From commensalism to parasitism in Carapidae (Ophidiiformes): heterochronic modes of development?

Specimen collection

Specimens were obtained by snorkelling, scuba diving or trawling. The DNA retrieved from carapids (Carapidae family) represents 21 specimens, nine species and four genera, of which seven species are of the tribe Carapini. Except Echiodon drummondi Thompson, 1837, which was trawled at the Bay of Biscay, all specimens were hand-collected from inside their hosts by snorkelling or scuba diving in three locations: Madagascar (Toliara), French Polynesia (Moorea) and Corsica (STARESO station, Calvi). The collected specimens were brought back to the various laboratories (Aqua-Lab Laboratory of the Institut Halieutique et des Sciences Marines in Madagascar, CRIOBE laboratory in French Polynesia and STARESO Laboratory in Corsica) and placed in seawater tanks. Fish were removed from their host by dissection, or by depleting the hosts of oxygen using containers where the hosts were confined until the fish exited. Samples of integument were preserved in 100% ethanol for DNA extraction at 4 °C. Besides the collected specimens, data for the species Carapus bermudensis (Jones, 1874) were available from GenBank ( AP004404) and included in the present study (Table 1).

DNA extraction, Polymerase Chain Reaction (PCR) and DNA sequencing

Genomic DNA was extracted with the commercial Invitek Spin Tissue Mini kit (Invisorb). One nuclear gene (18S rDNA) and five mitochondrial genes were analysed in the present study (Table 1). DNA fragments from the mitochondrial large ribosomal subunit (12S rDNA, ca. 770 bp; 16S rDNA, ca. 1,220 bp), ATP synthase (ATPase 8 and 6, ca. 840 nucleotides), cytochrome b (Cytb, ca. 1,005 nucleotides) and cytochrome oxidase I (COI, ca. 592 nucleotides) were amplified by PCR (see File S1). All new sequences were deposited in GenBank (Table 1).

Outgroup, alignment, assessment of saturation, and site selection

As Echiodon drummondi and Onuxodon fowleri are considered as a sister group of Carapini in cladistic analyses based on morphological characteristics (Markle & Olney, 1990; Parmentier et al., 2000a), their monophyly and placement as a sister group of the Carapini was first tested by choosing three species of fish, Bassozetus zenkevitchi Rass, 1955, Cataetyx rubrirostris Gilbert, 1890 and Diplacanthopoma brachysoma Günther, 1887 as the outgroup with the 21 collected individuals as the ingroup. These fishes in the outgroup are members of Ophidiiformes and their selection was based on their phylogenetic affinities with pearlfishes (Miya et al., 2003) and the availability of the sequences of interest in databases (these fish presented five of the six sequences of interest: 18S rDNA sequences did not exist for any of these fish). As these first analyses revealed that E. drummondi and O. fowleri do form a well-defined monophyletic sister group to the Carapini, they were taken as outgroup of the Carapini in some subsequent analyses.

12S rDNA, 16S rDNA and 18S rDNA sequences were aligned using default parameters in the automatic multiple alignment program MUSCLE (Edgar, 2004); http://www.ebi.ac.uk/Tools/msa/muscle) and compared to the alignment obtained using default parameter settings in ClustalX (Thompson et al., 1997) . The two alignments differed by only 1 bp for the data set, a character that was consequently removed from the analyses. ATPase, Cytb and COI sequences were aligned according to the corresponding amino acid alignment. All alignments were inspected by eye for any obvious misalignments. The incongruence length difference (ILD) test (Farris et al., 1994) was implemented in PAUP*4.0b4a (Swofford, 2002).

The obtained matrix included 5,169 characters that were then analysed in MetaPIGA 2.1.3 (Helaers & Milinkovitch, 2010). Five sequences (two Carapus boraborensis (Kaup, 1856) from Madagascar and Tahiti; one Encheliophis gracilis (Bleeker, 1856) from Madagascar, one Carapus homei (Richardson, 1844) from Madagascar and one C. mourlani (Petit, 1834) from Madagascar) that had more than 40% of ambiguous positions were removed from the phylogenetic analyses. The two Carapus acus (Brunnich, 1768) from Corsica had similar sequences, and only one was retained. Invariable characters were detected in MetaPIGA and trimmed using the Gappyout trimming option (Capella-Gutiérrez, Silla-Martínez & Gabaldón, 2009). Divergence amongst sequences did not show any signs of saturation even when genes were tested separately. The resulting matrix that was subsequently analysed included 19 taxa and 5,060 characters.

Phylogenetic analyses

The complete data set, separated genes and combinations of genes were analysed using maximum parsimony (MP), Bayesian inference (BI) and maximum likelihood (ML) methods. In all analyses, gaps were treated as missing data. MP analyses were carried out using PAUP*4.0b4a (Swofford, 2002), with all characters unordered and equally weighted. Heuristic searches were conducted using random stepwise addition of the terminals for 1,000 replicates with the tree bisection reconnection (TBR) permutation algorithm and with maximum zero-length branches collapsed. The resulting trees were summarised via strict consensus. Clade support was assessed using bootstrapping (Felsenstein, 1985) of sites on 1,000 replicates.

ML analyses were performed using the Metapopulation Genetic Algorithm (MetaGA) using the software MetaPIGA 2.1.3 (Helaers & Milinkovitch, 2010). The alignment was considered in the analyses as a whole or partitioned according to the genes (six partitions). The probability consensus pruning was used among four populations of four individuals each, and the best-fitting ML nucleotide substitution model was selected on the basis of the Akaike Information Criterion implemented in MetaPIGA. The best models that sorted from the analyses were the GTR model with rate heterogeneity without invariant for the 12S, 16S and 18S and the HKY85 for the others. As MetaPIGA (in contrast to MrBayes) does not allow applying different models to different partitions, we tested both models on the non-partitioned and partitioned alignment. To generate an estimate of the posterior probability distribution of possible trees, we performed replicated MetaGA searches and stopped automatically when a series of mean relative error values among ten consecutive consensus trees remained below 5%, with a minimum of 100 replicates (Helaers & Milinkovitch, 2010).

Bayesian analyses were performed with MrBayes v3.0b4 (Ronquist & Huelsenbeck, 2003) on the gene-partitioned alignment using the models selected on the basis of the Akaike Information Criterion implemented in MetaPIGA. We also tested the GTR +I +Γ on the non-partitioned data set as it was the best model proposed during a Modeltest analysis (Posada & Crandall, 1998). In addition to the analyses made on the exclusive DNA matrix, we added a seventh partition consisting of 42 morphological characteristics. Some of the morphological characteristics had already been used previously by one of us (Parmentier et al., 2000a) in a cladistic analysis retracing the phylogeny of carapids; the others are new and come from new morphological observations (see File S2). For BI analyses, four independent Markov chain Monte Carlo (MCMC) were run simultaneously for 10 × 10⁶ generations and phylogenetic trees were sampled every 100 generations. The first 25,000 trees were discarded as burn-ins. The remaining trees were used to compute Bayesian posterior probabilities (BPP) for each clade of the consensus tree. The run was repeated twice to ascertain convergence towards the same posterior parameter distribution (Huelsenbeck et al., 2002).

Alternative phylogenetic hypotheses were compared statistically by means of the Kishino–Hasegawa ML ratio test implemented in PAUP* (Kishino & Hasegawa, 1989). Also, out of the 13 existing carapin species, the molecular analyses made here only apply to eight of them. The last five species, C. dubius (Putnam, 1874), C. sluiteri (Weber, 1913), E. chardewali (Parmentier, 2004), E. vermicularis Müller, 1842 and E. sagamianus (Tanaka, 1908), are rare species. We thus also performed MP analyses on the matrix including the 42 morphological characteristics (File S2). The analyses were done either without constraint or with a backbone constraint where we imposed the topology sorted out from the molecular phylogenetic analyses. Clade supports were also assessed using bootstrapping (1,000 replicates).

Mesquite 3.04 (Maddison & Maddison, 2015) was used to reconstruct the evolution of characters associated with symbiosis and with the sound apparatus (swim bladder) both under maximum likelihood criteria. These characters were chosen because they provide interesting information on the phylogenetic history of the Carapini as the communication (sonic mechanism) and the habitat (host) can both have a function in prezygotic barriers (Cocroft & Ryan, 1995; Colleye et al., 2011; Grant & Grant, 1996). Likelihood methods find the ancestral state(s) that maximizes the probability of the observed states (at terminal nodes) evolving under a defined stochastic model of evolution (Pagel, 1999). The Markov k-state 1-parameter model (Lewis, 2001) was used for the ML reconstruction. The character states considered in the symbiotic status were: free living (FL), commensal of bivalves (CB), commensal of holothuroids (CH), commensal of ascidiaceans (CA), opportunistic commensal (i.e., more than two classes as hosts; OC), parasite of holothuroids (PH) and opportunistic parasite (i.e., more than two classes as hosts; OP). These character states were mapped on the backbone constrained tree that included 14 species (see Fig. 3) to which the Pyramodontinae Snyderidia canina was added as outgroup. The character states considered in the sound apparatus (swim bladder) were the insertion with a long tendon (LT), the insertion with a short tendon (ST) and the presence of the tendon hook system (THS). These character states were mapped on the phylogenetic tree obtained with DNA and morphological data (9 carapid species; see Fig. 1) as the character states of the other species were unknown.

Common developmental features of the species

In all the examined specimens, the heads of the larvae differ from those of the adults in the proportions and positions of their components (Fig. 4). These differences are more or less important depending on the species, probably because they do not have the same larval duration in open ocean (Parmentier, Lo-Yat & Vandewalle, 2002b). The neurocranium of the larvae appears proportionally higher, the eye is larger and more posterior on the skull, the opercle is smaller, the suspensorium is less extensive. The lower jaws are shorter, but more elevated, the ratio between the height of the coronoid process and the length of the lower jaw being more important than in adults. In all the studied larvae, development seems principally linked to two ontogenic trajectories.

1. The posteroventral displacement of the quadratomandibular articulation. This transformation is accompanied by the posterior movement of the ventral part of the hyomandibula. Moreover, by the same apparent movement, the preopercle, the symplectic and the opercle become upright (Fig. 4). The posterior movement of the quadrate is also accompanied by lower jaw lengthening and later by the formation of a larger gape, as in adults.

2. The lengthening of the otic region is more important than that of the orbital region and the hyomandibula becomes proportionally wider.

Other development patterns are as follows:

• Initially, the metapterygoid appears isolated and later joins the ventral margin of the hyomandibula. On the other hand, the mesopterygoid is initially in contact with the palatine and the quadrate only. The backwards growth of its posterior and dorsal margins then brings the mesopterygoid in contact with the metapterygoid and the hyomandibula. The latter grows ventrally and forward, while the hyomandibula thickens and the preopercular grooves broaden (Fig. 4).

• At the level of the opercle, the subopercle progressively develops an elongated dorsal spine and denticles on its posterior margin. The V-like shape of the opercle is already present in the youngest collected larvae. In the course of development, the ventral part of the V elongates.

• The number of teeth on the palatine, dentary and premaxilla increase during the development of the tenuis stage. The larvae first have poorly developed dentition. Depending on the species, teeth are missing, emerging or developing (Fig. 4). Carapus and Encheliophis larvae bear a cardiform tooth on the premaxilla (Fig. 4).

The most intriguing characteristics are found in Encheliophis gracilis larvae, whose morphology is closer to that of larval and adult Carapus than to adult E. gracilis (Fig. 4). Numbers following in parentheses refer to the numbered arrow in Fig. 4. In Encheliophis and Carapus larvae, (1) the maxilla and the premaxilla are free from one another, (2) the posterior part of the maxilla is broadened distally and has a rounded end, (3) the ascending process of the premaxilla is more developed than the articular process, (4) the premaxilla bears several small conical teeth, (5) the subopercle presents the outline of a dorsal spine, (6) the lower jaw is relatively robust and (7) the three gill rakers (not shown in figure) of the first ceratobranchial are well developed (Parmentier, Lo-Yat & Vandewalle, 2002b). All these features also characterise adult Carapus (Parmentier et al., 2000a; Parmentier et al., 1998) but are not found in E. gracilis adults in which (1) the maxilla and the premaxilla are attached along their entire lengths by short connective fibres, (2) the maxilla is relatively thin and tapers to a point distally, (3) the ascending process of the premaxilla is less developed than the articular process, (4) the premaxilla bears only cardiform teeth, (5) the subopercle does not display a developed dorsal spine, (6) the mandible is slender and (7) the three gill rakers of the first ceratobranchial are not developed.

Discussion

One of the fascinating aspects of the life history of carapids lies in the understanding of their evolutionary transformation from a free living mode to a parasitic way of life. The initial relationship between carapid and host is thought to be accidental (Trott, 1970) since carapid species (e.g., Carapus dubius) and some species from the sister family Ophidiidae show a tendency to enter crevices and can assume tail-standing posture (Herald, 1953). The way some carapids (Schwarz et al., 2012) and Ophidion (Greenfield, 1968) enter their hosts and sand respectively is tail first. Among clownfishes, the first symbiotic fish was likely host specialists and subsequent speciation events led to a combination of generalist and specialist groups (Elliott et al., 1999). Carapus dubius is the only Carapini found in bivalves (Castro-Aguirre, Garcia-Dominguez & Balart, 1996; Parmentier, Castro-Aguirre & Vandewalle, 2000) as is the case for the Onuxodon species belonging to the sister taxon Echiodontini (Markle & Olney, 1990; Williams, 1984), supporting the assumption that molluscs were the first hosts of Carapidae. This species is, however, restricted to the North American coast (Markle & Olney, 1990). Another Carapini, C. sluiteri, was found in ascidians but it is difficult to discuss this species because to date only one specimen is known (Weber, 1913). The results from phylogenetic analysis also permit establishing a hypothesis of the relationships between Carapini and their hosts. The highest ML probabilities support other Carapus first opportunistic commensals (Fig. 3) as is the case for Carapus mourlani. This species inhabits gastropods, holothuroids and asteroids (Glynn et al., 2008). It has also the widest distribution of all Carapini, and is found from the East African to the West American coast (Markle & Olney, 1990; Parmentier, Mercier & Hamel, 2006). Carapidae were most probably first free living before having commensal relationships with hosts (bivalves and ascidians) deprived of chemical ways of defence. They were then able to enter into echinoderms and resist their chemical defences (Eeckhaut et al., 2015). Some species within this group finally developed parasitic relationships with these echinoderms.

Carapini show two morphotypes (commensal and parasite) that are closely tied to their ecological niche (Parmentier, Chardon & Vandewalle, 2002). The parasitic morphotype appeared once during the history of the fishes from the commensal morphotype. This is in accordance with the ecological niche of parasites being smaller, with a restricted number of hosts (Parmentier, Chardon & Vandewalle, 2002) and a narrower geographic distribution (Markle & Olney, 1990). Different studies have dealt with the carapini species that would be included in either Carapus or Encheliophis. Various studies have placed commensal morphotypes in Carapus: C. homei, C. boraborensis, C. acus, C. mourlani, C. bermudensis, C. dubius and C. sluiteri (Arnold, 1956; Parmentier et al., 2000a; Trott, 1970; Williams, 1984). All parasitic morphotypes have always been placed in Encheliophis. According to the molecular and consensus trees, Encheliophis is still monophyletic but Carapus appears to be paraphyletic. On the basis of the phylogenetic reconstruction and ontogenetic data, it is possible to draw a comprehensive scenario of the evolution of the tribe.

Adaptive radiation of a taxon results from morphological, physiological or behavioural modifications (Schluter, 2001). According to Mayr (1989), the entrance of a phyletic lineage to a new adaptive zone is linked to a morphological reorganisation leading to a new morphotype called a “rank”. Once a new rank is reached, minor morphological modifications allow the species to specialise for different ecological niches, promoting speciation (Schluter, 2009). The term “new adaptive zone” refers to a set of ecological niches that may be occupied by a group of species that exploit the same resources in a similar manner after the acquisition of morphological and/or physiological characteristics (Dumont et al., 2012; Mitter, Farrell & Wiegmann, 1988; Schluter, 2000). Morphological modifications that allow entry into a new adaptive zone can result from one or several changes of an ancestral plan (Zelditch & Fink, 1996) or from the emergence of novelties. In Carapini, the evolutionary jump from the commensal to the parasitic morphotype gave access to a new adaptive zone with which a new morphological type has appeared and radiated in different species. These morphological modifications usually appear at the end of ontogenetic development because the genetic developmental programme has no way of eliminating the reminiscent ancestral stages and is thus forced to modify them during the subsequent steps of development (Alberch et al., 1979; Mayr, 2001). The more phylogenetically closely related the species are, the later the phenotypic differentiations in development appear (Maglia, Pugener & Truer, 2001). Our data on the development of the head skeleton fulfils this assertion and concur with Haeckel’s views: new forms correspond to terminal modifications of features at the end of the ontogeny (Adriaens & Verraes, 2002; Lovtrup, 1978). The comparison between larval and adult Carapini shows that the Encheliophis gracilis larva is morphologically closer to the commensal morphotype than the parasitic morphotype (Fig. 4). During its ontogeny, the parasitic form goes beyond the adult stage reached by commensal form, i.e., one stage appears to have been added at the end of development (Fig. 4). These results confirm the phylogenetic results: the parasitic morphotype (=Encheliophis species) evolved from one of the commensal morphotypes and developed adaptations to a zone that differs from the ecological zone of the ancestral taxon. Some characteristics such as conical teeth on the upper jaw or gill rakers on the ceratobranchials 1 in the branchial basket could be the result of paedomorphosis: their development stops before the adult stage (Vandewalle et al., 1998). However, all modifications do not appear to be ontogenetic timing in Carapus but instead novel modifications of the shape trajectories. Modifications involving the upper jaws, lower jaws and opercle seem to be non-heterochronic modes of developmental reprogramming. More precisely, these modifications correspond to examples of “allometric repatterning”: ancestor and descendant differ in the trajectory of ontogenetic shape (Webster & Zelditch, 2005).

Some phylogenies reconstructed with acoustic signals were shown to be congruent with phylogenies based on morphological and molecular data (Malavasi, Collatuzzo & Torricelli, 2008). An interesting result of the present study is the position of Carapus homei which seems to represent a missing link between commensal and parasitic morphotypes. This result from the phylogenetic analysis (Fig. 1) is fully supported by morphological and behavioural features. Carapus homei shows all the characteristics associated with a commensal way of life. The morphology of the buccal and pharyngeal jaws clearly fits a diet based on elusive prey such as fish or crustaceans (Parmentier et al., 1998; Parmentier & Das, 2004; Vandewalle et al., 1998), however, the inner sound-producing apparatus is closer in structure to that of parasites. In all examined Carapus (C. boraborensis, C. mourlani, C. acus) it was recently shown that the primary sonic muscles (PSM) terminate in a complex tendon, the “tendon–hook” system (THS), which includes a “hook” that fits over a tubercle on the swim bladder (Parmentier et al., 2008). The sonic muscles contract slowly, pulling rostrally the anterior bladder. Sound is generated when the tension trips the THS and causes the bladder to snap back to its resting position (Parmentier et al., 2006b). Carapus homei, Encheliophis gracilis and E. vermiops lack the THS and, consequently have direct insertion of their PSM onto the swim bladder, meaning there are differences in the way of sounds are produced (Parmentier et al., 2008). Variation in acoustic signals could be involved in isolating mechanisms (Cocroft & Ryan, 1995; Slabbekoorn & Smith, 2002). Once again, the way the sound-producing muscles are inserted on the swim bladder supports the assumption that some characters are the result of paedomorphosis: the configuration of the short tendon is simpler than the tendon hook system.

The sound-producing system and resulting calls cannot explain the transition from the commensal to the parasitic morphotype. However, it has the advantage of providing a feature that does not seem to be under environmental constraints. Adaptive traits alone do not permit retracing the evolutionary history of a group; they only show the evolutionary stages. Orientation of evolutionary history needs the presence of characteristics apparently independent of a niche (Parmentier & Vandewalle, 2003). This seems to be the case for the sonic system in Carapini.