Life-history data

We searched the literature for data on relevant life-history variables for all pelobatoid species (Appendix S1), starting from the summary provided in Gomez-Mestre & Buchholz [24]. Whenever possible, we used only data measured in the field under natural conditions. For those few species lacking field-based data for a specific variable, data from the lab were used instead. We obtained data on larval period (from hatching of the eggs to approximately Gosner stage 42) and hatching time (from egg deposition to hatching). For each variable, when multiple records were available for one species, we obtained the maximum and minimum values from the available records and calculated the midpoint. Data were available for most scaphiopodid, pelobatid, and pelodytid species, but relevant data were available for only two megophryid species. We note that megophryids are geographically and climatically distinct from the other pelobatoid families, occurring primarily in mesic tropical and subtropical areas of Asia [21].

Data on genome sizes were obtained from T. R. Gregory's database (http://www.genomesize.com/). Data were available for only eight pelobatoid species, but these species included two from each genus for all the genera in the Pelobatidae, Pelodytidae, and Scaphiopodidae. Summary life-history data for each species for each variable are provided in Table S1.

Climatic data

To obtain climatic data for each species, we first obtained georeferenced locality data. We used species distribution data from HerpNET (www.herpnet.org) and the Global Biodiversity Information Facility (GBIF; www.gbif.org; Version 1.2.6). HerpNET and GBIF both provide a frequently updated database of museum specimen records. We searched each database for each species, and then combined the records into a set of unique localities for each species. Sample sizes ranged from 3 to 322 localities per species (mean  = 89.1).

Localities for each species were visualized using DIVA-GIS version 7.5.0.0 and compared to species distribution maps from the IUCN Red List of Threatened Species, version 2012.2 [30]. The known distributions of these species are relatively stable and agreed upon by different sources [11], [20], [30]. Localities falling outside the IUCN distribution map were excluded. We also excluded localities with estimated elevations (from WorldClim, see below) that fell outside the range of reported elevations from IUCN.

Using this carefully vetted set of georeferenced localities, we then obtained climatic data from the WorldClim (version 1.3) database [31]. Data are based on averages from weather stations from the years ∼1950–2000 (with spatial interpolation to localities between weather stations), with a spatial resolution of ∼1 km². For each locality, we extracted data on annual precipitation (Bio12), precipitation seasonality (Bio15), and precipitation of the wettest quarter (Bio16). We expect these variables to have the strongest influence on rainfall available for filling temporary ponds for anuran reproduction. Maximum and minimum values across localites within the species range were obtained, as well as the midpoint of these two values, and the mean value for the species averaged across all localities.

We also used a measure of aridity per se, following Oufiero et al. [32]. For each species, this was calculated as where P is annual precipitation (mm; mean across localities across species range), Tmax is the maximum value of Bio5 (maximum temperature of warmest month) across the species range, and Tmin is the lowest value of Bio6 (minimum temperature of the coldest month). Arid environments have a lower Q [33]. We also used logQ2, in which Tmax is the mean value of Bio10 (mean temperature of warmest quarter) across localities and Tmin is the mean value of Bio11 (mean temperature of coldest quarter), but this gave similar results. A summary of the climatic data for each species is provided in Table S1.

Time-calibrated phylogeny

We estimated a time-calibrated phylogeny for pelobatoid frogs, since one including all relevant taxa was not available at the time we initiated our study (and one available now has some issues, see below). We first reduced the data matrix compiled by Pyron & Wiens [22] to include only the 16 pelobatoid species for which life-history data were available. However, these data were available for only two megophryid species, and life-history data were available for Leptobrachium nigrops but not sequence data. Therefore, rather than exclude this species and genus, we used sequence data from Leptobrachium chapaense to represent L. nigrops in the tree. We also excluded genes sampled for fewer than 4 included species. The resulting data matrix included 16 species and data from the mitochondrial ribosomal genes 12S and 16S, the mitochondrial protein-coding gene cytochrome b, and the nuclear protein-coding genes H3A, RAG1, RHO, SIA, and SLC8A3 (total length of combined alignments  = 9,355 base pairs). Data were not available for all 8 genes for all 16 species, and some species were therefore missing data for some genes. However, both simulations and empirical studies suggest that including some missing data need not lead to inaccurate estimates of phylogeny, especially when a large number of characters is sampled overall (review in [34]).

The time-calibrated tree was estimated using the Bayesian uncorrelated lognormal approach in BEAST 1.5.4 [35], [36]. We used the GTR + I + Γ model (following [22]) with 4 rate categories for Γ, and estimated base frequencies. We used a clock model with the uncorrelated, lognormal approach, and an estimated rate. The starting tree was based on a Yule speciation prior.

We initially used two fossil calibration points. For each point, we identified a fossil that represented the oldest taxon that could be confidently assigned to a clade of extant species. A fossil can be used to determine the minimum age of a clade, but the clade can be older than this oldest known fossil. We therefore used a lognormal prior distribution for each calibration point, with an offset equal to the minimum age of the oldest fossil (the youngest age of the oldest stratum in which it is found) and a mean of 5 and a standard deviation of 1 Myr. This combination of mean and standard deviation yields a 95% prior distribution that extends from just slightly older than the minimum age of the fossil to approximately 15 Myr older (an arbitrary but seemingly plausible range), with the highest probability slightly older than the age of the fossil.

We initially used the crown-group age of Pelobatoidea as being at least 50.3 Myr old, given a fossil scaphiopodid (Scaphiopus guthriei) from the Wind River formation (lower Eocene Wasatchian 50.3–55.4 Mya) following Rocek & Rage [37]. The 95% interval on the prior is 50.9–66.0 Myr.

We initially treated the crown-group age of Pelobatidae and Megophryidae as being at least 33.9 Myr old, given the fossil Eopelobates grandis which appears to be closely related to Pelobates [37], [38], from the Chadron formation (33.9–38 Mya). The 95% interval on the prior for Pelobatidae+Megoprhyidae is 34.5–49.6 Myr. This clade may be older if the undescribed “Green River pelobatid” can be assigned to it (Wasatchian, 50.3–5.4 Mya; [37]). Further, the most recent common ancestor of Pelodytidae, Pelobatidae, and Megophryidae is also at least 33.9 Mya, given fossil Pelodytes from the late Eocene (37.2–33.9 Mya; [37]), but we did not use this fossil calibration (given that the pelobatid + megophryid calibration ensures that this clade is at least this old).

Initial analyses yielded family-level topologies that did not match those of Pyron & Wiens [22], most likely due to the lack of outgroups. Given that these family-level relationships are generally well supported when outgroups are included (e.g. [22], [39]–[41]), we constrained these relationships. Specifically we constrained the clade: Pelodytidae+Pelobatidae+Megophryidae. Given that the pelobatid+megophryid clade is a fossil constraint, these two constraints enforce the Pyron & Wiens [22] topology for families. Importantly, the same set of family-level relationships is also found in other previous analyses of pelobatoid relationships, including those based on mitochondrial data only [39], nuclear data only [40], and combined nuclear and mitochondrial data [41].

We performed two independent runs each with 50,000,000 generations sampled every 1,000 generations. We used the maximum clade credibility trees with mean node heights. The first 10% of generations sampled were discarded as burn-in using TreeAnnotator version 1.5.4 and viewed using FigTree version 1.3.1 [35]. We confirmed that the two independent runs gave effective sample sizes (ESS) greater than 200 for the likelihood and selected clade ages, and that they converged on similar topologies and divergence dates. Trees from the two analyses were combined to yield a majority-rule consensus tree with mean branch lengths.

This initial analysis yielded estimated ages for Pelobatoidea and the family-level clades within it that were considerably younger than those estimated in previous studies (e.g. [40]–[42]). For example, the pelobatoid crown group was 53 Myr old in this tree, and ∼130, 170 and 150 Myr old (respectively) in these previous studies. Therefore, we reran the analyses as above, but making two changes. First, for the crown age of Pelobatoidea, we used a normal prior distribution with a mean age of 150 Mya and a standard deviation of 10 Myr (95% prior interval: 133.6–166.4). This prior interval roughly corresponds to the range of estimated ages in previous studies. Second, we used the fossil calibration for Scaphiopus guthrei for the crown-group age of Scaphiopodidae, rather than the stem-group age (this choice is less conservative about the placement of this fossil but more in line with previous age estimates).

This second set of results gave an identical topology and similar relative branch lengths to the first analysis, but with absolute branch lengths (ages) similar to those estimated in previous studies. We used this second BEAST tree for our phylogenetic comparative analyses. The topology was very strongly supported, with only one node with a posterior probability <0.95 (Fig. 1). Therefore, we did not incorporate uncertainty in the phylogeny into our comparative analyses. This topology is available in nexus/newick format in Appendix S2.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Time-calibrated phylogeny of pelobatoid frogs used in comparative analyses.

Numbers at nodes indicate Bayesian posterior probabilities of clades.

https://doi.org/10.1371/journal.pone.0096637.g001

Note that another study has recently estimated a large-scale time-calibrated tree for amphibians, including pelobatoids [43]. However, given the large number of taxa included in that study, the use of a somewhat suboptimal method for estimating divergence dates was necessary (penalized likelihood; [44]). Furthermore, that study [43] relied entirely on secondary calibration points (from [42]). Therefore, we prefer our estimate of divergence dates. Nevertheless, these estimates are actually quite similar (and similar to estimates in other recent studies [7], [40]–[42]), and are based on nearly identical molecular datasets [22].

Phylogenetic comparative analysis

We tested the relationship between pairs of variables using phylogenetic generalized least squares (PGLS; [45]) as implemented in the R package caper, version 0.5 [46]. Prior to conducting these analyses, we found the best-fitting evolutionary model for each variable using the R packages ape [47] and geiger [48]. We compared the fit of the models using the estimated likelihood and Akaike information criterion (AIC), with an AIC difference of 4 or greater indicating support for alternative models [49]. We compared the Brownian motion (BM; perfect fit of a character to the phylogeny), Ornstein-Uhlenbeck (OU; equivalent to stabilizing selection around a single optimum), and estimated lambda (level of phylogenetic signal is estimated) models (see Table S2). We found that in most cases, model fit was similar between the OU and lambda models (AIC difference <4), with the exception of two climatic variables for which OU was strongly favored (annual precipitation, wettest quarter precipitation). We therefore used the lambda model in PGLS, given that this model was either favored and/or alternate models were not. We also performed a set of analyses using the OU model. Specifically, we repeated the PGLS analyses after using geiger to transform the tree based on the OU model and the estimated value of alpha. To estimate alpha, the selected variable was fitted to the OU model 10 times and the alpha with the minimum deviance (−2 * log-likelihood) was applied in the transformation [50]. PGLS results were generally similar using the lambda and OU models, and we present the results using the lambda model as our primary results. We present the OU results as supplementary information (Table S4).

We used PGLS to test the following specific hypotheses. (1) We predicted that overall hatching times and larval periods of species will be related to their mean values for climatic variables (annual precipitation, precipitation of the wettest quarter, aridity), assuming that species reduce their hatching times and larval period to allow them to metamorphose before temporary breeding ponds dry in more arid climates. For these analyses, we summarized variation in hatching times and larval periods within species based on midpoint values (midpoint between lowest and highest values within species). (2) We predicted that the minimum (shortest) hatching times and larval periods within species will be related to the lowest values for annual precipitation related variables across the range of each species. This second set of analyses was intended to address the possibility that overall species values might not reflect variation within species, and that we expect shorter larval periods and hatching times in parts of the species range with lower rainfall. (3) We predicted that hatching times and larval period would be related, assuming that species occurring in drier environments will evolve to minimize both simultaneously. We examined both midpoint and minimum values for these variables within species. (4) We predicted that genome sizes would be smaller in species with more rapid development (shorter hatching times and larval periods).

We acknowledge that the methods described above could lead to a potential mismatch between developmental traits and climatic variables for specific localities (e.g. for a given species, the locality with the shortest recorded larval period may not correspond to the driest locality where the species occurs). Therefore, we performed an additional analysis in which both developmental and climatic values for each species were based on a single locality with the shortest recorded field-based larval period for that species (focusing on annual precipitation). However, there were some issues in this analysis. First, the data on larval period for four species could not be traced to specific localities (i.e. Pelodytes caucasica, Pelobates syriacus, Pelobates varaldii, Megophrys nasuta). For these species, we used data on the minimum recorded larval period (from the field) and the lowest annual precipitation across sampled localities. In several other species, it was not possible to trace the shortest recorded larval period known for that species to a specific locality. In these cases, we simply used the shortest larval period for a given species that could be traced to a specific locality. Although these larval periods were sometimes longer than the shortest larval periods recorded for that species, they may also be more reliable, and should provide a strong overall test of how climate and larval period are related. The data and references used are summarized in Appendix S3 and Table S3.

We also tested each variable for phylogenetic signal using lambda [51]. Given that the phylogenetic results were very strongly supported (Fig. 1) and similar to previous estimates (see above), we did not test the robustness of the results of the comparative analyses to alternative trees.