ABSTRACT
Theory predicts that sexually dimorphic traits under strong sexual selection, particularly those involved with intersexual signaling, can accelerate speciation and produce bursts of diversification. Sexual dichromatism (sexual dimorphism in color) is widely used as a proxy for sexual selection and is associated with rapid diversification in several animal groups, yet studies using phylogenetic comparative methods to explicitly test for an association between sexual dichromatism and diversification have produced conflicting results. Sexual dichromatism is rare in frogs, but it is both striking and prevalent in African reed frogs, a major component of the diverse frog radiation termed Afrobatrachia. In contrast to most other vertebrates, reed frogs display female-biased dichromatism in which females undergo color transformation, often resulting in more ornate coloration in females than in males. We produce a robust phylogeny of Afrobrachia to investigate the evolutionary origins of sexual dichromatism in this radiation and examine whether the presence of dichromatism is associated with increased rates of net diversification. We find that sexual dichromatism evolved once within hyperoliids and was followed by numerous independent reversals to monochromatism. We detect significant diversification rate heterogeneity in Afrobatrachia and find that sexually dichromatic lineages have double the average net diversification rate of monochromatic lineages. By conducting trait simulations on our empirical phylogeny, we demonstrate our inference of trait-dependent diversification is robust. Although sexual dichromatism in hyperoliid frogs is linked to their rapid diversification and supports macroevolutionary predictions of speciation by sexual selection, the function of dichromatism in reed frogs remains unclear. We propose that reed frogs are a compelling system for studying the roles of natural and sexual selection on the evolution of sexual dichromatism across both micro- and macroevolutionary timescales.
INTRODUCTION
In The Descent of Man and Selection in Relation to Sex, Darwin (1871) observed that many closely related taxa differed primarily in secondary sexual characters and suggested that sexual selection plays a role in the diversification of species. The concept of speciation through sexual selection was later developed into a theory that links the coevolution of secondary sexual traits and mating preferences to premating reproductive isolation (Lande 1981, 1982; Kirkpatrick 1982; West-Eberhard 1983). This conceptual framework predicts that both the strength of sexual selection and the prevalence of sexually selected traits have a positive association with speciation rate (Lande 1981, 1982; West-Eberhard 1983; Barraclough et al. 1995). If divergence in secondary sexual characters and sexual selection are indeed major drivers of speciation, then clades exhibiting elaborate sexually dimorphic traits and/or elevated sexual selection should display higher species richness and increased diversification rates at macroevolutionary scales. Empirically, these predictions have mixed support across a range of sexually dimorphic traits that serve as proxies for sexual selection (reviewed in Kraaijeveld et al. 2011), indicating that macroevolutionary trends for traits involved with intersexual signaling and mate-choice may differ from those more strongly influenced by intrasexual or natural selection. For example, traits under ecological selection such as body size dimorphism have no consistent relationship with diversification (Kraaijeveld et al. 2011), whereas sexual dichromatism, a form of sexual dimorphism in which the sexes differ in color, often functions as a mate recognition signal and is associated with diversification in several taxonomic groups (Misof 2002; Stuart-Fox & Owens 2003; Alfaro et al. 2009; Kazancıo⍰lu et al. 2009; Wagner et al. 2012). Accordingly, sexual dichromatism has become a widely used proxy for studying the effects of sexual selection on speciation rate. However, phylogenetic comparative analyses explicitly testing for an association between sexual dichromatism and diversification have produced conflicting results, even in well-studied groups such as birds (Barraclough et al. 1995; Owens et al. 1999; Morrow et al. 2003; Phillimore et al. 2006; Seddon et al. 2013; Huang & Rabosky 2014) where sexual dichromatism plays an important role in signaling and mate-choice (Price 2008). Differences in methodologies and the spatial, temporal, and taxonomic scales among studies may partially explain these contrasting results. In particular, recent studies have highlighted concerns about the ability to distinguish between trait-dependent and trait-independent diversification scenarios using phylogenetic comparative methods (Rabosky & Goldberg 2015; Beaulieu & O’Meara 2016). In a broader sense, however, this disparity among studies may reflect more nuanced or novel mechanisms underlying the evolution of sexual dichromatism such that it does not consistently fit the conceptual framework of speciation by sexual selection. For instance, the striking sexual dichromatism in Eclectus parrots results from intrasexual competition to attract mates and intersexual differences in exposure to visual predators (Heinsohn et al. 2005). Likewise, the dynamic sexual dichromatism in frogs that form large breeding aggregations may serve to identify other competing males rather than to attract mates (Sztatecsny et al. 2012; Kindermann & Hero 2016; Bell et al. 2017b). Consequently, investigating the evolution of secondary sexual characters like dichromatism across ecologically diverse taxonomic groups is essential if we aim to generalize about the function of sexually dimorphic traits and better understand the roles of natural and sexual selection in biological diversification.
Secondary sexual traits are diverse and prevalent among anurans (frogs and toads), and include sexual size dimorphism, structures for acoustic signaling, nuptial pads, spines, elongated fingers, glands, and sexual dichromatism (Duellman & Trueb 1986). Sexual size dimorphism is common and occurs in >90% of species (Shine 1979; Han & Fu 2013), and body size evolution of the sexes is often attributed to fecundity for females and intrasexual competition, energetic constraints, agility, and predation for males (Salthe & Duellman 1973; Wells 1977; Shine 1979; Woolbright 1983; De Lisle & Rowe 2013; Han & Fu 2013; Nali et al. 2014). Male frogs of most species produce acoustic or vibrational signals to attract females, and these signals are essential for mate recognition and relaying social information (Ryan 1980; Gerhardt 1994; Gerhardt & Huber 2002). Structures like spines and tusks, which are present in males of many species, are used in male-male combat (e.g. McDiarmid 1975) whereas the diverse assortment of glands and nuptial pads, which are widespread in male frogs, are likely involved in courtship and amplexus (Duellman & Trueb 1986). In contrast to these widespread secondary sexual characters, anuran sexual dichromatism is rare, occurring in only ~2% of frog species (Bell & Zamudio 2012; Bell et al. 2017b). Behavioral studies in a handful of frog species indicate that these sexual color differences may be subject to natural selection as well as inter- and intrasexual selection (Mann & Cummings 2009; Sztatecsny et al. 2012). Sexual selection on coloration has historically been dismissed in frogs with the assumption that communication is predominately acoustic (reviewed in Starnberger et al. 2014); however, several studies document the importance of color signals in courtship behavior and mate-choice, even in nocturnal species (Gomez et al. 2009; Gomez et al. 2010; Jacobs et al. 2016; Yovanovich et al. 2017; Akopyan et al. 2018). If sexual dichromatism in frogs contributes to premating reproductive isolation, then sexually dichromatic clades may fit the conceptual framework of speciation by sexual selection and display higher species richness and increased diversification rates at macroevolutionary scales.
The Afrobatrachia frog radiation (Arthroleptidae, Brevicipitidae, Hemisotidae, Hyperoliidae) includes over 400 species distributed across sub-Saharan Africa that reflect much of the morphological, ecological and reproductive mode diversity present in anurans (Portik & Blackburn 2016). Afrobatrachian frogs display many unusual secondary sexual characters including the hair-like skin structures of male Hairy Frogs (Trichobatrachus), an elongate third digit in males (up to 40% of body length in Arthroleptis and Cardioglossa), extreme body size dimorphism (Leptopelis, Chrysobatrachus, Breviceps), and prominent pectoral glands (Leptopelis) or gular glands on the male vocal sac (Hyperoliidae). Afrobatrachian frogs also have the highest incidence of sexual dichromatism among anurans, which is striking and prevalent in many hyperoliid reed frogs (Hyperolius, Heterixalus). In sexually dichromatic hyperoliids, the difference in coloration is female-biased: both sexes exhibit a consistent juvenile coloration upon metamorphosis (termed Phase J), but at the onset of maturity sex steroids trigger a color and/or color pattern change in females (Phase F) whereas males typically retain the juvenile color pattern (Fig. 1) (Schiøtz 1967; Richards 1982; Hayes 1997; Hayes & Menendez 1999). In many dichromatic reed frog species adult males can also display the Phase F coloration, but this generally occurs in lower frequency than the Phase J morph (Fig. 1) (Schiøtz 1967, 1999; Amiet 2012; Kouamé et al. 2015; Portik et al. 2016a). The function of the ontogenetic color shift in female reed frogs is poorly understood (Bell & Zamudio 2012), but it may be similar to female-biased sexual dichromatism in other vertebrates, which can result from a reversal in mating system in which females compete for males (Andersson 1994) or sexual niche partitioning in which males and females use different habitats (Shine 1989; Heinsohn et al. 2005). Alternatively, distinct female color patterns in reed frogs may play a role in courtship and mate recognition at breeding sites where upwards of eight hyperoliid species congregate in a single night (Drewes & Vindum 1994; Kouamé et al. 2013; Portik et al. 2018). The link between sexual dichromatism and rapid speciation across disparate animal clades (Misof 2002; Stuart-Fox & Owens 2003; Alfaro et al. 2009; Kazancıo⍰lu et al. 2009; Wagner et al. 2012) demonstrates that when sexual dichromatism functions primarily as an intersexual signal under strong sexual selection, there are predictable outcomes on diversification rate. Therefore, as a first step towards understanding the potential function of dichromatism in hyperoliids, including the plausibility of intersexual signaling, we aim to assess whether this trait fits the predictions of speciation by sexual selection on a macroevolutionary scale.
In this study, we reconstruct the evolutionary history of sexual dichromatism within Afrobatrachia and investigate whether diversification rate shifts in this continental radiation are associated with dichromatism. We produce a well-resolved species tree of Hyperoliidae from genomic data and greatly increased taxonomic sampling relative to previous studies (Wieczorek et al. 2000; Veith et al. 2009; Portik & Blackburn 2016). To explore the evolution of sexual dichromatism within the broader phylogenetic and biogeographic context of African frogs, we incorporate all published sequence data of Afrobatrachia to produce a robust, time-calibrated phylogeny. Using methods that improve the accuracy of character state reconstructions by accounting for trait and diversification rate heterogeneity (King & Lee 2015; Beaulieu & O’Meara 2016), we test the hypothesis that sexual dichromatism evolved repeatedly within Afrobatrachia (Veith et al. 2009). Finally, we estimate net diversification rates from our time-calibrated phylogeny using the hidden state speciation and extinction framework (HiSSE; Beaulieu & O’Meara 2016) and examine if diversification rate shifts occur within Afrobatrachia and if so, whether they are associated with the occurrence of sexual dichromatism. Given recent concerns raised about the ability to distinguish between trait-dependent and trait-independent diversification scenarios (Rabosky & Goldberg 2015; Beaulieu & O’Meara 2016), we assess the performance of available state-dependent diversification methods with a trait simulation study conducted using our empirical phylogeny.
MATERIALS & METHODS
Species Tree Estimation of Family Hyperoliidae
Taxonomic sampling
We included 254 hyperoliid samples in our sequence capture experiment with multiple representatives per species when possible. Although there are 230 currently recognized hyperoliid species (AmphibiaWeb, 2018), this family is in a state of taxonomic flux with recent studies recommending both the synonymy of species names and the splitting of species complexes (Rödel et al. 2002; Rödel et al. 2003; Wollenberg et al. 2007; Rödel et al. 2009; Rödel et al. 2010; Schick et al. 2010; Conradie et al. 2012; Dehling 2012; Channing et al. 2013; Conradie et al. 2013; Greenbaum et al. 2013; Liedtke et al. 2014; Loader et al. 2015; Portik et al. 2016a; Bell et al. 2017a; Conradie et al. 2018). We estimate that our sampling represents approximately 143 distinct hyperoliid lineages including 12 of 17 described hyperoliid genera. The five unsampled genera are either monotypic (Arlequinus, Callixalus, Chrysobatrachus, Kassinula) or species poor (Alexteroon, 3 spp.). Our sampling of the remaining genera is proportional to their recognized diversity and includes several known species complexes with lineages not reflected in the current taxonomy. We also sampled outgroup species from the following families: Arthroleptidae (7 spp), Brevicipitidae (1 sp), Hemisotidae (1 sp), and Microhylidae (1 sp). Museum and locality information for all specimens is provided in Table S1.
Sequence capture data and alignments
The full details of transcriptome sequencing, probe design, library preparation, sequence captures, and bioinformatics pipelines are described in Portik et al. (2016b), but here we outline major steps of the transcriptome-based exon captures. We sequenced, assembled, and filtered the transcriptomes of four divergent hyperoliid species and selected 1,260 orthologous transcripts for probe design. We chose transcripts 500–850 bp in length that ranged from 5–15% average pairwise divergence. Five additional nuclear loci (POMC, RAG-1, TYR, FICD and KIAA2013) were also incorporated based on published sequence data (Portik & Blackburn 2016). The final marker set for probe design included 1,265 genes from four species and 5060 individual sequences, with a total of 995,700 bp of target sequence. These sequences were used to design a MYbaits-3 custom bait library (MYcroarray, now Arbor Biosciences) consisting of 120 mer baits and a 2X tiling scheme (every 60 bp), which resulted in 60,179 unique baits. Transcriptomes, target sequences, and probe designs are available on Dryad (Portik et al. 2016c).
Genomic DNA was extracted using a high-salt extraction method (Aljanabi & Martinez 1997) and individual genomic libraries were prepared following Meyer & Kircher (2010) with modifications described in Portik et al. (2016b). Samples were pooled for capture reactions based on phylogenetic relatedness, and the combined postcapture libraries were sequenced on three lanes of an Illumina HiSeq2500 with 100 bp paired-end reads. Raw sequence data were cleaned following Singhal (2013) and Bi et al. (2012), and the cleaned reads of each sample were de novo assembled, filtered, and mapped as described in Portik et al. (2016b). The final filtered assemblies were aligned using MAFFT (Katoh et al. 2002, 2005; Katoh & Standley 2013) and trimmed using TRIMAL (Capella-Gutierrez et al. 2009). We enforced additional post-processing filters for alignments, including a minimum length of 90 bp and a maximum of 30% total missing data across an alignment, resulting in 1,047 exon alignments totaling 561,180 bp.
Species tree estimation
We used the sequence capture data set to estimate a species tree for Hyperoliidae using ASTRAL-III (Mirarab et al. 2014; Mirarab & Warnow 2015; Zhang et al. 2017), which uses unrooted gene trees to estimate the species tree. This method employs a quartet-based approach that is consistent under the multi-species coalescent process, and therefore appropriate for resolving gene tree discordance resulting from incomplete lineage sorting (Mirarab et al. 2014; Mirarab & Warnow 2015). This approach also allows for missing taxa in alignments, which were present in our sequence capture data and are problematic for other coalescent-based summary methods such as MP-EST (Liu et al. 2010). We kept samples with the most complete sequence data to collapse the alignments to a single representative per lineage and generated unrooted maximum likelihood (ML) gene trees with 200 bootstrap replicates for each locus using RAxML v8 (Stamatakis 2014) under the GTRCAT model. The set of 1,047 gene trees was used to infer a species tree with ASTRAL-III, and node support was assessed with 1) quartet support values, or local posterior probabilities computed from gene tree quartet frequencies, which also allows the calculation of branch lengths in coalescent units (Sayyari & Mirarab 2016), and 2) 200 replicates of multi-locus bootstrapping, where each of the 200 RAxML bootstrap trees per locus are used to infer a species tree and a greedy consensus tree is created from the 200 species trees to calculate percent support across nodes (Seo 2008).
Evolutionary Relationships of Afrobatrachian Frogs
DNA barcoding
We obtained sequence data from the mitochondrial marker 16S ribosomal RNA (16S) for all samples included in the sequence capture experiment and for additional species we were unable to include in our sequence capture experiment due to insufficient DNA yield. Polymerase chain reactions (PCRs) were carried out in 12.5 μl volumes consisting of: 1.25 μl Roche 10x (500 mM Tris/HCl, 100 mM KCl, 50 mM (NH4)2 SO4, 20 mM MgCl2, pH =?8.3), 0.75 μl 25 mM MgCl2, 0.75 μl 2 mM dNTPs, 0.25 μl 10.0 μM forward primer, 0.25 μl 10.0 μM reverse primer, 8.40 μl H2O, 0.10 μl Taq, and 0.75 μl DNA. Amplification involved initial denaturation at 94°C for 4 min, followed by 35 cycles of 95°C for 60 s, 51°C for 60 s, 72°C for 90 s, and a final extension at 72°C for 7 min, using the primer pairs 16SA and 16SB (Palumbi et al. 1991). The PCR amplifications were visualized on an agarose gel, cleaned using ExoSAP-IT, and sequenced using BigDye v3.1 on an ABI3730 sequencer (Applied Biosystems). Newly generated sequences are deposited in GenBank (Accession numbers: XXXXX).
GenBank data
To expand our taxonomic sampling we included all available published sequence data of Afrobatrachian frogs. We built a molecular data matrix using the five nuclear loci included in our captures (FICD, KIAA2013, POMC, TYR, and RAG-1) and the mtDNA marker 16S. This resulted in the inclusion of 30 additional hyperoliids and 130 arthroleptid, brevicipitid, and hemisotid species, though many are represented solely by 16S mtDNA data. Nuclear loci were aligned using MUSCLE (Edgar 2004), and 16S sequences were aligned using MAFFT with the E-INS-i strategy (Katoh et al. 2002; Katoh et al. 2005). The final concatenated alignment of the expanded taxonomic data set consisted of 283 taxa and 3,991 bp, with 36% total missing data.
Phylogenetic methods and divergence dating analyses
We reconstructed the phylogenetic relationships of Afrobatrachian frogs from the five nuclear gene and mtDNA data set using a maximum likelihood (ML) approach in RAXML v8 (Stamatakis 2014). To preserve the relationships inferred with our species tree analyses of sequence capture loci, the sequence capture hyperoliid species tree containing 153 taxa (143 ingroup and 10 outgroup taxa) was used as a partial constraint tree for the ML analysis of the 283-taxon alignment of six loci. A preliminary step for our divergence dating analyses in BEAST v1.8.1 (Drummond et al. 2012) involved transforming the ML Afrobatrachian frog tree to an ultrametric tree, and for this we used penalized likelihood with the ‘chronopl’ function of APE (Sanderson 2002; Paradis et al. 2004), setting age bounds to allow divergence times to be compatible with our BEAST calibration priors. We performed analyses using a fixed ultrametric starting tree topology by removing relevant operators acting on the tree model. We used four secondary calibration points with normal distributions to constrain the most recent common ancestors (MRCAs) of Afrobatrachia to 80 Ma ± 5 SD, (Hemisotidae + Brevicipitidae) to 50 Ma ± 5 SD, Arthroleptidae to 40 Ma ± 5 SD, and Hyperoliidae to 40 Ma ± 5 SD, based on published age estimates of Afrobatrachian frogs (Roelants et al. 2007; Kurabayashi & Masayuki 2013; Loader et al. 2014; Portik & Blackburn 2016). We used the Yule model of speciation as the tree prior, applied an uncorrelated relaxed lognormal clock, and ran two analyses for 30,000,000 generations sampling every 3,000 generations. Runs were assessed using TRACER v1.5.0 (Rambaut et al. 2013) to examine convergence, and a maximum clade credibility tree with median heights was created from 7,500 trees after discarding a burn-in of 2,500 trees.
Evolution of Sexual Dichromatism and State-Dependent Diversification
We scored the presence or absence of sexual dichromatism for hyperoliid species in our data set from multiple sources, including publications (Schiøtz 1967, 1999; Channing 2001; Channing & Howell 2006; Wollenberg et al. 2007; Rödel et al. 2009; Veith et al. 2009; Amiet 2012; Bell & Zamudio 2012; Channing et al. 2013; Conradie et al. 2013; Portik et al. 2016a), examination of museum specimens (Portik 2015), and the collective field observations from all authors. A species was considered sexually dichromatic if adult females and adult males exhibit distinct color patterns (Phase F and Phase J, respectively), which also includes species with variation in male color phase (males display Phase J and Phase F). A summary of the sexual dichromatism data is provided in Table S2.
We reconstructed the evolution of sexual dichromatism on the time-calibrated phylogeny of Afrobatrachia using Bayesian ancestral state reconstruction in BEAST (Drummond et al. 2012), and also with hidden state speciation and extinction analyses using the R package HISSE (Beaulieu & O’Meara 2016) (described below). We performed Bayesian ancestral state reconstructions using several combinations of clock and character models in BEAST v1.8.1 (Drummond et al. 2012) followingKing and Lee (2015). The topology and branch lengths were fixed by removing all tree operators, and sexual dichromatism was treated as a binary alignment. Because all outgroup families are monochromatic, we fixed the root state by adding a placeholder monochromatic taxon to the root with a zero branch length, creating a hard prior distribution on the root where P(monochromatic root) = 1, and P(dichromatic root) = 0. A stochastic Mk model of character evolution (Lewis 2001) was used with symmetrical (Mk1) or asymmetrical (Mk2) transition rates between states, and each character model was analyzed using a strict clock model (enforcing a homogenous trait rate) and a random local clock model (allowing for heterotachy), resulting in four analysis combinations. The analyses involving the random local clock allowed estimation of the number and magnitude of rate changes using MCMC. All analyses were run for 200 million generations with sampling every 20,000 generations, resulting in 10,000 retained samples. We used the marginal likelihood estimator with stepping stone sampling, with a chain length of 1 million and 24 path steps, to estimate the log marginal likelihood of each run (Baele et al. 2012, 2013). We performed five replicates per analysis to ensure consistency in the estimated log marginal likelihood, and subsequently compared the four different analyses using log Bayes factors, calculated as the difference in log marginal likelihoods, to select the best fit clock model and character model combination. We summarized transitions between character states and created consensus trees to estimate the posterior probabilities of a character states across nodes.
We performed hidden state speciation and extinction (HiSSE) analyses using the R package HISSE (Beaulieu & O’Meara 2016) to identify if sexual dichromatism in Afrobatrachian frogs is associated with increased diversification rates, and to reconstruct ancestral states while accounting for transition rate and diversification rate heterogeneity. The HiSSE model builds upon the popular binary-state speciation and extinction (BiSSE) model (Maddison et al. 2007) by incorporating ‘hidden states’ representing unmeasured traits that could impact the diversification rates estimated for states of the observed trait. The HiSSE model is therefore able to account for diversification rate heterogeneity that is not linked to the observed trait, while still identifying trait-dependent processes. The HiSSE framework also includes a set of null models that explicitly assume the diversification process is independent from the observed trait, without constraining diversification rates to be homogenous across the tree. The inclusion of these character-independent models circumvents a significant problem identified in the BiSSE framework, in which the simple ‘null’ model of constant diversification rates is typically rejected in favor of trait-dependent diversification when diversification rate shifts unrelated to the trait occur in the phylogeny (Rabosky & Goldberg 2015; Beaulieu & O’Meara 2016). The improved character-independent diversification models, referred to as CID-2 and CID-4, contain the same number of diversification rate parameters as the BiSSE and HiSSE models, respectively. We fit 26 different models to our sexual dichromatism data set (Table 1): six represent BiSSE-like models, four are variations of the CID-2 model, five are variations of the CID-4 model, nine are various HiSSE models with two hidden states, and two are HiSSE models with a single hidden state. Within each of these classes, the models vary mainly in the number of distinct transition rates (q), extinction fraction rates (ε), and net turnover rates (τ), and the most complex HiSSE model includes four net turnover rates, four extinction fraction rates, and eight distinct transition rates. We enforced a monochromatic root state for all models and evaluated the fit of the 26 models using AIC scores, ΔAIC scores, and Akaike weights (ωi) (Burnham & Anderson 2002). From the best-fit model, we estimated confidence intervals for relevant parameters and transformed ε and τ to obtain speciation (λ), extinction (μ), and net diversification rates using the ‘SupportRegion’ function in HISSE (Beaulieu & O’Meara 2016). We performed ancestral state estimations for each of the 26 models using the marginal reconstruction algorithm implemented in the ‘MarginRecon’ function of HISSE, again enforcing a monochromatic root state. Our final estimation and visualization of diversification rates and node character states on the Afrobatrachian phylogeny took model uncertainty into account by using the model averaging approach described by Beaulieu and O’Meara (2016), such that model contributions to rates and states were proportional to their likelihoods.
Simulated traits and state-dependent diversification
The identified bias towards the detection of trait-dependent diversification in the BiSSE framework (Rabosky & Goldberg 2015; Beaulieu & O’Meara 2016) prompted us to investigate if a similar outcome would be detected in our data set, and whether the HiSSE framework could improve our ability to distinguish whether the observed states are correlated with diversification rates. One concern raised by Beaulieu and O’Meara (2016) is that neutrally evolving traits simulated on trees generated from a complex heterogenous rate branching process can lead to false signals of trait-dependent diversification, signifying the HiSSE framework may be sensitive to particular types of tree shapes.
To evaluate the performance of these methods given our empirical tree topology, we simulated neutrally evolving traits on the Afrobatrachian frog phylogeny and tested for trait-dependent diversification using both the BiSSE and HiSSE frameworks. We conducted independent simulations of a binary trait with the ‘sim.history’ function in R package PHYTOOLS (Revell 2012), using the unequal rates q-matrices obtained from our empirical data, enforcing a root state of trait absence. We required a minimum of 10% of taxa to exhibit the derived state and conducted simulations until we obtained 1,500 replicates meeting this criterion. We used maximum likelihood to fit a BiSSE model and the typical ‘null’ model with equal speciation and extinction rates to each simulated trait using the R package DIVERSITREE (Maddison et al. 2007; FitzJohn et al. 2009; FitzJohn 2012). We performed likelihood ratio tests and calculated ΔAIC scores to determine if the constraint model could be rejected with confidence (ΔAIC > 2 or p-value < 0.05), and summarized the number of instances each of the two models was favored across the simulations. We analyzed the simulated data in the HiSSE framework as with our empirical data, but with a reduced set of five models representing each major category of model. This reduced model set included two BiSSE-like models that differed only in the constraint of τ, a CID-2 and CID-4 model, and a HiSSE model with two hidden states, three transition rates, equal τ, and distinct τ. We set a probability of one for trait absence at the root state to match the manner in which traits were simulated and evaluated the fit of the 5 models using ΔAIC scores and Akaike weights (ωi) for each simulation, using a threshold of ΔAIC greater than two to favor a model. Specifically, we were interested in whether the unconstrained BiSSE-like or HiSSE models were favored, resulting in the detection of a false pattern of trait-dependent diversification, or if the CID-2 or CID-4 models were selected, capturing the expected pattern where the diversification process was independent from trait evolution.
RESULTS
Phylogenetic Relationships
The sequence capture data set consisting of 1,047 loci and 561,180 bp produced a well-resolved species tree with a normalized quartet score of 0.877 and only eight of 150 nodes (5%) with quartet scores below 0.9 (Fig. S1). The multilocus bootstrapping analysis produced similar results with low support for only seven nodes, six of which also received low support in the species tree analyses (Fig. S1). The higher-level relationships recovered in the species tree are largely congruent with those recovered by Portik and Blackburn (2016), though we found strong support for a different placement of the genus Acanthixalus as sister to the clade containing Semnodactylus, Paracassina, Phlyctimantis, and Kassina, which together are sister to all other hyperoliids. Our improved sampling provides the first comprehensive assessment of evolutionary relationships within the speciose genera Afrixalus and Hyperolius. The monophyly of Hyperolius is supported, however Afrixalus is paraphyletic, and we found a sister relationship between the Ethiopian-endemic A. enseticola and the Malagasy-Seychelles species of Heterixalus and Tachycnemis, which are in turn sister to all remaining Afrixalus (Fig. S1). We found strong support for the southern African species Hyperolius semidiscus as sister to all other lineages in the genus, and furthermore recovered H. parkeri, H. lupiroensis and the H. nasutus complex as sister to two larger subclades of Hyperolius (Clades 1 and 2; Figs. 2, S1). In an effort to distinguish the main division within Hyperoliidae, we recognize the subfamilies Kassininae Laurent, 1972 and Hyperoliinae Laurent, 1943 and define the content within each on the basis of our species tree analysis as follows: 1) Kassininae: Acanthixalus, Kassina, Paracassina, Phlyctimantis, and Semnodactylus, 2) Hyperoliinae: Afrixalus, Cryptothylax, Heterixalus, Hyperolius, Morerella, and Opisthothylax (Fig. 2). We retain previous subfamily assignments for genera not sampled in our molecular study (Hyperoliinae: Alexteroon, Arlequinus, Callixalus, Chrysobatrachus, Kassinula), which should be included in future phylogenetic studies to confirm these placements.
The phylogenetic analyses of the Afrobatrachia supermatrix produced family-level relationships consistent with previous analyses (Pyron & Wiens 2011; Portik & Blackburn 2016; Feng et al. 2017), including a sister relationship between Hyperoliidae and Arthroleptidae, and between Hemisotidae and Brevicipitidae (Fig. S2). This expanded taxonomic data set also included improved sampling for the Malagasy hyperoliid genus Heterixalus and the Hyperolius nasutus complex, for which we recovered results consistent with Wollenberg et al. (2007) and Channing et al. (2013), respectively. We recovered an Eocene age for the time to most recent common ancestor (TMRCA) of the families Hyperoliidae, Arthroleptidae and Brevicipitidae, of approximately 42.4 Ma (95% HPD: 39.1–51.6 Ma), 45.4 Ma (95% HPD: 35.5–48.7 Ma), and 41.8 Ma (95% HPD: 33.9–49.8 Ma) (Fig. S2). These dates are younger than previous estimates (Roelants et al. 2007; Loader et al. 2014) but are consistent with more recent multilocus analyses (Portik & Blackburn 2016; Feng et al. 2017). We found diversification events began within Hyperoliinae approximately 35.8 Ma (95% HPD: 30.1–41.6 Ma) and within Kassininae approximately 38.3 Ma (95% HPD: 30.9–45.6 Ma). The majority of speciation events within Hyperolius occurred from the late Miocene to the Plio-Pleistocene (Fig. S2).
Evolution of Sexual Dichromatism and State-Dependent Diversification
We found sexual dichromatism occurs in 60 of the 173 (34%) hyperoliid frog species included in our analyses. The Bayesian ancestral state reconstructions using four model combinations produced different patterns of character reconstructions and rates of trait evolution (Fig. S3). Overall, we found models with asymmetric character transition rates (Mk2) outperformed those with symmetric rates (Mk1), regardless of the clock model used (log-transformed Bayes factor range: 3.40–10.39; Table 2). The analysis with the highest log-marginal likelihood incorporated the relaxed local clock (RLC) Mk2 model, though it was not a significantly better fit than the simpler strict clock (SC) Mk2 model (log-transformed Bayes factor of 0.10) indicating low variation in lineage-specific evolutionary rates for this trait. The ancestral character reconstructions of the SC Mk2 indicated a median of two transitions from monochromatism to dichromatism. However, the character reconstructions revealed that this estimate is due to low posterior probability estimates for two critical nodes within Hyperoliinae (Fig. S3), and the presence of dichromatism at either one or both of these nodes would result in a single origin of sexual dichromatism. By contrast, the SC Mk2 model demonstrated approximately 27 independent losses of sexual dichromatism within hyperoliids, including losses in entire groups (Afrixalus, H. nasutus complex) and numerous reversals within Heterixalus and Hyperolius (notably in Clade 2) (Fig. S4).
Our hidden state speciation and extinction analyses revealed that a HiSSE model with four net turnover rates, equal extinction fraction rates, and three distinct transition rates was the best-fit model (Model 19, Table 1). The second and third ranked models (ΔAIC of 1.0, 2.5) were also HiSSE models that varied in the number of extinction fraction rates or the number of transition rates. Together these three HiSSE models accounted for 84.1% of the model weight (Table 1) and support a signal of state-dependent diversification in which sexual dichromatism and a hidden state are associated with diversification rates. The net diversification rates inferred using the best-fit model were nearly twice as high in sexually dichromatic lineages (0.157) as compared to monochromatic lineages (0.091) in the absence of the hidden state (Fig. 3), and the combination of the hidden state and dichromatism or monochromatism resulted in much lower net diversification rate estimates (0.02 and <0.001, respectively). The trait reconstructions for all four state combinations indicated that the dichromatism plus hidden state combination was only present in the MRCA of two genera (Cryptothylax, Morerella) (Fig. S5), and was associated with a markedly lower diversification rate (Fig. 3). The model-averaged ancestral state reconstructions demonstrated strong support for a single origin of sexual dichromatism and 25 independent reversals to monochromatism within Hyperoliinae, with reversal patterns similar to the Mk2 analyses (Figs. 2, S5).
Simulated traits and state-dependent diversification
Analyzed in the BiSSE framework, we found that many of the trait simulations on the phylogeny of Afrobatrachia resulted in the rejection of the ‘null’ model of equal diversification rates across character states. Based on the significance of likelihood ratio tests, we rejected the ‘null’ model in favor of trait-dependent diversification for 565 (37.6%) of our 1500 comparisons. We recovered similar results using a delta AIC cutoff value of two, for which we found support for trait-dependent diversification in 546 (36.4%) of the simulations (Fig. 4). In many of these cases we found unexpectedly strong support for the BiSSE model, and 160 (10.6%) of the comparisons resulted in delta AIC values ranging from 10–40.
In contrast to the BiSSE analyses, the addition of the character independent models (CID-2 and CID-4) in HiSSE dramatically reduced the detection of a false association between simulated traits and diversification rates by providing appropriate null models. In addition to these two CID models, our set of five models also included two BiSSE-like models and a typical HiSSE model. Based on a delta AIC cutoff value of two, out of the 1500 analyses performed the CID-4 model was selected 1,132 times (75.4%), the HiSSE model was chosen 125 times (8.3%), and the remaining 243 analyses (16.2%) showed equivocal support (ΔAIC = 0–2) for either the CID-4 or HiSSE model (Fig. 5). In the cases of equivocal support, the CID-4 and HiSSE models were always the top two models, which should be interpreted as a lack of support for trait-dependent diversification. Our error rate with the HiSSE model being favored in only 8.3% of our simulations was substantially lower than the Beaulieu and O’Meara (2016) ‘difficult tree’ scenario in which the HiSSE model was favored in 29% of the data sets. These simulation results strongly suggest the branching pattern of our empirical phylogeny is not inherently problematic for the investigation of trait-dependent diversification using these available methods, adding support to our empirical analyses in which we detected an association between diversification rates and sexual dichromatism.
DISCUSSION
Hyperoliid Relationships and the Origin of Sexual Dichromatism
Afrobatrachian frogs account for more than half of all African amphibians and this continental radiation exhibits incredible variation in ecomorphology, reproductive mode, and other life history traits (Portik & Blackburn 2016). Within Afrobatrachia, the family Hyperoliidae is the most species-rich (~230 species) with surprisingly little diversity in ecomorphology and reproductive mode (Schiøtz 1967, 1999; Portik & Blackburn 2016), but with incredible variation in coloration and sexual dichromatism. These phenotypic characteristics have generated considerable taxonomic confusion within hyperoliids (e.g. Ahl 1931), hindering a clear understanding of species diversity and the evolutionary history of this radiation. Here, we have produced the most comprehensive hyperoliid species tree to date and found support for an Eocene origin of two subfamilies representing a major division withiin Hyperoliidae: Kassininae (26 species) and Hyperoliinae (~200 species). Within Hyperoliinae, we clarified relationships within the hyperdiverse genus Hyperolius (~150 species), which largely consists of two clades (Figs. 2, S1, S2). These clades represent parallel rapid radiations with species distributed across a variety of habitats and altitudes throughout sub-Saharan Africa, which showcases Hyperoliidae as a rich comparative framework for future biogeographic research.
While both Kassininae and Hyperoliinae include colorful species, sexual dichromatism only occurs within Hyperoliinae where it is present in the genera Tachycnemis, Cryptothylax and Morerella, several species of Heterixalus, and more than half of the Hyperolius species we sampled (Fig. 2, Table S2). A previous study hypothesized that sexual dichromatism evolved multiple times within Hyperoliidae (Veith et al. 2009); however, we found overwhelming support for a single origin of sexual dichromatism (Figs. 2, S3) and over twenty independent reversals to monochromatism that range from early transitions that characterize entire genera (e.g., Afrixalus) to recent reversals within Heterixalus and Hyperolius (Figs. 2, S4). This transition bias from dichromatism to monochromatism also occurs in birds (Price & Birch 1996; Omland 1997; Burns 1998; Kimball et al. 2001; Hofmann et al. 2008; Dunn et al. 2015; Schultz & Burns 2017), in which secondary monochromatism can evolve as the result of a color change in either sex (Kimball & Ligon 1999; Johnson et al. 2013; Price & Eaton 2014; Dunn et al. 2015; Schultz et al. 2017). Similar transitional pathways to secondary monochromatism occur in hyperoliids, in which females may lose the ability for color transformation at sexual maturity (both sexes retain the Phase J juvenile coloration) or males undergo obligatory ontogenetic color change at maturity (both sexes develop Phase F coloration) (Figs. 1, 5). Characterizing juvenile coloration across species, which is undocumented in many hyperoliids, and documenting ontogenetic color change, including the prevalence of male color phases (Portik et al. 2016a), will be essential steps toward differentiating between these two forms of monochromatism. In experimental settings, the hormone estradiol induces color transformation in both sexes in the dichromatic species H. argus and H. viridiflavus, yet the effects of testosterone differ across species (Richards 1982; Hayes & Menendez 1999) suggesting that evolutionary transitions from dichromatism to monochromatism result directly from differences in hormone sensitivities among species. Reversals to monochromatism are especially prominent in Hyperolius Clade 2 (twelve independent events; Figs. 2, S4), highlighting the potential for future research in this group to identify the physiological basis and molecular underpinnings of these transitions. For example, three monochromatic species of the H. cinnamomeoventris complex that are endemic to the islands of São Tomé and Príncipe (H. drewesi, H. molleri, H. thomensis; both sexes with Phase F coloration) are derived from a mainland clade with sexually dichromatic (H. cinnamomeoventris, H. olivaceus; females Phase F, males Phase J) and monochromatic species (H. veithi; both sexes Phase J) (Fig. 5). Variation among these closely related species is well suited for investigating both the evolutionary and ecological contexts underlying transitions from sexual dichromatism to monochromatism.
Sexual Dichromatism is Linked to Increased Diversification Rates
Sexual dichromatism is an important predictor of diversification in several taxonomic groups including cichlids (Wagner et al. 2012), labrid fish (Alfaro et al. 2009; Kazancıo⍰lu et al. 2009), agamid lizards (Stuart-Fox & Owens 2003), and dragonflies (Misof 2002); yet in birds, the relationship between dichromatism and species richness/speciation rate varies among studies that differ in methodology and evolutionary scale (Barraclough et al. 1995; Owens et al. 1999; Morrow et al. 2003; Phillimore et al. 2006; Seddon et al. 2013; Huang & Rabosky 2014). Though some state-dependent speciation and extinction model sets, such as the BiSSE method, are no longer considered adequate for robustly detecting trait-dependent diversification (Rabosky & Goldberg 2015; Beaulieu & O’Meara 2016), the HiSSE method contains an expanded model set that can link diversification rate heterogeneity to observed traits, hidden traits, or character independent processes. In particular, the inclusion of appropriate null models (character-independent models) reduces the inference of trait-dependent diversification when a phylogeny contains diversification rate shifts unrelated to the focal trait (Beaulieu & O’Meara 2016). Our trait simulation study recapitulated this result (Fig. 4) while also demonstrating that the branching pattern of our afrobatrachian phylogeny is unlikely to drive false inferences of trait-dependent diversification using HiSSE (e.g., the “worst-case” scenario of Beaulieu & O’Meara 2016). We found that sexually dichromatic hyperoliid lineages have nearly double the average diversification rate of monochromatic lineages, and that these diversification rates are not inflated by a hidden trait. Arboreal oviposition and an associated shift to breeding in pond systems both evolved within hyperoliids and were previously suggested as potential factors influencing diversification (Portik & Blackburn 2016). In our analyses these traits may have represented hidden factors co-distributed with dichromatism; however, the shift to the sexual dichromatism plus hidden state character combination occurred only once in the common ancestor of two genera (Cryptothylax, Morerella; Fig. S5) and is actually associated with lower diversification rates (Fig. 3), indicating that these additional traits had little impact on our analyses. Together, our results demonstrate that diversification rate heterogeneity occurs within Afrobatrachia and that the origin and persistence of sexual dichromatism in hyperoliid frogs is linked to their rapid diversification across sub-Saharan Africa.
How Does Sexual Dichromatism Influence Diversification Rate?
Sexual dichromatism is a common proxy for sexual selection in macroevolutionary studies (reviewed in Kraaijeveld et al. 2011), especially for testing the prediction that clades with variation in secondary sexual characters under strong sexual selection exhibit higher diversification rates and species richness (Lande 1981, 1982; West-Eberhard 1983; Barraclough et al. 1995). While our trait-dependent diversification analyses strongly support a faster rate of net diversification associated with sexual dichromatism, we currently lack behavioral and natural history data in hyperoliids to establish whether sexual dichromatism is actually under strong sexual selection. At more recent timescales, speciation by sexual selection can result in a group of closely related species with high ecological similarity but differing almost exclusively in signaling traits (West-Eberhard 1983; Dominey 1984; Panhuis et al. 2001; Turelli et al. 2001; Andersson 2004; Mendelson & Shaw 2005; Ritchie 2007; Safran et al. 2013). Frogs are generally opportunistic gape-limited predators (Duellman & Trueb 1986), and in hyperoliids food partitioning is strongly dictated by body size (Luiselli et al. 2004). In one well-characterized community site, the females of seven syntopic sexually dichromatic Hyperolius species display minimal body size differences (Portik et al. 2018) but show exceptional divergence in color (Portik et al. 2016a). Five of these species occur in the predominantly dichromatic Hyperolius Clade 1, and within this clade there are typically striking interspecific color differences in females, but not males, among species (Fig. 5). This pattern of interspecific variation in secondary sexual characters in the absence of ecological divergence is consistent with the predictions of speciation by sexual selection, and by inference, would imply that dichromatism in hyperoliids – specifically female color – is an essential mate recognition signal (sensu Mendelson & Shaw 2012). In frogs, male calls are well-established mate recognition signals that can be under strong sexual selection (Ryan 1980; Gerhardt 1994), and these acoustic signals are demonstrably important for hyperoliid frogs. Males form dense breeding aggregations and choruses (Bishop et al. 1995), calls differ between closely-related species (Schiøtz 1967, 1999; Gilbert & Bell 2018), and females prefer conspecific calls over calls of syntopic heterospecifics (Telford & Passmore 1981). However, visual displays can also serve as important courtship signals in frogs, typically in conjunction with acoustic signaling (Gomez et al. 2009; Gomez et al. 2010; Starnberger et al. 2014; Jacobs et al. 2016; Yovanovich et al. 2017; Akopyan et al. 2018). Although most studies have documented female preference for male coloration, the recent discovery of female displays during nocturnal phyllomedusine treefrog courtship highlights the possibility of male mate-choice and intersexual selection of female coloration (Akopyan et al. 2018). The lack of heterospecific matings among dichromatic species at high-diversity breeding sites (Portik et al. 2018) strongly suggests behavioral isolation may be linked to divergent mate recognition signals, including male call (Telford & Passmore 1981), gular gland compounds (Starnberger et al. 2013), or female coloration. Our knowledge of reproductive behavior in hyperoliids is largely based on a single dichromatic species (H. marmoratus; Dyson & Passmore 1988; Telford & Dyson 1988; Dyson et al. 1992; Jennions et al. 1995), and as such there may be an overlooked role for mutual mate-choice in hyperoliids, in which females locate males by call and/or pheromones and males assess color patterns of approaching females.
Although it can be tempting to equate sexually dimorphic traits such as dichromatism with sexual selection, several alternative mechanisms may also contribute to female-biased dichromatism in hyperoliids. For instance, aposematism is a widespread anti-predator mechanism in frogs that is typically accompanied by the presence of skin toxins (reviewed in Toledo & Haddad 2009; Rojas 2017), commonly alkaloids (Daly 1995). Sex-specific differences in chemical defense occur in some frogs (Saporito et al. 2010; Jeckel et al. 2015); however, both males and females of four dichromatic hyperoliids lacked alkaloids in their skin (Portik et al. 2015) suggesting that either aposematism is an unlikely explanation for ornate coloration or that hyperoliids have evolved novel compounds for chemical defense. Female-biased dichromatism has also been tied to sex-role reversal in fishes and birds (Roede 1972; Oring 1982; Berglund et al. 1986a, 1986b; Eens & Pinxten 2000), in which females compete more intensely than males for access to mates. This mechanism seems unlikely for hyperoliids because males in both monochromatic and dichromatic species form dense choruses and compete intensely to attract females, often engaging in combat (Telford 1985; Backwell & Passmore 1990). Finally, sexual niche partitioning (Selander 1966; Shine 1989) can result in sexual dichromatism when the sexes use different microhabitats, and as a consequence are subject to different selective pressures and predation regimes (Heinsohn et al. 2005; Bell & Zamudio 2012). During the breeding season, male hyperoliids are exposed on calling sites and Hayes (1997) proposed that more cryptic male coloration may reduce predation pressure, a hypothesis that is consistent with greater observations of predation events on females of dichromatic species (Grafe 1997; Portik et al. 2018). Quantifying differences in predation rates between the sexes, between monochromatic and dichromatic species, and between Phase F and Phase J males within dichromatic species would address whether this aspect of natural selection is also shaping the evolution of sexual dichromatism. Although these mechanisms and other differences in natural selection pressures may influence the evolution of sexual dichromatism in hyperoliids, they are generally not expected to elevate rates of reproductive isolation or drive diversification rate shifts comparable to the effects of sexual selection. Therefore, we propose that hyperoliids are a compelling system for disentangling the roles of sexual selection and natural selection in the evolution of sexual dichromatism, and how these mechanisms have promoted diversification at both microevolutionary and macroevolutionary timescales.
ACKNOWLEDGEMENTS
Laboratory work conducted by DMP was funded by a National Science Foundation DDIG (DEB: 1311006), a National Science Foundation grant (DEB: 1202609) awarded to D.C. Blackburn, an Ecological, Evolutionary, and Conservation Genomics Research Award presented by the American Genetic Association, the Museum of Vertebrate Zoology, and by R.C. Bell and J.A. McGuire. This work used the Vincent J. Coates Genomics Sequencing Laboratory at UC Berkeley, supported by NIH S10 Instrumentation Grants S10RR029668 and S10RR027303. We thank the following institutions for accessioning field collections and for facilitating loan access: California Academy of Sciences, Cornell University Museum of Vertebrates, Institut national de Recherche en Sciences Exactes et Naturelles, Museum für Naturkunde, Berlin, Muséum National d’Histoire Naturelle, Museum of Comparative Zoology, Museum of Vertebrate Zoology, National Museum, Prague, Natural History Museum, London, North Carolina Museum of Natural Sciences, Port Elizabeth Museum, Senckenberg Natural History Collections Dresden, South African Institute for Aquatic Biodiversity, South African National Biodiversity Institute, The Field Museum, Trento Museum of Science, University of Texas at El Paso Biodiversity Collections, and the Zoological Natural History Museum, Addis Ababa University. All authors express thanks to the many government agencies, ministries, and departments that issued research permits and provided the access necessary to conduct their individual field work in sub-Saharan Africa.