Institution: NIH Library Sign In as Member / Individual

This Article Full Text Services Alert me to new issues of the journal Request Copyright Permission

Phylogenetic Analyses. Phylogenetic analyses (by using parsimony and Bayesian methods) were based on combined data matrices consisting of reproductive mode (Data Set 1), published morphological data [23 characters (refs. 1–6; R. Etheridge, personal communication; Data Set 2)], and published and previously unpublished mitochondrial DNA sequences (Data Set 3) totaling 1,026 parsimony informative characters (pic). The preferred estimate of liolaemid relationships was based on combined analyses of the molecular and morphological data. Before combining the data sets, each was analyzed separately by using equally weighted parsimony and Bayesian analysis, and these methods gave similar results (i.e., clades either were congruent or were only weakly supported if incongruent). Results from the three separately analyzed data sets [NADH dehydrogenase subunit (ND) 2, 12S ribosomal subunit, and morphology) were visually inspected for clades that were in conflict between any pair of data sets and strongly supported in each (7). Our criterion for strong support was a bootstrap value ³ 70% (8) or a Bayesian posterior probability ³ 95% (9). No strongly supported incongruence was found, and the preferred estimate of phylogeny was therefore based on the combined data tree (7). The approach for integrating data sets used in this study, in which data sets are combined but topological areas of strongly supported incongruence are considered ambiguously resolved, may be highly advantageous relative to combined, consensus, or conditional combination approaches (7). Combined parsimony analyses used equal weighting of all characters, but one meristic character (character 7; number of precloacal pores) was gap-weighted (10) and scaled (between-character scaling; ref. 11) to maintain equivalence among all morphological and molecular characters.

The models for each molecular data set in the combined Bayesian analysis were initially chosen based on likelihood ratio tests of the separate data sets . Best-fitting models (i.e., GTR + I + G for ND2 and Trn + I + G for 12S [effectively equivalent to GTR + I + G ]) were selected by using MODELTEST 3.06 (12). Bayesian analysis of morphological data assumed a likelihood model (Mk; ref. 13) in which gains and losses of a given character state were equally likely, and differences in rates of change among characters were described by using a gamma distribution. We also performed Bayesian analysis of the combined data by using several alternative modeling strategies (Table 1) and compared the goodness-of-fit of these models to the data by using Bayes factors (14). These analyses support the combination of models selected by using likelihood ratio tests as the best strategy (Table 2). Parsimony and Bayesian analyses of the combined data also gave similar results.

For all Bayesian analyses, we used four chains and uniform priors (i.e., parameter values for each model were not specified a priori). Plots of likelihood values over time were examined for stationarity, and trees generated before achieving stationarity were discarded as burn-in. Exploratory Bayesian analyses (comparing model fit) used 1.0 × 10⁶ generations (sampled every 100 generations), whereas the final combined analysis used two replicate searches each with 2.0 × 10⁶ generations. Almost all analyses reached stationarity before 100,000 generations. Parsimony analyses used heuristic searches with 50 random taxon-addition sequence replicates, and bootstrap analyses with 200 pseudoreplicates with five random taxon-addition sequence replicates each.

Some simulation studies have suggested that Bayesian posterior probability values may be inappropriately high in some cases (9, 15, 16). However, most of the clades in our analysis that are strongly supported by Bayesian analysis are also well supported by parsimony bootstrapping (see Fig. 1), suggesting that this issue is not a serious problem in our study.

Origins of Herbivory in Liolaemids. For parsimony reconstructions, each species was assigned one of three character states for diet based on categories assigned in Data Set 1. Evolution of diet was reconstructed as an ordered (insectivore ® omnivore ® herbivore) three-state character by using parsimony with MACCLADE 4.0 (17). We estimated the number of origins of herbivory in liolaemids as the number of transitions from omnivory or insectivory to herbivory. Diet evolution was also analyzed as a continuous character in a likelihood framework by using linear generalized least squares (18) (GLS-linear) with COMPARE 4.5 (http://compare.bio.indiana.edu). We assumed equal branch lengths (1.0) for this analysis because no single molecular data set was available for all of the taxa included. We reconstructed the proportion of plant matter in the diet at each node, classified ancestors as herbivores, omnivores, or insectivores (by following the diet categories defined in the text), and tallied the number of origins of herbivory.

Supertree Analysis. Our phylogenetic analysis included 87 of the nearly 170 currently recognized species of Liolaemidae. To estimate the number of origins of herbivory for virtually all Liolaemidae, we obtained data on diet from 161 taxa of Liolaemidae (also from previous studies and from dissections of museum specimens as described in Data Set 1) and constructed a phylogenetic supertree for these species. Our Bayesian phylogeny based on combined data (Fig. 1) was used as the backbone for this supertree. Taxa that had previously been assigned to phylogenetically defined species groups (4) were added to those taxonomic groups that were supported as monophyletic in the combined-data tree. Because the placement of these additional taxa was not tested by our analyses, relationships within these groups were collapsed into a polytomy when these species were added to the supertree. Many subspecies of Liolaemus may prove to warrant specific recognition (some also have different diets), so all subspecies were initially treated as separate taxa in this analysis. In the absence of contradictory phylogenetic information, these taxa were placed in a clade with conspecific subspecies. Because a recent morphology-based phylogenetic analysis of the chiliensis group (19) included many species not sampled by us, and because many groupings from that study are not concordant with our findings, we treated relationships within the chiliensis group as largely unresolved. MACCLADE (17) was used to create 1,000 trees for which polytomies were resolved randomly (assuming all resolutions are equally likely) to estimate the mean number of origins of herbivory among these trees. Results of these analyses are based on the number of origins of herbivory among all equally parsimonious reconstructions for a given tree and then averaged among the 1,000 trees that were evaluated.

Our decision to treat subspecies as separate units in the analysis had little impact on the results. We performed an analysis in which conspecific subspecies with identical diets were deleted, which is analytically equivalent to treating polytypic species that varied in diet as a single polymorphic terminal taxon. This analysis gave results that were almost identical to those from the first analysis (mean number of origins of herbivory among 1,000 trees is 18.5 before deleting identical subspecies, and is 18.3 after deleting them).

Origins of Herbivory in Nonliolaemid Lizards. The number of independent origins of herbivory in nonliolaemid squamates was reconstructed by using MACCLADE (17) based on previous studies (20, 21). Monophyly and relationships among squamate families followed recent revisions and summaries (6, 22). Family (number of independent origins of herbivory, herbivorous taxa): Iguanidae (sensu ref. 6) (1, all genera); Agamidae (2, Hydrosaurus and Uromastyx); Lacertidae (1, Gallotia simonyi); Scincidae (4, Corucia, Egernia, Macroscincus, and Tiliqua); Gerrhosauridae (1, Angolosaurus); and Teiidae (1–2, Cnemidophorus arubensis, C. murinus, and C. sp.). Note that not all species within each genus are necessarily herbivorous. Those genera lacking reference to specific species contain a single origin of herbivory.

Body sizes. Maximum body sizes of herbivorous lizards were assembled from previous studies: Iguanidae (23) except Iguana (24), Hydrosaurus (25), Uromastyx (26), Gallotia (27), Australian Scincidae (28), Corucia (29), Macroscincus (30), Angolosaurus (31), Cnemidophorus (32–35), and Liolaemidae (Data Set 1).

Distribution. Latitudinal and elevational ranges of each species were taken from Data Set 1.

Body temperatures. Body temperatures used in the analyses are listed in Data Set 1.