Further Analysis and a Reply

Home , Arrhyton, Atractaspis corpulenta, Bogertophis, Boiga cynodon, Boiga drapiezii, Cemophora

Herpetologica, 58(2), 2002, 270-275 ? 2002 by The Herpetologists'League, Inc.

SNAKE RELATIONSHIPS REVEALED BY SLOWLY-EVOLVING PROTEINS: FURTHER ANALYSIS AND A REPLY

RICHARD HIGHTON', S. BLAIR HEDGES2, CARLA ANN HASS2, AND HERNDON G. DOWLING3 'Department of Biology, University of Maryland, College Park, MD 20742, USA 2Department of Biology, The Pennsylvania State University, University Park, PA 16802, USA 3Rendalia Biologists, 1811 Rendalia Motorway, Talladega, AL 35160, USA

ABSTRACT: A reanalysis of our allozyme data (Dowling et al., 1996) for four slowly-evolving loci in 215 species of snakes by Buckley et al. (2000) concluded that because of ties in genetic distances our published UPGMA tree had "little resolution, indicating that these data are highly ambiguous regarding higher-level snake phylogeny." They also concluded that "the high degree of resolution in the published phenogram is an analytical artifact." Our study was intended to obtain information on lower-level relationships for the snake species that we had available, and it provided support for some current hypotheses of snake relationships at that level. Buckley et al. (2000) reached their conclusions because in their analysis they used only strict consensus trees and did not randomize the order of their input data. By randomizing data input order and using a majority-rule consensus tree, we show that there is considerable phylogenetic signal in our data. Key words: Allozymes; Genetic distances; Phylogeny; Serpentes; UPGMA trees

FIVE years ago, we published an allo- tle phylogenetic signal in our data set. zyme study of 215 species of snakes, about Thus, they claimed that the allozyme data 8% of living species (Dowling et al., 1996). are "highly ambiguous regarding higher- We selected four protein loci that ap- level snake taxonomy" and that the "high peared to be evolving relatively slowly be- degree of resolution in our published cause they had the fewest alleles (25-43) phenogram is an analytical artifact" of our among the loci that we initially surveyed. failure to consider alternate trees "implied With only four loci, our primary objective by tied distance values". We disagree with was not to obtain a robust phylogeny of these claims and will show that there is snakes (that would be a remarkable feat phylogenetic information in the allozyme for any allozyme data set surveying 215 data set. species). Instead, our purpose was to ob- In their reanalysis, Buckley et al. (2000) tain information on lower-level relation- removed 92 species because, in their ships among the species that we had avail- words, "identical taxa (D = 1, I = 0) [they able. Because of this, we entitled our study meant D = 0, I = 1] provide no additional a "preliminary survey". Although we point- information about the structure of the ed out a number of protein similarities tree." Actually, adding or deleting identical among higher-level taxa, we concluded taxa may affect tree topology. If identical that the interfamily-level topology of our OTUs are removed from a study, the re- UPGMA tree was "not well supported". maining OTUs may join more basal We also noted that we were unable to dis- branches in a different order. Moreover, tinguish relationships at the species level. the removal of 43% of our species resulted However, at the levels of genera, tribes, in the deletion of the most important re- subfamilies, and families, this approach sults of our study, which focused primarily provided some valuable information. We on lower-level relationships, not on higher- discussed the many cases of concordance level relationships as claimed by Buckley with current snake taxonomy at those lev- et al. (2000). Closely related species (and els. perhaps closely related genera) might be Recently, Buckley et al. (2000) reana- expected to be similar to each other at lyzed our data, constructed strict consen- slowly evolving loci because few substitu- sus trees, and concluded that there was lit- tions that result in changes in electropho-

270 June 2002] HERPETOLOGICA 271 retically detectable proteins are expected ogy. If Buckley et al. (2000) were interest- among recently diverged species. The re- ed in the effect of ties on the topology of sults confirmed our expectations. At the UMPGA trees based on our snake data lowest taxonomic level, there was no with- set, randomly reordering the taxa would in-group variation in 19 groups ranging in have clearly revealed whether or not the size from 2-30 species (identical alleles at tree topology was influenced by the pres- all four loci), and many other closely-re- ence of ties. lated species had identical genotypes at The majority-rule consensus method three of the four loci. However, all species should have been used instead of the strict were not clustered within their own lower- consensus method. As its name implies, level phylogenetic group; we pointed out the strict consensus method only resolves that at least five species were likely mis- groups that appear in all sampled trees placed on the tree (Dowling et al., 1996). (Nei and Kumar, 2000). Thus, if the par- Ties can be a problem with some meth- ticular group of species clustered at a node ods of phylogenetic analysis,including the is different in only one tree out of 10,000, one that we used (UPGMA). Tree-building the strict consensus tree will reflect this algorithms often break ties arbitrarilyand disagreement, resulting in a polytomy for therefore the order in which taxa occur in all involved branches. We believe that a the input file of the data used to calculate rare occurrence should not completely ne- the tree may influence the topology of the gate an overwhelming agreement among tree (Hedges et al., 1992; Hendy et al., the majority of the trees. For example, 1988). Buckley et al. (2000) acknowledged among the 39 species of natricines in our the possibility of this taxon-orderbias, yet study, I-values vary from 1.0 to 0 with they did not randomize the input order of many ties at all values (1.0, 0.75, 0.50, taxa. Instead they used a taxon order that 0.25, 0). The same is true for the reduced was similar to the one in our study to cal- number of eight species of natricines that culate their 9999 "equivalent"trees, the Buckley et al. used, except that there are maximum number for their software. no I-values of 1.0 since they removed all Their method did not alwaysdetect ties, species that were identical at all four loci. even with 9999 trees. An example is the By using their method of analysis, the na- arrangementof three genera (Carphophis, tricines form a polytomy with most or all Diadophis, and Farancia) on their consen- of the remaining groups of snakes in both sus trees. These genera clustered together of their consensus trees (70 of 73 groups in our tree in the arrangement (Carpho- in their Fig. 2 and all 31 groups in their phis (Diadophis, Farancia)) and in the Fig. 3). In contrast, UPGMA trees will al- same way in of all their 9999 UPGMA most always cluster all 39 species of natri- trees (their Fig. 2) since they do not show cines (or their reduced sample of eight a polytomy for the group. In their Fig. 3, species) as a monophyletic group. This is their 9999 trees all had a different topol- because all natricines in our study have an ogy ((Carphophis, Diadophis) Farancia) I-value to at least one other natricine indicating the input order had been re- -0.75, while none has an I-value to any versed (the I-value between the pair Car- non-natricine snake in the study >0.50 (all phophis and Diadophis is the same as that natricines differ from all species in other between Diadophis and Farancia). Using groups at two or more of the four loci). their methodology,they did not find an in- However, by using a strict consensus tree, dication of the tie involved in either of if only one tree in 10,000 clusters a single their two sets of 9999 trees in spite of the species of natricine in some other group fact that each of the two arrangements (or a single species of another group with- would be expected half the time. In our in the natricines), both groups will be majority-rule consensus tree (discussed placed in a polytomy, indicating no reso- below), the tie was no longer present be- lution of the groups. cause we included two species of Faran- Buckley et al. (2000) noted that many of cia; thus all 50 trees had the same topol- their trees agreed well with "traditional 272 HERPETOLOGICA [Vol. 58, No. 2

Clharina bottae Tropidophis haetianus Arrhyton funereus* Dartirsgtonia haetiana Tihamnodynastes strigilis Phitodryas burmeisteri Philodryas viridis Lamprophis fuliginosus 8 4 - | - Alsophis cantherigerus Antillophis parvidons 8 0 Arthyton exiguum Alsophis portoricensis Hydrodynastes gigas HHelico s angulatus _ 9 0 1 00 Helicops leopardinus 8 0 | -- Arrhyton taeniatum 84 so -C= Arrhytoncalli/aemum 58 100 Arrhytonlandoi Dipsas catesbyi Rhadinaeaflavlata _ 7 4 Liophis miliaris 100 F- ~~~~~~~~~~~~~Waglerophismerrami Lys/rophis dorbingi 74 n Xenodonseverus 10 Hypsirhynchusferox 1 8 4 _ ~ -- -- Uromacercatesbyi b _ 10 0 Uromacer frenatus la/tris dorsalis 1 0 0 t .. ,,- -Uromacer catesbyi a A ------Liophis poecilogyrus 84 Clelia rustica 9 0 1- - - Liophisviridis 90 Micruroides euryxanthus Naja naja 100 Ophiophagus hannah Bungarus fasciatus 100 Micrurus annellatus Micrurus dias/ema Dendrelaphis caudolineata Amphiesma sto/ata Reginarigida Reginaseptemvittata i 00^ w v - ~~~~~ ~Sinonarrix- annularis Chlonophis kirtlandii Nerodia cyclopion 100 Nerodia erythrogaster 100 Nerodia fasciata Nerodia harteri Nerodia compressicauda Nerodia rhombifer Nerodia sipedon Nerodia taxispilota Regina alleni Seminatrix pygaea Storeria occipitomaculata Thamnophis brachystoma Thamnophis but/eri Thamnophis chrysocepha/us 100 Thamnophis couchi Thamnophis elegans 62 Thamnophis eques Thamnophis marcianus Thamnophis melanogaster Thamnophis mendax Thamnophis ordinoides Thamnophis proximus Thamnophis radix Thamnophis rufipunctatus Thamnophis sauritus Thamnophis sirtalis Thamnophis sumicrasti Virginia striatula Virginia va/eriae Rhabdophis chrysargus Rhabdophis sp. 100 Xenochrophis flavipunctatus 76 - - Xenochrophispiscator 100 enohrophis punctulatus Pseustes poecilonotus 6845 _Tretanorhinus nigroluteus* O68 6 8 - Leptophis ahaetulla Dasypeltis scaber r - | ~~~~~Chironiuscinnamomeus :10 0 Chitoniusmulutiventris Mastigodryas bifossatus 68 Pseustes sulphurus Ptyas korros Salvadora grahami Zaocys carinatus Spalerosophis cliffordi 100 ---1001 nn -- Coluberven/romaculatus 10 0 | I - Gonyosoma oxycephalum Drymarchon corais Drymoluber dichrous Entechinus semicarinatus Leptophis mexicanus 50 I Chironius carinatus I 1~~~50 ~ ~ ~I hrnu p 1 Senticolis triaspis l 68 - I r Pituophis melanoleucus l l 68 | I I | Arizona elegans 100 Bogertophis subocularis 100 Cemophora coccinea Elaphe bimaculata Elaphe climacophora 100 Elaphe flavirufa Elaphegut/ata Elaphe moeliendorffi

FIG. 1.-Majority rule consensus tree based on 50 UPGMA trees using Cavalli-Sforza and Edward's chord distances with randomized data input. At each node, the percentage of the 50 trees that include all the species of the group is indicated. Five species that we indicated (Dowling et al., 1996) were misplaced are indicated by asterisks. The 50 trees were calculated using Joseph Felsenstein's PHYLIP program, and the consensus tree was modified to reduce nodes occurring in <50% of the trees to polytomies. June 2002] HERPETOLOGIQA 273

Elaphe obsoleta Elaphe quadrivirigata Elaphe quatuorlineata 100 Elaphe radiata Elaphe schrencki Elaphe situla Elaphe taeniura 100 Elaphe vulpina Lampropeltis getulus Lampropeltis mexicanus Rhinocheilus lecontei 66 56 100 Lampropeltis triargulum 0 Lampropeltis calligaster Elaphescalaris Ptyas mucosus Coluber constrictor LiochJorophis vernalis 100 lateralis 100 ~~~~~~~~~MasticophisElaphe dione Elaphe flavolineata Elaphe longissima Elaphe rufodorsata Tantilla coronata Opheodrys aestivus CaJamaria gervasii Rhamphiophis oxyrhynchus 8$4 10 0 Psammophis condenarus A -Carphophis amoena 84rD s g -- Diadophis punctatus luu_0 0 10 0 Farancia bacura L ioo~~~~~~~~0 1 00 Farancisaerytrogramma 56 Lycodon laoensis 5i 10B0 Boiga jaspida Philothamnus sp. Chrysopelea ornata 8 6 - Dinodon--- semicarinatum _ Dryocalamus trivirgatus 6 6 8 6 Ahaetullaprasins 100 S~~~~~~~oigacyanea 100 Boiga cynodon 8 6 Boiga dendrophila Boiga drapiezii Oxybelis fulgidus Boiga nigriceps 10n n -Boiga multimaculata Psammodynastes pulverensis Telescopus sp. Acrochordusjavanicus Enhydris boucourti _6 6 - - Enhydris enhydris 60 ____jEnhydrisjagori 10 0 _ v 100 Enhydris plumbea Enhydris chinensis tentacutatum 100 ~~~~~~~~~~~~~ErpetonHomalopsus buccata 6Biis arietans* Python reticulatus 10UU0 1: Python regius ______Boa constrictor 10 0 Epicrates striatus Tropidophis canus* Atractaspis corpulenta' 9 _ _ . Heterodon platirhinos - O o 100 n -Heterodon simus 9 8 1 ------X - - - - Phytlorhynchus decurtatus 9 8 ------e - 1 Oligodonmodestus Atractus trilineatus* ______Agkistrodon bilineatus -*Agkistrodon 100 100 contortrix l l | ~~~~~~~~~~~~~~~~~~~~~10UU0Aqkistrodon, piscivorus5 g ~~~~~~~~~~~~~~~~~~~~~~~~Sistrurus catenatus 10 0 _ l I I - CroCtaus lepidus Sistrurus ravus | 6 8 1 1 I 1 58 1 Crotalus pusillus I I I . - I I 10~~~~~~~~~~~~~~~~~~~~~0I Crotalus willardi I00 I I I " nr I I ICrotalus mitchelli Crotalus horridus Crotalus molossus I00 _ ^ I I ! _ Crotatus durissus 58 Crotatus atrox 0 0 | | ? ~~~~~~~~~~~~~~~~~~~~~~~Crotalusruber Crotatus catalinensis 10 0 1 1 | Crotatus adamanteus 56 Sistrurusmiliarius ~~~~~~~~~~~~~~Crotaius cerastes Crotalus viridis 68 | Trimeresurus| | " - - okinavensis l Crotalusscututatus 100 ~~~~100 100 Bothropsatrox l ~~~~~~~~~~~~~~~~~~~~~~~~~10 o-Afropoides nummifer 100 1 Trimeresurus kanburiensis Trimeresurus flavoviridis 100 100 Trimeresurus tokarensis 46_8 1 Hypnate hypnale Tropidolaemus wagteri 10 0 | ----.-- , - ::., :,; . - : Trimeresurus etegans Trimeresurus atbolabris Calloselasma rhodostoma - WI ri -" - Cerastes vipera 100 100 russeiii l 10 0 ------~~~~~~~~~~~~1 | |~~~00Daboia Psevdocerastes persicus Causus rhombeatus -00 - -Atheris- #--- squamigera ----- s vv r Typhtopsjamaicensis ------Ii I Typhiops hypomethes i100 Typhlops platycephalus 0.01 10 0 Typhlops richardi

FIG. 1.-Continued. 274 HERPETOLOGICA [Vol. 58, No. 2 snake taxonomy", but they still concluded jority rule consensus tree resulting from that our data are not "informative about this analysis shows that there is phyloge- the higher level phylogeny of snakes". Two netic signal in the data set, in spite of the factors, (1) the pattern of genetic variation numerous ties in the distance data. Al- in the data, and (2) the clustering of tra- though we did not design our study to ditional taxonomic groups together on evaluate higher-level relationships and such a high proportion of trees, make it never claimed the tree to be robust, the obvious that there is phylogenetic signal allozyme data may provide some useful in- present in this data set. The clustering to- formation at that level. gether of all species of so many traditional The UPGMA majority rule consensus groups on our UPGMA tree is not a tree united numerous species into clades chance event. The probability of all 39 at tribe, subfamily, or family levels, which species of natricines clustering together by often corresponded to traditional taxo- chance in our single published tree is so nomic groups; these were also present in low as to be infinitesimal, as is the clus- our original phenogram. Some examples tering of many other lower-level groups on are (1) all 54 of the species in 26 colubrine the tree. As we will show below, the same genera of the tribe Colubrini (seen in 68% 39 species of natricines also cluster togeth- of the trees), although the group also in- er on all 50 new trees, as do most of the cluded one misplaced dipsadine species other lower-level groups that we identi- (see discussion below); (2) all 39 species in fied. A majority-rule consensus tree would 11 genera of the natricine tribe Natricini have indicated the proportion of trees in (100%); (3) all 16 species of nine colubrine which the arrangement of species at each genera of the tribes Boigini and Philo- node occurs and indicated which groups thamni (86%); (4) all seven species of were affected by ties. three homalopsine genera of the tribe We reanalyzed our entire data set by Homalopsini (this group also includes an randomizing the input of taxa 50 times and obviously misplaced viperid) (100%); (5) a obtained an UPGMA tree for each taxon group of 28 species in 18 genera of the input order using Cavalli-Sforza and Ed- subfamily Xenodontinae (four additional wards (1967) chord distances. A majority- members of this subfamily are likely mis- rule consensus tree is shown in Fig. 1. If placed in this analysis-two are discussed ties were a major problem in our data set, in more detail below) (84%); (6) all six spe- the resulting trees would not be in agree- cies of five genera of the family Elapidae ment with our original tree and there (100%); (7) both species in one genus of would be a large number of polytomies (as the family Pythonidae (100%); (8) both in the two strict consensus trees in Buck- species in two genera of the subfamily ley et al., 2000). The tree we published Boinae (100%); (9) all 30 species in 10 vi- (Dowling et al., 1996:Figs. 1-5) had 121 perid genera of the subfamily Crotalinae nodes (including the 19 nodes for the (100%); (10) all four species in one genus identical sets of species). The majority-rule of the family Typhlopidae (100%). Our consensus tree has 105 nodes, indicating UPGMA tree also showed that a few that randomization of input order and ties groups of uncertain or controversial rela- in distance values have little effect on the tionships clustered with taxa with which topology of the tree for this data set. The they sometimes had been associated [e.g., tree in Fig. 1 is likely a better estimate of Oligodon with Phyllorhynchus (98%); and the relationships of these species than the Acrochordus with the homalopsines one in Dowling et al. (1996) because rare (100%)]. The tree also indicated the ex- groupings of species are eliminated and pected basal position of the family Typhlo- the involved nodes are collapsed into po- pidae (the only family to have no alleles in lytomies. The strict consensus trees cal- common with any other family), and in- culated by Buckley et al. (2000) had fewer dicated little similarity in all paired com- nodes, leading them to conclude that there parisons among families of snakes. was little resolution in our tree. The ma- The agreement in topology among the June 2002] HERPETOLOGICA 275

50 trees is high; more than half (60 of 105) tion. They mention that the thousands of of the nodes include the same grouping of "equivalent" trees that they generated also species in all 50 trees. The agreement seem to be in agreement with traditional would have been even higher if it were not taxonomy, yet they claim that discrepan- for the few obviously misplaced taxa. For cies among their trees "sum to a consid- example, we noted (Dowling et al., 1996) erable loss of resolution". It is true that the that two dipsadines were misplaced on our strict consensus trees that they published tree: Tretanorhinus nigroluteus clustered have a large loss of resolution. If their with the colubrines, and Atractus trilinea- thousands of UPGMA and parsimony tus clustered with the crotalines. These trees are in general agreement with tradi- two species clustered as sister species in tional hypotheses of snake relationships, 16 of the 50 trees in our reanalysis. Had then their use of strict consensus trees as they not been included in our study, seven a method of analysis would appear to have additional nodes on the path from Tretan- led them to the conclusion, which we dis- orhinus nigroluteus to Atractus trilineatus pute, that there is little resolution in our in Fig. 1 would have been in 100% agree- UPGMA tree. If they had used the major- ment on all 50 trees (instead of 68%), and ity-rule consensus tree approach and ran- another seven nodes on that same path domized taxon order for each tree to ad- would have much higher percentages of dress the problem of ties in genetic disagreement. We obtained only 50 trees be- tances, they would have retained the in- cause we had to randomize the input order formation that is present in the data set. separately for each tree since no available computer program can do this for 215 LITERATURE CITED taxa. The accuracy of the results should BUCKLEY, L., M. KEARNEY, AND K. DE QUEIROZ. not be greatly affected because of using 2000. Slowly evolving protein loci and higher-level snake phylogeny: a reanalysis. Herpetologica 56: only 50 trees. For example, the expected 324-332. agreement in the above example should be CAVALLI-SFORZA, L. L., AND A. W F. EDWARDS. 66.67% (there is a three-way tie of Tretan- 1967. Phylogenetic analysis: models and estimation orhinus nigroluteus to Atractus trilineatus, procedures. Evolution 21:550-570. DOWLING, H. G., C. A. HASS, S. B. HEDGES, AND R. Leptophis ahaetulla, and Pseustes poeci- HIGHTON. 1996. Snake relationships revealed by lonotus, but Tretanorhinus and Atractus slow-evolving proteins: a preliminary survey. Jour- will cluster together only when both are nal of Zoology, London 240:1-28. loaded consecutively, an event which HEDGES, S. B., S. KUMAR, K. TAMURA, AND M. STO- NEKING. 1992. Human origins and analysis of mi- would be expected one-third of the time). tochondrial DNA sequences. Science 255:737-739. Buckley et al. (2000) agreed that our HENDY, M. D., M. D. STEEL, D. PENNY, AND I. M. "published phenogram exhibits general HENDERSON. 1988. Pp. 355-362. In H. H. Bock agreement with traditional hypotheses (Ed.), Classification and Related Methods of Data about snake relationships". However, no Analysis. Elsevier, Amsterdam, The Netherlands. NEI, M., AND S. KUMAR. 2000. Molecular Evolution taxonomy of snakes has ever been widely and Phylogenetics. Oxford University Press, Ox- accepted because there is considerable ford, U.K. controversy in regard to the placement of Accepted: 2 October 2001 many groups at most levels of classifica- Associate Editor: Stephen Tilley

DATE OF PUBLICATION Herpetologica, Vol. 58, No. 1, was mailed 26 February 2002.