Most of the domain architectures of transmembrane and extracellular matrix-like SOMPs, except TSP-1 and VWA dcp, are comparable to those of non-calcified cnidarian proteins. The domain architecture of each SOMP is described in more detail below.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Number of proteins with specific domain architecture encoded in animal genomes.

a: Figures in parentheses indicate number of proteins without transmembrane domain.

https://doi.org/10.1371/journal.pone.0156424.g003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Domain architectures of SOMPs from Acropora digitifera and sea anemone proteins similar to the SOMPs.

(a) Cadherin. (b) Neurexin. (c) REJ domain-containing protein (dcp). (d) Zona pellucida dcp. (e) Multi-copper oxidase. (f) MAM and LDL receptor dcp. (g) TSP-1 and VWA dcp. (h) Mucin4-like protein. (i) CUB dcp. (j) Vitellogenin-like protein. Lengths of amino acid sequences are shown at the right.

https://doi.org/10.1371/journal.pone.0156424.g004

Annotation of transmembrane proteins

Proteins with one or more transmembrane domains were found in the skeletal proteome. Cadherin, neurexin, EGF and laminin G dcps, REJ dcp, zona pellucida dcp, MCO, MAM and LDL receptor dcp, TSP-1 and VWA dcp, and mucin4-like protein were categorized as transmembrane proteins (Fig 1).

The Acropora SOMP encoded by transcriptome contig adi_EST_assem_2804 and gene model aug_v2a.19518 had the domain structure typical of classical type-III cadherin [50], including 31 extracellular cadherin domains, 2 laminin G domains, 3 EGF-like domains, a single transmembrane domain, and a classical cadherin cytoplasmic domain (CCD) (Fig 4A, S3 Fig). This domain architecture is widely found in both protostomes and deuterostomes (Fig 3). The intracellular p120-catenin- and β-catenin-binding domains are also retained (S3 Fig). Thus, we annotated this SOMP as cadherin or CDH.

One coral skeletal matrix protein encoded in gene models aug_v2a.24512, had laminin G domains, EGF-like domains, and a transmembrane domain (Fig 4B, S4 Fig). Its overall domain structure was comparable to that of Drosophila neurexin IV (S4 Fig), which is the transmembrane component of septate junctions [51–53]. We named this SOMP following the suggestions of Chapman et al. [53], who described a putative cnidarian ortholog as “Neurexin.” The domain structure similar to neurexin is expanded in cnidarian species [54] (Fig 3). Two gene models (aug_v2a.06122 and aug_v2a.06123) encoded SOMPs with one or two laminin G domains and one EGF-like domain (S4 Fig). We named them “EGF and laminin G dcp -1 and -2”. In addition, one SOMP, encoded by aug_v2a.15580, was annotated as “Laminin G dcp” since it carried two tandem laminin G domains (S4 Fig). Although this protein does not have a signal peptide or a transmembrane domain, we tentatively classified it as an extracellular matrix-like protein based on its overall domain architecture. Alignment of these four SOMPs showed significant similarity to laminin G and EGF-like domain regions, as well as non-domain regions (S5 Fig), indicating a common origin of these genes.

Gene model aug_v2a.02830 encoded a multi-pass transmembrane protein with a Receptor for Egg Jelly or REJ domain (Fig 4C, S6 Fig), which is found in the polycystic kidney desease-1 (PKD1) gene family [55, 56]. The A. digitifera REJ domain-containing SOMP has a series of conserved domains, including WSC, PKD repeats, REJ, GPS, and a PLAT domain. PKD1 homologs have been reported in deuterostome animals, including sea urchins [55] and humans [56]. A functional domain search showed that sea anemones also have REJ domain-containing proteins and that the overall domain architecture is very similar to that of the coral SOMP (S6 Fig). It has been reported that the sea urchin SpREJ3 and human Polycystin-1 are cleaved at the G-protein-coupled receptor cleavage site (GPS) [57, 58]. Sequence alignment of the GPS domain showed that the cleavage site is highly conserved among cnidarians and deuterostomes (S6 Fig), suggesting that the coral REJ dcp may also undergo proteolytic processing.

Zona pellucida (ZP) domain proteins are extracellular matrix components found in a diverse array of tissues from many animals [59]. Genes encoding one ZP domain are extensively found in metazoan animals (Fig 3). The primary structure shows that the ZP domain includes consensus cysteine residues that are responsible for intra- and inter-molecular disulfide bridges and polymerization of the proteins (S7 Fig) [60]. In general, ZP proteins undergo proteolytic cleavage and the moiety that carries the ZP domain is released to the extracellular space [59, 61]. Anthozoan ZP proteins retain the conserved cleavage sequence (Arg-X-Lys/Arg-Arg) [62], which is located between the ZP and transmembrane domains (S7 Fig).

Multi-copper oxidase (MCO), with six cupredoxin domains, was also found in the coral skeletal proteome (Fig 4E). In vertebrates, MCOs are involved in iron efflux in various tissues [63], and the ceruloplasmin-type MCO family includes three genes: ceruloplasmin (Cp), hephaestin (Heph), and hephaestin-like (Hephl, also known as zyklopen) [63, 64]. Sequence alignment of mouse six-domain MCOs and the A. digitifera MCO in the skeleton proteome revealed that the coral MCO retains metal-binding residues in the cupredoxin domains (S8 Fig). Six-domain MCOs had not previously been described from non-chordate animals [65]. We found MCOs with six domains in various non-chordate genomes (Fig 3, S9 Fig). With the aim of clarifying evolutionary relationships among MCOs and of annotating the Acropora MCO SOMP precisely, a molecular phylogeny of MCOs was constructed (S10 Fig, S2 Table, S1 Data). The result clearly shows that cnidarian MCOs are clustered together and distinguished from chordate MCOs. Therefore, we annotated this SOMP as multi-copper oxidase or MCO. In addition, two Acropora MCOs, present in the skeleton proteome, are closely clustered.

Gene models aug_v2a.09968 and 09969 were tandemly arranged in the scaffold, and cDNA sequence adi_EST_assem_4944 encoded C-terminus of 09968 and N-terminus of 09969, respectively (S11 Fig). Therefore, it is concluded that these two gene models are actually a single gene. The protein product consists of a long repeat of an MAM domain with an LDL receptor repeat, as well as other extracellular matrix domains of fibronectin, EGF, and Trefoil (Fig 4, S12 Fig). The tandem MAM and LDL receptor domains are present in vertebrate transmembrane proteins such as Endotubin (also known as apical early endosomal glycoprotein) [66] and Diet1 [67]. Proteins with MAM, LDLr, and EGF domains were found in cnidarians and bilaterian animals (Fig 3, S12 Fig). Notably, Nematostella has two gene models (protein IDs, 198667 and 239043) that are tandemly arranged in scaffold 11, and the domain structures of the two proteins are highly similar to the N- and C- terminal halves of the A. digitifera SOMP, respectively. Thus, it is likely that N. vectensis also has homologs of coral MAM and LDLr domain-containing proteins.

The SOMP corresponding to gene model aug_v2a.05945 had three VWA domains, successive TSP-1 domains, and two F5_F8_type_C domains (Fig 4G, S13 Fig). Genome-wide domain searches showed no significant global similarity to available sequences in public databases including those from the N. vectensis, Aiptasia, and H. magnipapillata genomes, except for A. millepora SOMP (Fig 3, S13 Fig). Hence, we named this protein “TSP-1 and VWA domain-containing protein.”

The SOMP encoded by gene model aug_v2a.09809 was annotated as a mucin4-like protein, since it contained NIDO, AMOP, VWD, and EGF domains, which are typically present in mucin4 proteins [68] (Fig 4H, S14 Fig). We categorized this protein as a transmembrane protein based on its considerable sequence similarity to the A. millepora mucin-like SOMP [29], although the A. digitifera sequence lacked a signal peptide and a transmembrane domain, possibly because both ends of the sequence were truncated. Coral mucin4-like SOMPs have additional TSP1 domains at their N-termini (Fig 4H, S14 Fig). We surveyed public databases and there is no protein sequence with the same domain architecture in any other metazoan (Fig 3).

Annotation of extracellular matrix-like proteins

Two CUB domain-containing SOMPs, with signal peptides at their N-termini and no other known domains, were found in the skeletal proteome (Fig 4I). Proteins with a single CUB domain are widely distributed in animal genomes (Fig 3). A. digitifera CUB domain-containing protein 1 (aug_v2a.14283) and 2 (aug_v2a.10941) were closely similar to A. millepora CUB domain-containing protein and Threonin-rich protein (S15 Fig). These four proteins retain a threonine-rich sequence that is absent in other animal genomes.

A. digitifera skeletal proteome includes a protein with vitellogenin-like domain architecture, which is composed of vitellogenin N-terminus and VWD domains (Fig 4J). This domain architecture is common to metazoan animals (S16 Fig). However, the coral vitellogenin-like SOMP is unique since the protein carries an additional protein kinase domain (Figs 3 and 4J, S16 Fig).

Two gene models, aug_v2a.22918 and aug_v2a.00002, encode SOMPs similar to A. millepora USOMP-5 (S17 Fig). These Acropora SOMPs share a partial sequence that is significantly conserved in coral egg proteins (EPs) identified from Euphyllia ancora [69] and Galaxea fascicularis [70]. We therefore annotated these SOMPs as “EP-like” 1 and 2. It is notable that the conserved sequence is also encoded in the Nematostella genome (S17 Fig).

Annotation of galaxins and cysteine-rich proteins

Galaxin is a coral-specific secretory, skeleton matrix protein that has tandem repeat motifs containing di-cysteine sequences (Cys-Cys). Galaxin was originally identified from the coral, Galaxea fascicularis [71]. Two galaxins, Adi-galaxin and Adi-galaxin2, have been identified in the A. digitifera proteome (Fig 1). Galaxins are rich in di-cysteines, which appear 19 times in Adi-galaxin and 13 in Adi-galaxin2 (S3 Table). In addition, another cysteine-rich SOMP was also characterized. This Cys-rich SOMP contains 51 cysteine residues that comprise 10.49% of the mature protein sequence, comparable to the galaxins (12.7% for Adi-galaxin and 11.76% for Adi-galaxin2, respectively). The Cys-rich SOMP had tandem repeat motifs with conserved cysteine residues (S18 Fig), while only 4 di-cysteine blocks were found.

Annotation of acidic proteins

Genes encoding skeletal acidic proteins or SAPs were predicted in the A. digitifera genome sequence [34] and it is confirmed that SAP-1 and SAP-2 are components of the skeletal matrix of A. digitifera (Fig 1). Orthologs of their proteins have been identified in A. millepora, but not in S. pistillata [26, 29]. BLAST searches showed that cDNA sequences corresponding to SAP-1 and -2 were not found among ESTs of the corals, Montastraea faveolata and Porites astreoides; therefore, SAPs appear unique to Acropora species [34].

We have found two gene models that are highly similar to A. millepora secreted acidic Asp-rich protein (SAARP) 2. Alignment of these amino acid sequences deduced from the corresponding cDNAs (S19 Fig) indicated that adi_EST_assem6252 and 6253 may be transcripts of the 5’ and 3’ halves of the SAARP gene. In addition, gene models encoding these two cDNAs (aug_v2a.01441 and aug_v2a_01140, respectively) were located adjacently in the genome (S19 Fig). Therefore, we conclude that these two gene models are really a single gene that encodes Adi-SAARP2. The counterpart of Ami-SAARP1 was also detected in the A. digitifera proteome. In addition, we have identified another acidic protein encoded in the gene model aug_v2a.06327. We named the protein “Adi-SAARP3,” since its sequence is very similar to that of SAARP1 and 2, both in the acidic and non-acidic regions (S20 and S21 Figs). These Acropora SAARPs are also similar to CARP4 and 5 of S. pistillata [26]. Interestingly, SAARPs and CARPs share significantly similar sequences between acidic domains (S20 and S21 Figs), and the conserved amino acid sequence is also encoded in Nematostella, Hydra, and, unexpectedly, also in bivalve (Pinctada and Crassostrea) genomes (S21 Fig), while none of them possess the acidic region (data not shown).

Annotation of uncharacterized coral-specific SOMPs

The proteome of A. digitifera contained six unique proteins that showed no significant similarity to any proteins in public databases. Three of them that were comparable to Ami-USOMP-1, were named “Adi-USOMP-1a, b, and c” (S22 Fig). Adi-USOMP-9 displays partial similarity to USOMP-1 proteins (S22 Fig). Two SOMPs named “USOMP-10 and -11” were novel proteins, and there is no corresponding protein in A. millepora skeletal proteome. Signal peptide searches and secretome detection analysis indicated that all six USOMPs are extracellular proteins (Fig 1, S3 Table).

Tandem arrangement of SOMP genes in the A. digitifera genome

In A. digitifera, two laminin G and EGF dcp genes and two USOMP-1 genes were tandemly arranged in the genome scaffold (S23 Fig). These genes are likely to have emerged due to gene duplication events. High conservation of amino acid sequences (S5 and S22 Figs) and similar gene expression patterns (Fig 2) imply their functional redundancy.

Identification of coral SOMPs

Scrupulous attention needs to be paid to avoid contamination with proteins not associated with the skeleton, since cellular debris may remain if the coral skeleton is inadequately cleaned [27, 28]. We treated skeleton samples with extended bleaching to remove soft tissue attached to the skeleton surface [29]. We propose that the proteins identified here are veritable SOMPs, for several reasons. First, cell membrane ATPase, which has been detected in calicoblastic cells, but not in the skeleton [23], was not found in the A. digitifera skeletal proteome. Second, most proteins presented here were also detected in previous, independent experiments in different species, using different cleaning methods. Third, several transmembrane protein families contain various proteins, but only one specific member of the family was found in the present proteomic analysis. For example, the A. digitifera genome has 51 gene models that encode cadherin domains, but only one cadherin domain-containing protein, of which an ortholog is detected in the A. millepora and S. pistillata skeletal proteomes, is evident among A. digitifera SOMPs. Fourth, in the case of transmembrane proteins, only peptide sequences presumably located in extracellular side were identified in the skeletal proteome (S3 Table). Furthermore, while transmembrane neurexin is found among coral SOMPs, cytoplasmic components of the septate junction, including Gliotactin, Disc large, Scribble, and Coracle [53] were not detected in the coral skeletal proteome. This confirms that the coral skeleton proteome was not contaminated by intracellular components and that neurexin polypeptides are actually associated with the extracellular skeletal matrix. For all of these reasons, the proteins listed in Fig 1 are highly likely to be SOMPs and intimately incorporated into the coral skeleton.

Twenty-six proteins in the A. digitifera skeleton proteome are shared with A. millepora (Fig 1, S2 Fig). However, some A. millepora SOMPs, such as carbonic anhydrase, peptidase, and collagen, were not detected in A. digitifera. One possible explanation is that the gene gain/loss or change of protein function that occurred after the divergence of A. digitifera and A. millepora, resulted in different skeletal matrix proteomes in these two species. However, such a significant change is unlikely to have occurred after the separation of these closely related species. Therefore, we believe that the difference between the two Acropora proteomes is likely due to technical variability. Two A. digitifera SOMPs, vitellogenin and Adi-USOMP-11 were not identified in A. millepora SOMP (S2 Fig) and their presence in A. digitifera SOMP was supported by a relatively low number of peptides (18 and 9, respectively, S3 Table). It remains possible that these extracellular proteins are randomly incorporated into the skeleton. It is also probable that enzymes like carbonic anhydrase and peptidase, which were exclusively identified from A. millepora SOMP, function in soluble form and not in the solid skeleton. We speculate such enzymes are incidentally included in the coral skeleton, and that they are weakly associated with calcium carbonate crystals. Thus, these proteins are easily removed by strict sample cleaning. Further experiments with different sample cleaning and extraction conditions and studies of other coral species should confirm lineage-specific and conserved coral skeleton proteins.

The in silico prediction of protein localization suggests that all A. digitifera SOMPs are exposed in the subcalicoblastic space, where calcification occurs. Two gene models (aug_v2a.09809 and adi_v2a.15580) and their corresponding mRNA sequences predict no signal peptides at the N-termini and show low NN-scores, presumably due to truncated sequences. These proteins (laminin G dcp and mucin4-like protein) may also be located in the extracellular space, according to their functional domain architecture (S4 and S14 Figs).

Gene expression profile of SOMPs during development

Gene expression analysis of SOMPs throughout coral development provides new insights into the biological role of the proteins. Many transmembrane SOMP and laminin G dcp genes were expressed throughout developmental stages (Fig 2), indicating that these proteins have general cellular functions related to development, in addition to skeleton formation. Sixteen genes are expressed in both polyp and adult stages, suggesting that they are responsible for larval and adult skeleton formation. This group includes MAM and LDLr dcp, TSP1 and VWA dcp, extracellular matrix-like proteins, cysteine-rich proteins, acidic proteins except SAP-1, and uncharacterized protein -1 and -9. Other genes, encoding mucin4-like, Adi-SAP1, USOMP10, and 11, are expressed only in adults. Therefore, these proteins have a specialized role in adult calcification.

The differential expression of SOMP genes in polyps and adults (Fig 2A) indicates that the molecular basis of initial skeleton formation is not identical to that of the adult skeleton. This is concordant with SEM observations, showing that settled polyps generate disk-like structures on the substrate during the initial step in skeleton formation (Fig 2B), and subsequently, they begin to construct a network structure similar to adult skeleton (Fig 2C). Moreover, smaller and less ordered crystals in the primary skeleton [31, 33] may also be related to the lack of adult-specific SOMPs, and it is also possible that there are polyp-specific SOMPs that were not examined here. Further investigation including the larval skeleton proteome will be necessary in order to complete the inventory of coral skeleton SOMPs.

Transcripts of SOMP genes encoding CUB domain-containing protein, Galaxin, SAARP1, USOMP-1b, and -1c, were detected weakly in the egg and prawnchip stages (Fig 2). This may occur because maternal mRNAs are provided to the egg and are retained in early developmental stages in Acropora [72], but it is not certain that these SOMP gene transcripts are translated and functional in early development.

Putative function of coral SOMPs

In total, twelve transmembrane proteins were identified in the A. digitifera skeletal proteome (Fig 1, S3 Table). Interestingly, many of these, including cadherin, neurexin, REJ dcp, ZP dcp, MCO, MAM and LDLr dcp, were remarkably similar to transmembrane proteins of non-calcifying animals, such as sea anemones (Fig 4).

Cadherin is a cell-cell adhesion protein shared by all metazoans [50, 73, 74]. The overall domain structure of coral cadherin SOMP is highly similar to the counterpart found in other metazoans (Figs 3 and 4A, S3 Fig), and the conserved p120-catenin- and β-catenin-binding domains in the CCD suggest that the intracellular region is functional, interacting with the actin cytoskeleton, mediated by catenins [75] (S3 Fig). The gene expression profile showed that the A. digitifera cadherin gene is expressed throughout development (Fig 2). Therefore, the cadherin SOMP may have inherent functions of cell-cell adhesion and may function in a variety of developmental processes [73, 76, 77], later acquiring the additional function of skeleton formation. One possible role of the cadherin SOMP in skeleton formation is the link between cells and organic matrix in the skeleton, inferred from its primary role in cell adhesion. Detachment of the polyp from the skeleton was observed when the animal was subjected to low calcium seawater [78], possibly because cadherin, the adhesion function of which is calcium-dependent, mediates attachment of the animal. Besides, it is known that in some biological processes, the classical cadherin ectodomain can be released into the extracellular space by proteolytic cleavage [79, 80]. Coral cadherin SOMP may also be shed, and the extracellular peptides with cadherin domains may help to build up the organic matrix. It is possible that intact, cell surface cadherin molecules combine with the cleaved moiety that constitutes the organic matrix, forming a close association between ectodermal cells and skeleton. Similarly, REJ dcp, and ZP dcp carry a proteolytic cleavage site upstream of the transmembrane domain (S3 and S4 Figs). N-terminal polypeptides of these proteins may be released into the extracellular subcalicoblastic space and where they constitute the organic matrix. Cell membrane proteins exposed to the extracellular space may connect to the organic matrix, and participate in settlement on the substrate, as well as maintaining the attachment to the skeleton.

On the other hand, the remaining transmembrane SOMPs do not contain such specific cleavage domains. Ramos-Silva et al. [29] showed that extracellular proteases were detected in the A. millepora proteome and that those enzymes were responsible for digestion of transmembrane proteins. The same process may also explain the presence of peptides of transmembrane proteins in the A. digitifera skeletal proteome. In addition, it is also possible that transmembrane proteins are anchored to the cell membrane and interact with calcium carbonate crystals in vivo. For example, Adi-SAARP2 and 3 carry one transmembrane domain at their C-terminal ends while the aspartic acid-rich domains are positioned at the N-terminal ends. It is conceivable that the Asp-rich domains bind to the calcium carbonate surface and make close attachments between calicoblastic ectoderm and the crystal surface. The mammalian multi-copper oxidase, ceruloplasmin, has an affinity for calcium ions [81], giving rise to the idea that Adi-MCO can mediate the interaction between cell membranes and calcium carbonate crystals. The idea of cell-skeleton interaction mediated by transmembrane proteins is congruent with previous direct observations that calicoblastic ectoderm is intimately associated with the mineral phase [19, 22]. Desmocytes, which are the anchoring cells in calicoblastic ectoderm [82], maintain the attachment to the skeleton by extending fibrils into the organic matrix [83]. Desmocytes associated with the organic matrix are responsible for attachment to the substrate of the octocorallian soft coral, Dendronephthya hemprichi [84], suggesting that the primal role of the anthozoan organic matrix produced by aboral ectoderm is for settlement of the polyp, and that it was subsequently utilized for skeletal organization in the scleractinian lineage.

In A. digitifera SOMPs, there are many extracellular matrix protein domains, such as cadherin, EGF, laminin_G, VWA, MAM, LDLr, fibronectin, TSP-1, VWA, vitellogenin, and CUB. Proteins and other organic components of the extracellular matrix, such as polysaccharides, interact with each other via these domains and constitute the organic framework. Galaxins and Cys-rich SOMPs may also be involved in the organic matrix framework [71].

Acidic proteins are key components in biomineralization, since aspartic acid (Asp) and glutamic acid (Glu) are negatively charged at neutral pH and are able to interact with calcium ions [85, 86]. Many vertebrates and invertebrates employ acidic proteins for biomineral formation [87–91]. Previous studies have shown that aspartic acid and glutamic acid are major components of the amino acid fraction of the coral skeleton [11, 92, 93]. Our proteomic analysis detected many polypeptides derived from acidic proteins. For example, the spectral counts of Adi-SAP2 and Adi-SAARP1 are 4,703 and 3,310, respectively, and these are the most abundant A. digitifera SOMPs (S3 Table). This means the aspartic acid-rich proteins are abundant in the A. digitifera skeletal proteome. Therefore, SAPs and SAARPs are candidate proteins for major roles in CaCO₃ crystallization including nucleation, growth, and inhibition. Acidic transmembrane proteins (SAARP-2 and -3) may also mediate the close association between cells and skeletal crystals. In addition to acidic regions, SAARPs contain a conserved sequence that is present in Nematostella, Hydra, and two bivalve genomes (S21 Fig). Six of these non-coral proteins are secretory (data not shown), implying that the conserved sequence may have a similar function in the extracellular space.

Evolutionary origin of coral SOMPs

We classified A. digitifera SOMPs into three evolutionary phases, according to the similarity of protein domain structure to non-coral cnidarian (Nematostella and Aiptasia) proteins (Fig 5, Phases I, II, and III). Many transmembrane SOMPs have apparent homologs in cnidarian and bilaterian genomes (Fig 4, S3, S4, S6, S7, S9, S10 and S12 Figs), indicating that these proteins were present in uncalcified cnidarian ancestors (Fig 5, Phase I).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Evolutionary and stepwise acquisition processes for coral skeleton formation.

In Phase I, pre-existing proteins that functioned in cellular activity and developmental processes in non-skeletal cnidarians, may have acquired additional functions, presumably for attachment to the substrate and to constitute the initial extracellular organic matrix. In Phase II, novel proteins emerged after gene duplication, domain shuffling, and rapid molecular evolution, and became specifically involved in constructing the organic matrix framework. Finally, unique proteins emerged in the coral lineage (Phase III). They may have interacted with other organic matrix proteins to play important roles in crystal formation.

https://doi.org/10.1371/journal.pone.0156424.g005

TSP-1 and VWA dcp, mucin4-like, and vitellogenin-like protein contain unique domain architecture that is absent in genomes of non-skeletal cnidarians and other metazoans (Figs 3 and 4, S13, S14 and S16 Figs). We defined these as Phase II proteins (Fig 5). The TSP1, VWA, and F5/F8 type C domains are widely distributed among phylogenetically disparate organisms; however, these three domains do not normally occur in the same protein (Fig 3, S13 Fig). Coral mucin4-like SOMPs have repeated TSP1 domains at their N-termini, in addition to typical mucin4 domain structure, and the overall domain structure is unique to corals (Figs 3 and 4H, S14 Fig). Likewise, Acropora vitellogenin-like SOMP is composed of similar domain architecture to vitellogenin, and specifically has a protein kinase domain (Fig 4J, S16 Fig). It is hypothesized that these three SOMPs appeared in the coral lineage by domain shuffling after divergence of corals from other anthozoans.

Phase II proteins also include CUB dcps, EP-like proteins, galaxins, Cys-rich, and SAARPs. The CUB dcps and the EP-like proteins retain partial sequences that are comparable to non-coral cnidarian proteins (Fig 4I, S17 Fig). The restricted sequence similarity indicates these proteins have been modified to engage coral-specific, skeletal matrix formation process. Proteins similar to galaxin, with repeated motifs containing di-cysteine, are found in non-skeletal cnidarians such as Hydra and Nematostella [29]. Detectable sequence similarity suggests that coral cysteine-rich SOMPs and non-coral cnidarian proteins may share an evolutionary origin. Therefore, we put galaxin and Cys-rich SOMP in the Phase II category (Fig 5). Galaxin-like genes have been reported in Acropora species [34, 94], while protein products of these genes are not present in the skeleton proteome. Cysteine-rich proteins may originally have been responsible for biological processes unrelated to biomineralization, then utilized for skeletal formation after rapid evolution of this gene family in the coral lineage. SAARPs include a conserved region that is quite similar to those of non-coral cnidarian and bivalves (S21 Fig). The conserved sequence may be derived from a metazoan ancestor, and may have an unknown function. The acidic amino acid regions that are thought to be important for crystallization processes were specifically acquired by SAARPs. In addition, lineage-specific expansion of SOMP genes seems to have occurred among CUB dcp, EP-like, galaxin and SAARP genes. Gene duplication events are also evident in matrix proteins of molluscs [95] and vertebrates [96], reflecting the rapid molecular evolution of biomineralization-related genes.

Finally, SAPs and uncharacterized SOMPs are classified as Phase III (Fig 5), since they have unique sequences such as an acidic domain that is not present in non-coral cnidarians. The evolutionary origin of these proteins is unclear, but it is possible they emerged by rapid molecular evolution, which expunges detectable sequence similarity to non-coral animal proteins.

Stepwise evolution of coral biomineralization

In biomineralization, organic matrix provided by cells plays a pivotal role in controlling hard tissue formation. The present and previous studies confirm that more than 30 proteins are present in the coral skeletal proteome.

Based on thorough gene annotation of coral SOMPs, stepwise evolution of coral biomineralization is hypothesized. We found that many genes are closely related to those of the sea anemones, N. vectensis and Aiptasia, indicating that coral genes were derived from genes of a non-calcified anthozoan ancestor (Fig 5, Phase I). The majority of transmembrane proteins are involved in this phase. The coral skeletal organic matrix is also composed of newly innovated proteins that manifest reduced similarity to other cnidarian proteins (Fig 5, phase II). Partial sequence conservation in domains indicates that these proteins can be traced to those of non-calcified cnidarians or common metazoan ancestors. These proteins are thought to have emerged by domain shuffling or rapid molecular evolution, and to have been recruited to function in the skeletal organic matrix. Finally, skeletal acidic proteins and uncharacterized proteins in the organic matrix are thought to have emerged after coral divergence from non-skeletal anthozoans (Fig 5, Phase III). To examine the hypothesis, experimental functional analysis such as gene knockdown, identification of SOMPs from different coral species, and comparative genomics using more cnidarian species are needed.