We collected sequence data for 18S rRNA, 28S rRNA, H3 and COI genes in 33 galeommatoidean species belonging to 25 genera and two families, five putative galeommatoidean species belonging to Peregrinamor and Basterotia, and eight outgroup species for the molecular phylogenetic analysis (Table 1). To examine whether Peregrinamor and Basterotia do belong to Galeommatoidea, we selected eight outgroup species representing a range of lineages within Heterodonta, to which Galeommatoidea belongs. We also included three non-heterodont species representing each of three non-heterodont orders to root the entire heterodont tree. All sequences were newly obtained in this study except for the sequences of three non-heterodont species (Additional file 1). Ingroup specimens included 13 free-living species and 25 symbiotic species associated with the crustacean, echinoid, holothurian, cnidarian, sipunculan, and echiuran animal groups (Figure 1; Table 1).

Taxonomic classification follows Bieler et al. (2010).

Phylogenetic analysis based on the combined data (18S + 28S + COI + H3; Additional file 2) suggests that the currently circumscribed Galeommatoidea is not monophyletic and includes Peregrinamor and Basterotia (Figure 2). The expanded Galeommatoidea with the above two genera was strongly supported as a monophyletic group and consists of at least six major clades, most of which are supported by high clade support (Figure 2). However, the relationships among these clades received low support and remained obscure (Figure 2). Notably, the branches leading to Neaeromya and the three species of Basterotia are especially long, which likely further complicates the recovery of branching order among these major lineages. Thus, while we consider that the monophyly of each of the six major clades is supported, the relationships among these clades must be viewed with caution. The seemingly high Bayesian posterior probability values at higher nodes should also be taken with caution because Bayesian posterior probabilities often produce overcredible results when compared with bootstrap analyses 21 22 23 .

Our molecular phylogenetic analysis suggests that Galeommatoidea includes Basterotia and Peregrinamor, which were previously assigned to Cyamioidea and Mytiloidea, respectively 18 . Basterotia bivalves differ from other galeommatoideans in having the inhalant siphon located posteriorly 20 , while Peregrinamor bivalves were included in Mytilidae on the sole basis of superficial resemblance of their shells to those of true Mytilidae 19 . However, our result firmly places these enigmatic genera within Galeommatoidea (Figure 2). Our analysis further reveals that the expanded Galeommatoidea is composed of six major clades. Below we detail major discrepancies between the current classification of Galeommatoidae and phylogenetic relationships obtained in this study. Detailed accounts of the morphology and ecology of each of the major clades are provided in Additional file 3.

Recent provisional classification suggests that Galeommatoidea comprises two families, Galeommatidae and Lasaeidae 17 . However, our results did not support this taxonomic circumscription (Figure 2) and suggested that Galeommatidae and Lasaeidae are polyphyletic (Figure 2). Except for Scintillona stigmatica, the members of the family Galeommatidae were grouped into Clade 1, which also included several lasaeid species (Figure 2), whereas the members of Lasaeidae were divided into six major clades (Figure 2). Lasaeidae was previously divided into several families, namely Lasaeidae, Kellidae, and Montacutidae 18 . However, these previously proposed families were also each non-monophyletic and included genera that were separated into different major clades (Figure 2). These results suggest that the previous family-level classification of Galeommatoidea does not correspond to any of the well-supported clades recovered in our molecular phylogenetic analysis. Furthermore, according to our results, some of the previous genus-level classifications were also not supported. For example, Scintilla and Nipponomysella are polyphyletic (Figure 2). Yet, the discrepancies between the traditional classification and the results of our molecular phylogenetic analysis are not unexpected, considering that many members of Galeommatoidea have reduced or highly specialized morphologies associated with symbiotic life, which potentially obscures the historical information contained in their morphology.

To investigate the evolutionary pattern of symbiotic lifestyles in Galeommatoidea, we mapped information on the lifestyle (symbiotic or free-living), host taxon, and host association mode of each galeommatoidean species onto the phylogenetic tree (Figure 3). The host organisms we recorded in this study were consistent with those reported in previous studies (Additional file 4), except for Pseudopythina aff. nodosa and Nipponomontacuta actinariophila. The host organism of the latter species was misidentified in the original paper 24 ; we obtained N. actinariophila from the sea anemone Telmatactis sp., which conforms to more recent records.

Clades 1 and 2 each comprise both free-living and symbiotic species, whereas Clades 3–6 consist of only symbiotic species (Figure 3). The approximately unbiased (AU) test 25 rejected the hypothesis that free-living or symbiotic species are monophyletic in Galeommatoidea (P ≪ 0.0001, Additional file 5b), suggesting that transitions between free-living and symbiotic lifestyles occurred multiple times in this bivalve group. Associations with crustaceans, holothurians, sipunculans, and echiurans are each found in multiple major clades (Figure 3). The AU tests also rejected the hypothesis that species with same host taxa are monophyletic (all P values ≪ 0.0001, Additional file 5c–f), suggesting that host specialization to these animal groups likely occurred multiple times independently in Galeommatoidea. In addition, species having very similar modes of physical attachment to the hosts were seen in different major clades (Figure 3). For example, the bivalves attached to the abdomen of thalassinidean shrimps were divided into Clades 4 and 5 (Figure 3), and those attached to the body of intertidal crabs were also divided into Clades 1 and 4. These results were also supported by AU tests (all P values ≪ 0.0001, Additional file 5g–h). The hypothesis that the species symbiotic with crustaceans in Clade 1 are monophyletic was also rejected (P ≪ 0.0001, Additional file 5i), whereas this was not the case in Clade 4 (P = 0.065; Additional file 5j). These results suggest that the gain and/or loss of symbiotic association with crustaceans likely occurred multiple times at least in Clade 1. Although the AU test does not account for uncertainty in phylogenetic estimation, all the strongly rejected tests concern species belonging to different major clades, each with high statistical support, indicating that our conclusions are unlikely affected by phylogenetic uncertainty.

Clades 2, 4 and 6 each includes species utilizing hosts from different animal phyla (Figure 3), suggesting that interphylum host switches occurred repeatedly even within these major clades. In contrast, Clades 1, 3 and 5 each includes species symbiotic with a single phylum (Figure 3).

The ecological modes of host associations are diverse in the Galeommatoidea 11 12 13 20 . We categorized host association modes into four types: ectosymbiotic on the host body, living inside the host burrow, endosymbiotic inside the host esophagus, and living inside the shells carried by host hermit crabs (Table 1). The former two are the predominant host use patterns in Galeommatoidea (Table 1), whereas the latter two are each found in only a single species. According to the phylogenetic tree (Figure 3), the ectosymbiotic species belonged to five major clades, whereas burrow associates belonged to four major clades. The AU tests rejected the hypothesis that ectosymbiotic species or burrow associates are monophyletic (both P values ≪ 0.0001, Additional file 5k, l), suggesting that there have been repeated shifts between these two modes of host utilization in this bivalve superfamily.

Symbiotic associations with crustaceans, holothurians, sipunculans, and echiurans occurred multiple times independently within Galeommatoidea (Figure 3). Below, we compare the groups that independently established symbiotic associations with the same animal groups to better understand the evolution of host associations in Galeommatoidea.

In most parasitic organisms, phylogenetic tests demonstrate that host associations are generally conserved across the phylogeny and that switches between distantly related hosts are relatively infrequent 38 39 40 41 42 . In contrast, we found multiple possible instances of host switching between different phyla by galeommatoidean bivalves (Figure 3), raising the question as to what factors promote such a pattern of host switching in this superfamily.

The predominant mode of host switching in parasitic organisms is that between closely related hosts 38 39 40 41 42 . This is because parasites are under strong pressure to overcome host defenses, and switches to phylogenetically distant hosts likely require entirely new sets of anti-defense adaptations. Parasitic life also involves exploitation of nutrients from host organisms. Thus, parasites undergoing host shift must adapt to drastic changes in the nutrients they derive from the new host. These and other ecological conditions probably constrain the host range of parasitic organisms and make it difficult to switch between distantly related hosts (e.g., across phyla). In contrast, the symbiotic galeommatoideans are predominantly filter feeders and do not depend directly on their hosts for nutrients, although they indirectly benefit from the water currents created by the host 12 . Furthermore, while symbiotic galeommatoideans benefit greatly from being associated with their host organisms, they usually cause little or no harm to their hosts. In fact, there are no known host defense mechanisms that specifically target these symbionts, although the bivalves themselves have evolved adaptations to stay cryptic on the host body or inside the host burrow. Thus, the commensal lifestyle of these bivalves probably allows them to switch hosts without a need to adapt to host physiology or evolve other defense mechanisms. Overall, these ecological attributes of galeommatoidean bivalves probably made them frequent colonizers of various marine invertebrate hosts.

Our results suggest that host switching played an important role in the diversification of Galeommatoidea. However, this superfamily also includes a large number of free-living species 15 , whose diversification obviously did not occur by host switching. In addition, several well-supported clades of symbiotic bivalves include species that share closely related hosts, suggesting that mechanisms other than host switching have also been important in the evolution of these bivalves. For example, the closely-related Anisodevonia ohshimai, Entovalva lessonothuriae and Devonia semperi in Clade 6 all utilize sea cucumbers as hosts, and thus, cospeciation (i.e., parallel speciation of hosts and parasites) and/or shift in the mode of host utilization may have been the major driver of speciation. Similarly, Basterotia species (Calde 3) commonly share the same echiura host species 34 , indicating that host switches are unlikely to have occurred in this genus. Clarifying alternative mechanisms of speciation other than host switching is therefore needed to gain a full understanding of the diversification history of Galeommatoidea.

Overall, our molecular phylogenetic analysis has greatly progressed our understanding of the phylogenetic relationships within Galeommatoidea. However, the relationships among the major clades, as well as those within each clade, are still poorly resolved. Inclusion of more species in the analysis and better marker choice are probably needed to obtain higher resolved molecular phylogenies of this bivalve superfamily. Such well-resolved phylogenies can then be used to further test questions such as (1) the direction and frequency of transitions between free-living and symbiotic lifestyles in Galeommatoidea, (2) the relative importance of host switching as compared to other speciation mechanisms, as discussed above, and (3) patterns of morphological, behavioral and physiological specialization/reduction associated with symbiosis with diverse invertebrate hosts.

We collected 38 specimens from 33 galeommatoidean and five putative galeommatoidean species belonging to 27 genera and three families. For outgroups, we sampled five species belonging to five non-galeommatoidean families within Heterodonta (Table 1). All specimens were collected in southwestern Japan with the exception of Neaeromya rugifera, which was collected in Oregon, USA (see Additional file 1). We also included the sequence data of three non-heterodont outgroup species that were available in GenBank (Additional file 1) to root the entire heterodont phylogeny.

Total DNA was isolated following a previously described method 43 . Adductor muscle tissue was homogenized in 800 μl lysis buffer and incubated at 55°C overnight, after which 80 μl saturated potassium chloride was added to the lysate. This solution was incubated for 5 min on ice and then centrifuged for 10 min. The supernatant (700 μl) was transferred to a new tube, cleaned once with a phenol/chloroform solution, and precipitated with an equal volume of 2-propanol. The DNA pellet was rinsed with 70% ethanol, vacuum-dried, and dissolved in 100 μl TE buffer.

We sequenced the fragments of the nuclear 18S and 28S ribosomal RNA (rRNA), H3 and the mitochondrial COI genes. Polymerase chain reactions (PCRs) were used to amplify ~1700 bp of 18S rRNA, ~1000 bp of 28S rRNA, ~350 bp of H3 and ~700 bp of COI. Amplifications were performed in 20 μl mixtures consisting of 0.4 μl of forward and reverse primers (primer sequences are provided in Additional file 6), 1.6 μl of dNTP, 2.0 μl of ExTaq buffer, 0.1 μl of ExTaq polymerase (TaKaRa, Otsu, Japan), and 15.1 μl of distilled water. Thermal cycling was performed with an initial denaturation for 3 min at 94°C, followed by 30 cycles of 30 s at 94°C, 30 s at a gene-specific annealing temperature (Additional file 6), and 2 min at 72°C, with a final 3 min extension at 72°C. The sequencing reaction was performed using the PCR primers and internal primers (Additional file 6) and the BigDye Terminator Cycle Sequencing Ready Reaction Kit (Applied Biosystems, Foster City, CA) and electrophoresed on an ABI 3130 sequencer (Applied Biosystems). The obtained sequences have been deposited in the DDBJ/EMBL/GenBank databases with accession numbers AB714745–AB714907 (Additional file 1).

Sequences of the 18S, 28S, H3 and COI genes were aligned using Muscle 44 as implemented in the software Seaview 45 46 under default settings. The alignments of H3 and COI sequences did not require insertion of gaps and was therefore unambiguous. We used the software Gblocks v0.91b 47 48 49 to delimit ambiguously aligned regions in the 18S and 28S alignments (Additional file 7), and the following phylogenetic analyses were conducted with and without alignment-ambiguous regions. The full 18S and 28S alignments contained 474 and 282 variable sites, respectively, and 453 and 249 variable sites when alignment-ambiguous regions were excluded. The H3 and COI alignments contained 87 and 297 variable sites, respectively, indicating that despite their short sequence lengths, they contain comparable amount of information as the 18S and 28S partitions. Because the initial phylogenetic analyses of individual genes did not produce essentially different results (Additional file 8), we focused our analyses on the combined four-gene data set, which will be described below.

Phylogenetic trees were obtained by the maximum-likelihood (ML) and Bayesian methods. For the ML analysis, model selection and tree search were conducted using the TreeFinder program 50 51 . The robustness of the ML tree was validated by bootstrap analysis with 1000 replications using the same program.

Bayesian analyses were performed using MrBayes 3.1.2 52 with substitution models chosen using MrModeltest 2.3 53 . In the combined data set, substitution parameters were estimated separately for each gene using the ‘unlink’ command. Two independent runs of Metropolis-coupled Markov chain Monte Carlo were performed simultaneously, sampling trees every 100 generations and calculating the average standard deviation of split frequencies every 1000 generations. Using the ‘stoprule’ option, analyses were continued until the average standard deviation of split frequencies dropped below 0.01, at which point the two chains were considered to have achieved convergence. Because the average standard deviation of split frequencies was calculated based on the last 75% of the samples, we discarded the initial 25% of the sampled trees as burn-in. We confirmed that analyses reached stationarity well before the burn-in period by plotting the ln-likelihood of the sampled trees against generation time.

To evaluate the evolutionary history of symbiotic life in Galeommatoidea, we mapped information on the lifestyle (symbiotic or free-living), host taxon, and mode of host utilization of each galeommatoidean species onto the phylogenetic tree. Host information was based on our sampling data. We checked our sampling data against previously available information on host association for each bivalve species (see Additional file 4) to confirm the validity of our observations of their lifestyle and host taxon.

We tested the hypothesis of monophyly of species sharing the same lifestyle (free-living and symbiotic), modes of host utilization, or host taxa using the AU test 25 . The analyses were done using the combined four-gene data set. The alternative trees were obtained by ML heuristic search under the topological constraint (Additional file 5), and the AU test was conducted based on 1,000,000 replications using Treefinder 50 51 .