Evolution of diverse host infection mechanisms delineates an adaptive radiation of lampsiline freshwater mussels centered on their larval ecology

Sample collection

Our sampling strategy, for both ingroup and outgroup taxa, was primarily guided by the Zanatta & Murphy (2006) study, although we were not successful in obtaining, and/or genotyping, all of the species they included. Tissues samples from a total of 84 species were collected from the field (N = 13) as well as obtained from various research collections (N = 71) including the Illinois Natural History Survey, The University of Florida, North Carolina Museum of Natural Sciences, and from the Alabama Aquatic Biodiversity Center. Our final dataset consisted of 109 sequenced individuals representing 54 species across 22 different genera (Table 1).

Among the Zanatta & Murphy (2006) taxa that we were unable to source was the genus Popenaias, that positioned within the Amblemini in mitochondrial gene trees (Campbell et al., 2005; Zanatta & Murphy, 2006). However more recent studies, using data from the large nuclear ribosomal gene in addition to mt sequences (Pfeiffer et al., 2019), and from an anchored hybrid phylogenomic approach (Pfeiffer, Breinholt & Page, 2019) recovered this genus as members of a newly recognized Mesoamerican and Rio Grande clade, Popenaiadini, sister to Lampsilini.

A non-lethal biopsy technique developed by Berg et al. (1995) was used to collect tissue samples from mussels in the field. Mussel species were categorized based on presence or absence of mantle lure and type of brood lure (simple, complex, or tethered). Mantle lures and brood lures were treated as separate variables because they are not mutually exclusive with many species having both mantle lures and brood lures. The wide spectrum of mantle lure phenotypes found across the clade (Barnhart, Haag & Roston, 2008; Haag, 2012) complicated discrete sub-categorization so this variable was scored simply into presence or absence states. Brood lures were broken down into four categories: absence of brood lure, simple/fragile brood lure, complex brood lure, and tethered brood lure. Information regarding primary hosts, and host infection strategies, for each mussel species (Table 1) was compiled from various literature sources. Reference literature used for each species listed and cited in Table S1.

ddRADseq data collection and bioinformatics

Genomic DNA was extracted from tissue samples using the E.Z.N.A. Mollusk DNA kit (Omega Bio-Tek, Norcross, GA) according to manufacturer’s instructions and then stored at −80 °C. The quality and quantity of DNA extractions were assessed using a Qubit 2.0 Fluorometer (Life Technologies, Carlsbad, CA) and ddRADseq libraries were prepared following the protocols of Peterson et al. (2012). We then used 200 ng of DNA for each library prep. This involved digestion with Eco-RI-HF and MseI (New England Biolabs, Ipswich, MA) restriction enzymes, followed by isolating 294–394 bp fragments using a Pippen Prep (Sage Science, Beverly, MA) following the manufacturer’s instructions. Prepared ddRADseq libraries then were submitted to the University of Michigan’s DNA sequencing core and run in three different lanes using 150 bp paired-end sequencing on an Illumina HiSeq 2500. Two control individuals of Lampsilis fasciola were run in each lane and reads for both individuals clustered together in every analysis with 100% bootstrap support, indicating no lane effects on clustering across individuals. Raw demultiplexed data were deposited at genbank under the bioproject ID PRJNA704566 with accession numbers SAMN18093783–SAMN18093865.

The alignment-clustering algorithm in ipyrad v.0.7.17 (Eaton, 2014; Eaton & Overcast, 2020) was used to identify homologous ddRADseq tags. Ipyrad is capable of detecting insertions and deletions among homologous loci which increases the number of loci recovered at deeper evolutionary scales compared to alternative methods of genomic clustering (Eaton, 2014). Demultiplexing was performed by sorting sequences by barcode, allowing for zero barcode mismatches (parameter 15 setting 0) and a maximum of five low-quality bases (parameter 9). Restriction sites, barcodes, and Illumina adapters were trimmed from the raw sequence reads (parameter 16 setting 2) and bases with low-quality scores (Phred-score < 20, parameter 10 setting 33) were replaced with an N designation. Sequences were discarded if they contained more than 5 N’s (parameter 19). Reads were clustered and aligned within each sample at two different similarity thresholds, 85 and 90% and clusters with a depth <6 were discarded (parameters 11 and 12). We also varied the number of individuals required to share a locus from ∼25% (N = 27) to ∼46% (N = 50). Ipyrad output files were used for further downstream analyses and are available on Dryad at the following DOI: https://doi.org/10.5061/dryad.c866t1g62.

Phylogenomic analyses

We analyzed the four concatenated ddRAD-seq alignment files (85% and 90% clustering similarity and 25% and 46% minimum samples per locus) using maximum likelihood in RAxML v8.2.8 (Stamatakis, 2014). A general time-reversible model (Lanave et al., 1984) was used for these analyses that included invariable sites and assumed a gamma distribution. Support was determined for each node using 100 fast parametric bootstrap replications. Due to the relatively deep phylogenetic scale comprised by our taxon sampling, we recovered many more loci with a minimum of 25% individuals per locus and 85% clustering threshold (4,725 loci) compared to runs that included 46% individuals per locus at the same clustering threshold (664 loci). The 90% clustering threshold produced even fewer loci and was not very useful for our phylogenomic analyses. Relationships were robust for most nodes with the 85% clustering threshold, and downstream analyses were performed using both of these datasets (85%–25% and 85%–46%).

The maximum likelihood phylogeny output from RAxML was trimmed to remove the outgroup taxa (Quadrula quadrula, Amblema plicata, Fusconaia flava, and Eurynia dilata) as well as all multiples of each species using the ‘ape’ package in R version 3.5.2 (R Core Team, 2018; Paradis & Schliep, 2019). This tree with a single individual of each species was used to create an ultrametric tree with two comparable methods using penalized maximum likelihood approaches (Sanderson, 2002; Kim & Sanderson, 2008); one implemented in R using the ‘ape’ package with a correlated rate model (Paradis & Schliep, 2019), and another using treePL (Smith & O’Meara, 2012).

Ancestral state reconstruction

We analyzed the evolution of mantle lures and brood lures separately because these host infection strategies are neither homologous characters, nor mutually exclusive with many species using both mantle lures and brood lures (Corey, Dowling & Strayer, 2006; Barnhart, Haag & Roston, 2008). For each species of mussel, we independently assessed the mantle lure and brood lure characters and categorized them into binary, present or absent, character states based on the current available data (Table S1). Ancestral State reconstructions for both mantle lures and brood lures were performed using the rerooting method (Yang, Kumar & Nei, 1995), implemented in the ‘Phytools’ package in R (Revell, 2012; Paradis & Schliep, 2019), and using both a one-rate model (ER; equal transition rates among all character states) and a symmetric model (SYM; rates can vary among different traits but forward and reverse transition are constrained) where rates are allowed to differ between transitions but are constrained between forward and reverse transitions.

Lampsiline diversification rates

Two different approaches were used to investigate the potential influence of host infection strategies on diversification rates in the Lampsilini. The first method used State Speciation and Extinction models to explicitly test the association between host infection strategies and diversification rates, the second method used BAMM to estimate diversification rates and evidence of rate shifts in the lampsiline phylogeny

Hidden State Speciation and Extinction models were implemented using the ‘hisse’ package in R (Beaulieu & O’Meara, 2016). Four models were performed independently for each trait (presence of mantle lure, presence of brood lure, broadcast release); a binary state-dependent model (BiSSE), a hidden state dependent model (HiSSE), a two-state character-independent model, and a four-state character independent model. The two-state and four state character-independent models were included as null models to compare to the BiSSE and HiSSE models. Rabosky & Goldberg (2015) found that BiSSE models tend to have a high type-1 error rate when compared to a null model that assumes homogenous diversification rates across the tree. The two-state and four-state character-independent models were proposed as an alternative null model which allows for rates to vary, independent of the trait value, and reduces type-1 error rates (Beaulieu & O’Meara, 2016). All models allowed extinction rates to vary independently for each character state, and transition rates between states were fixed to simplify the models. The revised freshwater mussel taxonomy by Williams et al. (2017) was used to estimate sampling frequency for each trait category. This analysis was performed with both the ultrametric tree derived from the ‘ape’ package as well as the one derived using TreePL. To further explore state-dependent models, the R package ‘Diversitree’ was used to estimate and visualize diversification rates using an MCMC approach (FitzJohn, 2012).

For the second method, we used Bayesian Analysis of Macroevolutionary Mixtures (BAMM) software package (v. 2.5) and the R package “BAMMtools” to estimate diversification rates in the Lampsilini phylogeny, (Rabosky, 2014; Rabosky et al., 2014). BAMM uses a reversible-jump Markov Chain Monte Carlo to automatically detect clades that share common evolutionary parameters of diversification (Rabosky et al., 2013). BAMM was performed using 10,000,000 generations, sampling every 5,000 generations. Priors for the model were selected using the setBAMMpriors function in R (Rabosky et al., 2014). To account for incomplete taxon sampling, we used previously published mitochondrial phylogenies for this group (Campbell et al., 2005; Zanatta & Murphy, 2006) and the revised list of freshwater mussels of the United States and Canada by Williams et al. (2017) to estimate clade-specific frequencies of sampling biases.

ddRADseq data collection and bioinformatics

Illumina sequencing returned raw reads ranging from 287,978 to 14,377,252 per individual across the 83 unionid samples included in the analyses. Mean coverage depth, for the 85% clustering threshold, ranged from 1.48 (Toxolasma lividum) to 5.25 (Lampsilis virescens) (Table 1, Table S2).

We identified between 4,745 and 664 homologous loci across the two best ddrad datasets (85%–25% and 85%–46%) and, in general, much higher numbers of loci were recovered for the core Lampsilini ingroup (>1,000 loci) relative to the outgroups (<100 loci). Although lowering clustering thresholds produced a much greater amount of missing data in the ddrad supermatrix, they also greatly increased the number of loci which could be used, e.g., for the 85% clustering threshold, 664 loci were recovered when a minimum of 46% individuals were included, whereas a 25% minimum yielded 4,745 loci. Simulation studies and empirical analyses both suggest that large amounts of missing data may be relatively unproblematic for phylogenetic reconstructions, especially if the total dataset is large (Rubin, Ree & Moreau, 2012; Huang & Knowles, 2016; Eaton et al., 2016). Datasets recovered from both the 25% and 46% minimum samples per locus clustering thresholds were used in all our phylogenomic analyses.

Phylogenomic analyses

The ddRADseq gene tree topologies we recovered were highly consistent across all of the parameter settings analyzed, with a few differences in placement of poorly supported nodes (Fig. 3 and Fig. S1). All our phylogenetic trees recovered the monophyletic genus Toxolasma as sister to the other members of the Lampsilini tribe included in the study. The latter formed four well-supported crown clades, each composed of members of >1 genus: a 2-species clade with Glebula and Cytronaias spp., a 10-species clade with Medionidus, Lemiox, and Pytchobranchus spp., a 5-species clade containing Leptodea, Potamilus and Truncilla spp., and a 33-species clade containing Ligumia, Epioblasma, Obovaria, Venustaconcha, Hamiota, Villosa, Sagittunio, Cambarunio, Leaunio and Lampsilis spp. Across our topologies, some genera were recovered as monophyletic (Toxoplasma (4 species), Obovaria (2 species), Venustaconcha (2 species) Hamiota (4 species), Lampsilis (14 species), Pytchobranchus (3 species), Sagittunio (2 species; see Watters, 2018), Leaunio (2 species; see Watters, 2018), but some others did not (Medionidus (6 species), Leptodea (2 species)). The new reclassification of Villosa suggested by Watters (2018) is supported in our analyses for the species we have included.

An ultrametic tree (Fig. 4) was created with TreePL from the 85% clustering similarity with 25% minimum samples per locus topology (Fig. 3) and manually pruned to one individual per species according to read count. The mussel species are color-coded according to their host infection strategy and their primary (most frequently used) host taxa are indicated. A striking feature of this topology is the high degree of conservation shown by ingroup mussels in their primary fish host taxa, e.g., the mantle-lure producing Toxolasma spp. clade with sunfishes (Lepomis spp.), the mixed strategy dominated 33-species clade primarily with bass (Micropterus spp.), the mantle lure or brood lure 10-species Medionidus/Lemiox/Ptychobranchus spp. clade with darters (Etheostoma spp.) and sculpins (Cottus spp.), and the 5-species Leptodea/Potamilus/Truncilla spp. clade–some broadcasting larvae, some with mantle lures–with freshwater drum (Aplodinotus grunniens).

Ancestral state reconstruction

Ancestral state reconstructions were performed for both mantle lures and for brood lures using two different models for transition rates (ER and SYM). The likelihood values for the mantle lure models are ER = −24.65 and SYM = −24.65. For brood lure reconstructions the likelihood values are ER = −26.14 and SYM = −23.81. Using the SYM model, estimated probabilities of character states at each node were plotted on the ultrametric tree (Fig. 5; Fig. S2). These results imply that mantle lures evolved early in the Lampsilini phylogeny, being present in the ingroup’s last common ancestor, with four to six subsequent losses. Brood lures are inferred to have independently evolved twice in this phylogeny.

Gain of a complex brood lure was coupled with loss of a mantle lure in Pytchobranchus (Fig. 5), although this transition was not associated with a change in primary host fishes (darters/sculpins; Fig. 4). Gain of a simple brood lure in ancestor of the 33-species, 10-genus, predominantly bass host specialist clade did not result in the loss of a mantle lure. However, within that clade, the subsequent evolution of a complex, tethered brood lure in Hamiota was associated with the loss of a mantle lure in 3/4 species (Fig. 5), but no change in primary host fishes (Fig. 4). Eleven of the 33 species in this clade have primary fish hosts other than bass (darters/sculpins (7), sunfishes (3), and walleye (1)) and, while all of them have retained mantle lures, 3 of the 7 species targeting darters/sculpins–Obovaria subrotunda, Obovaria choctawensis and Epioblasma triquetra–have lost simple brood lures, with the latter species physically capturing host fish to enable larval infection (Fig. 2C). The remaining cases of mantle lure loss are associated with the gain of broadcast larval release in two clades: one containing the gar specialist Cyrtonaias tampicoensis and the sunfish specialist Glebula rotundata, the other involving three members of the 5-species Leptodea/Potamilus/Truncilla spp. clade: the drum (Aplodinotus grunniens) specialists Truncilla macrodon and Potamilus ohiensis and the white perch (Morone americana) specialist Leptodea ochracea (Figs. 4 and 5).

Lampsiline diversification rates

Three traits were assessed independently (mantle lure, brood lure, and broadcast release) using four different models (BiSSE, HiSSE, 2 state character independent, and 4 state character independent; Table 2). The best-performing model (AICc) for the mantle lure trait was the two-state independent model, suggesting no relationship between mantle lures and net diversification rates. The BiSSE model was the best-performing model (AICc) for the brood lure trait by a small margin, suggesting an increase in net diversification rate for species with brood lures (estimated net diversification rate of 11.7 for species with brood lure versus 8.5 for those without) and largely similar estimates for extinction fraction, which is the ratio of extinction rate/speciation rate (0.38 versus 0.41 respectively). This result was consistent across both the 25% minimum samples per locus topology (Table 2) and the 46% topology (Table S3), regardless of how the ultrametric tree was derived. To explore these models further, we used an MCMC modeling approach, implemented in the R package ‘diversitree’ (FitzJohn, 2012) to estimate diversification rates for species with and without brood lures. The distributions for the parameter estimates have some overlap (Fig. 6) but display two distinct peaks and the species with brood lures have a higher estimated diversification rate. When analyzing the 85%–46% tree (Fig. S1), we found the BiSSE model was also the best-performing model (AICc) for broadcast release by a small margin (Table S3), hinting at a possible reduced diversification rate for broadcast releasers, but this was result was not corroborated in the 85%–25% tree (Table 2).

We tested for differences in speciation rates among the 54 species of lampsilines by performing BAMM analyses for 10,000,000 generations on the ultrametric tree (Fig. 4). The mean, model averaged diversification rates estimated along each branch are displayed in Fig. 7A and all four credible rate-shift sets recovered are displayed in Fig. 7B. The best rate-shift configuration (f = 0.44) suggests a static diversification rate across the entire ingroup topology, with no clade-specific differences in diversification rate (Fig. 7A). However, the second, third and fourth most sampled rate-shift configurations (f = 0.22, 0.21 and 0.13), comprising 56% of configurations sampled, indicate an increase in diversification rate on adjacent stem branches of the 33-species clade containing Ligumia, Epioblasma, Obovaria, Venustaconcha, Hamiota, Villosa, Sagittunio, Cambarunio, Leaunio, and Lampsilis spp. (Figs. 7Bii-7Biv).

Evolution of infection strategies in lampsiline mussels

Our genomic phylogeny of Lampsilini represents a robust and comprehensive inferred evolutionary history of this North American unionid tribe. In contrast with earlier mitochondrial phylogenies (Campbell et al., 2005; Zanatta & Murphy, 2006), nodal support throughout the topology (Fig. 3) was generally high: a large majority of nodes displayed support values of 100 and only 15% had values <90. Most of the latter were concentrated within the Lampsilis clade, with the exception of the placement of the Villosa and Hamiota clade, and may stem from either incomplete lineage sorting or hybridization processes (Maddison & Knowles, 2006), but this question requires further investigation. Nevertheless, it is important to emphasize that our genomic phylogeny agrees broadly with those of previous molecular studies both in regard to outgroup/ingroup (Campbell et al., 2005) and among-ingroup (Campbell et al., 2005; Zanatta & Murphy, 2006; Pfeiffer, Breinholt & Page, 2019) relationships.

Our phylogenomic analyses (Fig. 4) indicate that fish host use in the Lampsilini through time is characterized by a high degree of mussel clade specificity for both primary host type and host infection mechanism(s). This result corroborates Haag’s (2012) suggestion that host use is highly conserved in this group as well as Hewitt, Wood & Foighil’s (2019) finding of topological congruence between North American unionids and their hosts. It also implies that lure-based host infection mechanisms are adaptive in origin, being specialized for attracting suitable hosts, as has been observed in the wild for a subset of co-occurring mussels (Haag & Warren, 2003). There are numerous examples of such across our tree topology (Fig. 4), e.g., most Lampsilis species target bass (Micropterus spp. - predators that are highly piscivorous when large (Hickley et al., 1994)) as primary hosts using large, conspicuous mantle lures that typically resemble small fishes (Barnhart, Haag & Roston, 2008). Likewise, Toxolasma species have a worm-like mantle lure (Fig. 2ii) and predominantly target sunfishes in the genus Lepomis that are generalist predators with a diet that includes worms (Parsons & Robinson, 2007). Finally, the clade composed of Medionidus spp. (with small, cryptic mantle lures) and Ptychobranchus spp. (with small demersal brood lures that typically mimic insect or fish larvae) specialize in darters and sculpins (small, benthic predatory fishes (Haag, 2012; Cummings & Watters, 2017)).

Our ancestral state reconstruction results corroborated Graf & Ó Foighil’s (2000) and Zanatta & Murphy’s (2006) mt phylogeny-based inferences that lampsiline mantle lures evolved early in this clade, followed by multiple secondary losses. These inferred losses occurred across much of the ingroup topology, apart for the genus Toxolasma (characterized by its worm-like mantle lures), and mantle lure loss was associated with the de novo gain of either complex brood lures or of broadcast infection strategies (Fig. 5; Fig. S2). The former occurred independently in two genera [Ptychobranchus, and in 3/4 of the Hamiota species represented] and involved a change in mimetic lure type: from mimetic mantle lures to mimetic brood lures, although Hamiota altilis retains both. The latter cases of mantle lure loss, inferred separately for Cyrtonaias tampicoensis and Glebula rotundata, and for Leptodea ochracea, Potamilus ohiensis and Truncilla macrodon, were more radical in that they involved the abandonment of prey mimicry and host deception as a host infection strategy. Haag & Warren (1998) found that population densities of specialist mussels and their fish hosts were correlated for broadcasters, but not so for lure-producing mussels. The evolutionary loss of lures in host specialist mussels would therefore appear counterintuitive, especially for mussels with low-density fish hosts, but there are potentially mitigating life history traits in some of these taxa that may act to increase their rate of host infection.

One such life history trait is increased larval production: relative to other lampsilines, Glebula rotundata females release more larval broods per year (Parker, Hackney & Vidrine, 1984) and the genera Truncilla and Leptodea have higher fecundities and smaller-sized larvae (Haag, 2013). Another such trait may involve targeting mussel predators as larval hosts, e.g., adult Aplodinotus grunniens (freshwater drum) prey on mussels and at least some of the species that use it as a host may engage in a sacrificial strategy whereby infection occurs when gravid females (especially smaller specimens) are consumed (Barnhart, Haag & Roston, 2008; Haag, 2012). Four of five members of the Leptodea/Potamilus/Truncilla spp. clade (Fig. 4) are A. arunniens specialists [the fifth, L. ochracea, occurs outside of this fish’s range (Page & Burr, 2011)] and, until recently, it was assumed that these three mussel genera lacked mantle lures. However, Sietman, Hove & Davis (2018) documented the presence of cryptic, nocturnally displayed, mantle lures for one member of each of these genera (including Truncilla truncata and Leptodea fragilis). In light of these new data, we view the current categorization of Leptodea ochracea, Potamilus ohiensis and Truncilla macrodon as lacking mantle lures (Figs. 4 and 5) to be provisional. For the taxa included here, our topology corroborates that of Smith, Pfeiffer & Johnson (2020) and our data support their decision to reclassify Leptodea ochracea to Atlanticoncha ochracea.

Our ancestral state reconstruction of brood lures (Fig. 5B) is consistent with two origins (one each in the genera Ptychobranchus and Hamiota) of complex, mimetic brood lures, and one additional origin of simple, non-mimetic brood lures in the ancestor of the 33-species, 10-genus, predominantly bass host specialist clade (Fig. 5B). The latter clade contains Hamiota, implying that the complex tethered brood lure found in Hamiota species (Fig. 2H) may be derived from the simple brood lures found in most of this clade, including species of its sister genus Villosa (Fig. 5B). In contrast, the darter/sculpin specialist clade containing Ptychobranchus (Fig. 4) lacks simple, non-mimetic brood lures (Fig. 5B). The evolutionary origins of the Ptychobranchus demersal mimetic brood lure (Fig. 2F) may stem from a common ancestor with the genus Cyprogenia. Previous mt phylogenies (Campbell et al., 2005; Zanatta & Murphy, 2006) have placed the genus Cyprogenia, with its demersal, mimetic baited brood lures (Barnhart, Haag & Roston, 2008), sister to the genus Ptychobranchus. Unfortunately, we failed to extract sufficient genomic data for our Cyprogenia stegaria sample to corroborate this relationship.

The 10-genus, predominantly bass host specialist clade comprised 33 species (Fig. 4) of which 26 (in the genera Lampsilis, Villosa, Ligumia, Leaunio, Cambarunio, Sagittunio, and Venustaconcha) produce mantle lures as well as simple brood lures (Figs. 5A, 5B). Mantle lures are regarded as their primary method of infecting fish hosts (Haag & Warren, 2000; Barnhart, Haag & Roston, 2008; Gascho Landis et al., 2012) and a gravid female may display hers for weeks to months (Kraemer, 1970; Haag & Warren, 2003). During an elicited host fish attack on mantle lure-displaying Lampsilis spp. gravid females, glochidia are extracted (Barnhart, Haag & Roston, 2008) from only a subset of their ∼60 marsupium water tubes and displaying females often exhibit a mix of undischarged (i.e., containing larvae) and discharged water tubes for much of the spring/summer host infection season (Haag & Warren, 1999). Lampsiline mussels have evolved bradytictic life cycles in which spawning typically occurs in the late summer and the resulting larvae are brooded overwinter (Graf & Ó Foighil, 2000). Gravid females must therefore release the previous year’s brood to facilitate fertilization and retention of their new clutch of eggs and it was initially unclear if the release of simple brood lures in these species represented a default end-season emptying of marsupial water tubes (Barnhart, Haag & Roston, 2008), a stress response to captivity (Corey, Dowling & Strayer, 2006), or a supplementary host infection strategy (Corey, Dowling & Strayer, 2006; Barnhart, Haag & Roston, 2008; Haag, 2012). Gascho Landis et al. (2012) performed a detailed experimental study of mantle lure display and simple brood lure production in Ligumia subrostrata and concluded that the latter clearly represents a secondary bet-hedging infection strategy. Nevertheless, the relative attractiveness of simple brood lures as putative food items to host fishes remains to be established as does their durability in nature: they typically break up quickly after release (Barnhart, Haag & Roston, 2008).

Based on available data, we propose a hypothesized three-step bet-hedging host infection strategy in these mussels (Fig. 8). This would involve (A) host attraction and infection via prolonged maternal mantle lure display; (B) the secondary release of residual brooded larvae within simple brood lures prior to the onset of seasonal spawning; and (C) tertiary broadcast dispersal (in lotic habitats) of individual infective larvae following simple brood lure breakup, although the probability of broadcast larvae encountering a host is likely low (Jansen, Bauer & Zahner-Meike, 2001) unless the latter is locally abundant.

The genus Epioblasma is a notable exception to the modal host infection mechanism found in this 10-genus crown clade in that gravid females produce mantle lures only and specialize in darter hosts that they actively trap during the infection process using female-specific shell margin extensions (Barnhart, Haag & Roston, 2008). This genus is highly underrepresented in our study with only one member, E. triquetra, included; a shortcoming primarily due to the exceptionally intense extinction pressure the genus has been subjected to over the past century. Of the 28 currently recognized species of Epioblasma (Williams et al., 2017), 13 are listed as extinct on the IUCN Red List and most of the remainder are critically endangered.

Diversification rates

The BAMM and state-dependent speciation model analyses yielded new insights into lampsiline diversification rates albeit with some methodological and sampling (e.g., the genus Epioblasma) caveats. The most supported BAMM result—a single diversification rate regime across the entire Lampsilini clade (Fig. 7I)—needs to be treated with caution as this methodology is biased towards zero rate shifts in smaller trees that contain fewer than approximately 150 species (Rabosky, Mitchell & Chang, 2017; Kodandaramaiah & Murali, 2018). In contrast, the three next-most supported results (Figs. 7Bii–7Biv) identified inferred rate shift accelerations that were tightly clustered on adjacent stem nodes of the 10-genus/33-species crown clade. This collective topological placement bracketed the inferred origin of the mixed infection strategy predominant in this crown clade that combines the use of mantle lures, a plesiomorphic trait (Fig. 5A), with simple brood lures, a derived trait (Fig. 5B). That topological congruence is broadly consistent with the BiSSE (Table 2; Table S3) and MCMC (Fig. 6) modeling results that found evidence for increasing diversification rates among lampsiline species with brood lures. However, it must be emphasized that the majority of these species produce simple brood lures and are likely to rely on mantle lures as their primary host infection strategy (Gascho Landis et al., 2012). Barnhart, Haag & Roston (2008) suggested that species that use both mantle lures and brood lures (conglutinates) could potentially parasitize both large- and small-bodied hosts (the latter being less likely to attack mantle lures). Similarly, a hypothesized three-step bet-hedging strategy (Fig. 8) could potentially generate higher diversification rates by expanding the repertoire of potential host fishes and thereby decreasing the risk of extinction. However, testing such a hypothesis requires significantly better data on host infection processes in natural populations as well as a more comprehensive phylogeny of unionids. The latter is also required to adequately address another outstanding question: the relative diversification rates of broadcasters and lure-using mussel taxa. It is notable that in a broadly parallel case, the evolution of deceit pollination in orchids apparently did not increase their rate of net diversification (Givnish et al., 2015).

Adaptive radiation of lampsilini

Models of adaptive radiation predict that the availability of ecological niches within an environment, and the response of adapting lineages to occupy them, drive and modulate this important evolutionary process. (Schluter, 1996; Gavrilets & Losos, 2009; Losos, 2010; Arbour & López-Fernández, 2016). Our primary phylogenomic result—that lampsiline clades are highly specific in primary fish host type and in host infection mechanism—is consistent with adaptive radiation expectations in regard to their larval ecology, despite the relatively brief duration of this life history stage. Two factors may bear on this ostensibly surprising result. Once lampsiline mussels evolved a high degree of fish host specialization (Haag & Warren, 1998), the number of discrete larval ecological niches potentially available to them, in the form of local host fish species diversity, greatly increased. In addition, successful larval infection and metamorphosis (transformation) on a fish host is a necessary precondition for juvenile mussel recruitment, and therefore for ecological persistence, in wild populations.

Although our data support an adaptive radiation framework operating at the level of lampsiline clades, they lack the fine-grained resolution of specific host data needed to establish if it equally applies to within-clade diversification. For instance, it remains to be established to what degree sympatric, closely related lampsiline species preferentially target different species of host within the same host guild, consistent with a seamless adaptive radiation paradigm, or rather compete for the same host species, consistent with an evolutionary arms race paradigm (Van Valen, 1977). We anticipate that the balance of these two potential within-clade evolutionary processes may differ among lampsiline lineages according to the range of potential hosts available to them. For example, there are ∼200 species of North American darters, many with small ranges (Near et al., 2011), and there may be considerable evolutionary scope for a high degree of host exclusivity and within-clade adaptive radiation among the darter-specialist lampsiline genera such as Medionidus, Ptychobranchus and Epioblasma. In contrast, there are fewer (∼41; Roe, Harris & Mayden, 2002; Baker, Blanton & Johnston, 2013; Freeman et al., 2015) species of centrarchids in North America than of lampsiline centrarchid specialists (∼50; Williams et al., 2017). Although new centrarchids species continue to be described (Baker, Blanton & Johnston, 2013; Freeman et al., 2015), the lower number of potential centrarchid hosts implies that some of these mussel species are more likely to compete directly, when in sympatry, for the same hosts and thereby become entrained in an evolutionary arms race for lure effectiveness. In such cases, coexistence could be modulated by frequency-dependent selection processes (Endler, 1988), in which previously infected host fishes are more likely to engage with unfamiliar/rare lure phenotypes, a process that has also been implicated in the evolution of lure polymorphisms in some lampsiline species (Zanatta, Fraley & Murphy, 2007; Barnhart, Haag & Roston, 2008).