Horned lizard (Phrynosoma) phylogeny inferred from mitochondrial genes and morphological characters: understanding conflicts using multiple approaches

Molecular Phylogenetics and Evolution xxx (2004) xxx xxx MOLECULAR PHYLOGENETICS AND EVOLUTION www.elsevier.com/locate/ympev Horned lizard (Phrynosoma) phylogeny inferred from mitochondrial genes and morphological characters: understanding conflicts using multiple approaches Wendy L. Hodges a, * and Kelly R. Zamudio b a Section of Integrative Biology, University of Texas at Austin, USA b Department of Ecology and Evolutionary Biology, Cornell University, USA Received 15 April 2003; revised 31 October 2003 Abstract The genus Phrynosoma includes 13 species of North American lizards characterized by unique and highly derived morphologies and ecologies. Understanding interspecific relationships within this genus is essential for testing hypotheses about character evolution in this group. We analyzed mitochondrial ND4 and cytochrome b gene sequence data from all species of Phrynosoma in conjunction with a previously published dataset including 12S and 16S rrna gene sequences and morphological characters. We used multiple phylogenetic methods and diagnostic tests for data combinability and taxonomic congruence to investigate the data in separate and combined analyses. Separate data partitions resulted in several well-supported lineages, but taxonomic congruence was lacking between topologies from separate and combined analyses. Partitioned Bremer support analyses also reveals conflict between data partitions in certain tree regions. When taxa associated with well-supported clades were removed from analyses, phylogenetic signal was lost. Combined, our results initially suggest conflict between data partitions, but further tests show the data are only appropriate for phylogenetic reconstruction of those parts of the topology that were well resolved. Nonetheless, our data analyses reveal five well-supported clades: (1) Phrynosoma ditmarsi and Phrynosoma hernandesi, (2) P. ditmarsi, P. hernandesi, and Phrynosoma douglasii, (3) P. ditmarsi, P. hernandesi, P. douglasii, and Phrynosoma orbiculare, (4) Phrynosoma mcallii and Phrynosoma platyrhinos, and (5) Phrynosoma braconnieri and Phrynosoma taurus. Ó 2003 Elsevier Inc. All rights reserved. Keywords: Mitochondrial genes; Cytochrome b; ND4; Morphology; Data partitions; Phylogeny; Squamate; Phrynosomatidae 1. Introduction Horned lizards (genus Phrynosoma) form a monophyletic group of 13 species most closely related to sand lizards within the family Phrynosomatidae (Etheridge and de Queiroz, 1988; Frost and Etheridge, 1989; Reeder, 1995; Reeder and Wiens, 1996). Phrynosoma has a wide geographic distribution; species occur in a variety of habitats ranging from southwestern Canada to Guatemala. Horned lizards have been of substantial interest to herpetologists because they exhibit extreme morphological differentiation and because they display * Corresponding author. Present address: Department of Biology, University of California, Riverside, CA 92521, USA. Fax: 1-909-787-4286. E-mail address: wendyh@ucr.edu (W.L. Hodges). variation in ecological traits; consequently, Phrynosoma ecology and population biology has been widely studied (Heath, 1964; Howard, 1974; Montanucci, 1989; Pianka and Parker, 1975). More recently, understanding interspecific relationships within this group has become essential for testing evolutionary hypotheses about characteristics such as reproductive mode (Hodges, 2002; Zamudio and Parra-Olea, 2000) and defensive blood squirting (Sherbrooke and Middendorf, 2001). Four hypotheses have been proposed for interspecific relationships within Phrynosoma: Reeve (1952) and Presch (1969) examined the distribution of shared osteological characteristics between species; Montanucci (1987) published the first cladistic analysis of the genus based on skeletal and external morphology; Reeder and Montanucci (2001) combined genetic data from mitochondrial 12S and 16S rrna gene sequences with the 1055-7903/$ - see front matter Ó 2003 Elsevier Inc. All rights reserved. doi:10.1016/j.ympev.2003.11.005

2 W.L. Hodges, K.R. Zamudio / Molecular Phylogenetics and Evolution xxx (2004) xxx xxx morphological dataset, although two taxa (P. braconnieri and P. douglasii sensu stricto) were omitted from their analysis. Zamudio et al. (1997) divided P. douglasii sensu lato into two species (P. douglasii sensu stricto and Phrynosoma hernandesi) based on molecular evidence and supported by morphological features. Reeder and Montanucci (2001) included only P. hernandesi in their study. While Reeder and MontanucciÕs (2001) study is the most complete, generic level, systematic study to date, it does not include all species currently recognized in the genus. Their combined analysis resolved only three well-supported groups in their best estimate of phylogeny. Here, we expand on this study by incorporating the two remaining taxa, including genetic data from two additional mitochondrial genes, and using congruence tests and multiple phylogenetic methods to infer the best topology for all species of Phrynosoma. Considerable debate exists about how to analyze multiple data sets; specifically, whether all data should be combined regardless of differences in data structure, whether data partitions should be analyzed separately, or whether only those data sets that are congruent or compatible should be combined in a single analysis (Bull et al., 1993; Chippindale and Wiens, 1994; de Queiroz et al., 1995; Huelsenbeck et al., 1996; Miyamoto and Fitch, 1995). Rather than choosing one of these approaches arbitrarily, we adopted a more pragmatic method of applying diagnostic tests to characterize patterns of incongruence in the data, investigating the nature of the conflict, and making decisions about the best method for phylogenetic inference. 2. Materials and methods 2.1. Specimen collection We obtained all DNA extracts included in the previous molecular assessment of relationships in this genus (Reeder and Montanucci, 2001) and included all those individuals whenever possible. We supplemented these original samples with specimens and tissue samples for all Mexican Phrynosoma species collected in 1998 1999, a single P. mcallii specimen collected in 1994, and three P. douglasii tissues samples collected in 1990 1993. Multiple specimens for several species were used in sequencing, but after initial analyses showed low intraspecific polymorphism and consistent conspecific clades, duplicates were eliminated and a single specimen for each species was used for phylogenetic reconstruction. When we had to choose among samples, the sequences from the Reeder and Montanucci (2001) study were given precedence. We obtained the 12S and 16S rrna gene sequences for all taxa except P. douglasii (sensu stricto) and P. braconnieri from GenBank (Appendix A), and the morphological data matrix was downloaded from the website published in Reeder and Montanucci (2001). The morphological data do not distinguish characters assigned to any of the individual outgroup taxa; morphological characters for the outgroup taxa were combined into a single taxon referred to as the ancestor (Reeder, 1995, personal communication; Reeder and Montanucci, 2001). Also, the morphological data do not distinguish the two taxa, P. douglasii or P. hernandesi; these data represent P. douglasii sensu lato because characters from specimens representing both P. douglasii and P. hernandesi were scored and the data pooled in the data matrix (Montanucci, 1987 and personal communication). 2.2. DNA extraction, amplification, sequencing, and sequence alignment Genomic DNA was extracted using standard Chelex and phenol chloroform extraction methods from all tissue samples (Hillis et al., 1996; Walsh et al., 1991). Primers and amplification conditions for the cytochrome b gene sequence and part of trna-thr gene sequence followed those in Trepanier and Murphy (2001); those used to amplify the 3 0 end of the ND4 gene, adjacent trna-his, and part of trna-ser followed Zamudio et al. (1997). Polymerase chain reactions were carried out in 25 ll volumes and included: 1 8 ll DNA (at a concentration of 10 100 ng/ll), 2.5 ll 10 reaction buffer, 1.25 ll of each primer (10 mm concentration), 0.5 ll of dntps (at a concentration of 40 mm), 0.125 ll Taq polymerase (5 U/ll), and the remaining volume in sterile water. Thermocycling conditions were as follows: initial denaturation at 94 C for 5 min, 30 cycles of denaturation at 94 C for 1 min, annealing at 45 C for 45 s, and extension at 72 C for 1 min, followed by a final extension at 72 C for 5 min. Amplified products were confirmed by agarose electrophoresis and ethidium bromide staining. Cytochrome b PCR samples were purified from a 1.5% agarose gel using a QIA-quick gel extraction kit (QIAGEN) according to manufacturerõs instructions before terminator sequencing reactions. Amplicons of ND4 fragments were used directly in cycle sequencing. Terminator cycle sequencing reactions were carried out in 10 ll total volume including 2 ll Big Dye terminator mix (Applied Biosystems), 1 ll primer, 1 ll 5 reaction buffer, 1 6 ll PCR products and enough sterile, double distilled water to bring reactions to 10 ll. Excess dyes and non-incorporated primers were removed from cycle sequencing products using Centri-Sep columns (Princeton Separations) before labeled fragments were electrophoresed on an ABI Prism 377 DNA Sequencer. Cytochrome b sequences were aligned to sequences for Alligator mississippiensis (Accession No.: NC_001922), Chrysemys picta (Accession No.:

W.L. Hodges, K.R. Zamudio / Molecular Phylogenetics and Evolution xxx (2004) xxx xxx 3 NC_002073), and published Uma sequences (Trepanier and Murphy, 2001). ND4-tRNA fragments (herein referred to as ND4) were aligned against published sequences in Arevalo et al. (1994) using Se-Al (version 2.0a6, Rambaut, 2001). Protein coding regions of our two fragments were translated and checked for nonsense codon sequences. 12S and 16S rrna gene sequences downloaded from GenBank were aligned using Clustal W (Thompson et al., 1994) under varying gap costs as described in Reeder and Montanucci (2001). Regions that aligned differently under different gap costs were considered of ambiguous homology and therefore excluded from phylogenetic analyses (Gatesy et al., 1993). 2.3. Combinability and phylogenetic analyses We tested for the presence of phylogenetic signal in each partition and the combined data set using the skewness statistic (g 1 ) from 10 6 random trees in PAUP* (version 4.0b8a, Swofford, 2001) and compared this metric to tables in Hillis and Huelsenbeck (1992). The skewness statistic was used to determine whether our data had a higher probability of finding the correct tree compared to random data (Hillis, 1991; Hillis and Huelsenbeck, 1992; Huelsenbeck, 1991). We used the partition homogeneity test in PAUP* to test whether different characters could be combined according to type (nucleotide versus morphology), gene (12S, 16S, cytochrome b, and ND4), and codon position (first, second, and third) for protein coding regions. We used outgroup comparisons for phylogenetic analyses (Watrous and Wheeler, 1981). Our outgroup taxa included Callisaurus draconoides, Cophosaurus texanus, Holbrookia maculata, Sceloporus merriami, Urosaurus ornatus, and Uta stansburiana (Etheridge and de Queiroz, 1988; Frost and Etheridge, 1989; Reeder, 1995; Reeder and Wiens, 1996). Reeder and Montanucci (2001) do not report morphological characters for these outgroup taxa; rather, they used a computed ancestor to root their analyses. For consistency, and to permit combinability of data sets, we also followed this procedure and reconstructed a hypothetical ancestor in MacClade (Version 3.08a, Maddison and Maddison, 1992) to root our trees. However, outgroup choice or method had no effect on the analyses (Hodges, 2002). We performed weighted and unweighted maximum parsimony (MP) and maximum likelihood (ML) analyses of the molecular data with PAUP* using the heuristic search option with 1000 random addition sequences. Only MP analyses were performed when morphological characters were analyzed. Methods used to calculate weights and actual values used in weighted MP analyses are available in Hodges (2002). Modeltest (version 3.06, Posada and Crandall, 1998) was used to determine the most appropriate substitution model for ML analyses for each individual genetic data partition and for all genetic data combined. Modeltest chose the TrN + I + C model of substitution as the best model for all genetic data sets, and the following parameters were used in ML analyses: unequal base frequencies with rate matrices for each base in each data partition, among site rate variation with a gamma distribution with alpha (shape parameter) calculated for each gene, and the proportion of invariable sites was also calculated. For the ML heuristic searches, we treated gaps as missing characters, and stipulated TBR branch swapping. For MP and weighted MP runs, initial upper bounds were computed by stepwise addition of taxa, and gaps were also treated as missing characters, TBR branch swapping, ÔMultreesÕ option, and both ACCTRAN and DELTRAN character-state optimizations were in effect. Multiple state taxa were interpreted as uncertain and branches were collapsed if the maximum length was zero. We assessed clade support using non-parametric bootstrap values and decay indices (Bremer, 1994; Felsenstein, 1985). In MP, we used 100 bootstrap replicates with 1000 random addition sequences; for likelihood we used 100 bootstrap replicates each with 10 1000 random additions. Decay indices (Bremer support) and partitioned decay indices (partitioned Bremer support) were calculated using TreeRot (version 2, Sorenson, 1999). Decay and partitioned decay indices were used to assess relative contribution of different data partitions for a given clade on combined data phylogenies (Baker et al., 1998; Bremer, 1994) and to explore potential conflicts between data sets. 2.4. Taxonomic congruence We used two methods to assess taxonomic congruence (sensu Mickevich, 1978, and defined in de Queiroz et al., 1995) between separately analyzed data sets. For each dataset we estimated strict consensus trees to assess congruence between MP and ML analyses and to assess congruence between data sets. We performed another test for taxonomic congruence using Monte Carlo (parametric bootstrap) simulations on alternate topologies (Goldman et al., 2000). Topologies from each data set and different analyses were designated as null hypotheses and a test statistic was calculated based on differences between tree scores from the topology produced in the original MP analysis of a data set and the null topology used to constrain MP analysis with the same data set. One hundred data sets were generated by Monte Carlo simulation in Seq-Gen (version 1.2.5, Rambaut and Grassly, 1997). Model parameters used to simulate data sets were estimated by PAUP* under a GTR + I + C model of evolution, the model chosen from nested likelihood ratio tests. Using MP

4 W.L. Hodges, K.R. Zamudio / Molecular Phylogenetics and Evolution xxx (2004) xxx xxx optimization in PAUP*, tree length scores were generated from the simulated data and used to create a null distribution of differences in scores between topologies. For example, using the cytochrome b data, the shortest tree was found under MP criteria and the topology from ML analysis of the same data was used as a null hypothesis. Other topological tests included trees produced in MP and ML analyses of ND4 data and 12S/16S rrna, MP analyses of morphological data, MP and ML analyses of all genetic data combined, and MP analysis of all data (genes and morphology). Similar tests were performed using the ND4 data set and all the molecular data combined. 3. Results We generated 2542 morphological and nucleotide characters for analyses: 32 morphological characters and 2510 nucleotide characters including 1044 nucleotides from cytochrome b, 753 nucleotides from ND4 and associated trnas, and 713 nucleotides from 12S and 16S rrna. Fifty-nine nucleotide characters were excluded because of ambiguous alignment. Outgroup analyses using the molecular data only included a total of 976 variable sites and 719 informative sites. Analyses using all molecular and morphological data with a reconstructed ancestor included 858 variable sites and 590 informative sites. All genes displayed unequal base frequencies with low guanine content and high adenine or thymine content. Cytochrome b had the highest ratio of informative to variable (I/V) characters (0.705 0.773) followed by ND4 (0.69 0.765). The I/V ratio for 12S/ 16S rrna was considerably lower (0.518 0.56). Likewise, the overall number of variable sites was much lower in the 12S/16S rrna data set than either the cytochrome b or ND4 data sets. 3.1. Data combinability and phylogenetic analyses All data sets (including weighted partitions) contained significant ðp < 0:01Þ phylogenetic signal according to skewness statistics, indicating they were more structured than random noise and were appropriate for phylogenetic analysis. Results from partition homogeneity tests were mixed. These tests supported combining all the molecular data partitions together at the genetic level ðp ¼ 0:44Þ and at the codon level ðp ¼ 0:54Þ regardless of whether weights were applied. However, the tests did not support combining genes and morphology ðp < 0:01Þ. Therefore, the genetic data were analyzed separately and jointly. Morphological data were analyzed separately except for one analysis where it was combined with all molecular data to determine if partitioned Bremer values could shed light on the nature of potential conflict between data sets. The model of evolution chosen by Modeltest (TrN + I + CÞ for ML analyses indicated all genetic data exhibited among site rate variation. Transitions in third position codons exhibited the largest variation from other positions in both genes. Estimated standard rates of change for third positions were about an order of magnitude greater than second position rates (see Hodges, 2002). First positions also showed a tendency towards high rates of change, but not to the same degree as third positions (only about 3 4 times the rate of second positions). Rates of change were only slightly higher in cytochrome b than ND4. Topologies representing best estimates for phylogeny using ML optimization of mitochondrial data sets independently and combined are shown in Fig. 1. In all analyses, several relationships are present: P. mcallii and P. platyrhinos are well supported as sister taxa, likewise, Phrynosoma taurus and P. braconnieri form a clade (except in the 12S/16S analyses because P. braconnieri was not sequenced for these genes). The short-horned lizards, P. ditmarsi, P. hernandesi, P. douglasii, and P. orbiculare, form a monophyletic group. The remaining taxa (Phrynosoma cornutum, Phrynosoma coronatum, Phrynosoma asio, and Phrynosoma solare) moved around in the topologies of each analysis and showed no consistent affinities for placement in different trees. The combined DNA analysis had seven well-supported clades (Fig. 1: assuming bootstrap values of 70% or greater represented accurate clades; Hillis and Bull, 1993). Six of these clades had bootstrap support greater than 90% and supported a monophyletic short-horned lizard group with Phrynosoma modestum as the sister taxon to this clade, P. mcallii and P. platyrhinos are sister taxa, and P. braconnieri and P. taurus are sister taxa. Topologies resolved in MP analyses of the mitochondrial genes were very similar to ML topologies. The short-horned lizards (P. ditmarsi, P. hernandesi, P. douglasii, and P. orbiculare) are monophyletic in all but the ND4 analyses where they are rendered paraphyletic by P. modestum (albeit with low bootstrap support of 30 40%). Differential weighting had no effect on this result. P. modestum was often associated with the shorthorned group either as the sister taxon or paired with another taxon as part of the sister group. In all analyses, P. platyrhinos and P. mcallii were sister taxa as well as P. taurus and P. braconnieri, and the remaining taxa were again differentially placed throughout the trees. ML and MP analyses resolved the same well- and moderately supported clades. The primary difference between weighted MP, unweighted MP, and ML analyses was slightly reduced levels of bootstrap support for some clades. In general, the method of phylogenetic reconstruction did not change resolution of well-supported clades for a given data set. Likewise, although different data sets yielded different topologies overall, the

W.L. Hodges, K.R. Zamudio / Molecular Phylogenetics and Evolution xxx (2004) xxx xxx 5 Fig. 1. Maximum likelihood phylograms for mitochondrial genes. Phylograms resulting from ML analysis of each mitochondrial data set and the mitochondrial genes combined. Numbers near branches represent bootstrap support values. well-supported clades consistently appeared regardless of phylogenetic reconstruction method. Topologies derived from morphological data were quite different from the topologies based on genetic data (Fig. 2). Only one clade (P. taurus and P. braconnieri) strongly (98% bootstrap) supported in the morphological analysis was also strongly supported by the genetic data. The only other strongly (95 97% bootstrap) supported clade in the morphological analysis was P. hernandesi and P. orbiculare. In genetic analyses, these two taxa formed part of a monophyletic group with P. ditmarsi and P. douglasii, but morphological analyses weakly (53% bootstrap) supported P. ditmarsi as the sister taxon to P. taurus and P. braconnieri instead of placing P. ditmarsi with the other short-horned lizards. Morphological analyses supported several clades that were not found in any of the genetic data sets. The two data types (genes and morphology) resolved very different relationships overall, and corroborated the partition homogeneity test results suggesting these two data sets are not compatible. When all molecular and morphological data are combined in a single MP analyses, the five well-supported clades obtained from analyzing the molecular data are recovered (Fig. 2). 3.2. Taxonomic congruence Strict consensus trees from analyses of all DNA data combined reiterate the well-supported clades in the separate molecular analyses (Fig. 3). A strict consensus tree of morphological and molecular data only retained a single resolved clade (P. braconnieri and P. taurus) and preserved no other structure in the tree. Though topological elements were consistent among analyses of different data sets, results from parametric bootstrap simulations suggested incongruence was still present. Topologies from MP and ML analyses of the 12S/16S rrna data were rejected by both ND4 and cytochrome

6 W.L. Hodges, K.R. Zamudio / Molecular Phylogenetics and Evolution xxx (2004) xxx xxx Fig. 2. Maximum parsimony phylogenies for morphological characters and all data combined. More than one best tree was found and a consensus of those topologies is shown rather than each tree. Bootstrap support is shown above each branch with decay index (Bremer support) below. Bootstrap values and Bremer indices were assumed to confer similar levels of support for a given clade, assuming bootstrap values of 70% or greater represented accurate clades with respect to the true tree (Hillis and Bull, 1993), and Bremer values greater than 5 reflected strong evidence of branch stability (Frost et al., 2001; Flores-Villela et al., 2000; Pellegrino et al., 2001). Fig. 3. Strict consensus tree of all analyses for all molecular data combined. Strict consensus trees of MP and ML analyses for the combined analyses of genetic data are shown. Clades are identified by a letter corresponding to the table below the tree, which shows one example of partitioned Bremer support values for each data set and bootstrap values calculated for the best tree from MP analyses of all the genetic data. b data sets ðp < 0:02Þ. Topologies from analyses of the morphological data also were rejected by ND4 and cytochrome b ðp < 0:01Þ. Results comparing ND4 and cytochrome b, however, were less straightforward and showed almost reverse patterns in which they rejected alternative hypotheses with respect to each other, but both data sets did show incongruence with the 12S/16S rrna and morphology based topologies. When all molecular data were combined, the ML topology from cytochrome b, ML and MP topologies from 12S/16S rrna, and the morphological data topologies were all rejected in favor of the combined molecular data topology from MP analysis (p < 0:01). Topologies from ML analysis of the combined molecular data, ML and MP analyses of ND4 data, MP analysis of cytochrome b, and all data combined could not be rejected in favor of the combined molecular data topology from MP analysis ðp > 0:07Þ. Parametric bootstrap simulation results suggested that different molecular data sets may be incompatible, but these results were ambiguous. Previous tests for data combinability (partition homogeneity tests) supported combining the molecular data. Topologies from individual analyses of each molecular data suggested that regions of well-supported clades existed and were consistent among the different data sets. However, tests of taxonomic congruence were unable to fully support the combinability of all the molecular data, especially combining the 12S/16S rrna with data from protein coding genes. Combining the morphological data with the molecular data was not supported by any method. Partitioned Bremer support indices were added to the consensus tree of all phylogenetic analyses conducted on the combined mitochondrial data (Fig. 3). A negative value indicates the data partition did not support a clade. The partitioned Bremer values showed that the data set conflicting most often with the consensus topology was 16S rrna. Bremer values of 0 were common for the 12S rrna indicating this data partition did not support or conflict with the topology shown. Bremer support indices failed to show support by the

W.L. Hodges, K.R. Zamudio / Molecular Phylogenetics and Evolution xxx (2004) xxx xxx 7 Fig. 4. Two best topologies using all available data. Two best phylogenies based on MP analyses of all combined data and partitioned Bremer support and bootstrap support. Clade identification is given on the top of each clade and total Bremer support indices are given below. Clade identification corresponds to tables on the right of each phylogeny containing the partitioned Bremer support for each data partition. Bootstrap values are shown at the bottom of each table. ND4 data set for the P. ditmarsi P. hernandesi clade, which the data set did support when analyzed independently. Cytochrome b (and 16S rrna) data conflicted with the placement of P. modestum in the short-horned lizard clade. Bremer support values were also calculated for the two shortest trees derived from the MP analyses of all data combined (Fig. 4). This analysis showed that only a single clade, P. mcallii P. platyrhinos, is positively supported by all data partitions and has high (100%) bootstrap support. One other clade, P. taurus P. braconnieri was supported by all but the 12S/16S rrna data set (recall 12S/16S rrna data were unavailable for P. braconnieri) and also received high (100%) bootstrap support. Though MP analysis identified these topologies as the shortest trees, Bremer support suggested several clades were not supported by any data. The morphological data showed only positive (rather than negative) support for clades, and in five cases the data simply showed no positive or negative support for a clade. Data sets showing the most negative Bremer values in the combined analysis were the 12S and ND4 data sets. Partitioned Bremer indices in the combined data analysis were similar to the results from taxonomic congruence tests and analysis of separate data sets in that they supported well-resolved branches found in most of the other analyses and revealed points of conflict in poorly supported branches. We pruned taxa from strongly supported regions of the tree and recalculated the skewness statistic after each removal. Removing four or five taxa representing each of the strongly supported regions reduced the phylogenetic signal in the data to the point they were no more structured than random noise. Removal of fewer taxa or taxa from the poorly supported regions of the tree did not have this effect. 4. Discussion Analyses of multiple data sets can be problematic if data are incongruent (Bull et al., 1993; Miyamoto and Fitch, 1995). Initial tests for character congruence using the partition homogeneity test suggested that all molecular data were combinable, but morphological data were not combinable with the molecular data. Independent analyses of each data set and an analysis of the combined molecular data set qualitatively supported this result; well-supported clades in each molecular partition were present across multiple analyses and results of morphological analyses supported different

8 W.L. Hodges, K.R. Zamudio / Molecular Phylogenetics and Evolution xxx (2004) xxx xxx topologies compared to the molecular data. Combined molecular data analyses confirmed that well-supported clades from independent molecular analyses were supported by the combined molecular data. However, a strict consensus tree from all separate analyses left only one resolved clade, containing P. braconnieri and P. taurus. Quantitative tests for taxonomic congruence revealed possible conflict between these different data sets. Parametric bootstrap simulations indicated that some molecular data partitions were incongruent and that the morphological data were incongruent with all molecular data; however, some of these results were ambiguous because no clear patterns of rejection between the protein coding genes were found, and, these data conversely rejected alternative hypotheses. A conservative approach in light of these ambiguities would be to analyze all data partitions separately. However, after we combined all the data and looked at partitioned Bremer support indices, we found the morphological data positively contributed to well-supported clades in the combined analysis (Fig. 4) that were not supported in the independent analysis of morphological data alone. Based on this positive support in the combined data analyses, a surprising result of our separate analyses was the discordance between morphological and molecular data in our study. One potential problem in the current morphological data is that P. douglasii and P. hernandesi are combined into a single taxon, P. hernandesi. This has unknown effects on the phylogenetic structure of the morphological data, since molecular data indicated that these two taxa are not sister taxa and are rendered paraphyletic by P. ditmarsi. Zamudio et al. (1997) analyzed data from 38 populations of P. douglasii (sensu lato) showing that the taxa are closely related to one another and only in some of their analyses was P. douglasii rendered paraphyletic. Presch (1969) considered P. ditmarsi to be a derivative of P. douglasii. In morphological analyses shown in this study and by Montanucci (1987), P. ditmarsi is not included in the P. hernandesi P. orbiculare clade, rather it falls as the sister taxon to P. taurus and P. braconnieri. In all molecular analyses, P. taurus and P. braconnieri form a unique and wellsupported, independent clade distant from P. ditmarsi. Conflict between molecular and morphological data sets could arise due to the presence of convergence in morphological data, different evolutionary rates between characters, introgressive hybridization, lineage sorting, or non-independence of characters (de Queiroz et al., 1995; Joy and Conn, 2001, Sota and Vogler, 2001). In the separate morphological analysis, the placement of P. ditmarsi with P. braconnieri and P. taurus is supported by only two changes, only one of which is unique (a reduction in number of caudal vertebrae). Despite taxonomic and character-based conflicts, partitioned Bremer support values in the combined analyses showed morphological characters were not in conflict with the molecular data except in one clade. When the morphological traits are mapped on the mitochondrial DNA tree, the clade containing four species of short-horned lizards are supported by seven changes, two that are unique (supralabial margin shape and orientation of the posterior jugal). All other changes except one are features of the skull suggesting non-independence or homoplasy between these characters. Combined data analysis showed that despite lower numbers of informative characters and apparent incongruence with molecular data in separate analyses, morphological data contributed positive support in well-supported clades. We would not have seen this result if we had only followed the outcome of congruency tests and not considered how the characters contributed to a topology representing all the data. In the combined data topology, the molecular data partitions showed more points of conflict with each other than they did with the morphological data. The conflict between the molecular partitions was apparently a result of each data setõs contribution to phylogenetic signal. The 12S/16S rrna offer the least amount of signal to the combined tree; with only 58 informative sites and the highest proportion of invariable sites; therefore, these data provided little phylogenetic information and may be misleading especially in poorly supported tree regions (Nei et al., 1998; Yang, 1998). The cytochrome b and ND4 gene sequence data contained more informative characters and were more variable. Variation was highest in third position codons where substitution rates were an order of magnitude higher than second position codons. However, if mutational rates of cytochrome b and ND4 data are too high, they can also limit resolving power in the phylogeny (Yang, 1995). Nucleotide changes contribute significant noise to the data if the number of changes between nodes is sufficiently high to randomize character states (Hillis and Huelsenbeck, 1992). In other taxa, cytochrome b is known to exhibit problems in resolving deeper relationships (Graybeal, 1993; Halanych and Robinson, 1999; Meyer and Wilson, 1990). The ND4 dataset appeared to follow the same pattern in this study. A combination of low number of variable and informative sites in the 12S/16S rrna and noisy variation in the cytochrome b and ND4 yielded a partially resolved phylogeny. As expected, resolved clades in our topology were found at the tips of the phylogeny among closely related species, while unresolved portions of the phylogenies were at basal positions. In our analyses, clades with weak support like those representing basal relationships always represent areas of uncertainty in the phylogeny. Weakly supported groups in separate analyses were not improved by any alternative weighting scheme or data combination, and strongly supported relationships in unweighted, combined analyses were present in separate analyses and

W.L. Hodges, K.R. Zamudio / Molecular Phylogenetics and Evolution xxx (2004) xxx xxx 9 analyses using different weighting criteria. Conflicts between trees produced from different data sets were present in weakly supported clades where no data set resolved relationships. These results supported including all data in analyses without differential removal or weighting characters and leaving areas in conflict as areas of phylogenetic uncertainty to be resolved by additional data (Baker et al., 2001; Wiens, 1998; Yang, 1998). The trees from unweighted MP analyses of all combined data were chosen to represent the best phylogenetic hypotheses for relationships within Phrynosoma with the understanding that basal relationships remain unsupported by currently available data (Fig. 4). These trees are chosen because the relationships resolved in these analyses by well-supported branches are present in the majority of individual dataset analyses. When all data are combined, the morphological data actually do show positive support for three out of five clades that receive high bootstrap support. The two well-supported clades remaining are simply not supported by the morphological data and these clades include P. douglasii and P. hernandesi, two taxa that the morphological data do not discriminate. Our analysis supports a monophyletic short-horned lizard clade, which includes four species of montane, viviparous species (P. douglasii, P. orbiculare, P. ditmarsi, and P. hernandesi). Likewise, a clade of the two remaining viviparous species (P. taurus and P. braconnieri) and a clade containing two oviparous species (P. mcallii and P. platyrhinos) are also present in our preferred topologies. These conclusions support Reeder and MontanucciÕs (2001) results, but our trees add resolution to the short-horned lizard group. While three of these taxa were supported in their analyses as a monophyletic group, their preferred tree provided weakly supported internal structure and placed P. ditmarsi as the basal taxon in this clade. Our analysis places P. orbiculare as the basal taxon and P. ditmarsi as a terminal taxon. This arrangement was suggested by Reeve (1952) and in part by Presch (1969), although he considered P. orbiculare to be the most basal Phrynosoma and the remaining short-horned lizards as members of a clade of northern derivatives. Our analyses also suggest that the viviparous species do not form a monophyletic group because viviparous taxa are always placed on two separate and well-supported clades (Hodges, 2002; Zamudio and Parra-Olea, 2000). The sister relationship of P. mcallii and P. platyrhinos is also present in the combined data and molecular analyses in Reeder and Montanucci (2001) and one of three alternative topologies proposed by Montanucci (1987). Neither Presch (1969) nor Reeve (1952) considered P. mcallii and P. platyrhinos to be sister taxa, but did place them in a group of closely related taxa with P. modestum. Future systematic work on Phrynosoma needs to address the basal relationships among the clades. Our study illustrates the benefit of multiple analyses on multiple data sets, with analyses based on independent partitions as well as combined. A single a priori approach limiting analyses only to total evidence or consensus analyses would cause us to overlook information about the nature of conflicts and ultimately the nature of the data. By fully exploring the data and with all available analytical techniques, it became evident that certain patterns are present in the data, yet the data were insufficient in their power to resolve the basal relationships in Phrynosoma. Although progress has been made in our analyses, additional data are needed to tease apart basal relationships within this genus. Some of these data sources could be found in additional molecular sequences (perhaps nuclear genes), fossil data, or more detailed morphological analyses. Acknowledgments Funding was provided by the University of Texas at Austin Graduate School, Department of Zoology (to W.H.), an NSF Minority Postdoctoral Fellowship (BIR9628913), and the College of Arts and Sciences at Cornell University (to K.Z.). Thanks to many people who assisted W.H. in the lab and field, including: A. Price, W. Sherbrooke, J. Treanor, A. Gluesenkamp, M. Swartz, J. Kuhn, O. Flores Villela, M. Benabib, E. Perez Ramos and family, A. Adame and family, H. Rivas Garcia, F. Vargas Santa Maria, A. Narvaez, the Navarro Cruz Family, E. Pianka, A. Nieto Montes de Oca, L. Conseco Marquez, M. Badgett, and T. Wilcox. Special thanks to T. Reeder for generously providing samples for the taxa in his study. Fieldwork in Mexico was conducted under Mexican permits DOO.750.-5300 and FAUT-0015. Appendix A. Specimens and materials examined Specimen identification for tissues and DNA with GenBank Accession numbers Species Voucher 12S rrna 16S rrna Cyt b ND4 trna-his Callisaurus draconoides LSUMZ 4811 L40437 L41441 AY141099 AY141061 AY141080 Cophosaurus texanus LSUMZ 48758 L40438 L41442 AY141100 AY141062 AY141081 Holbrookia maculata LSUMZ 48805 L40440 L41445 AY141101 AY141063 AY141082 Phrynosoma asio RRM 2499 L40446 L41452 AY141086 AY141048 AY141067

10 W.L. Hodges, K.R. Zamudio / Molecular Phylogenetics and Evolution xxx (2004) xxx xxx Appendix A (continued) Species Voucher 12S rrna 16S rrna Cyt b ND4 trna-his P. braconnieri WLH 01111 NA NA AY141098 AY141060 AY141079 P. cornutum LSUMZ 48807 L40447 L41454 AY141087 AY141049 AY141068 P. coronatum RRM 2479 AF346839 AF346846 AY141088 AY141050 AY141069 P. ditmarsi RRM 2459 AF346845 AF346852 AY141089 AY141051 AY141070 P. douglasii BJ 961/t011 NA NA AY141090 AY141052 AY141071 P. hernandesi RRM 2470 L40448 L41454 AY141091 AY141053 AY141072 P. mcallii ROM 13876/ WLH 10059 a AF346840 AF346847 AY141092 AY141054 AY141073 P. modestum LSUMZ 48831 L41455 L40449 AY141093 AY141055 AY141074 P. orbiculare RRM 2480 AF346841 AF346848 AY141094 AY141056 AY141075 P. platyrhinos ASU 15250 AF346842 AF346849 AY141095 AY141057 AY141076 P. solare ROM 15044 AF346843 AF346850 AY141096 AY141058 AY141077 P. taurus NA b AF346844 AF346851 AY141097 AY141059 AY141078 Sceloporus merriami LSUMZ 48844 L41418 L41468 AY141102 AY141064 AY141083 Urosaurus ornatus LSUMZ 48828 L41436 L41487 AY141103 AY141065 AY141084 Uta stansburiana LSUMZ 48840 L41438 L41489 AY141104 AY141066 AY141085 a P. mcallii specimen used in amplification of cytochrome b only. b See Reeder and Montanucci (2001). ASU: Appalachian State University, LSUMZ: Louisiana State University, RRM: Richard R. Montanucci, ROM: Royal Ontario Museum (Robert W. Murphy), WLH: Wendy L. Hodges. References Arevalo, E., Davis, S.K., Sites Jr., J.W., 1994. Mitochondrial DNA sequence divergence and phylogenetic relationships among eight chromosome races of the Sceloporus grammicus complex (Phrynosomatidae) in central Mexico. Syst. Biol. 43, 387 418. Baker, R.H., Wilkinson, G.S., DeSalle, R., 2001. Phylogenetic utility of different types of molecular data used to infer evolutionary relationships among stalk-eyed flies (Diopsidae). Syst. Biol. 50, 87 105. Baker, R.H., Yu, X., DeSalle, R., 1998. Assessing the relative contribution of molecular and morphological characters in simultaneous analysis trees. Mol. Phylogenet. Evol. 9, 427 436. Bremer, K., 1994. Branch support and tree stability. Cladistics 10, 295 304. Bull, J.J., Huelsenbeck, J.P., Cunningham, C.W., Swofford, D.L., Waddell, P.J., 1993. Partitioning and combining data in phylogenetic analysis. Syst. Biol. 42, 384 397. Chippindale, P.T., Wiens, J.J., 1994. Weighting, partitioning, and combining characters in phylogenetic analysis. Syst. Biol. 43, 278 287. de Queiroz, A., Donoghue, M.J., Kim, J., 1995. Separate versus combined analysis of phylogenetic evidence. Annu. Rev. Ecol. Syst. 26, 657 681. Etheridge, R., de Queiroz, K., 1988. A phylogeny of Iguanidae. In: Estes, R., Pregill, G. (Eds.), Phylogenetic Relationships of Lizard Families: Essays Commemorating Charles L. Camp. Stanford Univ. Press, Stanford, pp. 283 368. Felsenstein, J., 1985. Confidence limits on phylogeny: an approach using the bootstrap. Evolution 39, 783 789. Flores-Villela, O., Kjer, K.M., Benabib, M., Sites Jr., J.W., 2000. Multiple data sets, congruence and hypothesis testing for the phylogeny of basal groups of the lizard genus Sceloporus (Squamata, Phrynosomatidae). Syst. Biol. 49, 713 739. Frost, D.R., Etheridge, R., 1989. A phylogenetic analysis and taxonomy of Iguanian lizards (Reptilia: Squamata). Univ. Kansas Mus. Nat. Hist. Miscel. Publ. 81, 1 65. Frost, D.R., Rodriques, M.T., Grant, T., Titus, T.A., 2001. Phylogenetics of the lizards genus Tropidurus (Squamata: Tropiduridae: Tropidurinae): direct optimization, descriptive efficiency and sensitivity analysis of congruence between molecular data and morphology. Mol. Phylogenet. Evol. 21, 352 371. Gatesy, J., DeSalle, R., Wheeler, W., 1993. Alignment-ambiguous nucleotide sites and the exclusion of systematic data. Mol. Phylogenet. Evol. 2, 152 157. Goldman, N., Anderson, J.P., Rodrigo, A.G., 2000. Likelihood-based tests of topologies in phylogenetics. Syst. Biol. 49, 652 670. Graybeal, A., 1993. The phylogenetic utility of cytochrome b: lessons from bufonid frogs. Mol. Phylogenet. Evol. 2, 256 269. Halanych, K.M., Robinson, T.J., 1999. Multiple substitutions affect the phylogenetic utility of cytochrome b and 12S rdna data: examining a rapid radiation in leporid (Lagomorpha) evolution. J. Mol. Evol. 48, 369 379. Heath, J.E., 1964. Temperature regulation and diurnal activity in horned lizards. Univ. California Publ. Zool. 64, 97 136. Hillis, D.M., 1991. Discriminating between phylogenetic signal and random noise in DNA sequences. In: Miyamoto, M.M., Cracraft, J. (Eds.), Phylogenetic Analysis of DNA Sequences. Oxford Univ. Press, Oxford. Hillis, D.M., Bull, J.J., 1993. An empirical test of boot strapping as a method for assessing confidence in phylogenetic analysis. Syst. Biol. 42, 182 192. Hillis, D.M., Huelsenbeck, J.P., 1992. Signal, noise, and reliability in molecular phylogenetic analysis. J. Heredity 83, 189 195. Hillis, D.M., Moritz, C., Mable, B.K., 1996. Molecular Systematics, second ed. Sinauer Associates, Inc, Sunderland, MA. Hodges, W.L., 2002. Phrynosoma Systematics, Comparative Reproductive Ecology, and Conservation of a Texas Native. Ph.D. Dissertation. Univ. Texas Austin. Howard, C.W., 1974. Comparative reproductive ecology of horned lizards (Genus Phrynosoma) in southwestern United States and northern Mexico. J. Arizona Acad. Sci. 9, 108 116. Huelsenbeck, J.P., 1991. Tree-length distribution skewness: an indicator of phylogenetic information. Syst. Zool. 3, 257 270. Huelsenbeck, J.P., Bull, J.J., Cunningham, C.W., 1996. Combining data in phylogenetic analysis. Trends Ecol. Evol. 11, 152 158. Joy, D.A., Conn, J.E., 2001. Molecular and morphological phylogenetic analysis of an insular radiation in Pacific black flies (Simulium). Syst. Biol. 50, 18 38. Maddison, W.P., Maddison, D.R., 1992. MacClade: Analysis of Phylogeny and Character Evolution. Version 3.08a. Sinauer Associates, Sunderland, MA. Meyer, A., Wilson, A.C., 1990. Origin of tetrapods inferred from their mitochondrial DNA affiliation to lungfish. J. Mol. Evol. 31, 359 364.

W.L. Hodges, K.R. Zamudio / Molecular Phylogenetics and Evolution xxx (2004) xxx xxx 11 Miyamoto, M.M., Fitch, W.M., 1995. Testing species phylogenies and phylogenetic methods with congruence. Syst. Biol. 44, 64 76. Mickevich, M.F., 1978. Taxonomic congruence. Syst. Zool. 27, 143 158. Montanucci, R.R., 1987. A phylogenetic study of the horned lizard genus, Phrynosoma, based on skeletal and external morphology. Contrib. Sci. Nat. Hist. Mus. Los Angeles Co. 113, 1 26. Montanucci, R.R., 1989. The relationship of morphology to diet in the horned lizard genus Phrynosoma. Herpetologica 45, 208 216. Nei, M., Kumar, S., Takahashi, K., 1998. The optimization principle in phylogenetic analysis tends to give incorrect topologies when the number of nucleotides or amino acids used is small. Proc. Natl. Acad. Sci. USA 95, 12390 12397. Pellegrino, K.C.M., Rodrigues, M.T., Yonenaga-Yassuda, Y., Sites Jr., J.W., 2001. A molecular perspective on the evolution of microteiid lizards (Squamata, Gymnophthalmidae), and a new classification for the family. Biol. J. Linnean Soc. 74, 315 338. Pianka, E.R., Parker, W.S., 1975. Ecology of horned lizards: a review with special reference to Phrynosoma platyrhinos. Copeia 1975, 141 162. Posada, D., Crandall, K.A., 1998. Modeltest: testing the model of DNA substitution. Bioinformatics 14, 817 818. Presch, W., 1969. Evolutionary osteology and relationships of the horned lizards genus Phrynosoma (family Iguanidae). Copeia 1969, 250 275. Rambaut, A., 2001. Se-Al Sequence Alignment editor. Version 2.0. Univ. Oxford. Available from: <http://evolve.zoo.ox.ac.uk/software/se-al/se-al.html>. Rambaut, A., Grassly, N.C., 1997. Seq-Gen: an application for the Monte Carlo simulation of DNA sequence evolution along phylogenetic trees. Comput. Appl. Biosci. 13, 235 238. Reeder, T.W., 1995. Phylogenetic relationships among Phrynosomatid lizards as inferred from mitochondrial ribosomal DNA sequences: substitutional bias and information content of transition relative to transversion. Mol. Phylogenet. Evol. 4, 203 222. Reeder, T.W., Montanucci, R.R., 2001. Phylogenetic analysis of the horned lizards (Phrynosomatidae: Phrynosoma): evidence from mitochondrial DNA and morphology. Copeia 2001, 309 323. Reeder, T.W., Wiens, J.J., 1996. Evolution of the lizard family Phrynosomatidae as inferred from diverse types of data. Herpetol. Monog. 10, 43 84. Reeve, W.L., 1952. Taxonomy and distribution of the horned lizard genus Phrynosoma. Univ. Kansas Sci. Bull. 34, 817 960. Sherbrooke, W.C., Middendorf III, G.A., 2001. Blood-squirting variability in horned lizards (Phrynosoma). Copeia 2001, 1114 1122. Sorenson, M.D., 1999. TreeRot, version 2. Boston Univ., Boston, MA. Sota, T., Vogler, A.P., 2001. Incongruence of mitochondrial and nuclear gene trees in the carabid beetles Ohomopterus. Syst. Biol. 50, 39 59. Swofford, D.L., 2001. PAUP*. Phylogenetic Analysis Using Parsimony (*and Other Methods). Version 4.0b8a. Sinauer Associates, Sunderland, MA. Thompson, J.D., Higgins, D.G., Gibson, T.J., 1994. Clustal W: improving the sensitivity of progressive multiple sequence alignment through sequence weighting, position-specific gap penalties and weight matrix choice. Nucleic Acids Res. 22, 4673 4680. Trepanier, T.L., Murphy, R.W., 2001. The Coachella Valley fringetoed lizard (Uma inornata): genetic diversity and phylogenetic relationships of an endangered species. Mol. Phylogenet. Evol. 18, 327 334. Walsh, S.P., Metzger, D.A., Higuchi, R., 1991. Chelex 100 as a medium for simple extraction of DNA for PCR-based typing from forensic material. Biotechniques 10, 506 513. Watrous, L.E., Wheeler, Q.D., 1981. The outgroup comparison method of character analysis. Syst. Zool. 30, 1 11. Wiens, J.J., 1998. Combining data sets with differential phylogenetic histories. Syst. Biol. 47, 568 681. Yang, Z., 1995. Evaluation of several methods for estimating phylogenetic trees when substitution rates differ over nucleotide sites. J. Mol. Evol. 40, 689 697. Yang, Z., 1998. On the best evolutionary rate for phylogenetic analysis. Syst. Biol. 47, 125 133. Zamudio, K.R., Jones, K.B., Ward, R.H., 1997. Molecular systematics of the short-horned lizards: biogeography and taxonomy of a widespread species complex. Syst. Biol. 46, 284 305. Zamudio, K.R., Parra-Olea, G., 2000. Reproductive mode and female reproductive cycles of two endemic Mexican horned lizards (Phrynosoma taurus and Phrynosoma braconnieri). Copeia 2000, 222 229.