Slowly-Evolving Protein Loci and Higher-Level Snake Phylogeny A Reanalysis

Larry Buckley Maureen Kearney Kevin de Queiroz

Herpetologica Vol 56 No 3 (Sep 2000) pp 324-332

Stable URL

httplinksjstororgsicisici=0018-08312820000929563A33C3243ASPLAHS3E20CO3B2-Z

Herpetologica is currently published by Herpetologists League

Your use of the JSTOR archive indicates your acceptance of JSTORs Terms and Conditions of Use available athttpwwwjstororgabouttermshtml JSTORs Terms and Conditions of Use provides in part that unless you have obtainedprior permission you may not download an entire issue of a journal or multiple copies of articles and you may use content inthe JSTOR archive only for your personal non-commercial use

Please contact the publisher regarding any further use of this work Publisher contact information may be obtained athttpwwwjstororgjournalsherpetologistshtml

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printedpage of such transmission

The JSTOR Archive is a trusted digital repository providing for long-term preservation and access to leading academicjournals and scholarly literature from around the world The Archive is supported by libraries scholarly societies publishersand foundations It is an initiative of JSTOR a not-for-profit organization with a mission to help the scholarly community takeadvantage of advances in technology For more information regarding JSTOR please contact supportjstororg

httpwwwjstororgThu Feb 21 144611 2008

324 HERPETOLOGICA [Vol56 No 3

ronn SantiugoRio Piuntza 1830 n ~ KU 147071-72 249976 (cleared and stained) 249977-79 Bosque 147074 14707677 147079 de MazBn QCAZ 1666 Lagunas de Cajas 3850 n ~

Atelopzis carrikeri-COLOMBIA Departanlento MHNG 22584142 Cirbn Quimsacocha (p5ramo Magdalenu Sierra Nevada de Santa Marta Cuchilla sur del Cajas) 3700 m QCAZ 8835 Cebolleta ICN 32429 (CampS) Departa~netzto Guajira Atelopti8 igncscens-ECUADOR Prm-inamp Iirha-

Santa Marta mountains PBramo de Macotama hura Lagunas de Mojanda QCAZ 185M6 Prouitldu

(24303570 m elevation) UMMZ 48273 (paratype) Napo yiacachi QCAZ 27576 Pmciru-ia Pichirdw

Departumer~to del Chat Rio Guatapuri Sierra Ne- Pharno de Guamani La Virgen 3800-4200 in MHNG 241014 240995-100 227348 227361 227381-82 vada de Santa Marta USNM 123561-62 22738497 QCZ 266 (cleared and stained) Prmrit~c-ia Atelopus exiguus-ECUADOR Prouincia Azuay Gotopan-i Laguna de Lmpiopungo 4000 m QCAZ

Zunicuchu IV of Cuenca 3250 m SMF 4046 (lec- 254 385-236 8797 Zumbagua MHNG 238497-100 totype) 4047-51 3170-71 3916 (paralectotypes) 23851-10 Prouincius Gotopan-i-Napo ca 20 Knl (air- Laguna de Zunlcuchu (= Laguna Ltaviuco) 3200 in line) southeast of Latacunga on road from San hliguel KU 120381 (cleared and stained) 120385 (cleared de Salcedo to Lagnnas de Anteojos and towards the and stained) 120387 QCAZ 3744 10 krn W Cuenca east 01 01 S and 78 25 M approximately between QCAZ 4957 (cleared and stained) Rio Quinuas 15 3 2 W 8 0 0 m QCAZ 702 (neotype) 703-16 185758 km W Cuenca 3150 m QCAZ 3668 (12 tadpoles) Pruvincicz Chinamporux) 20 km N Riobamba QCAZ 9 QCAZ 3669 (1tadpole) Parqne Nacional Cajas 15- Pmindu Bolloar ca Chimborazo (4mbato-Guaranda 20 km W Cnenca 3200 m MHNC 249972-75 road) QCAZ 641

Herpetologrcn 56(3) 2000 324332 O 2000 b) The IIerpetologists League Inc

SLOWLY-EVOLVING PROTEIN LOCI AND HIGHER-LEVEL SNAKE PHYLOGENY A REANALYSIS

LARRYBUCKLEY~-lt KEARNEYMAUREEN AND KEVINDE QUEIROZ~ Bzology Department Rochester Instztute of Techncllogy Rochester NY 14623 USA

2Departnzent of Bzologzcal Sczences George Cliashzngton Unzzerzty Wnshzngton DC 20052 USA

lDcparttiwnt of 12rtebrate Zoology hTuttonal Museunz of Watural Hzstoy Smzthsot~zanInstitutzon Washt~~gtot~ DC 20560 USA

ABSTRACT IVe reanalyzed data from a recently published study of higher-level snake relation- ships based on four slowly-evolving protein loci The original study used phenetic clustering of genetic similarities and presented a single highly resolved tree Our reanalyses of these data reveal that the single published phenogram is only one of at least 10000 equivalent UPGMA phenograms the consensus of which is largely unresolved Additive distance analysis and character-based parsi- mony analysis of the data also yield little resolution indicating that these data are highly ambiguous regarding higher-level snake phylogeny The high degree of resolution in the published phenogram is an analytical artifact resulting from the failure to consider alternative trees implied by tied distance values which are numerous in the distance matrix derived from this particular data set Although the published phenogram exhibits general agreement with traditional hypotheses about snake re- lationships the same appears to be tnie for the thousands of equivalent phenograms discrepancies among which sum to a substantial loss of resolution Although the four loci sampled are evolving slowly relative to other commonly sunreyed protein loci they are nevertheless evolving too rapidly to be informative about the higher level phylogeny of snakes

Key words Allozymes Genetic distances Minintutn evolution Parsimony Phenetic clustering Phylogeny Serpentes

INa recent study of the higher-level re- the unweighted pair group method using lationships among snakes Dowling et al arithmetic averages ( U P G M A ) Their re- (1996) analyzed genetic similarities based sult was a single highly resolved tree (Fig on four protein-coding loci using avera e 1)that delineated 103 groups and exhib- linkage phenetic ~Iustering specifical fy ited general agreement with traditional

325 Sentember 20001 HERPETOLOGICA

Plesla3 pOBCll~ne T ~ e l a n o r ~ ~ n u s aoroiureus L e p I ~ h i s ahaelulla C h m n i ~ lcannafus cnronius n o n a m m e w Dwrnarchm coras

~~s~aYS LeptaphiE merlcaour

2p~gSodlerOSODhh Clifford Co i~ber ConSlilnOr Mdslicmnrs ia ls ial~s Plyas mucorus ~ i a p h s scalans opnscdrys aastms Tallla cwonala anrooa sieganr Lamproosit$ Callltparer Piiuoomrnsianaievcor SenhcoI lnaspls RnaMmnlS cnryrsrgur Rhabdaphs so XBnoOlrophls llapnCrafus Ampisrma rlorala ClonqhlE hlilldndi Sinonalnx anoulam Reglna riglda ReQloa seorsmwnara Chry~omlea m a l a Dendrelapnscaudornsara Caiamara ~ e m a s Rhamonioohs o ~ m y n c n v scalilhmnis amoena

DlddophlS pvoclalm

FalaOCIaabacYmPsamrnmynasrerouus saga mumaCiara Tewopus $4 Raga nigocspr OxyUBB fUlQdul Pmlenarno~sSO DlnMon semicannaturn ~ w m a l a m u s rrfvmarus Ahaelulfa oiarlna Bmgaipaiaspda L p o d m b-mnse Acrahordos jsvaoisus Enhydn~ amovnr Eonydnr sohydiir Ennydns lagori Enhydiis nlacuMrumonechinensis

HmalopSD bumala Lt ls ariatans LarnprWhrr ful io~nmur AlSopdS canmeiigeiur Anllllaphn pawlhonr ~ n h p o n exieuorn Helreapl dnpi ldlm Hydrmyoarier glga ~ ~ s w h r smnoncenrir Arihilon cariasrnum mnhyion laentarurn Philodryas bvrrnelslsrl Philodryas dS Dlosv caresby R h l d t m ~ a fldwlara ~ l r n n l s mlilaiir L y ~ l r w h l s doming1 WdpIBrWhlS memm Thamowrnarrsr rmotls Cislia rurlica LiDpOIS v~~~~~~Lh

Hyosiinyncnur rsrox Uromacerox~m~nchurUromdcei lienalvs

ldlllld dorsdl8~ Urmacsr carerby1 DdiliOQIOnlahaelmma

Nw oqaWqhawr nanmh

M I ~ i ~ i ~ ~ d s seuryxanhus W h o n iegivs P P o n rerlcujarus Boa consnctor Epicrates SdI AtraOlEOa comulenta Hereromn plarirhiom ~ s l s r o d mrimur TapampphS canvs

~ D m o oymmeshirAl raC l~ I I r l l inQdl~I

taxonomic groupings The authors argued that the slowly evolving nature of the loci and their ability to obtain traditionally rec- ognized groups supported the systematic utility of such data Here we show that the highly resolved tree presented by Dowling et al (1996) is an analytical artifact We reanalyze their data and show using sev- era1 different analytical approaches that those data are highly ambiguous support- ing thousands of trees equally well Our results call attention to the artifactual res- olution that can result from failing to con- sider ties in a distance matrix as well as the ambiguity of data consisting of rela- tively few loci each with numerous alleles Our results also call into question the slowly evolving nature of the four protein loci sampled relative to the time interval occupied by the diversification of snakes

METHODSAND RESULTS

The Data Dowling et al (1996) collected data on

four protein-coding loci (Acp-2 Ldh-2 Mdh-1 Pgm) each of which exhibited nu- merous alleles (42 43 29 and 25 respec- tively) They sampled 216 snake species each represented by a single specimen (two alleles) Degrees of resemblance among taxa were estimated using Neis (1972) coefficient of genetic identity (I) Because the similarity matrix was not pub- lished and because some of the programs we used required distances we first cal-

Hwnare hyonsie Tnmeresvrur sreQanr Trlmeissurur aloolabns

Trtmeiesurur kanbunsorlr A ~ ~ D ~ O ~ ~ B SnummiDi BMnrms arrox Ciolalus rcvlviafur Trmemsus Sblrwur ravvr Sbrrurur cslsnafur clolalus Isoldus C r ~ l i u sCBrasel CloldiuS indi l Cdllmeld~ma rhamplomd PTsUdoCBrd~les OemCUS C e r ~ l e smsra Ddb ld rYSSe arnem squammra c a m u mmbeam Armyion lunsreu9 Ulailna W d e Trmldophir ~ ~ B L ~ O U S Typhlop amaicenrir T ~ o n l w s ~h~di

F I G 1-Published phenogram o f Dowling et al (1996) with identical taxa removed and with Uroma- cer catesbyi (b) replaced by Uromacer oxyrhynchus in agreement with Table IV o f Dowling et al (1996) This dendrogram represents only the topology and not the branch lengths o f the published tree Ac-cording to the TAXAN output (supplied by C Hass)

culated Nei distance values from the allele frequency data (Dowling et al 1996 their Tables I 11 111 IV and V) using the GENDIST program in PHYLIP version 3 5 ~ (Felsenstein 1993) We then con-

t

three o f the polytomies in this tree are actually re- solved ( 1 ) Darlingtonia haetiana clusters with the el- apids (Naja naja through Micruroides euyxanthus) before clustering with the other xenodontines (Lam- prophis fuliginosus through Uromacer catesbyi) ( 2 ) Sistrurus ravus clusters with S catenatus and Cro- talus lepidus before clustering with C cerastes and C oiridis and (3) the group composed o f all taxa from Pseustes poecilonotus through Causus rhomhea- tus clusters first with Arrhyton funereus second with Trgihphis haetianus and third with Charina bottae

326 HERPETOLOGICA [I7ol 56 30 3

TABLE1-Summary of distance ( D = 1- I)values and their frequencies of occurrence for the snake allozyme data of Dowling et al (1996)Values in bold correspond with the five fractioils of alleles shared by specimens

that are homozygous at all four loci (ie 04 14 24 34 44)

Number of Number of Fract~onuf total occurrences Fractlon of total occ~~rrencesin ~ C C ~ I ~ ~ C I I L C S

111 matrix occurrences lnatnx ~~drn t l ca l (1dentic11~LYI

Distance value (all taxa) (all t a d twa ellrninatetl) elirnmatrd~

1OOOO 7716 0518618 4357 0571335 07500 4450 0299099 2019 0264752 05000 1199 0080589 4 70 0061631 08664 539 0036228 362 0047469 02500 403 0027087 105 0013769 05991 165 0011090 106 0013900 07327 140 0009410 102 0013373 00000 131 0008805 0 0000000 03318 35 0002352 22 0002885 00646 22 0001479 9 0001180 09286 14 0000941 14 0001836 04655 9 0000605 9 0001180 08557 8- 0000538 8 0001049 08571 I 0000470

I 0000918 02783 6 0000403 4 0000525 07113 3 0000336 5 0000656 05670 5 0000336 4 0000525 03571 3 0000202 3 0000393 01339 3 0000202 2 0000262 04286 3 0000202 3 0000393 07143 2 0000134 2 0000262 06429 2 0000134 2 0000262 03828 2 0000134 2 0000262 01982 2 0000134 2 0000262 02857 1 0000067 1 0000131 06914 1 0000067 1 0000131 04226 1 0000067 1 0000131 02285 1 0000067 1 0000131 00742 1 0000067 1 0000131 05714 1 0000067 1 0000131 00714 1 0000067 1 0000131

Totals 14878 1000000 7626 1000000

verted the Nei distances to Nei identities atus was corrected from 15 to 22 (original and the Nei identities to their complement data provided by C Hass and R Highton) distances using the equation D = 1 - Z Table 1 summarizes the distance values which yielded distances from 0 (all alleles and their frequencies of occurrence Of identical) to 1 (no alleles shared) for all the 14878 values (painvise comparisons) painvise comparisons We used D = 1 - in the matrix 14871 (999) are tied with Z rather than the standard Neis D (= -In at least one other value The great majority I)because distance clustering on D = 1 - of these ties (934 of the total values) I values (but not necessarily on Neis D represent values corresponding with the values) should give the same results as five possible fractions of alleles shared by similarity clustering on Neis I values Be- specimens that are homozygous for all loci fore calculating the identities and distanc- that is 4 4 (09) 34 (27) 24 (81) es we corrected two typographical errors 114 (299) and 04 (519) The remain- in the data tables published by Dowling et ing values (66) are accounted for by al (1996) the allele for Acp-2 in Tropi- comparisons involving heterozygotes Be- dophis canus was corrected from 03 to 42 cause identical taxa (D = 1I = 0) provide and the allele for P g m in Sistmrus caten- no additional information about the struc-

September 20001 HEHPEI

- - ~ l h s n i rquamtgsra camus nmbedl LlmprWhlr Iullginosvs A1smnr canmerlgeiur Anfilophir paiwtons annyton smguum nm~cops a o g ~ i a i ~ s Hwmynailes g9gas AIsWhls P ~ ~ O O C B ~ S ~ S Anhyton caiiiasmum Anhyton laenla Pniiodryar buimsisrsn Phlodiyar virldiJ Dlpras calesbp ~ h a d t n a ~ anaviiata LlOPhi miars LYI i iWhi l dorbmgl Waplsraphri merism TnamnWynartei rritgils Cislid NSliCd LlOPh3 Ylnds LiWhiS poeaiogyrus nypr~rnyncnvr rerox uiomacer Oxyrhyocnus UrOmdCB liendlus ldllns dYiEdli5 Uromacsr calesbyi Dalgl0na naeiana Naia nslaOphlrnhagus n m M h

MiCrYroideE BYrYXanthYS moon iegsur Python relicviarur Boa consrndor Eplcrarer srnalur AhdCldiplS COrPYlsnld neteimon p~arirhmos nseOmn rmur T i w i m o n a canus m g m m modsrrm Phyoynchus aecunatus I ~ C I U I I~~~EYI AOk1511040 PlSCiYOrY 1 L ~ Z t r M o nbinealr nypnaie nyonaie TnmeiesvrYS siegans TnmersSYiYS dlbOidbii Trwldaiaemus waphri Trimeiesurus r ~ k d i ~ n s ~ s TnmBrBiurY3 hanbuilensis AiiouoldB1 nYmmllei BDRiwS dlmX Crolalus ICYIY I I~Y I TiimeiesYrYS OkindvsnslS SiSlri r a Y Y I Sislwrus calenaluS crora1us Iapldus crOfa1I Ceraslsi CrOlaiY vindi Cufoseimma mMoStoma PIeUdoCerallBSPSrSfCYI C8relBC VipBrd Daaota i u ~ s e l i Annylon funereus T r w i d o o l r naslfanvr Cnanns bonas ~ y p h i w r bmaicenrr TyPhlaPs chard

FIG 2-Strict consensus tree for equivalent UPGMA trees resulting from alternative clustering pathways caused by ties in the distance matrix The consensus tree is based on 9999 equivalent UPGMA trees found by systematic tie-breaking using NTSYSpc An asterisk indicates an additional node that is col la~sed when the ~ubl ished tree of Dowling n

et al (1996)is included in the set of alternative trees

LOGICA 327

ture of the tree they were excluded from the remainder of our analyses The re-duced matrix has 124 taxa 7626 ~ainvise

I

comparisons and still contains 7619 values (999) that are tied with at least one oth- er value (Table 1) Once again the great majority of the total values (911) cor- respond with the five possible fractions of alleles shared by homozygotes

Phenetic Clustering Identical distance values (ie ties) im-

ply the existence of equivalent pheno- grams (trees generated by phenetic clus- tering) which are roughly analogous to equal minimum-length trees that are com- monly found in parsimony analyses (Hart 1983 de Queiroz and Good 1997) Be-cause there are numerous ties in the snake distance matrix we performed a UPGMA analysis with NTSYSpc 201d (Rohlf 1996) using the FIND option to search for tied (equivalent) trees with the tie toler- ance (TOL) set to the default value of 10-lo The search ended at 9999 trees the program maximum which was increased from 999 at our request by F J Rohlf The strict consensus of these tied trees (Fig 2) has only 34 nodes in contrast with the 103 nodes in the published tree (Fig 1) Given that the program reached its limit for tied trees the existence of additional equiva- lent trees and an even less resolved con- sensus tree is likely

Of the 34 nodes in the consensus tree (Fig 2) most (85) unite relatively small numbers of taxa (2-10) and reflect rela- tively shallow divergences relationships corresponding with early cladogenetic events are largely unresolved Based on the topology of the consensus tree none of the 9999 trees corresponds exactly with the published tree of Dowling e t al (1996) specifically our analysis failed to reproduce the placement of the group composed of Causus and Atheris in their tree (other seeming discrepancies between

which is the same additional node that is collapsed in the consensus tree for the 1000 equivalent UPGMA trees found by random tie-breaking using PAUP

328 HERPETOLOGICA [Vol56 No 3

our consensus tree and their published tree-involving the positions of Darling- tonia Tropidophis haetianus and Chari- nu-result from false poly-tomies in their tree see Fig 1 legend) Assuming that their tree represents an additional equiv- alent UPGMA tree then including it in the consensus would result in the loss of an additional node (marked with an aster- isk in Fig 2) Dowling et al (1996) used a different order of taxa in their datafile (supplied by R Highton) than that pub- lished in their tables we used the order in their tables This difference combined with the fact that our analysis did not find all of the equivalent UPGMA trees pre- sumably accounts for the minor discrep- ancy between the consensus of our 9999 UPGMA trees and their published tree

Because NTSYS generates equivalent phenograms by systematically exploring the alternative clustering pathways result- ing from ties we also used the random tie- breaking procedure in PAUP version 40b2 to examine alternative phenograms We performed 1000 UPGMA analyses each with a different random number as a tie-breaking seed producing 1000 equiv- alent UPGMA trees The strict consensus of these trees had 33 resolved nodes one node fewer than in the consensus of 10 times as many trees resulting from system- atic tie-breaking using NTSYS (Fig 2 ) and differing from that tree in the absence of the node uniting Causus plus Atheris with Acrochordus and the homalopsines Be- cause the number of nodes in the consen- sus tree for the analysis using random tie- breaking is less than the number of nodes in the consensus tree for the analysis using systematic tie-breaking there must be more than 9999 equivalent UPGMA trees In addition absence of the node uniting Causus and Atheris with Acrochordus and the homalopsines removes the only dis- crepancy between our UPGMA consensus tree and the tree obtained by Dowling et d (1996)

Additive Distance Analysis

When used to estimate phylogeny phe- netic clustering methods such as UPGMA carry an implicit assumption that rates of evolution among lineages are roughly con- stant (reviewed by de Queiroz and Good 1997) Moreover those methods do not employ an optimality criterion making evaluation of alternative topologies impos- sible Therefore we also analyzed the data using two alternative approaches additive distances and parsimony (reviewed by Swofford et al 1996) For the additive distance analysis we used the minimum evolution criterion in PAUP test version 40b2 A heuristic search was performed on the distance matrix obtaining the start- ing tree by neighbor-joining and searching for shorter trees using tree-bissection-re- connection (TBR) branch swapping We saved all trees shorter than the initial neighbor-joining tree (length = 2854687) which was itself shorter than the published UPGMA tree (length = 2887127) The search was terminated at 233363 trees but even more such trees presumably ex- ist because the search was not completed The strict consensus of these 233363 trees (not shown) has only seven resolved nodes each of which unites only two terminal taxa Thus under the minimum evolution criterion there are hundreds of thousands of trees shorter than both the neighbor joining tree and the published tree one or more of which contradict all but seven of the nodes in the published tree

A second minimum evolution analysis was performed once again obtaining the starting tree by neighbor-joining and using TBR branch swapping but this time keep- ing only the best trees This analysis yield- ed six optimal trees (length = 2817102) the strict consensus of which (not shown) is highly resolved (120 nodes) Much of that resolution however is of questionable significance The smallest distance in the data matrix (00646 Table 1)corresponds

Frc 3-Strict consensus of six optimal adchtive distance trees (minimum evolution criterion) with branches of length less than 00646 collapsed (see text for explanation)

329 Sentember 20001 HERPETOLOGICA

Opheodrys aerl iws Tanllfla lacoronala Anlona ele~eegans Lampropeals calllgaster PIUOP~Smelanolaucus Senlicons Inaspis Rhabdophls chryrargus Rhabdophls sp Xe00~hioph1sllaWpUnCla1US Amphresma slolala Clonophis klnland Sinonalrix annofaris Rspna ngrda Regina seplemvirtata

- - - Psammodynasles pulvarensls Boiga muntmaculaia Tmesc~possp

Enhydris bccourlr Enhydris enhydns Hmalops1s buccala BIIS anelans

-~ n h y t b nexiouvm

I Al~ophlscanthengeru~Hsdcop angulalos Hydrwnastes glgas

I Alsophis pononcensis Anhyion callilaemum Armyton laenlarum Ph8lodryas burmetsterl Ph i lod iya~ vrndls

1 Dlpsas caresby8Rhadlnaea Rawlala Llophls rnlbarts Ly~trophlsdorbinoi Wagleroph8s msriemi Lrophls p o e o l ~ y r ~ ~ Thamnodynasles slrigilis

I CleBa wsrica Llophis vrrldis Hypsrrhynchus lsrox Uromacer oxyrtiynchus Urornacer frenalus talliis dorsals Uiomacer caresbyr Naia oaiaI Ophqhagus hannah -I Mlcrurordes suryxanlhus Darltnglonia haet~ana Python regus Pyihon rettcufarus Boa conslncror Epncrales srnalus Helerodon plalirhinosHeteiWon sirnus Tropide9hls canus a i g m m modssfus PhyllOrhynCh~S deCUifalUS Agkisfrodon P ~ C W O N S Agh~slrodonbtiinealus Hypnale hypnaisTnmeresorus eteoans TnrnereSurUS albolabns

Caflwelasma f i o d m l m a PsaudoCerastespemicus Dabom msselii CerasIeS Yipera Alhens squamrgera C~USUSrhmbealus Aiihyian funereus Channa bonae Tropdoph~haellanus Typhlms iarnaicenscs Typhiops nchaidl

with the difference between taxa that pos- sess the same single allele at each locus except for the presence of a second allele in one of the taxa at one of the four loci (because the taxa are represented by single specimens this difference corresponds with the difference between an individual that is homozygous for all four loci and one that is homozygous for the same allele for three of those loci but heterozygous for the fourth with one of the two alleles of the heterozygote shared with the allele in the homozygote) If this distance (00646) is taken as the smallest distance that has a meaningful interpretation in terms of al- lelic change and all branches shorter than it are collapsed the resulting tree (Fig 3) is considerably less resolved (55 nodes) and exhibits little deep structure

Parsimony Analysis We performed a parsimony analysis of

the data using PAUP 40b2 treating the loci as characters and alleles as unordered states (eg Buth 1984 Mickevich and Mitter 1983) Multistate taxa (heterozy-gotes) were treated as polymorphic In contrast to the phenetic clustering and ad- ditive distance analyses in which only identical taxa were eliminated and the data sets consisted of 124 taxa the parsimony analysis was performed on a reduced data set of 95 taxa Both identical taxa and taxa that differed from others in the set only by possessing unique (and thus parsimony uninformative) alleles were eliminated Because of the large numbers of character states (gt32) in two of the characters we used a computer with a 64 bit Alpha pro- cessor to perform a heuristic search ob- taining the starting tree using simple step- wise addition searching for shorter trees using TBR branch swapping with the maximum number of trees set to 150000 The two species of Typhlops were desig- nated as outgroups The heuristic search yielded the maximum of 150000 equally most parsimonious trees (length = 144) the strict consensus of which (not illus- trated) was completely unresolved

DISCUSSION Our results indicate that the allozyme

data published by Dowling et al (1996)

)LOGICA IVol 56 No 3

are highly ambiguous concerning the high- er-level phylogeny of snakes This ambi- guity exists regardless of whether the data are analyzed using phenetic clustering ad- ditive distance methods or character-based ~arsimonv methods Under all three approamphes thousands of trees explain the data equally well or nearly so and the consensus of those trees exhibits little to no resolution particularly concerning the deeper nodes Given the ambiguity of the data it is worthwhile to consider first how Dowling et al (1996) obtained a single highly resolved tree and second why that tree exhibits general agreement with tra- ditional snake taxonomy

A single highly resolved tree presum- ably was obtained because the original analvsis did not reveal the existence of al- ternative (equivalent) trees Identical val-

L

ues in a distance matrix imply the exis- tence of alternative clustering pathways and thus alternative trees However many software im~lementations of UPGMA and related methods including the one in the TAXAN package used by Dowling et al (1996) do not explicitly reveal the exis- tence of alternative trees Instead ties are broken arbitrarily During each cycle in the clustering sequence if two or more taxa are equally (and minimally) distant from a third taxon the algorithm arbitrari- ly clusters one of these equally distant pairs first typically based on the input or- der of the taxa (Hart 1983 see also Farris et al 1996) The alternative clustering se- quence which may result in a different branching pattern (ie a different tree) is not explored Therefore the tree pro-duced mav be onlv one member of a set of equivalknt tree For the data analyzed in this paper there are both thousands of ties and thousands of equivalent trees

The reason that the single tree of Dowl- ing et al (1996) exhibits general agree- ment with traditional ideas about snake taxonomy is that the equivalent trees do not differ radically from one another so that any arbitrarily selected member of the set would probably agree more or less with traditional taxonomv To demonstrate

i

this we measured the dissimilarity be- tween the published tree and the 1000

- -

September 20001 HERPETOLOGICA 33 1

equivalent UPGMA trees found by ran-domly breaking ties using the symmetric difference metric of Penny and Hendy (1985) as implemented in PAUP version 40b2 Symmetric dfferences ranged from 77-141 (0318-0583) We then found the tree that differed most from the ~ublished tree and compared it with the hublished tree in terms of congruence with the tra- ditional taxonomy of snakes as described by Dowling et al (1996) and Dowling and Duellman (1978)

Even this maximallv different tree agreed reasonably closeamp with traditional snake taxonomy and more importantly it did not seem to agree any less well than the published tree Dowling et al (1996 their Table VI) recognized 30 groups based on their tree In 17 cases the groups were identical in composition to groups on the maximally different tree which there- fore matched tradtional taxonomv eauallv

i l i

well (or poorly) in three cases the group on their tree agreed better with traditional taxonomy in five cases the group on the most different tree agreed better and in

u

five other cases relative agreement was equivocal either because both trees matched traditional taxonomy poorly (but not identicallv) or because each tree

2

matched better in some resDects and worse in others Because the tree that is most different from the published tree fits traditional snake taxonomy reasonably well it seems reasonable to expect that the same will be true of most if not all of the equivalent trees everth he less even minor dilfferences among individual trees can sum to a high lampel of ambiguity when there are many equivalent trees This ap- pears to be the situation with the snake allozvme data for which there are thou- sands of equivalent trees and the consen- sus tree (Fig 2) is poorly resolved

The ambiguity of the snake allozyme data reanalyzed here is presumably related to the rate of evolution of the four loci surveyed which appears to be too rapid for the question being addressed Those loci have been characterized as slowly-evolving and they may indeed be evolving slowly relative to other loci commonly sur- veyed using protein electrophoresis~ev-

ertheless they are evolving relatively rap- idly in the context of the higher-level phy- logeny of snakes Rapid evolution is man- ifested in the large numbers of alleles (25 to 43) at each locus and in the fact that more than half of the entries in the dis- tance matrix exhibit the maximum possible distance value of 100 (Table l)indicating that the taxa share no alleles at any of the four loci The loci in question may be use- ful for analyzing relationships within small- er clades of snakes but they are highly am- biguous concerning the relationships with- in snakes as a whole particularly those re- lated to the deeper divergences in the history of that clade

Acknotc1edgments-We thank K Highton and C A Hass for generously providing original data and results F J Kohlf for assistance with NTSYSpc in- cluding increasing the tied-tree limit of the program D L Swofford for providing access to test versions of PAUP and J C LVilgenbusch and J A McGuire for help nith analyses performed on 64 bit compnt- ers

BUTH D G 1984 The application of electrophoretic data in systematic studies Annual Review of Ecol- ogy and Systematics 15501-522

DE QUEIROZK XD D A GOOD 1997 Phenetic clustering in biology a critique Quarterly Review of Biology 723-30

UOWLINGH G AND w7E DUELLMAN 1978 Sys- tematic herpetology a synopsis of families and higher categories HISS Publications New York New York USA

~~OVLISG AND RH G C A HASS S B HEDGES HIGHTON 1996 Snake relationships revealed by slow-evolving proteins a prelimirlap suwey Jol~r- nal of Zoolog) London 240 1-28

FRRISJ S L7 A ALBERT hl KXLLERSJOD LIP- SCOblB AND A G KLUGE 1996 Parsimony jack- knifing outperforms neighbor-joining Claclistics 12 99-124

FELSEXSTEIN 1993 PHYLIP ~ e r 3 5 Depart- J ment of Genetics Universit) of Vashington Seat- tle Washington USA

HART G 1983 The occurrence of multiple UPGhIA phenograms Pp 254-258 In J Felsenstein (Ed) Ni~~nericalTaxonomy Springer-L7erlag Berlin Ger- many

~IIcKEICHM F AND C MITTER 1983 Evolution- ary patterns in allozyme data a systematic ap-proach Pp 169-176 In N Platnick and A Funk (Eds) Advances in Cladistics L701 2 Columbia University Press Nev York New York USA

NEIM 1972 Genetic distance between populations American Naturalist 106283-292

PEKNYD AKD M D HENDY 1985 The use of tree comparison metrics Systematic Zoolog 3475-82

326 HERPETOLOGICA [I7ol 56 30 3

TABLE1-Summary of distance ( D = 1- I)values and their frequencies of occurrence for the snake allozyme data of Dowling et al (1996)Values in bold correspond with the five fractioils of alleles shared by specimens

that are homozygous at all four loci (ie 04 14 24 34 44)

Number of Number of Fract~onuf total occurrences Fractlon of total occ~~rrencesin ~ C C ~ I ~ ~ C I I L C S

111 matrix occurrences lnatnx ~~drn t l ca l (1dentic11~LYI

Distance value (all taxa) (all t a d twa ellrninatetl) elirnmatrd~

1OOOO 7716 0518618 4357 0571335 07500 4450 0299099 2019 0264752 05000 1199 0080589 4 70 0061631 08664 539 0036228 362 0047469 02500 403 0027087 105 0013769 05991 165 0011090 106 0013900 07327 140 0009410 102 0013373 00000 131 0008805 0 0000000 03318 35 0002352 22 0002885 00646 22 0001479 9 0001180 09286 14 0000941 14 0001836 04655 9 0000605 9 0001180 08557 8- 0000538 8 0001049 08571 I 0000470

I 0000918 02783 6 0000403 4 0000525 07113 3 0000336 5 0000656 05670 5 0000336 4 0000525 03571 3 0000202 3 0000393 01339 3 0000202 2 0000262 04286 3 0000202 3 0000393 07143 2 0000134 2 0000262 06429 2 0000134 2 0000262 03828 2 0000134 2 0000262 01982 2 0000134 2 0000262 02857 1 0000067 1 0000131 06914 1 0000067 1 0000131 04226 1 0000067 1 0000131 02285 1 0000067 1 0000131 00742 1 0000067 1 0000131 05714 1 0000067 1 0000131 00714 1 0000067 1 0000131

Totals 14878 1000000 7626 1000000

verted the Nei distances to Nei identities atus was corrected from 15 to 22 (original and the Nei identities to their complement data provided by C Hass and R Highton) distances using the equation D = 1 - Z Table 1 summarizes the distance values which yielded distances from 0 (all alleles and their frequencies of occurrence Of identical) to 1 (no alleles shared) for all the 14878 values (painvise comparisons) painvise comparisons We used D = 1 - in the matrix 14871 (999) are tied with Z rather than the standard Neis D (= -In at least one other value The great majority I)because distance clustering on D = 1 - of these ties (934 of the total values) I values (but not necessarily on Neis D represent values corresponding with the values) should give the same results as five possible fractions of alleles shared by similarity clustering on Neis I values Be- specimens that are homozygous for all loci fore calculating the identities and distanc- that is 4 4 (09) 34 (27) 24 (81) es we corrected two typographical errors 114 (299) and 04 (519) The remain- in the data tables published by Dowling et ing values (66) are accounted for by al (1996) the allele for Acp-2 in Tropi- comparisons involving heterozygotes Be- dophis canus was corrected from 03 to 42 cause identical taxa (D = 1I = 0) provide and the allele for P g m in Sistmrus caten- no additional information about the struc-

September 20001 HEHPEI

- - ~ l h s n i rquamtgsra camus nmbedl LlmprWhlr Iullginosvs A1smnr canmerlgeiur Anfilophir paiwtons annyton smguum nm~cops a o g ~ i a i ~ s Hwmynailes g9gas AIsWhls P ~ ~ O O C B ~ S ~ S Anhyton caiiiasmum Anhyton laenla Pniiodryar buimsisrsn Phlodiyar virldiJ Dlpras calesbp ~ h a d t n a ~ anaviiata LlOPhi miars LYI i iWhi l dorbmgl Waplsraphri merism TnamnWynartei rritgils Cislid NSliCd LlOPh3 Ylnds LiWhiS poeaiogyrus nypr~rnyncnvr rerox uiomacer Oxyrhyocnus UrOmdCB liendlus ldllns dYiEdli5 Uromacsr calesbyi Dalgl0na naeiana Naia nslaOphlrnhagus n m M h

MiCrYroideE BYrYXanthYS moon iegsur Python relicviarur Boa consrndor Eplcrarer srnalur AhdCldiplS COrPYlsnld neteimon p~arirhmos nseOmn rmur T i w i m o n a canus m g m m modsrrm Phyoynchus aecunatus I ~ C I U I I~~~EYI AOk1511040 PlSCiYOrY 1 L ~ Z t r M o nbinealr nypnaie nyonaie TnmeiesvrYS siegans TnmersSYiYS dlbOidbii Trwldaiaemus waphri Trimeiesurus r ~ k d i ~ n s ~ s TnmBrBiurY3 hanbuilensis AiiouoldB1 nYmmllei BDRiwS dlmX Crolalus ICYIY I I~Y I TiimeiesYrYS OkindvsnslS SiSlri r a Y Y I Sislwrus calenaluS crora1us Iapldus crOfa1I Ceraslsi CrOlaiY vindi Cufoseimma mMoStoma PIeUdoCerallBSPSrSfCYI C8relBC VipBrd Daaota i u ~ s e l i Annylon funereus T r w i d o o l r naslfanvr Cnanns bonas ~ y p h i w r bmaicenrr TyPhlaPs chard

FIG 2-Strict consensus tree for equivalent UPGMA trees resulting from alternative clustering pathways caused by ties in the distance matrix The consensus tree is based on 9999 equivalent UPGMA trees found by systematic tie-breaking using NTSYSpc An asterisk indicates an additional node that is col la~sed when the ~ubl ished tree of Dowling n

et al (1996)is included in the set of alternative trees

LOGICA 327

ture of the tree they were excluded from the remainder of our analyses The re-duced matrix has 124 taxa 7626 ~ainvise

I

comparisons and still contains 7619 values (999) that are tied with at least one oth- er value (Table 1) Once again the great majority of the total values (911) cor- respond with the five possible fractions of alleles shared by homozygotes

Phenetic Clustering Identical distance values (ie ties) im-

ply the existence of equivalent pheno- grams (trees generated by phenetic clus- tering) which are roughly analogous to equal minimum-length trees that are com- monly found in parsimony analyses (Hart 1983 de Queiroz and Good 1997) Be-cause there are numerous ties in the snake distance matrix we performed a UPGMA analysis with NTSYSpc 201d (Rohlf 1996) using the FIND option to search for tied (equivalent) trees with the tie toler- ance (TOL) set to the default value of 10-lo The search ended at 9999 trees the program maximum which was increased from 999 at our request by F J Rohlf The strict consensus of these tied trees (Fig 2) has only 34 nodes in contrast with the 103 nodes in the published tree (Fig 1) Given that the program reached its limit for tied trees the existence of additional equiva- lent trees and an even less resolved con- sensus tree is likely

Of the 34 nodes in the consensus tree (Fig 2) most (85) unite relatively small numbers of taxa (2-10) and reflect rela- tively shallow divergences relationships corresponding with early cladogenetic events are largely unresolved Based on the topology of the consensus tree none of the 9999 trees corresponds exactly with the published tree of Dowling e t al (1996) specifically our analysis failed to reproduce the placement of the group composed of Causus and Atheris in their tree (other seeming discrepancies between

which is the same additional node that is collapsed in the consensus tree for the 1000 equivalent UPGMA trees found by random tie-breaking using PAUP

328 HERPETOLOGICA [Vol56 No 3

our consensus tree and their published tree-involving the positions of Darling- tonia Tropidophis haetianus and Chari- nu-result from false poly-tomies in their tree see Fig 1 legend) Assuming that their tree represents an additional equiv- alent UPGMA tree then including it in the consensus would result in the loss of an additional node (marked with an aster- isk in Fig 2) Dowling et al (1996) used a different order of taxa in their datafile (supplied by R Highton) than that pub- lished in their tables we used the order in their tables This difference combined with the fact that our analysis did not find all of the equivalent UPGMA trees pre- sumably accounts for the minor discrep- ancy between the consensus of our 9999 UPGMA trees and their published tree

Because NTSYS generates equivalent phenograms by systematically exploring the alternative clustering pathways result- ing from ties we also used the random tie- breaking procedure in PAUP version 40b2 to examine alternative phenograms We performed 1000 UPGMA analyses each with a different random number as a tie-breaking seed producing 1000 equiv- alent UPGMA trees The strict consensus of these trees had 33 resolved nodes one node fewer than in the consensus of 10 times as many trees resulting from system- atic tie-breaking using NTSYS (Fig 2 ) and differing from that tree in the absence of the node uniting Causus plus Atheris with Acrochordus and the homalopsines Be- cause the number of nodes in the consen- sus tree for the analysis using random tie- breaking is less than the number of nodes in the consensus tree for the analysis using systematic tie-breaking there must be more than 9999 equivalent UPGMA trees In addition absence of the node uniting Causus and Atheris with Acrochordus and the homalopsines removes the only dis- crepancy between our UPGMA consensus tree and the tree obtained by Dowling et d (1996)

Additive Distance Analysis

When used to estimate phylogeny phe- netic clustering methods such as UPGMA carry an implicit assumption that rates of evolution among lineages are roughly con- stant (reviewed by de Queiroz and Good 1997) Moreover those methods do not employ an optimality criterion making evaluation of alternative topologies impos- sible Therefore we also analyzed the data using two alternative approaches additive distances and parsimony (reviewed by Swofford et al 1996) For the additive distance analysis we used the minimum evolution criterion in PAUP test version 40b2 A heuristic search was performed on the distance matrix obtaining the start- ing tree by neighbor-joining and searching for shorter trees using tree-bissection-re- connection (TBR) branch swapping We saved all trees shorter than the initial neighbor-joining tree (length = 2854687) which was itself shorter than the published UPGMA tree (length = 2887127) The search was terminated at 233363 trees but even more such trees presumably ex- ist because the search was not completed The strict consensus of these 233363 trees (not shown) has only seven resolved nodes each of which unites only two terminal taxa Thus under the minimum evolution criterion there are hundreds of thousands of trees shorter than both the neighbor joining tree and the published tree one or more of which contradict all but seven of the nodes in the published tree

A second minimum evolution analysis was performed once again obtaining the starting tree by neighbor-joining and using TBR branch swapping but this time keep- ing only the best trees This analysis yield- ed six optimal trees (length = 2817102) the strict consensus of which (not shown) is highly resolved (120 nodes) Much of that resolution however is of questionable significance The smallest distance in the data matrix (00646 Table 1)corresponds

Frc 3-Strict consensus of six optimal adchtive distance trees (minimum evolution criterion) with branches of length less than 00646 collapsed (see text for explanation)

329 Sentember 20001 HERPETOLOGICA

Opheodrys aerl iws Tanllfla lacoronala Anlona ele~eegans Lampropeals calllgaster PIUOP~Smelanolaucus Senlicons Inaspis Rhabdophls chryrargus Rhabdophls sp Xe00~hioph1sllaWpUnCla1US Amphresma slolala Clonophis klnland Sinonalrix annofaris Rspna ngrda Regina seplemvirtata

- - - Psammodynasles pulvarensls Boiga muntmaculaia Tmesc~possp

Enhydris bccourlr Enhydris enhydns Hmalops1s buccala BIIS anelans

-~ n h y t b nexiouvm

I Al~ophlscanthengeru~Hsdcop angulalos Hydrwnastes glgas

I Alsophis pononcensis Anhyion callilaemum Armyton laenlarum Ph8lodryas burmetsterl Ph i lod iya~ vrndls

1 Dlpsas caresby8Rhadlnaea Rawlala Llophls rnlbarts Ly~trophlsdorbinoi Wagleroph8s msriemi Lrophls p o e o l ~ y r ~ ~ Thamnodynasles slrigilis

I CleBa wsrica Llophis vrrldis Hypsrrhynchus lsrox Uromacer oxyrtiynchus Urornacer frenalus talliis dorsals Uiomacer caresbyr Naia oaiaI Ophqhagus hannah -I Mlcrurordes suryxanlhus Darltnglonia haet~ana Python regus Pyihon rettcufarus Boa conslncror Epncrales srnalus Helerodon plalirhinosHeteiWon sirnus Tropide9hls canus a i g m m modssfus PhyllOrhynCh~S deCUifalUS Agkisfrodon P ~ C W O N S Agh~slrodonbtiinealus Hypnale hypnaisTnmeresorus eteoans TnrnereSurUS albolabns

Caflwelasma f i o d m l m a PsaudoCerastespemicus Dabom msselii CerasIeS Yipera Alhens squamrgera C~USUSrhmbealus Aiihyian funereus Channa bonae Tropdoph~haellanus Typhlms iarnaicenscs Typhiops nchaidl

with the difference between taxa that pos- sess the same single allele at each locus except for the presence of a second allele in one of the taxa at one of the four loci (because the taxa are represented by single specimens this difference corresponds with the difference between an individual that is homozygous for all four loci and one that is homozygous for the same allele for three of those loci but heterozygous for the fourth with one of the two alleles of the heterozygote shared with the allele in the homozygote) If this distance (00646) is taken as the smallest distance that has a meaningful interpretation in terms of al- lelic change and all branches shorter than it are collapsed the resulting tree (Fig 3) is considerably less resolved (55 nodes) and exhibits little deep structure

Parsimony Analysis We performed a parsimony analysis of

the data using PAUP 40b2 treating the loci as characters and alleles as unordered states (eg Buth 1984 Mickevich and Mitter 1983) Multistate taxa (heterozy-gotes) were treated as polymorphic In contrast to the phenetic clustering and ad- ditive distance analyses in which only identical taxa were eliminated and the data sets consisted of 124 taxa the parsimony analysis was performed on a reduced data set of 95 taxa Both identical taxa and taxa that differed from others in the set only by possessing unique (and thus parsimony uninformative) alleles were eliminated Because of the large numbers of character states (gt32) in two of the characters we used a computer with a 64 bit Alpha pro- cessor to perform a heuristic search ob- taining the starting tree using simple step- wise addition searching for shorter trees using TBR branch swapping with the maximum number of trees set to 150000 The two species of Typhlops were desig- nated as outgroups The heuristic search yielded the maximum of 150000 equally most parsimonious trees (length = 144) the strict consensus of which (not illus- trated) was completely unresolved

DISCUSSION Our results indicate that the allozyme

data published by Dowling et al (1996)

)LOGICA IVol 56 No 3

are highly ambiguous concerning the high- er-level phylogeny of snakes This ambi- guity exists regardless of whether the data are analyzed using phenetic clustering ad- ditive distance methods or character-based ~arsimonv methods Under all three approamphes thousands of trees explain the data equally well or nearly so and the consensus of those trees exhibits little to no resolution particularly concerning the deeper nodes Given the ambiguity of the data it is worthwhile to consider first how Dowling et al (1996) obtained a single highly resolved tree and second why that tree exhibits general agreement with tra- ditional snake taxonomy

A single highly resolved tree presum- ably was obtained because the original analvsis did not reveal the existence of al- ternative (equivalent) trees Identical val-

L

ues in a distance matrix imply the exis- tence of alternative clustering pathways and thus alternative trees However many software im~lementations of UPGMA and related methods including the one in the TAXAN package used by Dowling et al (1996) do not explicitly reveal the exis- tence of alternative trees Instead ties are broken arbitrarily During each cycle in the clustering sequence if two or more taxa are equally (and minimally) distant from a third taxon the algorithm arbitrari- ly clusters one of these equally distant pairs first typically based on the input or- der of the taxa (Hart 1983 see also Farris et al 1996) The alternative clustering se- quence which may result in a different branching pattern (ie a different tree) is not explored Therefore the tree pro-duced mav be onlv one member of a set of equivalknt tree For the data analyzed in this paper there are both thousands of ties and thousands of equivalent trees

The reason that the single tree of Dowl- ing et al (1996) exhibits general agree- ment with traditional ideas about snake taxonomy is that the equivalent trees do not differ radically from one another so that any arbitrarily selected member of the set would probably agree more or less with traditional taxonomv To demonstrate

i

this we measured the dissimilarity be- tween the published tree and the 1000

- -

September 20001 HERPETOLOGICA 33 1

equivalent UPGMA trees found by ran-domly breaking ties using the symmetric difference metric of Penny and Hendy (1985) as implemented in PAUP version 40b2 Symmetric dfferences ranged from 77-141 (0318-0583) We then found the tree that differed most from the ~ublished tree and compared it with the hublished tree in terms of congruence with the tra- ditional taxonomy of snakes as described by Dowling et al (1996) and Dowling and Duellman (1978)

Even this maximallv different tree agreed reasonably closeamp with traditional snake taxonomy and more importantly it did not seem to agree any less well than the published tree Dowling et al (1996 their Table VI) recognized 30 groups based on their tree In 17 cases the groups were identical in composition to groups on the maximally different tree which there- fore matched tradtional taxonomv eauallv

i l i

well (or poorly) in three cases the group on their tree agreed better with traditional taxonomy in five cases the group on the most different tree agreed better and in

u

five other cases relative agreement was equivocal either because both trees matched traditional taxonomy poorly (but not identicallv) or because each tree

2

matched better in some resDects and worse in others Because the tree that is most different from the published tree fits traditional snake taxonomy reasonably well it seems reasonable to expect that the same will be true of most if not all of the equivalent trees everth he less even minor dilfferences among individual trees can sum to a high lampel of ambiguity when there are many equivalent trees This ap- pears to be the situation with the snake allozvme data for which there are thou- sands of equivalent trees and the consen- sus tree (Fig 2) is poorly resolved

The ambiguity of the snake allozyme data reanalyzed here is presumably related to the rate of evolution of the four loci surveyed which appears to be too rapid for the question being addressed Those loci have been characterized as slowly-evolving and they may indeed be evolving slowly relative to other loci commonly sur- veyed using protein electrophoresis~ev-

ertheless they are evolving relatively rap- idly in the context of the higher-level phy- logeny of snakes Rapid evolution is man- ifested in the large numbers of alleles (25 to 43) at each locus and in the fact that more than half of the entries in the dis- tance matrix exhibit the maximum possible distance value of 100 (Table l)indicating that the taxa share no alleles at any of the four loci The loci in question may be use- ful for analyzing relationships within small- er clades of snakes but they are highly am- biguous concerning the relationships with- in snakes as a whole particularly those re- lated to the deeper divergences in the history of that clade

Acknotc1edgments-We thank K Highton and C A Hass for generously providing original data and results F J Kohlf for assistance with NTSYSpc in- cluding increasing the tied-tree limit of the program D L Swofford for providing access to test versions of PAUP and J C LVilgenbusch and J A McGuire for help nith analyses performed on 64 bit compnt- ers

BUTH D G 1984 The application of electrophoretic data in systematic studies Annual Review of Ecol- ogy and Systematics 15501-522

DE QUEIROZK XD D A GOOD 1997 Phenetic clustering in biology a critique Quarterly Review of Biology 723-30

UOWLINGH G AND w7E DUELLMAN 1978 Sys- tematic herpetology a synopsis of families and higher categories HISS Publications New York New York USA

~~OVLISG AND RH G C A HASS S B HEDGES HIGHTON 1996 Snake relationships revealed by slow-evolving proteins a prelimirlap suwey Jol~r- nal of Zoolog) London 240 1-28

FRRISJ S L7 A ALBERT hl KXLLERSJOD LIP- SCOblB AND A G KLUGE 1996 Parsimony jack- knifing outperforms neighbor-joining Claclistics 12 99-124

FELSEXSTEIN 1993 PHYLIP ~ e r 3 5 Depart- J ment of Genetics Universit) of Vashington Seat- tle Washington USA

HART G 1983 The occurrence of multiple UPGhIA phenograms Pp 254-258 In J Felsenstein (Ed) Ni~~nericalTaxonomy Springer-L7erlag Berlin Ger- many

~IIcKEICHM F AND C MITTER 1983 Evolution- ary patterns in allozyme data a systematic ap-proach Pp 169-176 In N Platnick and A Funk (Eds) Advances in Cladistics L701 2 Columbia University Press Nev York New York USA

NEIM 1972 Genetic distance between populations American Naturalist 106283-292

PEKNYD AKD M D HENDY 1985 The use of tree comparison metrics Systematic Zoolog 3475-82

September 20001 HEHPEI

- - ~ l h s n i rquamtgsra camus nmbedl LlmprWhlr Iullginosvs A1smnr canmerlgeiur Anfilophir paiwtons annyton smguum nm~cops a o g ~ i a i ~ s Hwmynailes g9gas AIsWhls P ~ ~ O O C B ~ S ~ S Anhyton caiiiasmum Anhyton laenla Pniiodryar buimsisrsn Phlodiyar virldiJ Dlpras calesbp ~ h a d t n a ~ anaviiata LlOPhi miars LYI i iWhi l dorbmgl Waplsraphri merism TnamnWynartei rritgils Cislid NSliCd LlOPh3 Ylnds LiWhiS poeaiogyrus nypr~rnyncnvr rerox uiomacer Oxyrhyocnus UrOmdCB liendlus ldllns dYiEdli5 Uromacsr calesbyi Dalgl0na naeiana Naia nslaOphlrnhagus n m M h

MiCrYroideE BYrYXanthYS moon iegsur Python relicviarur Boa consrndor Eplcrarer srnalur AhdCldiplS COrPYlsnld neteimon p~arirhmos nseOmn rmur T i w i m o n a canus m g m m modsrrm Phyoynchus aecunatus I ~ C I U I I~~~EYI AOk1511040 PlSCiYOrY 1 L ~ Z t r M o nbinealr nypnaie nyonaie TnmeiesvrYS siegans TnmersSYiYS dlbOidbii Trwldaiaemus waphri Trimeiesurus r ~ k d i ~ n s ~ s TnmBrBiurY3 hanbuilensis AiiouoldB1 nYmmllei BDRiwS dlmX Crolalus ICYIY I I~Y I TiimeiesYrYS OkindvsnslS SiSlri r a Y Y I Sislwrus calenaluS crora1us Iapldus crOfa1I Ceraslsi CrOlaiY vindi Cufoseimma mMoStoma PIeUdoCerallBSPSrSfCYI C8relBC VipBrd Daaota i u ~ s e l i Annylon funereus T r w i d o o l r naslfanvr Cnanns bonas ~ y p h i w r bmaicenrr TyPhlaPs chard

FIG 2-Strict consensus tree for equivalent UPGMA trees resulting from alternative clustering pathways caused by ties in the distance matrix The consensus tree is based on 9999 equivalent UPGMA trees found by systematic tie-breaking using NTSYSpc An asterisk indicates an additional node that is col la~sed when the ~ubl ished tree of Dowling n

et al (1996)is included in the set of alternative trees

LOGICA 327

ture of the tree they were excluded from the remainder of our analyses The re-duced matrix has 124 taxa 7626 ~ainvise

I

comparisons and still contains 7619 values (999) that are tied with at least one oth- er value (Table 1) Once again the great majority of the total values (911) cor- respond with the five possible fractions of alleles shared by homozygotes

Phenetic Clustering Identical distance values (ie ties) im-

ply the existence of equivalent pheno- grams (trees generated by phenetic clus- tering) which are roughly analogous to equal minimum-length trees that are com- monly found in parsimony analyses (Hart 1983 de Queiroz and Good 1997) Be-cause there are numerous ties in the snake distance matrix we performed a UPGMA analysis with NTSYSpc 201d (Rohlf 1996) using the FIND option to search for tied (equivalent) trees with the tie toler- ance (TOL) set to the default value of 10-lo The search ended at 9999 trees the program maximum which was increased from 999 at our request by F J Rohlf The strict consensus of these tied trees (Fig 2) has only 34 nodes in contrast with the 103 nodes in the published tree (Fig 1) Given that the program reached its limit for tied trees the existence of additional equiva- lent trees and an even less resolved con- sensus tree is likely

Of the 34 nodes in the consensus tree (Fig 2) most (85) unite relatively small numbers of taxa (2-10) and reflect rela- tively shallow divergences relationships corresponding with early cladogenetic events are largely unresolved Based on the topology of the consensus tree none of the 9999 trees corresponds exactly with the published tree of Dowling e t al (1996) specifically our analysis failed to reproduce the placement of the group composed of Causus and Atheris in their tree (other seeming discrepancies between

which is the same additional node that is collapsed in the consensus tree for the 1000 equivalent UPGMA trees found by random tie-breaking using PAUP

328 HERPETOLOGICA [Vol56 No 3

our consensus tree and their published tree-involving the positions of Darling- tonia Tropidophis haetianus and Chari- nu-result from false poly-tomies in their tree see Fig 1 legend) Assuming that their tree represents an additional equiv- alent UPGMA tree then including it in the consensus would result in the loss of an additional node (marked with an aster- isk in Fig 2) Dowling et al (1996) used a different order of taxa in their datafile (supplied by R Highton) than that pub- lished in their tables we used the order in their tables This difference combined with the fact that our analysis did not find all of the equivalent UPGMA trees pre- sumably accounts for the minor discrep- ancy between the consensus of our 9999 UPGMA trees and their published tree

Because NTSYS generates equivalent phenograms by systematically exploring the alternative clustering pathways result- ing from ties we also used the random tie- breaking procedure in PAUP version 40b2 to examine alternative phenograms We performed 1000 UPGMA analyses each with a different random number as a tie-breaking seed producing 1000 equiv- alent UPGMA trees The strict consensus of these trees had 33 resolved nodes one node fewer than in the consensus of 10 times as many trees resulting from system- atic tie-breaking using NTSYS (Fig 2 ) and differing from that tree in the absence of the node uniting Causus plus Atheris with Acrochordus and the homalopsines Be- cause the number of nodes in the consen- sus tree for the analysis using random tie- breaking is less than the number of nodes in the consensus tree for the analysis using systematic tie-breaking there must be more than 9999 equivalent UPGMA trees In addition absence of the node uniting Causus and Atheris with Acrochordus and the homalopsines removes the only dis- crepancy between our UPGMA consensus tree and the tree obtained by Dowling et d (1996)

Additive Distance Analysis

When used to estimate phylogeny phe- netic clustering methods such as UPGMA carry an implicit assumption that rates of evolution among lineages are roughly con- stant (reviewed by de Queiroz and Good 1997) Moreover those methods do not employ an optimality criterion making evaluation of alternative topologies impos- sible Therefore we also analyzed the data using two alternative approaches additive distances and parsimony (reviewed by Swofford et al 1996) For the additive distance analysis we used the minimum evolution criterion in PAUP test version 40b2 A heuristic search was performed on the distance matrix obtaining the start- ing tree by neighbor-joining and searching for shorter trees using tree-bissection-re- connection (TBR) branch swapping We saved all trees shorter than the initial neighbor-joining tree (length = 2854687) which was itself shorter than the published UPGMA tree (length = 2887127) The search was terminated at 233363 trees but even more such trees presumably ex- ist because the search was not completed The strict consensus of these 233363 trees (not shown) has only seven resolved nodes each of which unites only two terminal taxa Thus under the minimum evolution criterion there are hundreds of thousands of trees shorter than both the neighbor joining tree and the published tree one or more of which contradict all but seven of the nodes in the published tree

A second minimum evolution analysis was performed once again obtaining the starting tree by neighbor-joining and using TBR branch swapping but this time keep- ing only the best trees This analysis yield- ed six optimal trees (length = 2817102) the strict consensus of which (not shown) is highly resolved (120 nodes) Much of that resolution however is of questionable significance The smallest distance in the data matrix (00646 Table 1)corresponds

Frc 3-Strict consensus of six optimal adchtive distance trees (minimum evolution criterion) with branches of length less than 00646 collapsed (see text for explanation)

329 Sentember 20001 HERPETOLOGICA

Opheodrys aerl iws Tanllfla lacoronala Anlona ele~eegans Lampropeals calllgaster PIUOP~Smelanolaucus Senlicons Inaspis Rhabdophls chryrargus Rhabdophls sp Xe00~hioph1sllaWpUnCla1US Amphresma slolala Clonophis klnland Sinonalrix annofaris Rspna ngrda Regina seplemvirtata

- - - Psammodynasles pulvarensls Boiga muntmaculaia Tmesc~possp

Enhydris bccourlr Enhydris enhydns Hmalops1s buccala BIIS anelans

-~ n h y t b nexiouvm

I Al~ophlscanthengeru~Hsdcop angulalos Hydrwnastes glgas

I Alsophis pononcensis Anhyion callilaemum Armyton laenlarum Ph8lodryas burmetsterl Ph i lod iya~ vrndls

1 Dlpsas caresby8Rhadlnaea Rawlala Llophls rnlbarts Ly~trophlsdorbinoi Wagleroph8s msriemi Lrophls p o e o l ~ y r ~ ~ Thamnodynasles slrigilis

I CleBa wsrica Llophis vrrldis Hypsrrhynchus lsrox Uromacer oxyrtiynchus Urornacer frenalus talliis dorsals Uiomacer caresbyr Naia oaiaI Ophqhagus hannah -I Mlcrurordes suryxanlhus Darltnglonia haet~ana Python regus Pyihon rettcufarus Boa conslncror Epncrales srnalus Helerodon plalirhinosHeteiWon sirnus Tropide9hls canus a i g m m modssfus PhyllOrhynCh~S deCUifalUS Agkisfrodon P ~ C W O N S Agh~slrodonbtiinealus Hypnale hypnaisTnmeresorus eteoans TnrnereSurUS albolabns

Caflwelasma f i o d m l m a PsaudoCerastespemicus Dabom msselii CerasIeS Yipera Alhens squamrgera C~USUSrhmbealus Aiihyian funereus Channa bonae Tropdoph~haellanus Typhlms iarnaicenscs Typhiops nchaidl

with the difference between taxa that pos- sess the same single allele at each locus except for the presence of a second allele in one of the taxa at one of the four loci (because the taxa are represented by single specimens this difference corresponds with the difference between an individual that is homozygous for all four loci and one that is homozygous for the same allele for three of those loci but heterozygous for the fourth with one of the two alleles of the heterozygote shared with the allele in the homozygote) If this distance (00646) is taken as the smallest distance that has a meaningful interpretation in terms of al- lelic change and all branches shorter than it are collapsed the resulting tree (Fig 3) is considerably less resolved (55 nodes) and exhibits little deep structure

Parsimony Analysis We performed a parsimony analysis of

the data using PAUP 40b2 treating the loci as characters and alleles as unordered states (eg Buth 1984 Mickevich and Mitter 1983) Multistate taxa (heterozy-gotes) were treated as polymorphic In contrast to the phenetic clustering and ad- ditive distance analyses in which only identical taxa were eliminated and the data sets consisted of 124 taxa the parsimony analysis was performed on a reduced data set of 95 taxa Both identical taxa and taxa that differed from others in the set only by possessing unique (and thus parsimony uninformative) alleles were eliminated Because of the large numbers of character states (gt32) in two of the characters we used a computer with a 64 bit Alpha pro- cessor to perform a heuristic search ob- taining the starting tree using simple step- wise addition searching for shorter trees using TBR branch swapping with the maximum number of trees set to 150000 The two species of Typhlops were desig- nated as outgroups The heuristic search yielded the maximum of 150000 equally most parsimonious trees (length = 144) the strict consensus of which (not illus- trated) was completely unresolved

DISCUSSION Our results indicate that the allozyme

data published by Dowling et al (1996)

)LOGICA IVol 56 No 3

are highly ambiguous concerning the high- er-level phylogeny of snakes This ambi- guity exists regardless of whether the data are analyzed using phenetic clustering ad- ditive distance methods or character-based ~arsimonv methods Under all three approamphes thousands of trees explain the data equally well or nearly so and the consensus of those trees exhibits little to no resolution particularly concerning the deeper nodes Given the ambiguity of the data it is worthwhile to consider first how Dowling et al (1996) obtained a single highly resolved tree and second why that tree exhibits general agreement with tra- ditional snake taxonomy

A single highly resolved tree presum- ably was obtained because the original analvsis did not reveal the existence of al- ternative (equivalent) trees Identical val-

L

ues in a distance matrix imply the exis- tence of alternative clustering pathways and thus alternative trees However many software im~lementations of UPGMA and related methods including the one in the TAXAN package used by Dowling et al (1996) do not explicitly reveal the exis- tence of alternative trees Instead ties are broken arbitrarily During each cycle in the clustering sequence if two or more taxa are equally (and minimally) distant from a third taxon the algorithm arbitrari- ly clusters one of these equally distant pairs first typically based on the input or- der of the taxa (Hart 1983 see also Farris et al 1996) The alternative clustering se- quence which may result in a different branching pattern (ie a different tree) is not explored Therefore the tree pro-duced mav be onlv one member of a set of equivalknt tree For the data analyzed in this paper there are both thousands of ties and thousands of equivalent trees

The reason that the single tree of Dowl- ing et al (1996) exhibits general agree- ment with traditional ideas about snake taxonomy is that the equivalent trees do not differ radically from one another so that any arbitrarily selected member of the set would probably agree more or less with traditional taxonomv To demonstrate

i

this we measured the dissimilarity be- tween the published tree and the 1000

- -

September 20001 HERPETOLOGICA 33 1

equivalent UPGMA trees found by ran-domly breaking ties using the symmetric difference metric of Penny and Hendy (1985) as implemented in PAUP version 40b2 Symmetric dfferences ranged from 77-141 (0318-0583) We then found the tree that differed most from the ~ublished tree and compared it with the hublished tree in terms of congruence with the tra- ditional taxonomy of snakes as described by Dowling et al (1996) and Dowling and Duellman (1978)

Even this maximallv different tree agreed reasonably closeamp with traditional snake taxonomy and more importantly it did not seem to agree any less well than the published tree Dowling et al (1996 their Table VI) recognized 30 groups based on their tree In 17 cases the groups were identical in composition to groups on the maximally different tree which there- fore matched tradtional taxonomv eauallv

i l i

well (or poorly) in three cases the group on their tree agreed better with traditional taxonomy in five cases the group on the most different tree agreed better and in

u

five other cases relative agreement was equivocal either because both trees matched traditional taxonomy poorly (but not identicallv) or because each tree

2

matched better in some resDects and worse in others Because the tree that is most different from the published tree fits traditional snake taxonomy reasonably well it seems reasonable to expect that the same will be true of most if not all of the equivalent trees everth he less even minor dilfferences among individual trees can sum to a high lampel of ambiguity when there are many equivalent trees This ap- pears to be the situation with the snake allozvme data for which there are thou- sands of equivalent trees and the consen- sus tree (Fig 2) is poorly resolved

The ambiguity of the snake allozyme data reanalyzed here is presumably related to the rate of evolution of the four loci surveyed which appears to be too rapid for the question being addressed Those loci have been characterized as slowly-evolving and they may indeed be evolving slowly relative to other loci commonly sur- veyed using protein electrophoresis~ev-

ertheless they are evolving relatively rap- idly in the context of the higher-level phy- logeny of snakes Rapid evolution is man- ifested in the large numbers of alleles (25 to 43) at each locus and in the fact that more than half of the entries in the dis- tance matrix exhibit the maximum possible distance value of 100 (Table l)indicating that the taxa share no alleles at any of the four loci The loci in question may be use- ful for analyzing relationships within small- er clades of snakes but they are highly am- biguous concerning the relationships with- in snakes as a whole particularly those re- lated to the deeper divergences in the history of that clade

Acknotc1edgments-We thank K Highton and C A Hass for generously providing original data and results F J Kohlf for assistance with NTSYSpc in- cluding increasing the tied-tree limit of the program D L Swofford for providing access to test versions of PAUP and J C LVilgenbusch and J A McGuire for help nith analyses performed on 64 bit compnt- ers

BUTH D G 1984 The application of electrophoretic data in systematic studies Annual Review of Ecol- ogy and Systematics 15501-522

DE QUEIROZK XD D A GOOD 1997 Phenetic clustering in biology a critique Quarterly Review of Biology 723-30

UOWLINGH G AND w7E DUELLMAN 1978 Sys- tematic herpetology a synopsis of families and higher categories HISS Publications New York New York USA

~~OVLISG AND RH G C A HASS S B HEDGES HIGHTON 1996 Snake relationships revealed by slow-evolving proteins a prelimirlap suwey Jol~r- nal of Zoolog) London 240 1-28

FRRISJ S L7 A ALBERT hl KXLLERSJOD LIP- SCOblB AND A G KLUGE 1996 Parsimony jack- knifing outperforms neighbor-joining Claclistics 12 99-124

FELSEXSTEIN 1993 PHYLIP ~ e r 3 5 Depart- J ment of Genetics Universit) of Vashington Seat- tle Washington USA

HART G 1983 The occurrence of multiple UPGhIA phenograms Pp 254-258 In J Felsenstein (Ed) Ni~~nericalTaxonomy Springer-L7erlag Berlin Ger- many

~IIcKEICHM F AND C MITTER 1983 Evolution- ary patterns in allozyme data a systematic ap-proach Pp 169-176 In N Platnick and A Funk (Eds) Advances in Cladistics L701 2 Columbia University Press Nev York New York USA

NEIM 1972 Genetic distance between populations American Naturalist 106283-292

PEKNYD AKD M D HENDY 1985 The use of tree comparison metrics Systematic Zoolog 3475-82

328 HERPETOLOGICA [Vol56 No 3

our consensus tree and their published tree-involving the positions of Darling- tonia Tropidophis haetianus and Chari- nu-result from false poly-tomies in their tree see Fig 1 legend) Assuming that their tree represents an additional equiv- alent UPGMA tree then including it in the consensus would result in the loss of an additional node (marked with an aster- isk in Fig 2) Dowling et al (1996) used a different order of taxa in their datafile (supplied by R Highton) than that pub- lished in their tables we used the order in their tables This difference combined with the fact that our analysis did not find all of the equivalent UPGMA trees pre- sumably accounts for the minor discrep- ancy between the consensus of our 9999 UPGMA trees and their published tree

Because NTSYS generates equivalent phenograms by systematically exploring the alternative clustering pathways result- ing from ties we also used the random tie- breaking procedure in PAUP version 40b2 to examine alternative phenograms We performed 1000 UPGMA analyses each with a different random number as a tie-breaking seed producing 1000 equiv- alent UPGMA trees The strict consensus of these trees had 33 resolved nodes one node fewer than in the consensus of 10 times as many trees resulting from system- atic tie-breaking using NTSYS (Fig 2 ) and differing from that tree in the absence of the node uniting Causus plus Atheris with Acrochordus and the homalopsines Be- cause the number of nodes in the consen- sus tree for the analysis using random tie- breaking is less than the number of nodes in the consensus tree for the analysis using systematic tie-breaking there must be more than 9999 equivalent UPGMA trees In addition absence of the node uniting Causus and Atheris with Acrochordus and the homalopsines removes the only dis- crepancy between our UPGMA consensus tree and the tree obtained by Dowling et d (1996)

Additive Distance Analysis

When used to estimate phylogeny phe- netic clustering methods such as UPGMA carry an implicit assumption that rates of evolution among lineages are roughly con- stant (reviewed by de Queiroz and Good 1997) Moreover those methods do not employ an optimality criterion making evaluation of alternative topologies impos- sible Therefore we also analyzed the data using two alternative approaches additive distances and parsimony (reviewed by Swofford et al 1996) For the additive distance analysis we used the minimum evolution criterion in PAUP test version 40b2 A heuristic search was performed on the distance matrix obtaining the start- ing tree by neighbor-joining and searching for shorter trees using tree-bissection-re- connection (TBR) branch swapping We saved all trees shorter than the initial neighbor-joining tree (length = 2854687) which was itself shorter than the published UPGMA tree (length = 2887127) The search was terminated at 233363 trees but even more such trees presumably ex- ist because the search was not completed The strict consensus of these 233363 trees (not shown) has only seven resolved nodes each of which unites only two terminal taxa Thus under the minimum evolution criterion there are hundreds of thousands of trees shorter than both the neighbor joining tree and the published tree one or more of which contradict all but seven of the nodes in the published tree

A second minimum evolution analysis was performed once again obtaining the starting tree by neighbor-joining and using TBR branch swapping but this time keep- ing only the best trees This analysis yield- ed six optimal trees (length = 2817102) the strict consensus of which (not shown) is highly resolved (120 nodes) Much of that resolution however is of questionable significance The smallest distance in the data matrix (00646 Table 1)corresponds

Frc 3-Strict consensus of six optimal adchtive distance trees (minimum evolution criterion) with branches of length less than 00646 collapsed (see text for explanation)

329 Sentember 20001 HERPETOLOGICA

Opheodrys aerl iws Tanllfla lacoronala Anlona ele~eegans Lampropeals calllgaster PIUOP~Smelanolaucus Senlicons Inaspis Rhabdophls chryrargus Rhabdophls sp Xe00~hioph1sllaWpUnCla1US Amphresma slolala Clonophis klnland Sinonalrix annofaris Rspna ngrda Regina seplemvirtata

- - - Psammodynasles pulvarensls Boiga muntmaculaia Tmesc~possp

Enhydris bccourlr Enhydris enhydns Hmalops1s buccala BIIS anelans

-~ n h y t b nexiouvm

I Al~ophlscanthengeru~Hsdcop angulalos Hydrwnastes glgas

I Alsophis pononcensis Anhyion callilaemum Armyton laenlarum Ph8lodryas burmetsterl Ph i lod iya~ vrndls

1 Dlpsas caresby8Rhadlnaea Rawlala Llophls rnlbarts Ly~trophlsdorbinoi Wagleroph8s msriemi Lrophls p o e o l ~ y r ~ ~ Thamnodynasles slrigilis

I CleBa wsrica Llophis vrrldis Hypsrrhynchus lsrox Uromacer oxyrtiynchus Urornacer frenalus talliis dorsals Uiomacer caresbyr Naia oaiaI Ophqhagus hannah -I Mlcrurordes suryxanlhus Darltnglonia haet~ana Python regus Pyihon rettcufarus Boa conslncror Epncrales srnalus Helerodon plalirhinosHeteiWon sirnus Tropide9hls canus a i g m m modssfus PhyllOrhynCh~S deCUifalUS Agkisfrodon P ~ C W O N S Agh~slrodonbtiinealus Hypnale hypnaisTnmeresorus eteoans TnrnereSurUS albolabns

Caflwelasma f i o d m l m a PsaudoCerastespemicus Dabom msselii CerasIeS Yipera Alhens squamrgera C~USUSrhmbealus Aiihyian funereus Channa bonae Tropdoph~haellanus Typhlms iarnaicenscs Typhiops nchaidl

with the difference between taxa that pos- sess the same single allele at each locus except for the presence of a second allele in one of the taxa at one of the four loci (because the taxa are represented by single specimens this difference corresponds with the difference between an individual that is homozygous for all four loci and one that is homozygous for the same allele for three of those loci but heterozygous for the fourth with one of the two alleles of the heterozygote shared with the allele in the homozygote) If this distance (00646) is taken as the smallest distance that has a meaningful interpretation in terms of al- lelic change and all branches shorter than it are collapsed the resulting tree (Fig 3) is considerably less resolved (55 nodes) and exhibits little deep structure

Parsimony Analysis We performed a parsimony analysis of

the data using PAUP 40b2 treating the loci as characters and alleles as unordered states (eg Buth 1984 Mickevich and Mitter 1983) Multistate taxa (heterozy-gotes) were treated as polymorphic In contrast to the phenetic clustering and ad- ditive distance analyses in which only identical taxa were eliminated and the data sets consisted of 124 taxa the parsimony analysis was performed on a reduced data set of 95 taxa Both identical taxa and taxa that differed from others in the set only by possessing unique (and thus parsimony uninformative) alleles were eliminated Because of the large numbers of character states (gt32) in two of the characters we used a computer with a 64 bit Alpha pro- cessor to perform a heuristic search ob- taining the starting tree using simple step- wise addition searching for shorter trees using TBR branch swapping with the maximum number of trees set to 150000 The two species of Typhlops were desig- nated as outgroups The heuristic search yielded the maximum of 150000 equally most parsimonious trees (length = 144) the strict consensus of which (not illus- trated) was completely unresolved

DISCUSSION Our results indicate that the allozyme

data published by Dowling et al (1996)

)LOGICA IVol 56 No 3

are highly ambiguous concerning the high- er-level phylogeny of snakes This ambi- guity exists regardless of whether the data are analyzed using phenetic clustering ad- ditive distance methods or character-based ~arsimonv methods Under all three approamphes thousands of trees explain the data equally well or nearly so and the consensus of those trees exhibits little to no resolution particularly concerning the deeper nodes Given the ambiguity of the data it is worthwhile to consider first how Dowling et al (1996) obtained a single highly resolved tree and second why that tree exhibits general agreement with tra- ditional snake taxonomy

A single highly resolved tree presum- ably was obtained because the original analvsis did not reveal the existence of al- ternative (equivalent) trees Identical val-

L

ues in a distance matrix imply the exis- tence of alternative clustering pathways and thus alternative trees However many software im~lementations of UPGMA and related methods including the one in the TAXAN package used by Dowling et al (1996) do not explicitly reveal the exis- tence of alternative trees Instead ties are broken arbitrarily During each cycle in the clustering sequence if two or more taxa are equally (and minimally) distant from a third taxon the algorithm arbitrari- ly clusters one of these equally distant pairs first typically based on the input or- der of the taxa (Hart 1983 see also Farris et al 1996) The alternative clustering se- quence which may result in a different branching pattern (ie a different tree) is not explored Therefore the tree pro-duced mav be onlv one member of a set of equivalknt tree For the data analyzed in this paper there are both thousands of ties and thousands of equivalent trees

The reason that the single tree of Dowl- ing et al (1996) exhibits general agree- ment with traditional ideas about snake taxonomy is that the equivalent trees do not differ radically from one another so that any arbitrarily selected member of the set would probably agree more or less with traditional taxonomv To demonstrate

i

this we measured the dissimilarity be- tween the published tree and the 1000

- -

September 20001 HERPETOLOGICA 33 1

equivalent UPGMA trees found by ran-domly breaking ties using the symmetric difference metric of Penny and Hendy (1985) as implemented in PAUP version 40b2 Symmetric dfferences ranged from 77-141 (0318-0583) We then found the tree that differed most from the ~ublished tree and compared it with the hublished tree in terms of congruence with the tra- ditional taxonomy of snakes as described by Dowling et al (1996) and Dowling and Duellman (1978)

Even this maximallv different tree agreed reasonably closeamp with traditional snake taxonomy and more importantly it did not seem to agree any less well than the published tree Dowling et al (1996 their Table VI) recognized 30 groups based on their tree In 17 cases the groups were identical in composition to groups on the maximally different tree which there- fore matched tradtional taxonomv eauallv

i l i

well (or poorly) in three cases the group on their tree agreed better with traditional taxonomy in five cases the group on the most different tree agreed better and in

u

five other cases relative agreement was equivocal either because both trees matched traditional taxonomy poorly (but not identicallv) or because each tree

2

matched better in some resDects and worse in others Because the tree that is most different from the published tree fits traditional snake taxonomy reasonably well it seems reasonable to expect that the same will be true of most if not all of the equivalent trees everth he less even minor dilfferences among individual trees can sum to a high lampel of ambiguity when there are many equivalent trees This ap- pears to be the situation with the snake allozvme data for which there are thou- sands of equivalent trees and the consen- sus tree (Fig 2) is poorly resolved

The ambiguity of the snake allozyme data reanalyzed here is presumably related to the rate of evolution of the four loci surveyed which appears to be too rapid for the question being addressed Those loci have been characterized as slowly-evolving and they may indeed be evolving slowly relative to other loci commonly sur- veyed using protein electrophoresis~ev-

ertheless they are evolving relatively rap- idly in the context of the higher-level phy- logeny of snakes Rapid evolution is man- ifested in the large numbers of alleles (25 to 43) at each locus and in the fact that more than half of the entries in the dis- tance matrix exhibit the maximum possible distance value of 100 (Table l)indicating that the taxa share no alleles at any of the four loci The loci in question may be use- ful for analyzing relationships within small- er clades of snakes but they are highly am- biguous concerning the relationships with- in snakes as a whole particularly those re- lated to the deeper divergences in the history of that clade

Acknotc1edgments-We thank K Highton and C A Hass for generously providing original data and results F J Kohlf for assistance with NTSYSpc in- cluding increasing the tied-tree limit of the program D L Swofford for providing access to test versions of PAUP and J C LVilgenbusch and J A McGuire for help nith analyses performed on 64 bit compnt- ers

BUTH D G 1984 The application of electrophoretic data in systematic studies Annual Review of Ecol- ogy and Systematics 15501-522

DE QUEIROZK XD D A GOOD 1997 Phenetic clustering in biology a critique Quarterly Review of Biology 723-30

UOWLINGH G AND w7E DUELLMAN 1978 Sys- tematic herpetology a synopsis of families and higher categories HISS Publications New York New York USA

~~OVLISG AND RH G C A HASS S B HEDGES HIGHTON 1996 Snake relationships revealed by slow-evolving proteins a prelimirlap suwey Jol~r- nal of Zoolog) London 240 1-28

FRRISJ S L7 A ALBERT hl KXLLERSJOD LIP- SCOblB AND A G KLUGE 1996 Parsimony jack- knifing outperforms neighbor-joining Claclistics 12 99-124

FELSEXSTEIN 1993 PHYLIP ~ e r 3 5 Depart- J ment of Genetics Universit) of Vashington Seat- tle Washington USA

HART G 1983 The occurrence of multiple UPGhIA phenograms Pp 254-258 In J Felsenstein (Ed) Ni~~nericalTaxonomy Springer-L7erlag Berlin Ger- many

~IIcKEICHM F AND C MITTER 1983 Evolution- ary patterns in allozyme data a systematic ap-proach Pp 169-176 In N Platnick and A Funk (Eds) Advances in Cladistics L701 2 Columbia University Press Nev York New York USA

NEIM 1972 Genetic distance between populations American Naturalist 106283-292

PEKNYD AKD M D HENDY 1985 The use of tree comparison metrics Systematic Zoolog 3475-82

329 Sentember 20001 HERPETOLOGICA

Opheodrys aerl iws Tanllfla lacoronala Anlona ele~eegans Lampropeals calllgaster PIUOP~Smelanolaucus Senlicons Inaspis Rhabdophls chryrargus Rhabdophls sp Xe00~hioph1sllaWpUnCla1US Amphresma slolala Clonophis klnland Sinonalrix annofaris Rspna ngrda Regina seplemvirtata

- - - Psammodynasles pulvarensls Boiga muntmaculaia Tmesc~possp

Enhydris bccourlr Enhydris enhydns Hmalops1s buccala BIIS anelans

-~ n h y t b nexiouvm

I Al~ophlscanthengeru~Hsdcop angulalos Hydrwnastes glgas

I Alsophis pononcensis Anhyion callilaemum Armyton laenlarum Ph8lodryas burmetsterl Ph i lod iya~ vrndls

1 Dlpsas caresby8Rhadlnaea Rawlala Llophls rnlbarts Ly~trophlsdorbinoi Wagleroph8s msriemi Lrophls p o e o l ~ y r ~ ~ Thamnodynasles slrigilis

I CleBa wsrica Llophis vrrldis Hypsrrhynchus lsrox Uromacer oxyrtiynchus Urornacer frenalus talliis dorsals Uiomacer caresbyr Naia oaiaI Ophqhagus hannah -I Mlcrurordes suryxanlhus Darltnglonia haet~ana Python regus Pyihon rettcufarus Boa conslncror Epncrales srnalus Helerodon plalirhinosHeteiWon sirnus Tropide9hls canus a i g m m modssfus PhyllOrhynCh~S deCUifalUS Agkisfrodon P ~ C W O N S Agh~slrodonbtiinealus Hypnale hypnaisTnmeresorus eteoans TnrnereSurUS albolabns

Caflwelasma f i o d m l m a PsaudoCerastespemicus Dabom msselii CerasIeS Yipera Alhens squamrgera C~USUSrhmbealus Aiihyian funereus Channa bonae Tropdoph~haellanus Typhlms iarnaicenscs Typhiops nchaidl

with the difference between taxa that pos- sess the same single allele at each locus except for the presence of a second allele in one of the taxa at one of the four loci (because the taxa are represented by single specimens this difference corresponds with the difference between an individual that is homozygous for all four loci and one that is homozygous for the same allele for three of those loci but heterozygous for the fourth with one of the two alleles of the heterozygote shared with the allele in the homozygote) If this distance (00646) is taken as the smallest distance that has a meaningful interpretation in terms of al- lelic change and all branches shorter than it are collapsed the resulting tree (Fig 3) is considerably less resolved (55 nodes) and exhibits little deep structure

Parsimony Analysis We performed a parsimony analysis of

the data using PAUP 40b2 treating the loci as characters and alleles as unordered states (eg Buth 1984 Mickevich and Mitter 1983) Multistate taxa (heterozy-gotes) were treated as polymorphic In contrast to the phenetic clustering and ad- ditive distance analyses in which only identical taxa were eliminated and the data sets consisted of 124 taxa the parsimony analysis was performed on a reduced data set of 95 taxa Both identical taxa and taxa that differed from others in the set only by possessing unique (and thus parsimony uninformative) alleles were eliminated Because of the large numbers of character states (gt32) in two of the characters we used a computer with a 64 bit Alpha pro- cessor to perform a heuristic search ob- taining the starting tree using simple step- wise addition searching for shorter trees using TBR branch swapping with the maximum number of trees set to 150000 The two species of Typhlops were desig- nated as outgroups The heuristic search yielded the maximum of 150000 equally most parsimonious trees (length = 144) the strict consensus of which (not illus- trated) was completely unresolved

DISCUSSION Our results indicate that the allozyme

data published by Dowling et al (1996)

)LOGICA IVol 56 No 3

are highly ambiguous concerning the high- er-level phylogeny of snakes This ambi- guity exists regardless of whether the data are analyzed using phenetic clustering ad- ditive distance methods or character-based ~arsimonv methods Under all three approamphes thousands of trees explain the data equally well or nearly so and the consensus of those trees exhibits little to no resolution particularly concerning the deeper nodes Given the ambiguity of the data it is worthwhile to consider first how Dowling et al (1996) obtained a single highly resolved tree and second why that tree exhibits general agreement with tra- ditional snake taxonomy

A single highly resolved tree presum- ably was obtained because the original analvsis did not reveal the existence of al- ternative (equivalent) trees Identical val-

L

ues in a distance matrix imply the exis- tence of alternative clustering pathways and thus alternative trees However many software im~lementations of UPGMA and related methods including the one in the TAXAN package used by Dowling et al (1996) do not explicitly reveal the exis- tence of alternative trees Instead ties are broken arbitrarily During each cycle in the clustering sequence if two or more taxa are equally (and minimally) distant from a third taxon the algorithm arbitrari- ly clusters one of these equally distant pairs first typically based on the input or- der of the taxa (Hart 1983 see also Farris et al 1996) The alternative clustering se- quence which may result in a different branching pattern (ie a different tree) is not explored Therefore the tree pro-duced mav be onlv one member of a set of equivalknt tree For the data analyzed in this paper there are both thousands of ties and thousands of equivalent trees

The reason that the single tree of Dowl- ing et al (1996) exhibits general agree- ment with traditional ideas about snake taxonomy is that the equivalent trees do not differ radically from one another so that any arbitrarily selected member of the set would probably agree more or less with traditional taxonomv To demonstrate

i

this we measured the dissimilarity be- tween the published tree and the 1000

- -

September 20001 HERPETOLOGICA 33 1

equivalent UPGMA trees found by ran-domly breaking ties using the symmetric difference metric of Penny and Hendy (1985) as implemented in PAUP version 40b2 Symmetric dfferences ranged from 77-141 (0318-0583) We then found the tree that differed most from the ~ublished tree and compared it with the hublished tree in terms of congruence with the tra- ditional taxonomy of snakes as described by Dowling et al (1996) and Dowling and Duellman (1978)

Even this maximallv different tree agreed reasonably closeamp with traditional snake taxonomy and more importantly it did not seem to agree any less well than the published tree Dowling et al (1996 their Table VI) recognized 30 groups based on their tree In 17 cases the groups were identical in composition to groups on the maximally different tree which there- fore matched tradtional taxonomv eauallv

i l i

well (or poorly) in three cases the group on their tree agreed better with traditional taxonomy in five cases the group on the most different tree agreed better and in

u

five other cases relative agreement was equivocal either because both trees matched traditional taxonomy poorly (but not identicallv) or because each tree

2

matched better in some resDects and worse in others Because the tree that is most different from the published tree fits traditional snake taxonomy reasonably well it seems reasonable to expect that the same will be true of most if not all of the equivalent trees everth he less even minor dilfferences among individual trees can sum to a high lampel of ambiguity when there are many equivalent trees This ap- pears to be the situation with the snake allozvme data for which there are thou- sands of equivalent trees and the consen- sus tree (Fig 2) is poorly resolved

The ambiguity of the snake allozyme data reanalyzed here is presumably related to the rate of evolution of the four loci surveyed which appears to be too rapid for the question being addressed Those loci have been characterized as slowly-evolving and they may indeed be evolving slowly relative to other loci commonly sur- veyed using protein electrophoresis~ev-

ertheless they are evolving relatively rap- idly in the context of the higher-level phy- logeny of snakes Rapid evolution is man- ifested in the large numbers of alleles (25 to 43) at each locus and in the fact that more than half of the entries in the dis- tance matrix exhibit the maximum possible distance value of 100 (Table l)indicating that the taxa share no alleles at any of the four loci The loci in question may be use- ful for analyzing relationships within small- er clades of snakes but they are highly am- biguous concerning the relationships with- in snakes as a whole particularly those re- lated to the deeper divergences in the history of that clade

Acknotc1edgments-We thank K Highton and C A Hass for generously providing original data and results F J Kohlf for assistance with NTSYSpc in- cluding increasing the tied-tree limit of the program D L Swofford for providing access to test versions of PAUP and J C LVilgenbusch and J A McGuire for help nith analyses performed on 64 bit compnt- ers

BUTH D G 1984 The application of electrophoretic data in systematic studies Annual Review of Ecol- ogy and Systematics 15501-522

DE QUEIROZK XD D A GOOD 1997 Phenetic clustering in biology a critique Quarterly Review of Biology 723-30

UOWLINGH G AND w7E DUELLMAN 1978 Sys- tematic herpetology a synopsis of families and higher categories HISS Publications New York New York USA

~~OVLISG AND RH G C A HASS S B HEDGES HIGHTON 1996 Snake relationships revealed by slow-evolving proteins a prelimirlap suwey Jol~r- nal of Zoolog) London 240 1-28

FRRISJ S L7 A ALBERT hl KXLLERSJOD LIP- SCOblB AND A G KLUGE 1996 Parsimony jack- knifing outperforms neighbor-joining Claclistics 12 99-124

FELSEXSTEIN 1993 PHYLIP ~ e r 3 5 Depart- J ment of Genetics Universit) of Vashington Seat- tle Washington USA

HART G 1983 The occurrence of multiple UPGhIA phenograms Pp 254-258 In J Felsenstein (Ed) Ni~~nericalTaxonomy Springer-L7erlag Berlin Ger- many

~IIcKEICHM F AND C MITTER 1983 Evolution- ary patterns in allozyme data a systematic ap-proach Pp 169-176 In N Platnick and A Funk (Eds) Advances in Cladistics L701 2 Columbia University Press Nev York New York USA

NEIM 1972 Genetic distance between populations American Naturalist 106283-292

PEKNYD AKD M D HENDY 1985 The use of tree comparison metrics Systematic Zoolog 3475-82

with the difference between taxa that pos- sess the same single allele at each locus except for the presence of a second allele in one of the taxa at one of the four loci (because the taxa are represented by single specimens this difference corresponds with the difference between an individual that is homozygous for all four loci and one that is homozygous for the same allele for three of those loci but heterozygous for the fourth with one of the two alleles of the heterozygote shared with the allele in the homozygote) If this distance (00646) is taken as the smallest distance that has a meaningful interpretation in terms of al- lelic change and all branches shorter than it are collapsed the resulting tree (Fig 3) is considerably less resolved (55 nodes) and exhibits little deep structure

Parsimony Analysis We performed a parsimony analysis of

the data using PAUP 40b2 treating the loci as characters and alleles as unordered states (eg Buth 1984 Mickevich and Mitter 1983) Multistate taxa (heterozy-gotes) were treated as polymorphic In contrast to the phenetic clustering and ad- ditive distance analyses in which only identical taxa were eliminated and the data sets consisted of 124 taxa the parsimony analysis was performed on a reduced data set of 95 taxa Both identical taxa and taxa that differed from others in the set only by possessing unique (and thus parsimony uninformative) alleles were eliminated Because of the large numbers of character states (gt32) in two of the characters we used a computer with a 64 bit Alpha pro- cessor to perform a heuristic search ob- taining the starting tree using simple step- wise addition searching for shorter trees using TBR branch swapping with the maximum number of trees set to 150000 The two species of Typhlops were desig- nated as outgroups The heuristic search yielded the maximum of 150000 equally most parsimonious trees (length = 144) the strict consensus of which (not illus- trated) was completely unresolved

DISCUSSION Our results indicate that the allozyme

data published by Dowling et al (1996)

)LOGICA IVol 56 No 3

are highly ambiguous concerning the high- er-level phylogeny of snakes This ambi- guity exists regardless of whether the data are analyzed using phenetic clustering ad- ditive distance methods or character-based ~arsimonv methods Under all three approamphes thousands of trees explain the data equally well or nearly so and the consensus of those trees exhibits little to no resolution particularly concerning the deeper nodes Given the ambiguity of the data it is worthwhile to consider first how Dowling et al (1996) obtained a single highly resolved tree and second why that tree exhibits general agreement with tra- ditional snake taxonomy

A single highly resolved tree presum- ably was obtained because the original analvsis did not reveal the existence of al- ternative (equivalent) trees Identical val-

L

ues in a distance matrix imply the exis- tence of alternative clustering pathways and thus alternative trees However many software im~lementations of UPGMA and related methods including the one in the TAXAN package used by Dowling et al (1996) do not explicitly reveal the exis- tence of alternative trees Instead ties are broken arbitrarily During each cycle in the clustering sequence if two or more taxa are equally (and minimally) distant from a third taxon the algorithm arbitrari- ly clusters one of these equally distant pairs first typically based on the input or- der of the taxa (Hart 1983 see also Farris et al 1996) The alternative clustering se- quence which may result in a different branching pattern (ie a different tree) is not explored Therefore the tree pro-duced mav be onlv one member of a set of equivalknt tree For the data analyzed in this paper there are both thousands of ties and thousands of equivalent trees

The reason that the single tree of Dowl- ing et al (1996) exhibits general agree- ment with traditional ideas about snake taxonomy is that the equivalent trees do not differ radically from one another so that any arbitrarily selected member of the set would probably agree more or less with traditional taxonomv To demonstrate

i

this we measured the dissimilarity be- tween the published tree and the 1000

- -

September 20001 HERPETOLOGICA 33 1

equivalent UPGMA trees found by ran-domly breaking ties using the symmetric difference metric of Penny and Hendy (1985) as implemented in PAUP version 40b2 Symmetric dfferences ranged from 77-141 (0318-0583) We then found the tree that differed most from the ~ublished tree and compared it with the hublished tree in terms of congruence with the tra- ditional taxonomy of snakes as described by Dowling et al (1996) and Dowling and Duellman (1978)

Even this maximallv different tree agreed reasonably closeamp with traditional snake taxonomy and more importantly it did not seem to agree any less well than the published tree Dowling et al (1996 their Table VI) recognized 30 groups based on their tree In 17 cases the groups were identical in composition to groups on the maximally different tree which there- fore matched tradtional taxonomv eauallv

i l i

well (or poorly) in three cases the group on their tree agreed better with traditional taxonomy in five cases the group on the most different tree agreed better and in

u

five other cases relative agreement was equivocal either because both trees matched traditional taxonomy poorly (but not identicallv) or because each tree

2

matched better in some resDects and worse in others Because the tree that is most different from the published tree fits traditional snake taxonomy reasonably well it seems reasonable to expect that the same will be true of most if not all of the equivalent trees everth he less even minor dilfferences among individual trees can sum to a high lampel of ambiguity when there are many equivalent trees This ap- pears to be the situation with the snake allozvme data for which there are thou- sands of equivalent trees and the consen- sus tree (Fig 2) is poorly resolved

The ambiguity of the snake allozyme data reanalyzed here is presumably related to the rate of evolution of the four loci surveyed which appears to be too rapid for the question being addressed Those loci have been characterized as slowly-evolving and they may indeed be evolving slowly relative to other loci commonly sur- veyed using protein electrophoresis~ev-

ertheless they are evolving relatively rap- idly in the context of the higher-level phy- logeny of snakes Rapid evolution is man- ifested in the large numbers of alleles (25 to 43) at each locus and in the fact that more than half of the entries in the dis- tance matrix exhibit the maximum possible distance value of 100 (Table l)indicating that the taxa share no alleles at any of the four loci The loci in question may be use- ful for analyzing relationships within small- er clades of snakes but they are highly am- biguous concerning the relationships with- in snakes as a whole particularly those re- lated to the deeper divergences in the history of that clade

Acknotc1edgments-We thank K Highton and C A Hass for generously providing original data and results F J Kohlf for assistance with NTSYSpc in- cluding increasing the tied-tree limit of the program D L Swofford for providing access to test versions of PAUP and J C LVilgenbusch and J A McGuire for help nith analyses performed on 64 bit compnt- ers

BUTH D G 1984 The application of electrophoretic data in systematic studies Annual Review of Ecol- ogy and Systematics 15501-522

DE QUEIROZK XD D A GOOD 1997 Phenetic clustering in biology a critique Quarterly Review of Biology 723-30

UOWLINGH G AND w7E DUELLMAN 1978 Sys- tematic herpetology a synopsis of families and higher categories HISS Publications New York New York USA

~~OVLISG AND RH G C A HASS S B HEDGES HIGHTON 1996 Snake relationships revealed by slow-evolving proteins a prelimirlap suwey Jol~r- nal of Zoolog) London 240 1-28

FRRISJ S L7 A ALBERT hl KXLLERSJOD LIP- SCOblB AND A G KLUGE 1996 Parsimony jack- knifing outperforms neighbor-joining Claclistics 12 99-124

FELSEXSTEIN 1993 PHYLIP ~ e r 3 5 Depart- J ment of Genetics Universit) of Vashington Seat- tle Washington USA

HART G 1983 The occurrence of multiple UPGhIA phenograms Pp 254-258 In J Felsenstein (Ed) Ni~~nericalTaxonomy Springer-L7erlag Berlin Ger- many

~IIcKEICHM F AND C MITTER 1983 Evolution- ary patterns in allozyme data a systematic ap-proach Pp 169-176 In N Platnick and A Funk (Eds) Advances in Cladistics L701 2 Columbia University Press Nev York New York USA

NEIM 1972 Genetic distance between populations American Naturalist 106283-292

PEKNYD AKD M D HENDY 1985 The use of tree comparison metrics Systematic Zoolog 3475-82

- -

September 20001 HERPETOLOGICA 33 1

equivalent UPGMA trees found by ran-domly breaking ties using the symmetric difference metric of Penny and Hendy (1985) as implemented in PAUP version 40b2 Symmetric dfferences ranged from 77-141 (0318-0583) We then found the tree that differed most from the ~ublished tree and compared it with the hublished tree in terms of congruence with the tra- ditional taxonomy of snakes as described by Dowling et al (1996) and Dowling and Duellman (1978)

Even this maximallv different tree agreed reasonably closeamp with traditional snake taxonomy and more importantly it did not seem to agree any less well than the published tree Dowling et al (1996 their Table VI) recognized 30 groups based on their tree In 17 cases the groups were identical in composition to groups on the maximally different tree which there- fore matched tradtional taxonomv eauallv

i l i

well (or poorly) in three cases the group on their tree agreed better with traditional taxonomy in five cases the group on the most different tree agreed better and in

u

five other cases relative agreement was equivocal either because both trees matched traditional taxonomy poorly (but not identicallv) or because each tree

2

matched better in some resDects and worse in others Because the tree that is most different from the published tree fits traditional snake taxonomy reasonably well it seems reasonable to expect that the same will be true of most if not all of the equivalent trees everth he less even minor dilfferences among individual trees can sum to a high lampel of ambiguity when there are many equivalent trees This ap- pears to be the situation with the snake allozvme data for which there are thou- sands of equivalent trees and the consen- sus tree (Fig 2) is poorly resolved

The ambiguity of the snake allozyme data reanalyzed here is presumably related to the rate of evolution of the four loci surveyed which appears to be too rapid for the question being addressed Those loci have been characterized as slowly-evolving and they may indeed be evolving slowly relative to other loci commonly sur- veyed using protein electrophoresis~ev-

ertheless they are evolving relatively rap- idly in the context of the higher-level phy- logeny of snakes Rapid evolution is man- ifested in the large numbers of alleles (25 to 43) at each locus and in the fact that more than half of the entries in the dis- tance matrix exhibit the maximum possible distance value of 100 (Table l)indicating that the taxa share no alleles at any of the four loci The loci in question may be use- ful for analyzing relationships within small- er clades of snakes but they are highly am- biguous concerning the relationships with- in snakes as a whole particularly those re- lated to the deeper divergences in the history of that clade

Acknotc1edgments-We thank K Highton and C A Hass for generously providing original data and results F J Kohlf for assistance with NTSYSpc in- cluding increasing the tied-tree limit of the program D L Swofford for providing access to test versions of PAUP and J C LVilgenbusch and J A McGuire for help nith analyses performed on 64 bit compnt- ers

BUTH D G 1984 The application of electrophoretic data in systematic studies Annual Review of Ecol- ogy and Systematics 15501-522

DE QUEIROZK XD D A GOOD 1997 Phenetic clustering in biology a critique Quarterly Review of Biology 723-30

UOWLINGH G AND w7E DUELLMAN 1978 Sys- tematic herpetology a synopsis of families and higher categories HISS Publications New York New York USA

~~OVLISG AND RH G C A HASS S B HEDGES HIGHTON 1996 Snake relationships revealed by slow-evolving proteins a prelimirlap suwey Jol~r- nal of Zoolog) London 240 1-28

FRRISJ S L7 A ALBERT hl KXLLERSJOD LIP- SCOblB AND A G KLUGE 1996 Parsimony jack- knifing outperforms neighbor-joining Claclistics 12 99-124

FELSEXSTEIN 1993 PHYLIP ~ e r 3 5 Depart- J ment of Genetics Universit) of Vashington Seat- tle Washington USA

HART G 1983 The occurrence of multiple UPGhIA phenograms Pp 254-258 In J Felsenstein (Ed) Ni~~nericalTaxonomy Springer-L7erlag Berlin Ger- many

~IIcKEICHM F AND C MITTER 1983 Evolution- ary patterns in allozyme data a systematic ap-proach Pp 169-176 In N Platnick and A Funk (Eds) Advances in Cladistics L701 2 Columbia University Press Nev York New York USA

NEIM 1972 Genetic distance between populations American Naturalist 106283-292

PEKNYD AKD M D HENDY 1985 The use of tree comparison metrics Systematic Zoolog 3475-82

332 HERPETOLOGICA [Vol56 No 3

ROHLF F J 1996 NTSYSpc version 201d Exeter D M HILLS 1996 Phylogenetic inference Pp Software Setauket New York USA 407514 In D M Hillis C Moritz and B K

SAITOUN AND M NEI 1987 The neighbor-joining Mable (Eds) Molecular Systematics 2nd ed Sin- method a new method for reconstructing phylo- auer Sunderland Massachusetts USA genetic trees Molecular Biology and Evolution 6 514-525 Accepted LO October 1999

SWOFFORDD L G J OLSEN ANDP J WADDELL Associate Editor Stephen Tilley

Herpetologica 56(3)2000 332-342 O 2000 by The Herpetolosts League Inc

MORPHOMETRIC VARIATION AMONG LARVAE O F F O U R SPECIES O F LUNGLESS SALAMANDERS

(CAUDATA PLETHODONTIDAE)

GAYLELIVINGSTONBIRCHFIELDAND RICHARDC BRUCE~ Highlands Biological Station Highlands NC 28741 USA and Department of Biology

Western Carolina University Cullowhee NC 28723 USA

ABSTRACTSeveral linear dimensions and body masses were recorded for larvae of four closely- related similar species of lungless salamanders (family Plethodontidae) that occur over a broad gradient of elevation and habitat in the southeastern United States The species were Gyrinophilus porphyriticus Pseudotriton ruber l montanus and Stereochilus nmrginatus The goals were to evaluate the usefulness of morphometrics in species identification and to examine trends in variation in relation to habitat utilization in the four species To remove the effect of variation in body size log-transformed values of the variables were plotted against log snout-vent length and the residuals generated by these plots were used as the dependent variables in all analyses Discriminant function analyses were only partially successful in classifying larvae by species Multivariate analyses of var- iance (MANOVA) failed to reveal any pronounced trends in morphology related to the stream- and pond-type categories of larval salamanders recognized by herpetologists However the slender hab- itus of larvae of the lowland species S marginatus may represent adaptation to sphagnum mats in ponds and sluggish streams in the Coastal Plain The slender body and reduced eyes of larvae of the montane 6porphyriticus are probably adaptations to a subsurface mode of life in springs and headwater streams It is proposed that lunglessness herein considered a larval adaptation to life in mountain streams in plethodontids is one factor that constrains adaptive diversification in this clade relegating larvae to a bottom-dwelling mode of life in lentic as well as lotic habitats

Key words Cauclata Gyrinophilus porphyriticus Larvae Morphology Pseudotriton montanus Pseudotriton ruher Salamanders Stereochilus nzarginatus

HEMIDACTYLIINEsalamanders of the and a lengthy larval phase rangng from genera Gyrinophilus Pseudotriton and 15-25 yr in Stereochilus marginatus Stereochilus are thought to constitute a (Bruce 1971) Pseudotriton montanus monophyletic lineage containing the least- (Bruce 1974 1978) and l mber (Bruce derived species of the family Plethodonti- 1972 1974 Semlitsch 1983) to as long as dae (Lombard and Wake 1986 Wake 4-5 yr in Gyrinophilus porphyriticus 1966) Putative plesiomorphic characters (Bruce 1980) The cavernicolous G pal- of this lineage include a biphasic life cycle leucus and G subterraneus are paedomor-

phic or nearly so (Besharse and Holsinger PRESENTADDRESSDivision of Biological Scienc- 1977 Brandon 1971) Thus the larval pe-

es University of Missouri Columbia MO 65211 riod is an important phase of the life cycle USA of all these species CORRESPONDING Department of Biol- ADDRESS ogy Western Carolina University Cullowhee NC In the southeastern United States G 28723 USA porphyriticus l mber P montanus and

You have printed the following article

Slowly-Evolving Protein Loci and Higher-Level Snake Phylogeny A ReanalysisLarry Buckley Maureen Kearney Kevin de QueirozHerpetologica Vol 56 No 3 (Sep 2000) pp 324-332Stable URL

httplinksjstororgsicisici=0018-08312820000929563A33C3243ASPLAHS3E20CO3B2-Z

This article references the following linked citations If you are trying to access articles from anoff-campus location you may be required to first logon via your library web site to access JSTOR Pleasevisit your librarys website or contact a librarian to learn about options for remote access to JSTOR

Literature Cited

The Application of Electrophoretic Data in Systematic StudiesDonald G ButhAnnual Review of Ecology and Systematics Vol 15 (1984) pp 501-522Stable URL

httplinksjstororgsicisici=0066-416228198429153C5013ATAOEDI3E20CO3B2-B

Phenetic Clustering in Biology A CritiqueKevin de Queiroz David A GoodThe Quarterly Review of Biology Vol 72 No 1 (Mar 1997) pp 3-30Stable URL

httplinksjstororgsicisici=0033-57702819970329723A13C33APCIBAC3E20CO3B2-I

Genetic Distance between PopulationsMasatoshi NeiThe American Naturalist Vol 106 No 949 (May - Jun 1972) pp 283-292Stable URL

httplinksjstororgsicisici=0003-0147281972052F06291063A9493C2833AGDBP3E20CO3B2-U

The Use of Tree Comparison MetricsDavid Penny M D HendySystematic Zoology Vol 34 No 1 (Mar 1985) pp 75-82Stable URL

httplinksjstororgsicisici=0039-79892819850329343A13C753ATUOTCM3E20CO3B2-P

httpwwwjstororg

LINKED CITATIONS- Page 1 of 1 -

You have printed the following article

Slowly-Evolving Protein Loci and Higher-Level Snake Phylogeny A ReanalysisLarry Buckley Maureen Kearney Kevin de QueirozHerpetologica Vol 56 No 3 (Sep 2000) pp 324-332Stable URL

httplinksjstororgsicisici=0018-08312820000929563A33C3243ASPLAHS3E20CO3B2-Z

This article references the following linked citations If you are trying to access articles from anoff-campus location you may be required to first logon via your library web site to access JSTOR Pleasevisit your librarys website or contact a librarian to learn about options for remote access to JSTOR

Literature Cited

The Application of Electrophoretic Data in Systematic StudiesDonald G ButhAnnual Review of Ecology and Systematics Vol 15 (1984) pp 501-522Stable URL

httplinksjstororgsicisici=0066-416228198429153C5013ATAOEDI3E20CO3B2-B

Phenetic Clustering in Biology A CritiqueKevin de Queiroz David A GoodThe Quarterly Review of Biology Vol 72 No 1 (Mar 1997) pp 3-30Stable URL

httplinksjstororgsicisici=0033-57702819970329723A13C33APCIBAC3E20CO3B2-I

Genetic Distance between PopulationsMasatoshi NeiThe American Naturalist Vol 106 No 949 (May - Jun 1972) pp 283-292Stable URL

httplinksjstororgsicisici=0003-0147281972052F06291063A9493C2833AGDBP3E20CO3B2-U

The Use of Tree Comparison MetricsDavid Penny M D HendySystematic Zoology Vol 34 No 1 (Mar 1985) pp 75-82Stable URL

httplinksjstororgsicisici=0039-79892819850329343A13C753ATUOTCM3E20CO3B2-P

httpwwwjstororg

LINKED CITATIONS- Page 1 of 1 -