Commensal associations and benthic habitats shape macroevolution of the bivalve clade Galeommatoidea

Abstract

The great diversity of marine life has been shaped by the interplay between abiotic and biotic factors. Among different biotic interactions, symbiosis is an important yet less studied phenomenon. Here, we tested how symbiotic associations affected marine diversification, using the bivalve superfamily Galeommatoidea as a study system. This superfamily contains large numbers of obligate commensal as well as free-living species and is therefore amenable to comparative approaches. We constructed a global molecular phylogeny of Galeommatoidea and compared macroevolutionary patterns between free-living and commensal lineages. Our analyses inferred that commensalism/sediment-dwelling is likely to be the ancestral condition of Galeommatoidea and that secondary invasions of hard-bottom habitats linked to the loss of commensalism. One major clade containing most of the free-living species exhibits a 2–4 times higher diversification rate than that of the commensals, likely driven by frequent niche partitioning in highly heterogeneous hard-bottom habitats. However, commensal clades show much higher within-clade morphological disparity, likely promoted by their intimate associations with diverse hosts. Our study highlights the importance of interactions between different ecological factors in shaping marine macroevolution and that biotic factors cannot be ignored if we wish to fully understand processes that generate marine biodiversity.

1. Introduction

Large-scale biodiversity patterns are shaped by complex interactions between geological, biological, and ecological processes [1–3]. Both abiotic and biotic factors modulate long-term evolutionary dynamics and undoubtedly contribute to the generation and maintenance of global biodiversity [2,4]. In marine ecosystems, there is ample evidence for abiotic drivers, such as major tectonic events [5], nutrient availability [6], and climate/sea-level-induced vicariant breakpoints [7]. The importance of biotic factors, though relatively much less studied, has also been emphasized in post-mass extinction faunal recoveries [8], in adaptive escalations [9,10], and in evolutionary radiations of symbiotic lineages [11–13].

One ubiquitous type of biotic interaction is symbiosis—a close and often long-term ecological association between different species (i.e. mutualism, commensalism, and parasitism). Compared with other types of biotic interactions (e.g. predation), symbiosis has long been considered an exceptional phenomenon, and it was not until recent decades that the prevalence of symbiosis in nature became recognized [14]. Symbiotic associations allow the partners to acquire new metabolic capacities [15], phenotypic characteristics [16], ecological functions, and even genomic compositions [17]. They are important sources of evolutionary novelty and offer opportunities for organisms to occupy otherwise inaccessible ecological niches [18]. The marine ecosystem encompasses diverse symbiotic associations, yet their macroevolutionary consequences are less well understood.

To help illuminate the influence of symbiosis on large-scale marine biodiversification, we consider the marine bivalve superfamily Galeommatoidea. This lineage makes for an attractive case study because (i) it has exceptional species diversity and phenotypic disparity, enabling statistical inference of macroevolutionary patterns, (ii) it has significant diversity in all major marine benthic habitats, and (iii) it includes both symbiotic and free-living species, enabling comparative studies.

Galeommatoidean bivalves are a hyperdiverse but poorly studied marine superfamily with a fossil record extending to the Cretaceous [19,20]. These small-bodied (often less than 1 cm) bivalves present exceptional morphological disparity and remarkable innovations (figure 1), including pronounced shell reduction/internalization [21] and elaborated soft tissue structures. The superfamily comprises approximately 500 described species [22], although many more species remain undescribed [23]. There is little consensus regarding the taxonomic and phylogenetic relationships within the superfamily [19,23], presumably owing to their small body sizes and fragile shells that often lack comparable, homologous features.

Figure 1.

(a) Time-calibrated molecular phylogeny of Galeommatoidea. Coloured tip labels indicate the lifestyle of each morphospecies, summary host information for commensal subclades are shown, and node labels indicate the posterior probability of each branching event. Pie charts on supporting branches represent the probability of commensal or free-living being the lifestyle at the respective nodes. Photos on the right show exemplars of representative galeommatoidean bivalves; the colour square at the bottom right of each photo indicates the lifestyle of the respective bivalve. (b) The same topology as (a), with branch length proportional to rate of speciation estimated using Bayesian analysis of macroevolutionary mixtures (BAMM). The clade labelled with a blue star corresponds to the star clade in A. (Photo credits: P. Maestrati and A. Anker.)

Over the past decade, the application of comprehensive sampling methodologies to marine ecosystems has catapulted Galeommatoidea from relative obscurity to the apex of Bivalvia biodiversity (electronic supplementary material, table S1) [23–25]. Galeommatids were found to be the most diverse bivalve family and the sixth most diverse molluscan family at an intensively studied coral reef site in New Caledonia [24]. Paulay [23] similarly found Galeommatidae s. l. (= Galeommatoidea) to be the most diverse bivalve group on Guam and speculated that its actual diversity is several times greater than any co-occurring bivalve family.

Galeommatoidea exhibits a striking ecological dichotomy in that some species are free-living, whereas others have obligate commensal relationships with diverse burrowing invertebrate hosts, including crustaceans, holothuroids, echinoids, and sipunculans, etc. [26,27]. A recent ecological synthesis [27] revealed that this lifestyle dichotomy is coupled with benthic habitat types. Free-living species are typically found in hard-bottom habitats, hidden in rock and coral head crevices, whereas commensal species are typically restricted to soft-bottoms habitats, where they occur within the oxygenated envelope produced by their bioturbating hosts. This is because intense predation in the top few centimetres of soft-bottom habitats exerts strong selection pressure on fragile, small-bodied galeommatoideans to maintain commensal associations, thereby achieving depth refuges. The pressure is much weaker in hard-bottom habitats as abiotic crevices greatly exceed the number of potential shelters created by burrowing invertebrates [27].

The prevalence of commensalism in Galeommatoidea raises the possibility that this trait has contributed to its remarkable diversity. Given that speciation by host shift occurs frequently in symbiotic lineages, we hypothesize that a major part of galeommatoidean diversity is driven by genetic isolation caused by host-switching and phenotypic adaptation to diverse hosts. We predict that commensal taxa exhibit higher rates of lineage diversification and higher morphological disparity compared with free-living species.

Galeommatoidea is globally distributed, and some species are known to have broad ranges [21,28,29]. Therefore, a meaningful macroevolutionary study requires a comprehensive phylogenetic framework based on a multi-basin, global sampling strategy. To date, our knowledge of galeommatoidean phylogeny and life-history evolution is limited. Here, we reconstructed a global-scale molecular phylogeny of Galeommatoidea taking advantage of several large-scale international biodiversity expeditions (electronic supplementary material, text and figure S1) as well as museum collections and published sources [26]. We gathered ecological and morphological information on the bivalves and compared patterns of lineage diversification and trait evolution between commensal and free-living species.

2. Material and methods

(a). Sampling and phylogenetic analyses

The majority of specimens were collected from extensive biodiversity expeditions in the Philippines, Vanuatu, Madagascar, Australia, and Mozambique. Additional specimens were loaned from 10 zoological museums (electronic supplementary material, dataset S1). Potential sampling bias was assessed by comparing our sampling records with documented species listed in the World Register of Marine Species (WoRMS) database (electronic supplementary material, text sections 1.1–1.2).

A time-calibrated Bayesian phylogeny of Galeommatoidea was reconstructed using four genes: 16S rRNA, 28S rRNA, histone H3, and adenine nucleotide translocator. The tree topology and divergence times were estimated simultaneously in Beast v. 1.7.3 [30]. The minimal age of the superfamily (crown group) was set based on the earliest documented appearance of Galeommatoidea in the fossil record (105.6 Ma [20]). Three independent Markov chain Monte Carlo (MCMC) analyses were run for 100 million iterations, respectively. Convergence and reliable effective sampling size values (more than 500) were ensured. A maximum credibility consensus tree and 1 000 evenly sampled post burn-in posterior trees (to accommodate for phylogenetic uncertainties, referred to as posterior trees hereafter) were used for further analyses. See electronic supplementary material, text section 1.3 for detailed methods on sequence generation and phylogenetic analysis.

(b). Evolutionary history of lifestyles

To estimate the phylogenetic signal of the lifestyles (free-living and commensal), mean Pagel's λ [31] was calculated with 95% confidence intervals from the 1 000 posterior trees using the R [32] package phytools 0.2.46 [33]. The signal was calculated twice where species with unknown lifestyles were treated as commensal and free-living, respectively. P-values were calculated using a likelihood ratio test, comparing the estimated model with a null model where λ was fixed to zero.

Ancestral lifestyles of four backbone nodes (figure 1a) over the posterior trees were estimated using the discrete and MultiState methods [34] implemented in BayesTraits v. 2.0. The nodes were specified using the ‘addMRCA’ option, and probabilities of the two lifestyles at the ancestral nodes were estimated using an MCMC approach. Two independent chains were run for 10 million iterations, respectively, and sampled every 1 000 iterations. Convergence of the two runs was confirmed, and results were combined after a 10% burn-in. Posterior probabilities of the ancestral states were visually represented as pie charts.

(c). Lineage diversification

Diversification rates in free-living and commensal lineages were estimated using the software BAMM [35]. Two MCMC chains were run for 10 million iterations and sampled every 10 000 iterations, assuming an estimated 75% missing taxa based on WoRMS. The two chains converged quickly and were combined with a 10% burn-in of each. Mean speciation rates of all branches were calculated and used to scale the branch lengths of the original phylogeny.

The more commonly used binary state speciation and extinction [36] approach for testing trait-dependent diversification was not selected because of its model inadequacies on traits with high phylogenetic signals [37]. The BAMM method does not assume any a priori classification of taxa and allows shifts of diversification parameters to occur along any branch in the tree, and therefore provides estimated diversification rates independent of ecological characters.

(d). Morphological analyses

Shell shape outlines of 174 species were captured using landmarks and semi-landmarks [38] (see electronic supplementary material, text section 1.4 for detailed procedures). Multiple individuals (2–7) per species were included whenever possible. All morphometric manipulations were conducted, using the R package geomorph v. 1.1.0 [39]. The final aligned Procrustes shape coordinates were used in the subsequent analyses. Log centroid sizes for all species were calculated and used as representations of shell sizes.

To assess the standing disparity of Galeommatoidea morphology, a principal component analysis (PCA) was performed on the aligned coordinates of all species. Scores of the first two PCs were plotted, and shell shapes on extreme axes were presented to show how general shell shape changes along PC1 and PC2. Standing disparities of the free-living and commensal taxa were compared using a multivariate homogeneity test of group dispersions [40] based on the first 20 PCs. PCA analyses were also performed on the free-living and commensal species separately to assess shape variations within each group. The results were plotted, respectively, and individuals were colour-coded based on the subclades to which they belong. Welch's two sample t-test was performed to test whether free-living and commensal taxa differ significantly in shell size.

Phylogenetic signals of both shape and size data were estimated in geomorph using the multivariate K statistic (K_mult) [41]. To test whether the two lifestyles exhibit significantly different K_mult, the difference between commensal and free-living K_mult was calculated over all posterior trees to obtain a distribution, and a one-sample t-test was used to determine whether the mean of this distribution differed significantly from zero.

To assess how shell shape disparities evolve across the phylogeny, disparity through time (DTT) analyses [42,43] for the shape coordinates of free-living and commensal taxa were conducted, respectively.

Null distributions of the DTT curve were generated from 1 000 Brownian simulations. The DTT analysis calculated the ratio between the average within-subclade disparity and the total disparity in the phylogeny (i.e. mean relative disparity) at all nodes in the chronogram, then compared the observed values with values simulated under a Brownian motion (BM) model. Deviations from the BM simulation were summarized as the morphological disparity index (MDI) [44]. MDI values close to zero indicate that most of the morphological variations are partitioned among subclades, whereas values close to one indicate high within-subclade disparity. To compare MDI values of the two lifestyles over all posterior trees, a one-sample t-test was again used to determine whether the differences were significant.

Lastly, we compared the patterns of morphological evolution for commensal and free-living taxa by fitting the BM model (random walk with rate parameter σ²) as a null model and three ecologically relevant modifications: Ornstein–Uhlenbeck (OU) [45], early burst (EB) [43], and speciational evolution (SE) [46]. The OU model constrains the random walk with a central tendency whose strength is proportional to α. In the EB model, the rate of trait evolution decreases over time with rate parameter a. The SE model allocates a portion of morphological divergence as step changes at speciation events, and the fraction of such change is represented by ψ.

Several possible scenarios may apply to commensal trait evolution. First, owing to the close attachments to the hosts' body surface, the bivalves may not exceed certain sizes and their shell shapes may conform to a certain type. In this case, an OU [45] model with a strong constraint force (high α-value) could be favoured. However, given that the commensal taxa are associated with diverse hosts, adaptations to different hosts may require different shell shapes. If host lineage fidelity is low (i.e. closely related bivalves occupy distantly related hosts) and shell morphologies change at host-switching events, then the morphological evolution could show very low phylogenetic conservatism. This is likely to produce a trait distribution pattern that resembles a weakly constrained OU process (low α-value) [43].

On the other hand, if the commensals show high host lineage fidelity (i.e. closely related bivalves occupy similar hosts), then most of the morphological divergence could occur at the initial divergence among commensal clades that occupy different hosts; trait evolution would slow down within each host-specific clade. In this case, the EB model could be the best fit.

The SE model serves as another alternative, where step morphological changes occur at each speciation event, regardless of host–bivalve interactions. We predict that the SE model or BM model would fit the free-living taxa better.

Model-fitting analyses were performed on free-living and commensal taxa, respectively. The EB model for size evolution was evaluated using geiger 1.99.2, and the same model for shape evolution (multi-variant) was evaluated using the R function fitContinuousMV written by G. Slater (http://fourdimensionalbiology.com). The remaining three models for both size and shape evolution were evaluated using the R package motmot v. 1.0.1 [47]. Model comparison was conducted based on the multivariate-corrected Akaike information criterion with correction for finite sample sizes (AICc) [48].

3. Results

(a). Phylogenetic relationships

We examined more than 1 000 galeommatoidean specimens from biodiversity surveys and museum collections. The final phylogeny includes 97 species currently considered valid and 120 unidentified morphospecies, spanning 39 known genera. Among the total 217 species (sensu lato), 67 are commensal, 135 are free-living, and 15 have unidentified lifestyles (electronic supplementary material, dataset S1).

Deep phylogenetic relationships within the superfamily were well-resolved with basal clades composed of mostly commensal species (figure 1a). Five well-supported clades were identified (figure 1a): clades a–d represent four major commensal clades, and clade e represents one major free-living clade, although it includes a commensal subclade (figure 1a, CS7). Seven commensal subclades (figure 1a, CS1–7) and nine free-living subclades (figure 1a, FS1–9) were further identified and their correspondence to existing taxonomical designations are discussed in the supplementary text (electronic supplementary material, figures S2 and S3).

Degrees of host lineage fidelity vary among different commensal subclades. Species in CS2 and CS7 are strictly restricted to echinoid and stomatopod hosts, respectively. Species in CS1 are almost all holothuroid commensals, except for one that is associated with a sipunculan host. Species in CS3–6, however, occupy a diverse spectrum of invertebrate hosts without apparent patterns of large-scale host lineage specialization (see electronic supplementary material, dataset S1 for host information).

(b). Evolutionary history of lifestyles

Bayesian ancestral state reconstructions strongly suggest that the plesiomorphic lifestyle of the galeommatoidean crown group is commensalism/sediment-dwelling. The distribution of commensal and free-living lifestyles on the phylogeny is highly clustered with a significant phylogenetic signal across all posterior trees: (95% confidence interval [0.5–1]) when unknowns are treated as commensal; when treated as free-living, (95% confidence interval [0.7–1]). Occasional lifestyle transitions occurred within both major commensal and free-living clades.

(c). Lineage diversification

Figure 1b shows the consensus Bayesian Galeommatoidea phylogeny with branch length proportional to estimated speciation rate on that branch. It is evident that species belonging to one clade (figure 1a, labelled by a blue star) predominantly exhibit higher speciation rates—estimated rates for each branch are typically 2–4 times higher than the rest of the tree (see all rates in electronic supplementary material, figures S2 and S3). All free-living subclades belong to the star clade, the exceptions being subclades FS1 and FS9 (figure 1a), which do not show significantly higher speciation rates than the commensal subclades.

(d). Morphological evolution

Major shape variations (92%) in Galeommatoidea are captured by the first two PCs (electronic supplementary material, table S2). Both PCs reflect variations in shell elongation and umbo (i.e. shell apex) position (figure 2a). Species on the positive extremes of the two PCs possess anteriorly positioned umbos, whereas the negative extremes reflect posteriorly positioned ones (figure 2a, umbos indicated by arrows). Free-living and commensal species tend to overlap in morphospace and their total shell shape disparities do not differ significantly (p = 0.16). However, the two groups achieved similar disparity in different ways. For free-living species, the disparity is achieved by multiple clades occupying different regions of the morphospace, whereas for commensal taxa, a single clade (e.g. figure 2c, clade CS7) has already achieved comparable levels of disparity as all free-living taxa combined.

Figure 2.

(a) Scatter plot of the first two PCs of the lateral shell shape (left valve) variance among all species. Lifestyles are colour coded. Four deformed grid plots represent shape changes along each axis to aid visualization. (b,c) Scatter plot of the first two PCs from PCA analysis of free-living and commensal species, respectively. Colour code corresponds to the subclade each species belongs to. Triangles and squares represent species belonging to the major commensal and free-living subclades, respectively. Species belonging to unresolved clades are coloured grey. (d,e). Diversity-through-time (DTT) plot for free-living and commensal species. The dashed line represent DTT under a Brownian motion model of trait evolution. 95% confidence intervals are shaded in grey.

The PCA was repeated for free-living and commensal taxa separately (figure 2b,c and electronic supplementary material, table S2) to compare their subclade distributions in morphospace. For free-living lineages, species in the same subclades tend to cluster with each other and the degree to which subclades overlap reflects phylogenetic relatedness. In contrast, commensal subclades mostly overlap with each other and within-clade disparity can be quite high (e.g. CS7). This discordance is supported by a higher phylogenetic signal (K_mult = 0.47) for free-living shell shapes and a lower phylogenetic signal (K_mult = 0.23) for the commensals. The K_mult difference is significant across all posterior trees (p < 0.001). A similar pattern is observed in the shell size data, where the phylogenetic signal is higher (K_mult = 0.23) for the free-living taxa and lower (K_mult = 0.15) for the commensals (p = 0.002). In addition, shell sizes of free-living species are significantly larger than those of commensal species (p < 0.001, electronic supplementary material, figure S4).

For the DTT analysis, the free-living DTT curve (figure 2d) generally falls within the 95% confidence interval of the BM simulations (MDI = 0.09). In contrast, the mean relative disparities for commensal species (figure 2e) remained above the BM 95% confidence interval throughout the phylogeny (MDI = 0.35). The difference is again significant across all posterior trees (p < 0.001). This implies that the within-subclade disparities in the commensal lineages are consistently higher than what would be expected from a random walk.

Fitting results for the EB model are not shown because they consistently converged on the simpler BM model (i.e. a = 0) during maximum-likelihood estimations for all datasets. Among the remaining three models, the SE model is strongly supported for free-living taxa morphological evolution, whereas the OU model is strongly favoured for commensal species (table 1).

Table 1.

Model-fitting results for shell shape and size evolution of Galeommatoidea. Models shown are Brownian motion (BM), Ornstein–Uhlenbeck (OU), and speciational evolution (SE).

		shell shape			shell size
		BM	OU	SE	BM	OU	SE
free-living	likelihood	615.4	644.3	693.5	−144.4	−85.4	−68.5
	ΔAICc	150	98	0	90	34	0
	parameter	σ2 = 0.0004	α = 0.03	ψ = 0.25	σ2 = 0.03	α = 0.16	ψ = 0.56
commensal	likelihood	167.3	201.3	189.1	−58.0	−35.9	−41.7
	ΔAICc	61	0	24	45	0	12
	parameter	σ2 = 0.0007	α = 0.09	ψ = 0.68	σ2 = 0.02	α = 0.21	ψ = 1

4. Discussion

(a). Ecological opportunities for elevated diversification

The BAMM analyses revealed that the majority of free-living taxa, belonging to the star clade (figure 1a) show elevated speciation rates. Contrary to our original prediction, the commensal taxa do not show higher diversification rates.

High speciation rates in marine lineages have been linked to numerous non-mutually exclusive environmental and life-history factors, such as high temperature, high spatial complexity, intense interspecies competition, and non-planktotrophic development, etc. [49,50]. The challenge here is to identify what mechanisms may have selectively led to high speciation in the star clade, but not in the commensals or other small free-living clades.

One important distinction between the two lifestyles is that almost all commensal taxa are found in soft-bottom habitats, whereas the great majority of free-living species are hard-bottom dwellers [27]. Only a few exceptions occur in the commensal clades, where some lineages transitioned to a free-living lifestyle yet remained sediment-dwelling (e.g. Mysella charcoti [51]). Our analysis suggests that ancestral galeommatoideans were likely sediment-dwelling commensals and that colonizations of hard-bottom habitats were coupled with losses of commensalism. This habitat transition was ecologically significant as it opened up previously unavailable niches in highly heterogeneous hard-bottom habitats. Geologically and biologically complex hard-bottom habitats, particularly in coral reef ecosystems, have been shown to promote high speciation in multiple marine groups through vicariance processes and niche partitioning driven by species interactions [11,52,53]. In Galeommatoidea, many species in the star clade are found in reef-associated habitats, especially in the Indo-Australian Archipelago (electronic supplementary material, figure S1). Interestingly, the diversification of many free-living subclades falls within the timeframe of modern coral reef expansions (40–23 Ma [54]). Further, the two free-living subclades that do not show higher speciation rates, FS1 (genera Lasaea and Arthritica) and FS9 (genus Kellia), are non-reef associated. These observations suggest that coral reef habitats have played key roles in driving the rapid divergence among free-living lineages. A comprehensive test of this hypothesis would entail additional taxon sampling from other significant coral reefs (e.g. Caribbean region) and detailed ecological studies of reef-associated galeommatoidean species. In addition, a more comprehensive calibration scheme is needed to better assess the timing of galeommatoidean diversification and the establishment of reef habitats. Owing to the lack of reliable internal fossil calibrations, analyses with additional external calibrations [55] are needed.

One, perhaps, surprising result of this study is that commensal lineages diversified more slowly than free-living ones. Generally, host-switching events in symbiotic systems are expected to provide additional opportunities for ecological divergence and promote speciation [56]. In Galeommatoidea, many commensal lineages lack clade-specific host fidelity, implying that host-switching is relatively common [26]. However, such processes do not seem to result in exceptionally high speciation rates. In a recent framework [57], Dynesius & Jansson pointed out three principal controls of speciation: rate of within-species lineage splitting; degree of persistence for split lineages; and time required for such lineages to become full species. Thus, the commensal bivalves could have high rates of within-species lineage splitting owing to host-switching events, but still have low speciation rates if the initially split lineages exhibit low degrees of persistence. Low persistence could be caused by either lineage merging owing to increased gene flow or by within-species lineage extinction [57], both of which are probable in Galeommatoidea. Microevolutionary studies on multiple commensal species [58,59] found no evidence of pronounced genetic differentiation among populations occupying different hosts, suggesting high levels of gene flow among these populations. In addition, because many commensal–host associations are obligate, host extinction events [58] can cause co-extinction of specialized commensal populations and thereby reduce rates of speciation. Compared with the commensals, free-living taxa not only have a higher probability of lineage splitting owing to the availability of heterogeneous hard-bottom habitats, but may also experience high degrees of lineage persistence owing to the stability of such habitats, resulting in the overall high speciation rate.

In addition to population-level extinction, high host dependence could also result in higher species-level extinction for the commensal taxa. The legitimacy of extracting information about extinction from patterns of molecular phylogenies is still highly controversial [60]. For now, we cannot confidently compare patterns of extinction without an extensive fossil record, which galeommatids do not have; not to mention the difficulty of assigning small, featureless shells to recent clades with certainty.

Because free-living and commensal lifestyles in Galeommatoidea show high phylogenetic conservatism (i.e. only one major free-living radiation), it is difficult to conclude that the higher speciation rate is promoted by hard-bottom habitats alone; other clade-specific life-history traits may also contribute significantly to the accelerated speciation in the star clade. For example, the average body size of free-living species is significantly larger than the commensals. Because body size is typically positively related to brood size and fecundity, larger body size can also increase the probability of within-species lineage persistence. Another important trait related to marine speciation is larval development, as it is generally assumed that the wide dispersal of planktotrophic larvae can suppress genetic divergence and reduce speciation rate [49]. While most galeommatoidean species possess planktotrophic larvae (indirect development), some taxa release crawl-away juveniles (direct development) [61]. Both developmental modes have free-living and commensal representatives. However, case studies on galeommatoideans have shown that direct developers can have extensive geographical ranges [29,61] and that indirect developers can be geographically quite restricted [62]. It is known that molluscan species with similar developmental modes can show drastically different population structures [63]. Therefore, the impact of larval ecology on galeommatoidean speciation could be complex and is unlikely to be the major driver of the observed patterns. Lastly, other potentially important ecological factors (e.g. sexual selection, predation, etc.) remain unexplored in this study and certainly call for further examination.

(b). Modes of morphological evolution

Multiple assessments of galeommatoidean morphology collectively support our second prediction that the commensal species show higher morphological disparity compared with the free-living taxa. For the free-living species, morphologies (lateral shell shape and size) of closely related species tend to resemble each other and among-clade disparity is higher than within-clade disparity throughout the phylogeny. Among the four trait evolution modes, the SE model provides the best fit to the free-living size and shape data, indicating that besides the gradual trait evolution occurring along the phylogeny, a fraction (ψ = 0.25) of the trait variation is contributed to by step changes at speciation events. This is consistent with the notion that ecological niche partitioning driven by structural and biological (e.g. predation) complexity is promoting the diversification of free-living taxa.

In contrast, most of the morphological disparities in the commensal species are explained by within-clade rather than by among-clade disparity, indicating that closely related species can be morphologically highly divergent and distantly related species sometimes resemble each other (convergence among clades). The OU model with a weak constraint (table 1) is strongly favoured for the commensal trait evolution. The combination of commensal DTT and model-fitting results suggests that within-clade disparities remain relatively constant through time and different clades overlap in morphospace. This requires species in each clade to quickly occupy the available morphospace after initial divergence. Such a pattern can be generated under two extreme conditions (or their combination): first, the available morphospace may be tightly constrained, and second, the species may be exploring the morphospace rapidly. Given that the shell shapes of extant commensal species are relatively diverse (e.g. even umbo orientation may differ among closely related taxa), it is unlikely that their shell shape morphospace is highly constrained. Therefore, we can infer that there is rapid morphological divergence among commensal species, regardless of phylogenetic relatedness.

The high level of morphological divergence among commensal species is likely driven by the host–commensal associations. Many commensal species are host-specific and they directly attach to the hosts' body walls or even occupy the hosts' body cavities [64]. Such obligate and specialized associations often require host-specific adaptations and the bivalve shells are usually shaped to fit the available attachment spaces. Therefore, attachment mechanisms/positions likely have great influence on the shell morphologies of commensal species (see electronic supplementary material, figure S5 for a striking example). For example, an obligate hermit crab commensal species possesses unique crescent-shaped shells that conform to the hosts' coiled snail shells [65]. Because closely related commensal species sometimes occupy very different host species, it is to be expected that their shell morphologies do not reflect phylogenetic affinity, but rather similarities among the microhabitats they occupy, which may result in different levels of morphological convergence.

Our analyses of galeommatoidean morphologies are based on shell shapes and sizes. However, many species also possess complex soft tissue structures, such as hypertrophied mantles that facultatively or permanently cover the shells [66–68]. The mantles can form expanded brood chambers [64], or are further elaborated into innervated, extendable papillae and tentacles. Functions of these soft tissue structures are poorly understood, but initial studies suggest that they serve autotomizing/secretory functions and are likely associated with defensive behaviours [67,69,70]. These structures could be especially important to the free-living species as they may be under much higher predation pressure than the commensals. Therefore, to further understand the impact of lifestyles on galeommatoidean morphological evolution, close examination of the evolution of soft tissue structures is also needed.

5. Conclusion

This study demonstrates that commensalism plays important roles in shaping galeommatoidean evolution, but free-living lineages also contribute significantly to the diversity of the superfamily. Given that the bivalves' lifestyles are determined by abiotic habitat types, their remarkable diversity is likely driven by a multi-level causal network that links abiotic and biotic factors together. This suggests that neither component can be neglected if we wish to fully understand large-scale marine diversification processes. In particular, the inclusion of biotic factors should be more widely applied to studies of neontological marine diversification.

Data accessibility

The datasets supporting this article have been uploaded as part of the electronic supplementary material. Newly sequenced genes were deposited in GenBank (KX361229-361311, KX375833-376246). Phylogenies and morphometric data were deposited in the Dryad Digital Repository: http://dx.doi.org/10.5061/dryad.8j36j.

Authors' contributions

J.L., D.Ó F., and E.E.S. designed the study. J.L. carried out the experiments and analyses. J.L., D.Ó F., and E.E.S. wrote the manuscript.

Competing interests

We declare we have no competing interests.

Funding

This study was supported by an NSF DEB award 1308457 to J.L. and a NSF OCE award 0850625 to D.Ó F.