Sysf. Biol. 49(3)592-612, 2000
Phylogenetic Relationships Among the Phrynosomatid Sand Lizards Inferred from Mitochondrial DNA Sequences Generated by Heterogeneous Evolutionary Processes
JAMES WILGENBUSCH^ AND KEVIN DE QUEIROZ^ ^Laboratory of Molecular Systematics, MRC 534, MSC, Smithsonian Institution, Washington, D.C. 20560, USA; E-mail: [email protected] ^Department of Vertebrate Zoology, National Museum of Natural History, Smithsonian Institution, Washington, D.C. 20560, USA; E-mail: [email protected]
Abstract.•Nucleotide sequences of the mitochondrial protein coding cytochrome b (cyt b; 650 bp) and small-subunit 12S ribosomal RNA (~350 bp) genes were used in analyses of phylogenetic relation- ships among extant phrynosomatid sand lizards, including an examination of competing hypotheses regarding the evolution of "earlessness." Sequences were obtained from all currently recognized species of sand lizards as well as representatives of the first and second outgroups and analyzed using both parsimony and likelihood methods. The cyt b data offer strong support for relationships that correspond with relatively recent divergences and moderate to low support for relationships reflecting more ancient divergences within the clade. These data support monophyly of the "earless" taxa, the placement of Uma as the sister taxon to the other sand lizards, and monophyly of all four taxa traditionally ranked as genera. All well-supported relationships in the 12S phylogeny are com- pletely congruent with weU-supported relationships in the cyt b phylogeny; however, the 12S data alone provide very Kttle support for deeper divergences. Phylogenetic relationships within species are concordant with geography and suggest patterns of phylogeographic differentiation, including the conclusion that at least one currently recognized species (Holbrookia macúlala) actually consists of more than one species. By independently optimizing likelihood model parameters for various subsets of the data, we found that nucleotide substitution processes vary widely between genes and among the structural and functional regions or classes of sites within each gene. Therefore, we compared competing phylogenetic hypotheses, using parameter estimates specific to those subsets, analyzing the subsets separately and in various combinations. The hypothesis supported by the cyt b data was favored over rival hypotheses in all but one of the five comparisons made with the entire data set, including the set of partitions that best explained the data, although we were unable to confidently reject (P < 0.05) alternative hypotheses. Our results highlight the importance of optimizing mod- els and parameter estimates for different genes or parts thereof•a strategy that takes advantages of the strengths of both combining and partitioning data. [Combining data; cytochrome b; maximum likelihood; mitochondrial DNA; phrynosomatidae; phylogeny; 12S RNA; sand lizards.]
The phrynosomatid sand lizards form a with which it shares the derived character clade of 8-10 currently recognized species of a concealed tympanic membrane, or of distributed in arid and semiarid regions of Callisaurus (Fig. 1, topology III and IV), to western North America (de Queiroz, 1989). which it bears a greater overall resemblance. This group has interested systematic biol- The second disagreement forms the basis of ogists because of a continuing controversy a related controversy about whether the con- about the phylogenetic relationships among cealed tympanic membrane ("earless" condi- four taxa traditionally recognized as gen- tion) of Cophosaurus and Holbrookia has been era: Callisaurus (zebra-tailed lizards, one inherited from a common ancestor or is the currently recognized species), Cophosaurus result of convergent or parallel evolution, (greater earless lizards, one species). Hoi- Most recent studies (e.g., Etheridge and de brookia (lesser earless lizards, three species), Queiroz, 1988; de Queiroz, 1989,1992; Porter and Uma (fringe-toed lizards, three to five et al., 1994; Changchien, 1996; Reeder and species). Among the four alternative hy- Wiens, 1996) have supported the early diver- potheses proposed for the relationships gence of Uma and the sister group relation- among these taxa (Fig. 1), the main differ- ship of Cophosaurus and Holbrookia (Fig. 1, enees concern first, whether the first taxon to topology I). Moreover, a review of all the diverge was Uma (Fig. 1, topologies I and III) data that had been presented up to 1989 (de or Holbrookia (Fig. 1, topology IV), and sec- Queiroz, 1989) revealed first, that most of ond, whether Cophosaurus is the sister group the similarties between Cophosaurus and Cal- oí Holbrookia (Fig. 1, topologies I and II), lisaurus are retained ancestral features, and
592 2000 WILGENBUSCH AND DE QUEIROZ•SAND LIZARD PHYLOGENY 593
.0* v^ 0, 4^ -J^
/ 0^00^°/ Topology I Topology II Topology Topology IV FIGURE 1. Alternative hypotheses of higher-level relationships among the sand Kzards. Topology I (Savage, 1958; Cox and Tanner, 1977; Etheridge and de Queiroz, 1988; de Queiroz, 1989, 1992; Reeder and Wiens, 1996; Changchien, 1996); topology II (Mittleman, 1942; Smith, 1946); topology III (Norris, 1958; Earle, 1961,1962); topology IV (Axtell, 1958; Clarke, 1965; Adest, 1978). second, that there were no known derived In contrast, likelihood has an explicit ba- characters shared by those two taxa that are sis for assessing the ability of different mod- not also shared by either Uma, Holbrookia, or els to explain the data (see Swofford et al., both of these taxa. Nevertheless, we wished 1996), because likelihood scores (the proba- to evaluate hypotheses about sand lizard bility of the data, given the hypothesis) are phylogeny with new data in the form of comparable across models. Moreover, statis- DNA sequences and new analytical capabil- tical tests can be used to assess the signifi- ities that have recently become available. cance of the differences in likelihood scores Like many recent molecular phylogenetic between models, which is desirable because studies, ours is based on data from struc- overly deterministic models, though they in- turally and functionally separate genes (i.e., evitably improve the likelihood score also in- cytochromeb [cyt b] and 12SRNA). Currently crease sampling variance and may ultimately there is disagreement about how multiple decrease accuracy (Cunningham et al., 1998). data sets, including those based on differ- For these reasons, we used an approach ent genes, should be analyzed•that is, sepa- based on likelihood to select a model from a rately or combined (reviewed by de Queiroz set of progressively more-restrictive models et al., 1995; Hillis et al, 1996). For the most (i.e., those with progressively greater num- part, opposition to combining data sets oc- bers of estimated parameters), using that curs when those data sets are thought to have model as the basis for subsequent phyloge- been generated by distinctly different pro- netic analyses. We also took advantage of cesses. Bull et al. (1993), for example, gener- the additive properties of likelihood scores ated simulated nucleotide sequence data on (Edwards, 1972) and the existence of previ- the same tree under two distinct models and ously specified alternative phylogenetic hy- showed that parsimony analysis of the com- potheses to assess the amount of support bined data sets resulted in a poor estimate among four rival hypothesis, using the en- of the true tree. In the context of parsimony, tire data set but allowing for heterogene- weighting for position (e.g., codon position, ity in the models applied to its various stems vs. loops) and transformation (e.g., subsets. transitions vs. transversions) has been used to accommodate different underlying pro- MATERIALS AND METHODS cesses (Chippindale and Wiens, 1994); how- ever, that approach lacks an explicit basis for Taxon Sampling assessing which weighting scheme provides Samples of fresh liver or muscle tis- the best explanation of the data (Huelsenbeck sue were collected from 26 phrynosomatid et al., 1994). sand lizards, representing all 10 currently 594 SYSTEMATIC BIOLOGY VOL. 49 TABLE 1. The taxonomic designation, code, museum number, and collecting locality of the specin-iens used in the study. Subspecies taxonon-iy oí Holbrookia macúlala follows Axtell (1958). Nonstandard abbreviations: KdQ = Kevin de Queiroz; RRM = Richard R. Montanucci (field catalogues). Taxon Code Museum no. Collecting locality Sand lizards Callisaurus draconoides bogerti CDBO ROM 14970 México; Sinaloa; vie. Mazatlan carmenensis CDCA ROM 14182 México; Baja California; vie. San Ignacio crinitis CDCR MVZ 214832 México; Baja California; vie. Guerrero Negro myurus CDMY MVZ 214751 USA; Nevada; Mineral Co. rhodostictus CDRH MVZ 214739 USA; California; Inyo Co. ssp. CDSS ROM 15034 México; Sonora; vie. Bah• Kino ventralis CDVE LSUMZ 48811 USA; New Mexico, Hidalgo Co. Cophosaurus texanus scitulus CTSC MVZ 214711 USA; Arizona; Pima Co. texanus CTTE USNM 315499 USA; Texas; Blanco Co. Holbrookia lacérala lacérala HLLA USNM 315500 USA; Texas; Concho Co. subcaudalis HLSU USNM 315503 USA; Texas; McMullen Co. propinqua propinqua HPRO MVZ 214863 USA; Texas; Nueces Co. macúlala bunkeri HMBU USNM 337740 USA; New Mexico; Luna Co. campi HMCA MVZ 214801 USA; New Mexico; McKinley Co. elegans HMEL ROM 14964 México; Sinaloa; vic. Mazatlan flavilenta HMFL MVZ 21814 USA; Arizona; Cochise Co. macúlala HMMA MVZ 214806 USA; New Mexico; Chaves Co. ruthveni HMRU USNM 337/52 USA; New Mexico; Otero Co. thermophila HMTH CAS 174377 USA; Arizona; Cochise Co. Uma exsul UEXS ROM 15315 México; Durango, vic. Hwy 47 inornata UINO USA; CaKfornia; Riverside Co. notata notata UNNO MVZ 214799 USA; CaKfornia; Imperial Co. notata rufopunctata UNRU ROM 13919 México; Sonora; vic. Puerto Peñasco paraphygas UPAR ROM 15089 México; Durango, vic. Mapimô scoparia USCl ROM 14637 USA; CaKfornia, San Bernardino Co. scoparia USC2 MVZ 214784 USA; CaKfornia; San Bernardino Co. First outgroup Phrynosoma hernandesi PHER RRM 2470 USA; Colorado; Weld Co. platyrhinos PPLA KdQ 057 USA; CaKfornia; no further data Second outgroup Sceloporus jarrovii SJAR KdQ 036 USA; Arizona; Cochise Co. Urosaurus ornatus UORN MVZ 214658 USA; Arizona; Yavapai Co. Uta stansburiana USTA CAS 178153 USA; Arizona; Pima Co. vic, in the vicinity of. recognized species (Table 1). An effort was samples in STE buffer (0.4 M NaCl, 10 mM made to sample geographic variation, par- Tris-HCl pH 7.5,10 mM EDTA), 20% sodium ticularly within polytypic species. In addi- dodecyl sulfate and proteinase K (10 /ig//il) tion, five specimens representing five dis- for at least 12 hours at 55 °C. DNA was puri- tantly related species within the first and fied by extraction with phenol/chloroform, second outgroups•based on the phyloge- precipitated with ethanol, and resuspended nies of Etheridge and de Queiroz (1988) and in 15-100 n\ of TMS-EDTA buffer (Hillis Reeder and Weins (1996)•were included in et al, 1990). the analysis. These taxa were used to root the The polymerase chain reaction (PCR) trees in the various phylogenetic analyses. (~50 ng DNA; 0.4 mM primer each; 0.15 mM Tissues were frozen in liquid nitrogen in the dNTP; 5/J1 of 10 X buffer; 1.5 mM MgCb; 1.25 field and maintained at • 80°C in the labora- to 2.5 units of Taq polymerase, and distilled tory until DNA was extracted. H2O up to 50 Ail) (96°C, 45 sec; 45-50°C, 45 sec; 72 °C, 60 sec) was used to amplify frag- ments of two mitochondria! genes, a 650-bp DNA Isolation, Amplification, fragment of the cyt h gene and a fragment and Sequencing of ~350 bp of the ribosomal 12S gene. Two Genomic DNA was extracted from ~ 10 mg equal sized (~400 bp) portions of the cyt of frozen liver tissue by incubating the h fragment were independently amplified 2000 WILGENBUSCH AND DE QUEIROZ•SAND LIZARD PHYLOGENY 595 using two pairs of primers: L14841: 5'-AAA (Huelsenbeck, 1995; Yang, 1996b). Further- AAGCTTCCATCCAACATCTCAGCATGA more, the ability to objectively evaluate evo- TGAAA-3', H15149: 5'-AAACTGCAGCCC lutionary models under likelihood is of great CTCAGAATGATATTTGTCCTCA3' and importance in light of the general recogni- L15066: 5'-AATAAGCTTTTAAAGAAACAT tion that the processes governing molecu- GAAA(T/C)ATTGGAGTA-3', H15498: 5'-A lar evolution are not uniform across taxa AACTGCAGGGAATAAAGTTATCTGGGT (Wolfe et al, 1989; Martin and Palumbi, 1993; CTC-3'; numbers in the primer names re- Simon et al., 1996; Nielsen, 1997) and that us- fer to positions in the human sequence ing oversimplified or arbitrary models to in- (Anderson et al., 1981). A single pair of 12S fer phylogeny has undesirable consequences primers was used: 5'-AAACTGGGATTA (Felsenstein, 1978; Yang, 1996b; Sullivan and GATACCCCACTAT-3' and 5'-GAGGGTGA Swofford, 1997; Cunningham et al, 1998; CGGGCGGTGTGT-3' (Reeder, 1995). The Naylor and Brown, 1998). double-stranded PCR products were pu- Phylogenetic analyses of the nucleotide se- rified with 20% polyethylene glycol, dye- quence data were conducted using test ver- labeled using a dye terminator cycle se- sions 4.0d64-65 of PAUP*, written by David quencing reaction kit (ABI PRISM; Amplitaq L. Swofford. We generated initial topolo- DNA polymerase), and sequenced on an gies for the cyt b and 12S sequences, using automated sequencer (models 373 a and heuristic searches (10 random stepwise addi- 373 stretch; Applied Biosystems, Inc.). The tions of taxa and tree bisection-reconnection absence of indels in the cyt b sequences made [TBR] branch swapping) under parsimony it possible to align these sequences unam- with equal weighting for both codon posi- biguously by eye. A preliminary alignment tions and classes of nucleotide substitutions. of the 12S consensus sequences was made We then evaluated the likelihood of several using CLUSTAL W (Thompson et al, 1994) nested models of sequence evolution (see be- under the default settings. Improvements low) on the parsimony topologies to deter- to the initial alignment were made by mine which model and associated parameter eye, using published models of secondary values yielded the highest probability for the structure for a variety of taxa (Sullivan sequence data. The model and associated pa- et al., 1995; Hickson et al., 1996; Richards rameters that maximized the likelihood were and Moore, 1996). Sequences are deposited next used in a heuristic search (as-is stepwise with GenBank under the accession numbers addition of taxa and TBR branch swapping) AF194215-AF194276. under the likelihood criterion. This process was reiterated with subsequent likelihood topologies until the same topology or set of Phylogenetic Analysis topologies was found in successive searches. We used maximum likelihood to select We explored four substitution models: models of sequence evolution, to refine pa- Jukes-Cantor 0C; Jukes and Cantor, 1969), rameter estimates used in phylogeny recon- Kimura two-parameter (K2P; Kimura, 1980), struction, and to select optimal trees (see Hasegawa-Kishino-Yano (HKY; Hasegawa Swofford et al., 1996). Processes governing et al., 1985), and the general time reversible DNA sequence evolution and their effects (GTR; the REV of Yang, 1994a) (see Swofford on the estimation of phylogenies have been et al. [1996] for model descriptions). Each well documented (e.g., Collins et al., 1995; substitution model was first evaluated by as- Sullivan et al., 1995; Sullivan, 1996; Wakely, suming no rate heterogeneity among sites 1996; Yang, 1996a). Consequently, explicit and then with three combinations of rate het- models of sequence evolution have been de- erogeneity parameters: (1) the I of Hasegawa veloped in an attempt to compensate for et al. (1985), in which a proportion of potentially confounding effects for use un- sites were considered invariable (variable der a likelihood approach to phylogeny re- sites were assumed to follow an equal rates construction. Likelihood models have been model); (2) the gamma (r) of Yang (1994b), tested under a diverse set of simulated con- in which all sites were assumed to follow ditions (Gaut and Lewis, 1995) and have a discrete gamma-distributed rates model; been shown to outperform other methods and (3) I + r (Gu et al, 1995; Waddel and when data are simulated under more com- Penny, 1996), in which some sites were con- plex (and arguably more realistic) conditions sidered invariable and the variable sites were 596 SYSTEMATIC BIOLOGY VOL. 49 assumed to follow a gamma-distributed For both cyt h and 12S bootstrap searches, the rates model. Rate heterogeneity parameters GTR + I + F model parameters were fixed to were optimized under each substitution the parameter values estimated on the tree model; therefore, 16 models were evaluated found in the final iteration of the successive (i.e., 4 substitution models X 4 combina- approximations procedure described above. tions of rate parameters) for each data set We also used our data to evaluate the rela- and topology. A likelihood ratio test (Cox tive merits of four previously proposed phy- and Hinkley, 1974) was used to determine logenetic hypotheses (Fig. 1)•one of which whether models differed significantly in their corresponds to our best estimate of sand likelihood scores. Because each model is a lizard phylogeny based on the cyt h se- special case of the most general GTR + I quences. The successive approximation ap- + r model, this amounted to determining proach described above was used to find whether increasing the number of param- optimal topologies, with the branching or- eters estimated from the data (i.e., relax- der among the four taxa traditionally recog- ing the model restrictions) significantly im- nized as genera constrained to correspond proved the ability of the model to explain the to each of the previously proposed phyloge- data. The test statistic is assumed to be ;);: ^ dis- nies. The cyt \) data were used to find the opti- tributed, with the degrees of freedom equal mal topologies under the various constraints to the difference in the number of parameters because these data provided greater resolu- between the two models. Theoretically, the tion for relationships reflecting deeper sand use of the i'^ distribution is valid only when lizard divergences. We then optimized mod- the likelihoods of the nested models are com- els and parameter estimates for both gene puted on the true topology (Yang et al., fragments together, for each gene fragment 1995a); however, Yang et al. (1995b) have sug- separately, and for structurally and function- gested that uncertainty about phylogeny is ally defined regions or classes of sites within not a practical problem because differences each gene fragment (i.e., stems and loops for between substitution models are most often 12S, codon positions for cyt V). greater than differences between topologies. In this approach, support for each topol- Because it was impossible for us to know ogy can be evaluated for each data partition. whether we were assessing the true topology, In addition, because the measure of support we compared the 16 nested models using (i.e., likelihood) is additive (see Edwards, several alternative topologies representing 1972), overall support for each hypothesis rival phylogenetic hypotheses (Fig. 1) to can be assessed as the sum of the estimates determine whether the preferred model was for the individual data partitions (Adachi consistent across the alternative topologies. and Hasegawa, 1992; Huelsenbeck and Bull, We assessed the support for the putative 1996). However, by estimating parameter clades identified by parsimony and likeli- values separately for each data partition, hood searches using nonparametric boot- we run the risk of overparameterizing the strap resampling (Felsenstein, 1985). Under model. It therefore becomes important to the parsimony criterion, 200 bootstrap repli- know whether the additional model param- cates were performed on the cyt h and 12S eters help to provide a better general fit to data sets using heuristic searches (simple the data, or just fit random variation. To this stepwise addition of taxa and TBR branch end, a likelihood ratio test was used to deter- swapping), both with equal weights for char- mine whether partitioning the data signifi- acter positions and with step matrices to cantly improved the overall likelihood score. down-weight transitions 10:5,10:4,10:3, and Here again the likelihood ratio \s y} dis- 10:2 (i.e., 2:1,2.5:1, -3.3:1, 5:1). The transver- tributed, the degrees of freedom being equal sion/transition (tv:ti) weighting schemes to the difference in the number of parameters were selected because they formed a narrow used to estimate the two likelihood scores. range around the transition bias estimated Because model parameters were optimized under maximum likelihood (i.e., ~2.8). Un- for each data partition, the total number of der the likelihood criterion, 100 bootstrap model parameters was equal to the number replicates were performed on the cyt h and of branch lengths (2T • 3, where T is the 12S data sets using heuristic searches (as-is number of terminal taxa) plus the number stepwise addition of taxa and TBR branch of substitution parameters, all multiplied by swapping) under the GTR + I + F model. the number of data partitions. Finally, we 2000 WILGENBUSCH AND DE QUEIROZ•SAND LIZARD PHYLOGENY 597 evaluated the significance of the observed (C = 0.042), whereas high T and low A and G differences between the likelihood scores frequencies at second positions resulted in of alternative phylogenetic hypotheses, us- an intermediate bias (C = 0.236), and high ing the method proposed by Kishino and A and low G frequencies at third positions Hasegawa (1989). All site likelihood scores resulted in the most bias (C = 0.293) among were generated using test version 4.0d64 of codon positions. General patterns of bias for PAUP*, and the standard errors of the dif- each codon position observed here are simi- ferences between topologies were generated lar to those reported for mitochondrial genes using SAS (SAS Institute, 1990). sequences from birds (Kornegay et al., 1993; In addition, the amino acid translation of Nunn and Cracraft, 1996; Nunn et al, 1996), cyt b nucleotide sequences was used to eval- mammals (Irwin et al., 1991), and coUem- uate the four competing hypothesis (Fig. 1) bolans (Frati et al., 1997). Corrected (likeli- under a likelihood framework. We used the hood under the GTR + I + F model) pairwise program PROTML in the MOLPHY pack- distances (Table 3) between and within cur- age (Adachi and Hasegawa, 1992) to eval- rently recognized sand lizard species ranged uate the likelihood of the four topologies between 0.009 and 0.508 and between 0.002 under the transition probability matrix of and 0.208, respectively. Jones et al. (1992), using as an estimate of the 12S rRNA.•Nucleotide sequence length amino acid frequencies those observed in the varied between 336 and 352 bp for the 31 cyt b fragment (JTT-F option in PROTML). Of taxa examined. After inserting 29 gaps to the alternative likelihood models available in accommodate the alignment of conserva- PROTML, we chose the JTT-F model because tive regions, 367 positions were aligned. A it provided the best fit to the amino acid- total of 19 positions could not be confi- coded data. PROTML was used to calculate dently aligned because of multiple indels, the standard error of the difference between and these positions were not considered in competing hypotheses and thus to compare subsequent analyses. Indels were only in- those hypotheses using the test of Kishino ferred in single-stranded regions (loops) of and Hasegawa (1989). the 12S molecule. Not including indels, 105 characters were variable, of which 76 were RESULTS parsimony-informative. More loop positions (37%) than stem (25%) positions were vari- Sequence Variation able. Alignment of complementary stem po- Cytochrome h.•In all, 650 nucleotide posi- sitions for each sequence resulted in 50% tions were aligned, of which 274 were vari- A-U, 38% C-G, 9% G-U, 1% A-C, and 2% able and 240 were parsimony-informative. miscellaneous base pair combinations. The No insertions or deletions (indels) were ob- relatively high number of noncanonical G-U served. 202 (93%) of the third codon posi- pairs in the stems of vertebrate ribosomal tions were variable compared with 60 (28%) RNA genes is not uncommon (Kraus et al., of first the codon and 12 (6%) of the second 1992; Dixon and Hillis, 1993; Kjer, 1995), codon positions. All of the third position sub- perhaps because they are thought to have stitutions were synonymous, whereas 60% the least effect on the destabilization of sec- of the first codon and 100% of the second ondary structure (James et al., 1988) and may codon substitutions resulted in amino acid even be selectively advantageous (Rousset substitutions. In all, 38 (17%) of the amino et al., 1991). Base composition did not vary acid positions were variable. Observed dif- significantly among taxa (loop x^ = 9.26, ferences in nucleotide base composition P = 1.0 z^) stem z = 4.02, P = 1.0; and all among taxa were not significant, as indi- sites x^ = 2.74, P = 1.0). However, the pat- cated by the x^ homogeneity test (first posi- tern of base compositional bias differed be- tion ;f^ = 5.71, P = 1.0; second position ;f^ = tween stem and loop regions (Table 2). In par- 0.35, P = 1.0; third position, ;t:2 ^ 29.34, P = ticular, the loop region was A-rich and G/T- 1.0; and all sites, z = 3.84, P = 1.0). On the poor (C = 0.301), whereas the stem region other hand, the frequencies of the four bases showed very little base compositional bias (C varied overall and for each codon position = 0.041). This pattern of base compositional (Table 2), as indicated by the base compo- bias for the 12S gene fragment is similar to sitional bias (Bias C) of Irwin et al. (1991). the pattern observed among Galapagos igua- First codon positions showed the lowest bias nas (Rassmann, 1997) and a diverse sample of 598 SYSTEMATIC BIOLOGY VOL. 49 rj t^ m LO CO O t^ 00 Xi u rt (N •* (N (N ^ (N rj H 0)ö Ö Ö Ö Ö Ö Ö Ö Ö ^ ID in m ON CO LO >£) O rH fN rH o rH rH CN i-H ^ > O Ö O O Ö o o o en en o % LO (N m ON (N (N (N ^ U (N (N (N (N (N (N (N CN y o O Ö O O Ö O O Ö CTs ^ ON m ^ ^o ^ rj < (N (N ^ '^ CO ^ (N CO O Ó O O Ó O O O rj o 00 t^ t^ ^ '^ o 00 LT) 00 ÍN rH ^ ON ^H O LO •!* •* t>. t>. t>. ^ ^ r-i (N o (N di di <ó TA tv t^ O ^ ^ 00 o .in CT^ rt t^ ON ON o t^ '^ > OJ LO VÛ t>. ^ ^ ^ LO •^ v^ tó o Ö o o Ö o o Ö bC ts O PQ ^- O -s r^ ^ r", X o m ^ ^ •* on rj o ^ CO ^ o \£) s >< LO o •* CO CO ^o LO o CO in ^ <:í O OJ N V 00 o LO LO CTs ^ (N m O O O CTN rj o m (N O LO ON CO rt ^ Os Ö ^ CO Ö '^ OJ y ^ O LO ^ ON o CO rj a^ LTi ON CO LOI CO ^ LO ^ 00 ^ LO 00 CO CTs Lfi t>l CO o CO rH '^ LO •*" s u Y CO •^ ^0 ON CO ^H 3 OJ [O (N O s o 00 LO ^ v£) LJ) ^ CO LO O LO LO ^ lO !-H ^ Î-H LO '^ LO rH (N Î-H (N ^H y LO LO ON LO ^ ^ ^ LO ^ LO O ON LO (N + I 6 o 00 ^ (N ON ^ LO 'il c¡í u ^ 00 CO •* o CO 00 CO H ^ a^ CO (N LO U 0) (tí ^ (N CO C M Oo 4-: s u =3 Codon Codon Codon Loop Stem AU u Cyt h and 12S ^ (N (N! ^t^ . o o, ^ ^_ tr^ o ON LT) ON CO 01 , , cnNtN^tNONOLn ^ ^ o ^ o o I u p p p p p p I '^^^^^^^^^^•. •.•.•.•.•.•.• 43 ó ó ó ó ó o Ln^^cO"*oot^oo(N(N(NoooaNaNLnt^^^o ^ ^ (N ON (N -s* ^^LOLOLn"*^'!i X (N (N (N •* LO o nj LD LT) rH •!l< ^ in -t-j CD o o o o o OOOÖÖOÖOOOOÖÖÖOÖOÖÖÖ Ö Ö Ö O ^tvtvtvtvtNLnrjooooooLTicn^^OLriaNtNaN fN ON o en ;-^ en en ir, ir, \o ^ bC o o o o o o C ÖÖÖOOÖÖÖÖÖÖÖÖÖÖÖÖÖÖÖ d d d d (N (N tv •!(< tv ^ CD ^ LT) 00 1 ^ o o 00 ^ •* •* CO O CO O ON O ON LT) . ON ON o o p ^ p p I ^ ^ rj fN CDiOiCDOOaoaaaaa CD c:i c:i a o ó ó ó > Ö Ö CD Ö •n c (N O ^ LO en in O o ^ ^ ^H ^H ^H tr^ ^ ^ O t^ ir, t^ in ON v^ 00 fvl ^ (N LO ^ '^ '^ ^ ^ (N (N ^ ^ 00 ON ON in (N o ON ON o o en <=> a a a T-l (N ^ ^ rj (N =i '0>'0>'0>'0>o^^'=>'=>'=>'=>'=>c:ic:i<=> 599 600 SYSTEMATIC BIOLOGY VOL. 49 phrynosomatid lizards (Reeder, 1995). Cor- lematic in that they do not take multiple sub- rected (likelihood: GTR + I + F) pairwise stitutions into consideration and therefore distances (Table 3) between and within cur- tend to underestimate transitions at rapidly rently recognized sand lizard species ranged evolving sites (see Wakely, 1996). However, between 0 and 0.177 and between 0 and 0.052, topology-based estimates using a model of respectively. substitution that accommodates among-site rate heterogeneity show that G<-^A transi- Nucleotide Substitution Patterns tion rates are roughly half of C<-^T transition Cytochrome b.•Nucleotide substitutions rates (Table 2), indicating that the recogni- are generally considered in terms of changes tion of only two classes of substitutions (i.e., within the two structural classes of nu- transitions and transversions) oversimplifies cleotides (purines and pyrimidines), that is, the substitution process. Furthermore, likeli- in terms of transitions and transversions. hood models with six substitution rate cate- However, pairwise estimates of nucleotide gories (GTR) as opposed to two (K2P or HKY) substitution patterns among phrynosomatid had markedly higher likelihood scores for lizards suggest that G<-^A transitions may all of the topologies examined (p < 0.0001) occur less frequently than in other verte- (Fig. 3a). brates (Reeder, 1995). The plot of observed US rRNA.•Both pairwise and topology- differences versus corrected distances (from based estimates of nucleotide substitution Table 3) suggests that G<-^A transitions do 5600 not match the pattern exhibited by C<-^T c^Gir 5600 ttT Hir transitions; instead, they more closely ap- (a) 5700 proximate the curves for the A<-»C and 5800 W ^'^ A<-^T transversions (Fig. 2a). Pairwise esti- 5300 mates of nucleotide substitutions are prob- 5P00 -6100 0) -6200 jj^Jir Q ob •6300 en (a) -6400 H -6500 K -5600 6700 -6800 -mnn J Gr G!r (b) Hr Hir Gl ff Kir HI 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Jir (b) o K Ü CO •I 1 1•I•I 1•I [ 1•I•I• 12 34 56 789 10 11 12 o Substitution Model Parameters FIGURE 3. The likelihood scores for the 16 models 0 of sequence evolution on the final ML tree for data 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 plotted as a function of the number of substitution Distance (GTR + I + r) model parameters. Substitution models are represented by single characters: J = Jukes-Cantor; K = Kimura FIGURE 2. Observed pairwise differences plotted as two-parameter; H = Hasegawa-Kishino-Yano; and G = a function of the maximum Kkelihood pairwise dis- general time reversible. Characters following the sub- tances under the GTR + I + F model. Lines were fitted stitution models represent among-site rate heterogene- by using a negative exponential function (SAS, 1990). ity parameters: I = proportion of invariable sites, F = (a) Cytochrome h. (b) 12S RNA. gamma, (a) Cytochrome b. (b) 12S. 2000 WILGENBUSCH AND DE QUEIROZ•SAND LIZARD PHYLOGENY 601 rates suggest that, as in the case of cyt b, of the three equal-weight MP topologies. classification of substitutions into two classes Nevertheless, all three MP topologies were (transitions and transversions) oversimpli- used to evaluate the 16 models of sequence fies the substitution process. Observed pair- evolution described above under the like- wise nucleotide differences plotted against lihood criterion. The GTR + I + F model corrected distances (from Table 3) show that had the highest likelihood score for each A<->G transitions occur at frequencies more of the three MP topologies. Moreover, de- similar to certain classes of transversions spite increased variance from the estimation (Fig. 2b) (Reeder, 1995). Similarly, topology- of additional parameters, the difference be- based estimates indicate that G<-^ A transition tween the GTR + I + r likelihood score rates are substantially lower than those for (-InL = 5544.36) and that of the next best C<-^T transitions, particularly in the loop re- model (GTR + F, -InL = 5582.22) was signif- gions (Table 2), and likelihood models with icant ix^ = 75.92, df = 1, P < 0.0001) for the six substitution rate categories (GTR) as op- most likely parsimony topology. A heuris- posed to two (K2P or HKY) have markedly tic search using the GTR + I + F model higher likelihood scores for all of the topolo- parameters optimized on the single most gies examined (p < 0.0001) (Fig. 3b). likely parsimony topology resulted in the same single tree (Fig. 5). Again, the GTR + I + F model had the highest likelihood Optimal Trees score (•InL = 5543.606), which was sig- Cytochrome b.•Parsimony analysis of the nificantly higher (x^ = 76.80, df = 1, P < cyt b data with equal weights for both codon 0.0001) than the score of the next best model positions and base transformations resulted (GTR + F, -InL = 5582.01) (Fig. 3a). Ad- in three most-parsimonious (MP) topologies ditional heuristic searches with parameters (length = 1,160, Consistency Index [CI] = optimized on successive maximum likeli- 0.3464 excluding uninformative characters) hood (ML) trees did not change the branch- (Fig. 4a). Parsimony searches using step ma- ing order from that of the original ML trices to downweight transitions favored one topology. Uta stansburiana • uta stansburiana - Uta stansburiana Sceloporus jarrovii • Urosaurus ornatus • Urosaurus ornatus Urosaurus ornatus • Sceloporus jarrovii - Sceloporus jarrovii P. hernandesi _r U. scoparia 1 j-H. /. lacerata -{ P. platyrhinos '- U. scoparia 2 T-H. /. subcaudalis r U. exsul - U. n. rufopunctata -H. p. propinqua '- U. paraphygas - U. inornata -H. m. flavilenta r U. scoparia 1 - U. n. notata j-H. m. bunkeri '- U. scoparia 2 r U. exsul T-H. m. ruthveni j- U. n. rufopunctata '- U. paraphygas • H. m. campi \r U. n, notata r P hernandesi -H. m. maculata '- U. inornata ""- P. platyrhinos J-H. m. elegans r C. d. crinitis _r C. t. texanus '-H. m. thermophila '- C. d. carmenensis ^ C. t. scituius rC. d. bogerti r C. d. bogerti _r C. d. bogerti I "-C. d. ssp. <- C. d. ssp. ~^ C. d. ssp. C. d. carmenensis p C. d. vent ralis • C. d, crinitis C. d. crinitis [r C. d. rhodostictus • C. d. carmenensis C. d. rhodostictus T- C. d. myurus C. d. rhodostictus N-C. d. myurus r C. t. texanus C. d. myurus '-C. d. ventralis '- C. t. scitulus C. d. ventralis _rP hernandesi 1- H. I. ¡acerata r H. I. lacérala '-P platyrhinos '- H. i. subcaudalis L H. I. subcaudalis _rC. f. texanus - H. m. elegans r- H. p. propinqua >-•. t. scitulus - H. m. tliermophila [r H. m. elegans _[- U. exsul - H. p. propinqua T- H. m. thermophila '- U. paraphygas • ¡-Í. m. campi H. m. flavilenta r U. scoparia 1 - H. m. maculata H. m. maculata H- U. scoparia 2 - H. m. flavilenta H. m. campi r-u.nU. n. notata - H. m. bunkeri H. m. bunkeri IrUn rufopunctata (a) - H. m. ruîhveni (b) H. m. ruthveni (C) i-U.irinornata FIGURE 4. The strict consensus trees for: (a) three MP trees for the cyt b data, (b) six MP trees for the 12S data, and (c) 15 ML trees for the 12S data. 602 SYSTEMATIC BIOLOGY VOL. 49 Urosaurus ornatus 65 Uta stansburiana 63-82 Sceloporus jarrovii 95 P. hernandesi 62-86 P. platyrhinos 100 U. exsul 87-91 U. paraphygas 95 100 U. scoparia 1 79 70-80 100 61-90 92 U. scoparia 2 96-99 98 U. n. rufopunctata 100 99 U. n. notata 71-83 U. inornata 64 95 C. d. bogerti 52-78 98-100 C. d. ssp. 100 95 C. d. crinitis 100 95-99 _ _68 C. d. carmenensis '<5 96 C. d. ventralis 50 92-96 90 C. d. rhodostictus 38-66 100 C. d. myurus 100 C. t texanus 100 C. t scitulus 52 100 H. I. ¡acerata 33-52 100 H. I. subcaudalis 63 80 H. m. elegans 41-54 84 52-83 H. m. thermophila 64-90 H. p. propincua 56 H. m. maculata 55-80 93 H. m. campi 99-100 75 <5 98 H. m. flavilenta 8 3-99 91 1 H. m. bunkeri 9 2-971 H. m. rutliveni FIGURE 5. Phylogenetic relationships among sand lizards based on likelihood and parsimony analyses of the cytochrome b data. The numbers above each node represent the proportion of 100 likelihood bootstrap replicates found under the GTR H- I H- F model of sequence evolution. Numbers below each node represent the range of parsimony bootstrap proportions for 200 replicates when using equal weights and transitions downweighted by 10:5, 10:4, 10:3, and 10:2. The tree is the majority rule bootstrap consensus tree for the likelihood analysis. The majority rule bootstrap consensus trees for the various parsimony analyses are identical except for the branches indicated by dashed Hnes. 2000 WILGENBUSCH AND DE QUEIROZ•SAND LIZARD PHYLOGENY 603 The estimation of model parameters used that of the next best model (-InL = 1714.81; in phylogeny reconstruction is influenced GTR + F) (Fig. 3b). Another heuristic search to some extent by the initial topology on with the refined GTR + I + F parameters which the parameters are optimized (Yang, yielded the same 15 ML topologies as the 1994b; Sullivan et al., 1996). Therefore, we previous search. optimized model parameters on MP topolo- The parsimony (equal weights and vari- gies constrained to conform to three al- ous differential tv:ti weighting schemes) and ternative phylogenetic hypotheses (Fig. 1, likelihood (GTR + I + F) bootstrap analy- topologies II-IV). In all cases, subsequent un- ses for the 12S data give roughly concordant constrained searches using the model param- amounts of support for the resolved nodes eters optimized on the constrained topolo- (bootstrap proportions >50%) in a gener- gies converged on the same tree that had been ally poorly resolved phylogeny (Fig. 6). Both found by the search using the model parame- parsimony and likelihood analyses of the ters optimized on the most likely parsimony 12S rRNA data show very little support for topology. the deepest nodes in the tree. Rather, those Bootstrap resampling of the cyt b data ac- data are more informative regarding the shal- cording to parsimony (equal weights and lower nodes. various ti:tv weighting schemes) and likeli- hood (GTR + I + r) criteria produced nearly concordant topologies (Fig. 5). In general, Alternative Phylogenetic Hypothesis increasing the weight of transversions in- We limited our evaluation of alternative creased the bootstrap values for nodes rep- hypotheses using partition-specific models resenting deeper divergences, although there to four previously proposed topologies is no objective criterion for assessing which (Fig. 1), one of which (Fig. 1, topology I) ti:tv weighting scheme best explains the data corresponds to the MP and ML trees found under parsimony. In addition, model se- in unconstrained analyses of our cyt b data lection under likelihood indicates that the (Fig. 5). The basal ingroup relationships data are explained best by a more compli- found in the unconstrained parsimony and cated model•one involving six substitution likelihood analysis of the 12S data (Fig. 4b classes as well as invariant sites and rate and 4c) were not included in this evaluation variation among variable sites. Therefore, al- because the bootstrap analyses (Fig. 6) though we present the range of bootstrap val- indicated that the 12S data provided little ues under parsimony, hereafter we empha- support for those relationships. We also size the results obtained under likelihood. evaluated both the 12S and cyt b topologies 12S rRNA.•Parsimony analysis of the 12S with the alternative data set, using the data with equal weights for both character Kishino-Hasegawa (K-H) test (Kishino positions and base transformations resulted and Hasegawa, 1989), and found that the in six equally MP trees (length = 262, CI = cyt b data rejected (P < 0.0001) the 12S 0.4784 excluding uninformative characters) topology but the 12S data could not reject (Fig. 4b). All six MP trees were used to eval- (P = 0.0975) the cyt b topology. Both gene uate the 16 models of sequence evolution fragments are physically linked on the under the likelihood criterion. The GTR + I + mitochondrial genome and therefore must r model had the highest likelihood score for share a common history. It follows that at each of the six MP topologies. The score for least one of the gene fragments is giving the GTR + I + r model (-InL = 1717.64) was an incorrect estimation of the phylogeny, significantly better (x^ = 8.48, P = 0.0036) and both bootstrap analyses and the K-H than that of the next best model (•InL = test suggest that the misleading fragment is 1721.88; GTR + T) for the most likely parsi- 12S. Although the structural and functional mony topology. A heuristic search using the constraints imposed on the 12S molecule GTR + I + r model parameters optimized on could be responsible for the incongruent the most likely parsimony topology resulted phylogenies (Simon et al., 1996; Sullivan, in 15 ML topologies (Fig. 4c). Exploring 1995, 1996), the relatively small size of the models of evolution given the 15 ML trees 12S gene fragment we sequenced may also again yielded a score for the GTR + I + F explain its apparent incongruence with the model (-InL = 1711.70) that was signifi- cyt b data as well as the poor support that it cantly better (x^ = 6.22, P = 0.0126) than provides for the basal ingroup relationships. 604 SYSTEMATIC BIOLOGY VOL. 49 • Sceloporus jarrovii 86 • Uta stansburiana 80-87 • Urosaurus ornatus 97 p. hernandesi 95-99 1• P. platyrhinosplatyrt 99 |• U. exsul 100 l_ U. paraphygas 85 |• U. scoparia 1 89-95 l_ U. scoparia 2 77 89-100 U. n. rufopunctata 66 U. inornata 76-83 U. n. notata 64 r C. d. bogerti 86-90 1• C. d. ssp. 97 C. d. crinitis 62 98-99 56-91 76 C. d. carmenensis 77-93 C. d. rhodostictus 56 C. d. myurus 48-56 C. d. ventralis 91 r C. t. texanus 93-99 1• C. t. scitulus 98 |• H. I. ¡acerata 95-98 l_ H. I. subcaudalis H. m. campi H. p. propinqua H. m. flavilenta 87 H. m. ruthveni 73-86 H. m. macúlala H. m. bunkerí 88 |• H. m. elegans 54-58 H. m. thermophila FIGURE 6. Phylogenetic relationships among the sand lizards based on likelihood and parsimony analyses of the 12S data. The numbers above each node represent the proportion of 100 likelihood bootstrap repKcates found under the GTR H- I H- F model of sequence evolution. Numbers below each node represent the range of parsimony bootstraps proportions for 200 replicates using equal weights and transitions downweighted by 10:5,10:4,10:3, and 10:2. The tree is the majority rule bootstrap consensus tree for the likelihood analysis. The majority rule bootstrap consensus trees for the various parsimony analyses are identical except for the resolution of three additional clades within Holbrookia. 2000 WILGENBUSCH AND DE QUEIROZ•SAND LIZARD PHYLOGENY 605 TABLE 4. Negative log Hkelihood scores of topologies I-IV (Fig. 1) under the GTR + I + F model for each of the data partitions and combinations thereof. Differences between the log Hkelihood scores of topology I and topologies II-IV are given in parentheses, followed by the SEs of the differences. The highest likelihood value for each data partition is given inbold. The results of Kkelihood ratio test, used to determine whether estimating model parameters on the individual data partitions significantly increases the model fit as measured by the log Kkelihood score, are given for the various combinations of the entire data set. Each combination of partitions for the entire data set is compared to the same data set with one less partition. Hypothesis Data set N Topi Top II Top m Top IV Cytochrome b Unpartitioned 650 5543.61 5546.99 (-3.38), 3.44 5546.97 (-3.36), 3.20 5552.02 (-8.41), 5.42 Codon pos. 1 (Cl) 217 1185.82 1186.15 (-0.33), 1.01 1186.59 (-0.77), 1.74 1187.07 (-1.25), 1.86 Codon pos. 2 (C2) 217 389.68 391.97 (-2.29), 2.32 386.47 (3.21), 2.63 386.18 (3.50), 3.34 Codon pos. 3 (C3) 217 3534.29 3537.64 (-3.35), 3.25 3536.34 (-2.05), 2.46 3541.16 (-6.87), 4.76 Cl + C2 + C3 650 5109.79 5115.76 (-5.07), 4.12 5109.4 (0.39), 4.00 5114.41 (-4.62), 6.10 Amino acids 216 1266.9 1272.0 (-5.1), 6.1 1268.7 (-1.8), 4.5 1269.0 (-2.1), 4.7 12S Unpartitioned 348 1733.78 1733.27 (0.51), 0.87 1733.95 (-0.17), 0.87 1732.11 (1.67), 2.85 Loop 180 977.13 977.13 (0.00), 0.01 980.22 (-3.09), 3.09 978.97 (-1.84), 3.09 Stem 168 697.26 697.26 (0.00), 0.00 697.26 (0.00), 0.00 697.26 (0.00), 0.00 Loop + stem 348 1674.39 1674.39 (0.00), 0.00 1677.48-(3.09), 2.98 1676.23 (-1.84), 3.91 AH data Unpartitioned 998 7358.06 7360.88 (-2.82), 3.13 7361.16 (-3.10), 3.03 7364.38 (-6.32), 5.62 Cyt b + 12S 998 7277.39 7280.26 (-2.87), 3.55 7280.92 (-3.53), 3.31 7284.13 (-6.74), 6.12 X¡y = 161.34 P < 0.00001 Cyt b + stem + loop 998 7218.00 7221.38 (-3.38), 3.44 7224.45 (-6.45), 4.36 7228.25 (-10.25), 6.68 X67 = 118-78 P = 0.0001 Cl + C2 + C3 + 12S 998 6844.47 6849.04 (-4.57), 4.21 6843.35 (1.12), 4.09 6846.51 (-2.04), 6.74 Xh = 747.06 P < 0.00001 Cl + C2 + C3 + loop + stem 998 6784.18 6790.15 (-5.97), 4.12 6786.88 (-2.70), 4.98 6790.64 (-6.46), 7.25 Xh = 120.58 P = 0.00008 We evaluated each of the four alterna- Based on results of the likelihood ratio test, five topologies (Fig. 1), using each separate separately optimizing branch lengths and data partition, combinations of partitions, model parameters for different data parti- and the amino acid sequence of the cyt b tions significantly improved the cumulative molecule (Table 4). Topology I {Uma {Cal- likelihood score for all combinations of data lisaurus {Cophosaurus, Holbrookia))) alone re- partitions relative to the unpartitioned data ceived the highest likelihood score for the un- (Table 4), and the data set with the most par- partitioned cyt b fragment, the first and third titions explained the data best. Because these codon positions, the amino acid sequences, comparisons were limited to topologies cho- and four of the five combinations of parti- sen a priori, the question remains whether tions using all of the data. Topology I also better topologies exist. Technological limita- received the tied highest likelihood score for tions prevented us from searching for op- the loop positions and the combination of timal trees using combined but partitioned loop and stem positions. Topology III was data. Although searching for trees using the favored by the combination of the Individ- combined but unpartitioned cyt b plus 12S ual cyt b codon positions and by the com- data sets under the GTR + I + F model bination of the individual codon positions yielded a novel optimal topology (Fig. 7a), plus the unpartitioned 12S fragment. Second the likelihood of this topology given the un- codon positions and the unpartitioned 12S partitioned data is only 1.74 units greater fragment favored topology IV. Topology II than that of topology I. In contrast, when was not uniquely favored by any of the data model parameters are optimized separately partitions, though it had the tied (with topol- for each of the partitions, the cumulative like- ogy I) highest likelihood score for loop posi- lihood score of the optimal tree for the unpar- tions and loop plus stem positions. The stem titioned data (i.e., •InL 6789.93) (Fig. 7a) is positions alone were unable to discriminate 5.75 units less than that of the optimal topol- among the alternative topologies. ogy (i.e., topology I). 606 SYSTEMATIC BIOLOGY VOL. 49 Uta stansburiana Uta stansburiana Urosaurus ornatus Sceloporus jarrovii Sceloporus jarroyii Urosaurus ornatus C. Î. texanus P. hernandesi £ C. t. scitulus £ P. platyrhinos H. I. ¡acerata U. exsul £ H. I. subcaudalis U. paraphygas H. m. elegans U. scoparia 1 H. m. thermophila |-[ U. scoparia 2 H. p. propincua U n. rufopunctata H. m. maculata U. n. notata H. m. campi U. inornata H. m. flavilenta C. d. bogerti H. m. bunkeri C. d. ssp. \ H. m. ruthveni C. d. crinitis C. d. bogerti C. d. carmenensis C. d. ssp. C. d. ventralis C. d. crinitis C. d. rhodostictus C. d. carmenensis C. d. myurus C. d. ventralis C. t. texanus C. d. rhodostictus C. t. scitulus C. d. myurus H. I. lacerata P. hernandesi H. I. subcaudalis £ P. platyrhinos H. m. elegans U. exsul H. m. thermophila U. paraphygas H. p. propinqua U. scoparia 1 H. m. campi U. scoparia 2 H. m. maculata U. n. rufopunctata H. m. flavilenta (a) U. n. notata (b) H. m. bunkeri U. inornata \ H. m. ruthveni FIGURE 7. Phylogenetic relationships among the sand Kzards based on the likelihood and parsimony analyses of the combined but unpartitioned cyt b and 12S data, (a) The single ML topology under the GTR + I + F model of sequence evolution (•InL = 7356.34). (b) The strict consensus of the two MP topologies under equal weights for both codon positions and classes of nucleotide substitutions (length = 1434). Because the data consist of partial gene needed to reject the alternative hypotheses is sequences, it is instructive to estimate the greater than the total number in the gene. We number of nucleotides necessary to confi- can roughly estimate the sample size needed dently (i.e., P < 0.05) reject the alternative because the mean of the site-likelihood hypotheses. In particular, we are interested in differences is proportional to the sample size knowing whether the number of nucleotides and the standard error is proportional to the 2000 WILGENBUSCH AND DE QUEIROZ•SAND LIZARD PHYLOGENY 607 square root of the sample size (Kishino and topology also has no nodes in common with Hasegawa, 1989). For 12S RNA, Simon et al. the favored topology (topology I). In addi- (1996) demonstrated large differences in sub- tion, this is the only hypothesis that might be stitutional pattern and evolutionary rates be- rejected with the entire nucleotide sequence tween the 3' and 5' ends of the molecule. of the cyt b gene. It is more difficult to discrim- Because our estimate of the number of sites inate between the favored hypothesis and necessary to confidently reject competing hy- the other two alternative hypotheses (topolo- potheses assumes roughly the same amount gies II and III), both of which have nodes in of variability as observed in the sequences common with the favored hypothesis. already collected, we confined our estimate Although topology IV is favored by the to the cyt b gene. We estimated that 2,586 bp second codon positions of the cyt b molecule, are necessary to confidently reject topology II the likelihood estimate is based on only 12 relative to topology I; 2,496 bp to reject topol- variable sites, all of which require an amino ogy III; and 1,037 bp to reject topology IV. acid substitution. Of these sites, only one fa- Based on these estimates, the entire cyt b gene vors topologies III and IV over topologies I (-1,130 bp, Kumazawa et al., 1998) might and II•and thus a sister group relationship be sufficient to reject topology IV, but it will between Cophosaurus and Callisaurus rather probably be necessary to collect data from an than between Cophosaurus and Holbrookia• additional gene to reject the other two com- on the basis of the individual site likeli- peting hypotheses. hood scores. All of the Cophosaurus and Cal- lisaurus specimens and a single Holbrookia specimen (HMCA) have thymine at this po- DISCUSSION sition; the other samples all have cytosine. Moreover, nonsynonymous substitutions at Higher-Level Relationships the first position of this codon result in a pat- Four alternative hypotheses have been tern of shared amino acids that does not cor- proposed to describe higher level ("inter- respond perfectly to the pattern of shared generic") phylogenetic relationships among nucleotides at the second position of the extant sand lizard taxa, referred to here as codon. Cophosaurus and the single Holbrookia topologies I-IV (Fig. 1). Our analyses of two specimen (HMCA) share valine, Callisaurus mitochondrial DNA gene fragments favor is characterized by isoleucine, a single Uma the phylogenetic relationships among sand specimen (UPAR) is characterized by threo- lizard "genera" first proposed by Savage nine, and all the other samples share alanine. (1958)•topology I (Fig. 1)•and later sup- Nevertheless, under the JTT transition model ported by Cox and Tanner (1977), Etheridge (Jones et al., 1992; see Methods), the single- and de Queiroz (1988), de Queiroz (1989, site likelihood scores for this codon still rank 1992), Changchien (1996), Reeder and Wiens topologies III and IV ahead of topologies I (1996), and in part Porter et al. (1994). In and II, in agreement with the site likelihood this hypothesis, Uma is the sister group of scores for the second position of this codon. all other sand lizards, and the "earless" The point is that this particular nucleotide lizards, Holbrookia and Cophosaurus, are sis- site is singularly responsible for the rank ter taxa. These relationships are favored order of the alternative topologies for the by four of the five analyses, on the ba- second position data, a ranking that is contra- sis of all of our sequence data, including dicted by both the nucleotide and amino se- the best estimate (i.e., Cl + C2 + C3 + quence data for the full cyt b fragment, which stem + loop), although the alternative hy- rank topology I highest. When this single potheses (topologies II-IV) cannot be re- character was removed and the four topolo- jected with a high level of confidence (i.e., gies were again evaluated with the second P < 0.05). The hypothesis that places Hol- position data, the rank order of the topologies brookia as the sister taxon to all other sand changed and the likelihood of the difference lizards (topology IV) is least favored by the between the three highest-ranking hypothe- mtDNA sequences examined in this study. ses decreased (topology I, •374.59; topology The tree constrained to conform to this II, -378.91; topology III, -373.99; and topol- hypothesis has the lowest likelihood score ogy IV,-375.19). in all but one of the comparisons that used The debate about whether Cophosaurus all of the data. The constrained part of the shares a more recent common ancestor 608 SYSTEMATIC BIOLOGY VOL. 49 with Callisaurus or with Holbrookia and relationships in the 12S tree (Eig. 6) involve thus whether the "earless" condition in taxa for which genetic distances do not ex- Cophosaurus and Holbrookia is synapomor- ceed 0.069. Within Uma, the cyt b and 12S phic versus homoplastic has a long history data sets recover nearly the same putative (Mittleman, 1942; Smith, 1946; Axtell, 1958; clades with similar bootstrap support. These Norris, 1958; Savage, 1958; Earle, 1961,1962; include a sister group relationship between Clarke, 1965; Cox and Tanner, 1977; Adest, U. exsul and Li. paraphygus (cyt b, 100%; 12S, 1978; Blackburn, 1978; Etheridge and de 99%); monophyly of a group composed of Queiroz, 1988; de Queiroz, 1989,1992). A sis- U. scoparia, U. notata, and U. inornata (cyt b, ter group relationship between Cophosaurus 92%; 12S, 77%); and monophyly of a group and Holbrookia occurs in two of the alterna- composed of Li. notata and Li. inornata (cyt b, tive hypotheses (Topologies I and II) and is 98%; 12S, 66%). The cyt b data also support favored by 12 of 15 data partitions or com- a sister group relationship between U notata binations thereof under likelihood (Table 4), notata and U. inornata (99%). In agreement including the one (Cl + C2 + C3 + stem + with Adest (1977) and de Queiroz (1989, loop) that best explains the data. It is also 1992), our analyses support a relatively large favored by parsimony analysis of the cyt b separation between the Uma of the Mojave data and by parsimony analysis of the com- and Sonoran deserts (Li. notata, U. inornata, bined cyt b plus 12S data (Fig. 7b). Given and Li. scoparia) and those of the Chihuahuan the agreement of our results with those of Desert {U. paraphygas and U. exsul). Within recent analyses based on morphological, al- the former (notata) group, authors have rec- lozymic, and DNA hybridization data (Cox ognized one (Schmidt, 1953; Adest, 1977), and Tanner, 1977, Etheridge and de Queiroz, two (e.g., Norris, 1958; Carpenter, 1963), or 1988; de Queiroz, 1989, 1992; Changchien, three (Mayhew, 1964; Pough, 1973, 1974, 1996; Reeder and Wiens, 1996), the sister 1977) species. The cyt b distances between group relationship between Cophosaurus and U. scoparia and U. notata-inornata (0.113- Holbrookia and the single evolution of the con- 0.127) are within the range of other between- cealed tympanic membrane must be consid- species comparisons in this study. In con- ered the best supported of the four alterna- trast, the cyt b distances between U. notata tive hypotheses (Eig. 1). and Li. inornata (0.009-0.020) are low even for within-species comparisons, though the taxa are morphologically distinguishable and are Lower-Level Relationships sometimes recognized as separate species Both cyt b and 12S sequences provide (e.g., Mayhew, 1964; Pough, 1973, 1974, support for lower-level relationships among 1977). Interestingly, our results indicate that sand lizards•that is, for relationships in- Li. notata is paraphyletic, with Li. notata volving monophyly of the taxa traditionally notata being more closely related to Li. inor- ranked as genera (or simply named as for- nata than to U. notata rufopunctata, and high- mal taxa) and particularly for relationships light the need for a detailed investigation of within those taxa. these three forms. Urna.•The cyt b bootstrap analyses in- Callisaurus.•We observed strong to sup- dicate strong support (95%) for a mono- port (cyt b, 100%; 12S, 97%) for a mono- phyletic Uma. Eor the 12S data, Uma is one phyletic Callisaurus, corroborating the results of two traditional genera for which mono- of an earlier allozyme study (de Queiroz, phyly is supported by <50% of the boot- 1992). The cyt b data support three major strap replicates under both parsimony and clades within the single currently recognized likelihood; however, this situation proba- species Callisaurus draconoides, two of which bly reflects the limited ability of the 12S are also supported by the 12S data, whereas data to resolve deeper divergences within the the third is not contradicted by them. Spec- sand lizard clade. In addition to the strong imens from Sonora and Sinaloa, represent- support provided by the cyt b data, mono- ing the taxa Callisaurus d. bogerti, brevipes, phyly of Uma is supported by at least eight or inusitatus (CDSS) and C. d. bogerti, form a unambiguous morphological characters (de strongly (cyt b, 95%) to weakly (12S, 64%) Queiroz, 1989). Moreover, genetic distances supported clade the placement of which between the two major clades within Uma are as sister to all other Callisaurus is weakly all > 0.075, whereas all strongly supported supported (cyt b, 68%; 12S, <50%) by the 2000 WILGENBUSCH AND DE QUEIROZ•SAND LIZARD PHYLOGENY 609 likelihood analyses but is contradicted by elegans and H m. thermophila), though sup- the cyt b parsimony analysis when transi- port is weak (56%). The 12S data are un- tions are downweighted 2:1: That is, 54% able to resolve the placement of H. propin- of the bootstrap replicates support the place- qua relative to the various populations of ment of the Baja California taxa (C. d. crini- H. maculata, in that neither likelihood nor tis and C. d. carmenensis) as sister taxa to parsimony bootstrap analyses find any res- all other Callisaurus. In the likelihood anal- olution in more than 50% of the replicates. ysis, the Baja California clade is strongly Moreover, the placement of H. propinqua in supported (95%) by the cyt b data but the likelihood analysis is incongruent with not by the 12S data. A clade of the three its placement in the parsimony analysis, in northern samples, representing C. d. myurus, that each of the 15 ML trees (Fig. 4c) places C. d. rhodostictus, and C. d. ventralis, is sup- H. propinqua outside of H. maculata, whereas ported strongly by cyt b (96%) and weakly the six MP trees (Fig. 4b) place it as sis- by 12S (56%). Within the later clade, the ter group of the southwestern populations Mojave sample of C. d. rhodostictus forms (H. m. elegans and H. m. thermophila). Mono- a clade with C. d. myurus that has strong phyly of the northeastern populations of H. (cyt b, 90%) to weak (12S, <50%) support. maculata is strongly supported by the cyt b At least some of the cyt b genetic dis- data (93%), but only the parsimony analy- tances between the three clades (0.093-0.117) ses of the 12S data show >50% bootstrap are within the range of between-species support for this putative clade. Within this comparisons observed among other sand clade, the cyt b data provide moderate to lizards, whereas all 12S distances (0.006- strong support (75-98%) for three nested 0.037) are within the range of within-species clades, though the most inclusive one is comparisons. contradicted by two of the weighted parsi- Cophosaurus.•Our data provide strong mony analyses: Downweighting transitions support for monophyly of the two by 10:3 and 5:1 under parsimony provides Cophosaurus samples, representing C. t. texa- weak support (52% and 56%, respectively) nus and C. t. scitulus. Because we did not for a clade composed of all of the taxa ex- sample the third Cophosaurus subspecies cept H. m. campi; the ML topology excludes (C. t. reticulatus), this result should not be H. m. maculata instead (Figure 5). Holbrookia taken as evidence for monophyly of the maculata elegans and H. m. thermophila also species as a whole. The distances between form a moderately (cyt b, 80%) to strongly the C. t. texanus and C. t. scitulus (cyt b, 0.055; (12S, 88%) supported clade. Genetic dis- 12S, 0.041) are similar to those observed tances between the southwestern and north- within other species in this study. eastern clades within the currently recog- Holbrookia.•The cyt b data provide weak nized species H. maculata (cyt b, 0.176-0.208; support (63%) for the monophyly of Hol- 12S, 0.037-0.052) are comparable with those brookia. The likelihood analysis of the 12S between each of these clades and H. propin- data did not support Holbrookia monophyly qua (cyt b, 0.161-0.200; 12S, 0.044-0.069) in >50% of the bootstrap replicates, though as well as other between-species compar- the taxon is monophyletic in all of the 15 ML isons. Those within the southwestern (cyt b, topologies (Fig. 4c) and all of the six MP 0.125; 12S, 0.019) and northeastern (cyt b, topologies (Fig. 4b). Within Holbrookia, our 0.014-0.063; 12S, 0.003-0.025) clades are com- analyses strongly support (cyt b, 100%; 12S, parable with other within-species compar- 98%) a monophyletic Holbrookia lacerata and isons. Of particular importance is the fact show moderate support (cyt b, 84%) to weak that the geographically proximate samples (12S, <50%) for its placement as the sister of H. m. thermophila and H. m. flavilenta from taxon of all other Holbrookia, that is, H. macu- Cochise Co., Arizona, exhibit considerably lata and H. propinqua. In contrast, monophyly greater distances from one another (cyt b, of H. maculata is contradicted by the cyt b 0.190; 12S, 0.044) than from geographically data. The northeastern populations of this more distant samples in the larger clades to taxon (H. m. maculata, H. m. campi, H m.flavi- which they belong (cyt b, 0.025-0.125; 12S, lenta, H. m. ruthveni, andH. m. bunkeri) appear 0.009-0.025). A more detailed study of geo- more closely related to Holbrookia propinqua graphic variation with Holbrookia is currently than to the southwestern populations (H. m. in progress. 610 SYSTEMATIC BIOLOGY VOL. 49 Heterogeneous Processes models and parameter estimates for different The molecular sequences presented here subsets of the data. Since then, the ability to provide useful data for reconstructing phylo- search for optimal trees under heterogeneous genetic relationships both among and within models has become available in PAML (Yang, the species of phrynosomatid sand lizards. 1999), and similar capabilities are being de- The cyt b sequences are useful for recon- veloped in PAUP* (D. L. Swofford, pers. structing both shallow and deep divergences comm.). Further development of such ca- within this clade, whereas the 12S sequences pabilities should greatly improve estimates are useful for reconstrucing shallow diver- of phylogeny based on large, heterogeneous gences. Although not decisive, these data data sets, including sequences from both sin- add further support to one of the four gle and multiple genes. alternative hypotheses (Fig. 1) that have been proposed concerning early cladoge- ACKNOWLEDGMENTS netic events within the phrynosomatid sand We thank J. O'Brien, J. Gauthier, R. Montanucci, lizards (topology I), as well as providing R. Seib, J. Losos, A. Bauer, D. Brown, G. Adest, D. new information on more recent cladoge- Morafka, R. Murphy, T. Reeder, D. Good, C. Phñlips, netic events and patterns of genetic and geo- D. Cannatella, J. Trochet, M. Morris, and S. Walker for providing or helping to coEect specimens. J. SuUivan, graphic differentiation within species. C. Sutton, J. Huelsenbeck, and D. Swofford provided Our results also highlight the heterogene- valuable assistance with the maximum likelihood anal- ity of molecular sequence evolution and the yses, and C. Gordon, T Reeder, and C. Huddleston con- concomitant importance of developing phy- tributed to the collection of the sequence data. We thank T. Reeder and J. Wiens for insightful comments on an ear- logenetic methods that allow analysis under lier version of this manuscript. This work was partially heterogeneous models. Optimization of like- supported by the Laboratory of Molecular Systematics lihood models for the different gene frag- of the Smithsonian Institution. ments (cyt b and 12S) and the structural and functional regions (stem and loop) or classes of sites (codon positions) within those gene REFERENCES fragments revealed pronounced differences ADACHI, J., AND M. HASEGAWA. 1992. MOLPHY: among the parameters used to characterize Programs for molecular phylogenetics I•PROTML: sequence evolution (e.g., base substitution Maximum likelihood inference of protein phylogeny. Computer Science Monograph 27. Institute of Statis- rates, proportion of invariant sites, and rate tics and Mathematics, Tokyo. heterogeneity among sites). When these dif- ADEST, G. A. 1977. Genetic relationships in the genus ferences were ignored (i.e., by not partition- Uma (Iguanidae). Copeia 1977:47-52. ing the data), estimates of model parame- ADEST, G. A. 1978. The relationships of the sand lizards Uma, Callisaurus, and Holbrookia (Sauria: ters yielded intermediate values that in some Iguanidae): An electrophoretic study. Ph.D. Disser- cases differed greatly from those estimated tation, Univ. California, Los Angeles. for one or more subsets of the data (Table 2). ANDERSON, S., A. T. BANKIER, B. G. BáRRELE, M. H. L. Consequently, attempting to characterize all DE BRUIJN, A. R. COULSON, J. DROUIN, I. C. EPERON, the available data with a single set of model D. P. NiERLicH, B. A. ROE, F. SäNGER, P. H. SCHREIER, A. J. H. SMITH, R. STADEN, AND I. G. YOUNG. 1981. parameters resulted in a considerably lower Sequence and organization of the human mitochon- value for the overall likelihood score. Bull drial genome. Nature 290:457-465. et al. (1993) cautioned against combining AXTELL, R. W. 1958. A monographic revision of the data sets where there is significant variation iguanid genus Holbrookia. Ph.D. Dissertation, Univ. in the underlying processes of sequence evo- Texas, Austin. BLACKBURN, D. G. 1978. The comparative, functional, lution. Our approach takes advantage of an and evolutionary myology of burrowing sceloporine important property of likelihood (i.e., the ad- lizards. M.S. Thesis, Cornell Univ., Ithaca, New York. ditivity of likelihood values from seperate BULL, J. J., J. P. HUELSENBECK, C. W. CUNNINGHAM, subsets of data), thereby permitting us to D. L. SWOFFORD, AND P. J. WADDELL. 1993. Parti- tioning and combining data in phylogenetic analysis. combine data sets while at the same time al- Syst. Biol. 42:384-397. lowing for variation in the processes acting CARPENTER, C. C. 1963. Patterns of behavior in three on them. In the present study, our use of this forms of the fringe-toed lizards (¡Jma-Iguanidae). approach was restricted to evaluating phylo- Copeia 1963:406-412. CHANGCHIEN, L.-L. 1996. A phylogenetic study of genetic hypotheses defined a priori because sceloporine lizards and their relationships with other available computer programs did not permit iguanid lizards based on DNA/DNA hybridization. searching for optimal trees by using different Ph.D. Dissertation, Univ. Wisconsin Madison. 2000 WILGENBUSCH AND DE QUEIROZ•SAND LIZARD PHYLOGENY 611 CHIPPINDALE,R T., AND J.J.WIENS. 1994. Weighting, for third domain of animal 12S rRNA. Mol. Biol. Evol. partitioning, and combining characters in phyloge- 13:150-169. netic analysis. Syst. Biol. 43:278-287. HILLIS, D. M., A. LARSON, S. K. DAVIS, AND E. A. CLARKE, R. F. 1965. An ethological study of the ZIMMER. 1990. Nulceic acids III: Sequencing. Pages iguanid lizard genera Callisaurus, Cophosaurus, and 318-370 in Molecular systematics, 1st edition (D. M. Holbrookia. Emporia State Res. Stud. 13:1-66. Hillis and C. Moritz, eds.). Sinauer, Sunderland, COLLINS, T. M., R H. WIMBERGER, AND G. J. P. NAYLOR. Massachusetts. 1995. Compositional bias, character-state bias, and HILLIS, D. M., B. K. MABLE, AND C. MORITZ. 1996. character-state reconstruction using parsimony. Syst. Applications of molecular systematics: The state of Biol. 43:482^92. the field and a look to the future. Pages 515-543 in Cox, D. R., AND D. V. HINKLEY. 1974. Theoretical Molecular systematics, 2nd edition (D. M. Hillis, C. statistics. Chapman and HaE, London. Moritz, and B. K. Mable, eds.). Sinauer, Sunderland, Cox, D. C, AND W. W. TANNER. 1977. Osteology Massachusetts. and myology of the head and neck regions of Cal- HUELSENBECK, J. P. 1995. Performance of phyloge- lisaurus, Cophosaurus, Holbrookia, and Uma (Reptilia: netic methods in simulation. Syst. Biol. 44:17^8. Iguanidae). Great Basin Nat. 37:35-56. HUELSENBECK,J. P., ANDJ. J. BULL. 1996. A likelihood CUNNINGHAM,C. W.,H.ZHU, AND D.M.HILLIS. 1998. ratio test to detect conflicting phylogenetic signal. Best-fit maximum-likelihoo d models for phylogenetic Syst. Biol. 45:92-98. inference: Empirical tests with known phylogenies. HUELSENBECK, J. P., D. L. SWOFFORD, C. W. Evolution 52:978-987. CUNNINGHAM, J. J. BULL, AND P. J. WADDELL. DEQUEIROZ, K. 1989. Morphological and biochemical 1994. Is character weighting a panacea for the prob- evolution in the sand lizards. Ph.D. Dissertation, Univ. lem of data heterogeneity in phylogenetic analysis? California, Berkeley. Syst. Biol. 43:288-291. DE QUEIROZ, K. 1992. Phylogenetic relationship and IRWIN, D. M., T. D. KOCHER, AND A. C. WILSON. 1991. rates of allozyme evolution among the lineages of Evolution of the cytochrome h gene of mammals. J. sceloporine sand Kzards. Biol. J. Linn. Soc. 45:333- Mol. Evol. 32:128-144. 362. JAMES, B. D., G. J. OLSEN, J. Liu, AND N. R. PACE. 1988. DE QUEIROZ, A., M. J. DONOGHUE, AND J. KIM. 1995. The secondary stiucture of ribonuclease P RNA, the Separate versus combined analysis of phylogenetic catalytic element of a ribonucleoprotein enzyme. Cell evidence. Annu. Rev. Ecol. Syst. 26:657-681. 52:19-26. DIXON, M. T., AND D. M. HILLIS. 1993. Ribosomal JONES, T. R., W. R. TAYLOR, AND J. M. THORNTON. 1992. RNA secondary structure: Compensatory mutations The rapid generation of mutation data matrices from and implications for phylogenetic analysis. Mol. Biol. protein sequences. Comput. Appl. Biosci. 8:25-282. Evol. 10:256-267. JUKES, T. H., AND C. R. CANTOR. 1969. Evolution of EARLE, A. M. 1961. The middle ear of Holbrookia and protein molecules. Pages 21-132 in Mammalian pro- Callisaurus. Copeia 1961:405-410. tein metabolism (H. N. Munro, ed.). Academic Press, EARLE, A. M. 1962. The middle ear of the genus Urm New York. compared to those of the other sand lizards. Copeia KIMURA, M. 1980. A simple method for estimating 1962:185-188. evolutionary rate of base substitution through com- EDWARDS, A. W. F. 1972. Likelihood. Cambridge parative studies of nucleotide sequences. J. Mol. Evol. Univ. Press, Cambridge, England. 16:111-120. ETHERIDGE, R., AND K. DE QUEIROZ. 1988. A phy- KiSHiNO, H., AND M. HASEGAWA. 1989. Evaluation of logeny of Iguanidae. Pages 283-367 in Phylogenetic the maximum likelihood estimate of the evolution- relationships of the Kzard families (R. Estes and G. ary tree topologies from DNA sequence data, and the PregiU, eds.). Stanford Univ. Press, Stanford, Califor- branching order in Hominoidea. J. Mol. Evol. 29:170- nia. 179. FELSENSTEIN, J. 1978. Cases in which parsimony and KjER, K. M. 1995. Use of rRNA secondary structure compatibility methods will be positively misleading. in phylogenetic studies to identify homologous posi- Syst. Zool. 27:401-410. tions: An example of alignment and data presentation FELSENSTEIN, J. 1985. Confidence limits on phy- from the frogs. Mol. Phylogenet. Evol. 4:314-330. logeny: An approach using the bootstrap. Evolution KORNEGAY, J. R., T. D. KOCHER, L. A. WILLIAMS, AND A. 39:783-789. C. WILSON. 1993. Pathways of lysozyme evolution FRATI, F, C. SIMON, J. SULLIVAN, AND D. L. SWOFFORD. inferred from the sequences of cytochrome h in birds. 1997. Evolution of the mitochondrial cytochrome ox- J. Mol. Evol. 37:367-379. idase II gene in Collemhola. J. Mol Evol. 44:145-158. KRAUS, F, L. JARECKI, M. M. MIYAMOTO, S. M. GAUT, B. S., AND P. O. LEWIS. 1995. Success of max- TANHAUSER, AND P. J. LAIPIS. 1992. Mispairing and imum Kkelihood phylogeny inference in the four- compensational changes during the evolution of mito- taxon case. Mol Biol. Evol. 12:152-162. chondrial ribosomal RNA. Mol. Biol. Evol. 9:770-774. Gu, X., Y.-X. Fu, AND W.-H. Li. 1995. Maximum like- KuMAZAWA, Y., H. OTA, M. NISHIDA, AND T. OZAWA. lihood estimation of the heterogeneity of substitution 1998. The complete nucleotide sequence of a snake rate among nucleotide sites. Mol. Biol. Evol. 12:546- (Dinodon semicarinatus) mitochondrial genome with 557. two identical control regions. Genetics 150:313-329. HASEGAWA,M.,H. KISHINO,ANDT.YANO. 1985. Dat- MARTIN, A. P., AND S.R. PALUMBI. 1993. Protein evo- ing the human-ape split by a molecular clock of mi- lution in different cellular environments: Cytochrome tochondrial DNA. J. Mol. Evol. 22:160-174. b in sharks and mammals. Mol. Biol. Evol. 10:873-891. HicKSON, R. E., C. SIMON, A. COOPER, G. S. SPICER, MAYHEW, W. W. 1964. Taxonomic status of California J. SULLIVAN, AND D. PENNY. 1996. Conserved se- populations of the lizard genus Uma. Herpetologica quence motifs, alignment, and secondary structure 20:170-183. 612 SYSTEMATIC BIOLOGY VOL. 49 MITTLEMAN, M. B. 1942. A summary of the iguanid SMITH, H. M. 1946. Handbook of lizards. Comstock, genus Urosaurus. Bull. Mus. Comp. Zool. Harvard Ithaca, New York. Uriiv. 91:106-176. SULLIVAN, J. 1996. Combining data with different dis- NAYLOR, G. J. P., AND W.M. BROWN. 1998. Amphioxus tributions of among-site rate variation. Syst. Biol. mitochondrial DNA, chordate phylogeny, and the 45:375-380. limits of inferencebased on comparisons of sequences. SULLIVAN, J., AND D. L. SwoFFORD. 1997. Are guinea Syst. Biol. 47:61-76. pigs rodents? The importance of adequate models NIELSEN, R. 1997. Site-by-site estimation of the rate of in molecular phylogenetics. J. Mammal. Evol. 4: substitution and the correlation of rates in mitochon- 77-86. drial DNA. Syst. Biol. 46:346-353. SULLIVAN, J., K. E. HOLSINGER, AND C. SIMON. 1995. NORRIS, K.S. 1958. The evolution and systematics of Among-site rate variation and phylogenetic analysis the iguanid genus Uma and its relation to the evolu- of 12S rRNA in sigmodontine rodents. Mol. Biol. Evol. tion of other North American desert reptiles. Bull. Am. 12:988-1001. Mus. Nat. Hist. 114:253-326. SULLIVAN, J., K. E. HOLSINGER, AND C. SIMON. 1996. NuNN, G. B., J. COOPER, P. JOUVENTIN, C. J. R. The effect of topology on estimates of among-site rate ROBERTSON, AND G.G. ROBERTSON. 1996. Evolu- variation. J. Mol. Evol. 42:308-312. tionary relationships among extant albatrosses (Pro- SwoFFORD,D.L.,G. J.OLSEN,P. J. WADDELL, AND D.M. ceUariiformes: Diomedeidae) established from com- HILLIS. 1996. Phylogenetic inference. Pages407-514 plete cytochrome-fo gene sequences. Auk 113:784-801. in Molecular systematics, 2nd edition (D. M. Hillis, C. NUNN,G.B.,ANDJ.CRACRAFT. 1996. Phylogeneticre- Moritz, and B. K. Mable, eds.). Sinauer, Sunderland, lationships among the major lineages of the birds-of- Massachusetts. paradise (Paradisaeidae) using mitochondrial DNA THOMPSON, J. D., D. G. HIGGINS, AND T. J. GIBSON. gene sequences. Mol. Phylogenet. Evol. 5:445-459. 1994. Clustal W: Improving the sensitivity of pro- PORTER, C. A., M. W. HAIDUK, AND K. DE QUEIROZ. gressive multiple sequence alignment through se- 1994. Evolution and phylogenetic significance of ri- quence weighting, position-specific gap penalties and bosomal gene location in chromosomes of squamate weight matrix choice. Nucleic Acids Res. 22:4673- reptiles. Copeia 1994:302-313. 4680. ROUGH, F. H. 1973. Uma inornata. Cat. Am. Amphib. WADDELL, P. J., AND D. PENNY. 1996. Evolutionary Reptiles 126:1-2. trees of apes and humans from DNA sequences. Pages ROUGH, F. H. 1974. Uma scoparia. Cat. Am. Amphib. 53-73 in Handbook of symbolic evolution (A. J. Lock Reptiles 155:1-2. and C. R. Peters, eds.). Clarendon Press, Oxford, Eng- ROUGH, F H. 1977. Uma notata. Cat. Am. Amphib. land. Reptiles 197:1-2. WAKELEY, J. 1996. The excess of transitions among RASSMANN, K. 1997. Evolutionary age of the Galapa- nucleotide substitutions: New methods of estimat- gos iguanas predates the age of the present Galapagos ing transition bias underscore its significance. Trends Islands. Mol. Phylogenet. Evol. 7:158-172. Ecol. Evol. 11:158-163. REEDER, T. W. 1995. Phylogenetic relationships WOLFE, A. C, P M. SHARP, AND W.-H. LI. 1989. Mu- among phrynosomatid lizards as inferred from mito- tation rates differ among regions of the mammalian chondrial ribosomal DNA sequences: Substitutional genome. NATURE 337:283-285. bias and information content of transitions relative to YANG, Z. 1994a. Estimating the pattern of nucleotide transversions. Mol. Phylogenet. Evol. 4:203-222. substitution. J. Mol. Evol. 39:105-111. REEDER, T.W., AND J.J.WIENS. 1996. Evolution of the YANG, Z. 1994b. Maximum likelihood phylogenetic lizard family Phrynosomatidae as inferred from di- estimation from DNA sequences with variable rates verse types of data. Herpetol. Monogr 10:43-84. over sites: Approximate methods. J. Mol. Evol. RICHARDS, C. M., AND W. S. MOORE. 1996. A phy- 39:306-314. logeny for the African treefrog family HyperoKidae YANG, Z. 1996a. Among-site rate variation and its im- based on mitochondrial rDNA. Mol. Phylogenet. pact on phylogenetic analyses. Trends Ecol. Evol. Evol. 5:522-532. 11:367-372. RoussET, F.,M. PELANDAKIS, AND M. SOLIGNOC. 1991. YANG, Z. 1996b. Phylogenetic analysis using parsi- Evolution of compensatory substitutions through G- mony and Kkelihood methods. J. Mol. Evol. 42:294- U intermediate state in Drosophila rRNA. Proc. Nati. 307. Acad. Sei. USA 88:10032-10036. YANG, Z. 1999. Phylogenetic Analysis by Maximum SAS INSTITUTE. 1990. SAS, version 6. SAS Institute, Likelihood (PAML), Version 2.0. University College Cary, North Carolina. London. SAVAGE, J. M. 1958. The iguanid lizard genera YANG, Z., N. GOLDMAN, AND A. FRIDAY. 1995a. Max- Urosaurus and Uta, with remarks on related groups. imum likelihood trees from DNA sequences: A pecu- Zoológica 43:41-54. liar statistical estimation problem. Syst. Biol. 44:384- SCHMIDT, K. P. 1953. A checklist of North American 399. amphibians and reptiles, 6th edition. American So- YANG, Z., I. J. LAUDER, AND H. J. LIN. 1995b. Molecu- ciety of Ichthyologists and Herpetologists, Chicago, lar evolution of the hepatitus B virus genome. J. Mol. Illinois. Evol. 41:587-596. SIMON, C, L. NIGRO, J. SULLIVAN, K. HOLSINGER, A. MARTIN, A. GRAPPUTO, A. FRANKE, AND C. MCINTOSH. 1996. Large differences in substitu- tional pattern and evolutionary rates of 12S ribosomal Received 12 January 1999; accepted 8 April 1999 RNA genes. Mol. Biol. Evol. 13:923-932. Associate Editor: }. Wiens