Phylogeny of squeakers (Arthroleptis)

21 Pages·2008·1.66 MB·English

MolecularPhylogeneticsandEvolution49(2008)806–826
ContentslistsavailableatScienceDirect
Molecular Phylogenetics and Evolution
journal homepage: www.elsevier.com/locate/ympev
Biogeography and evolution of body size and life history of African frogs: Phylogeny
of squeakers (Arthroleptis) and long-fingered frogs (Cardioglossa) estimated
from mitochondrial data
David C. Blackburn1
MuseumofComparativeZoology,DepartmentofOrganismicandEvolutionaryBiology,HarvardUniversity,Cambridge,MA02138,USA
a r t i c l e i n f o a b s t r a c t
Articlehistory: TheevolutionaryhistoryoflivingAfricanamphibiansremainspoorlyunderstood.Thisstudyestimates
Received7March2008 thephylogenywithinthefroggeneraArthroleptisandCardioglossausingapproximately2400basesof
Revised11August2008 mtDNAsequencedata(12S,tRNA-Valine,and16Sgenes)fromhalfofthedescribedspecies.Analyses
Accepted14August2008
are conducted usingparsimony,maximum likelihood,andBayesian methods.Theeffectofalignment
Availableonline30August2008
onphylogenyestimationisexploredbyseparatelyanalyzingalignmentsgeneratedwithdifferentgap
costsandaconsensusalignment.Theconsensusalignmentresultsinspeciesparaphyly,lownodalsup-
Keywords:
port, and incongruence with the results based on other alignments, which produced largely similar
Ancestralstatereconstruction
results.Mostnodesinthephylogenyarehighlysupported,yetseveraltopologiesareinconsistentwith
Arthroleptidae
previous hypotheses. The monophyly of Cardioglossa and of miniature species previously assigned to
Characterevolution
Crypticspecies Schoutedenella was further examined using Templeton and Shimodaira–Hasegawa tests. Cardioglossa
Directdevelopment monophylyisrejectedandC.aureoliistransferredtoArthroleptis.ThesetestsdonotrejectSchoutedenella
Miniaturization monophyly,butthishypothesisreceivesnosupportfromnon-parametricbootstrappingorBayesianpos-
Schoutedenella teriorprobabilities.Thisphylogenyprovidesaframeworkforreconstructinghistoricalbiogeographyand
analyzingtheevolutionofbodysizeandlifehistory.Directdevelopmentandminiaturizationappearat
thebaseofArthroleptisphylogenyconcomitantwitharangeexpansionfromCentralAfricatothroughout
mostofsub-SaharanAfrica.
(cid:2)2008ElsevierInc.Allrightsreserved.
1.Introduction Throughout Africa’s history, oscillating climatic regimes have
causedtherepeatedexpansionandcontractionofdifferenthabitat
Threeoftheworld’smoststudiedanimalspecieshavetheirori- types (e.g., Axelrod and Raven, 1978; Colyn et al., 1991; Stokes
ginsinsub-SaharanAfrica(e.g.,StraitandWood,1999;Evansetal., etal.,1997;Nichol,1999).Asthesechangesoccuroverbroadtime
2004; Keller, 2006). However, this interest in the fly Drosophila scales,thephylogeneticrelationshipsoforganismsthathavediver-
melanogaster,thefrogXenopuslaevis,andourselves,Homosapiens, sifiedanddispersedacrossthecontinentcanilluminatethehistor-
hasnotstimulatedsufficientstudyofthediversityanddiversifica- ical connections between various regions and thus the history of
tionofAfricananimalsonacontinent-widescale.Fourofthebio- the African continent. Such work moves a step beyond regional
diversityhotspotsdefinedasconservationprioritiesbyMyersetal. comparisons based on floristic or faunal surveys by explicitly
(2000)arelocatedincontinentalsub-SaharanAfricaandthereare including a historical component; both the order of cladogenic
numerous African regions with both many endemic species and eventsandbranchlengthscaninformpatternsofspatialandtem-
threatened species (Burgess et al., 2004). The diverse fauna and poraldiversification.By combininga phylogeneticapproachwith
floraofmanyAfricanregionsremainpoorlydocumentedandmost informationfromorganismalbiology,suchasecologyandlifehis-
regions have received little research attention from conservation tory,wecantesthypothesesregardingtheprocessesofdiversifica-
biologists(e.g.,Pimm,2007).Tobetterdevelopourunderstanding tionthatgeneratedAfrica’sendemicbiodiversity.
ofAfricanbiodiversity,itisnecessarytoincorporatespeciesfrom AmongAfricanvertebrates,amphibiansexhibitremarkablepat-
regions across sub-Saharan Africa and to integrate diverse forms ternsofgeographic,reproductive,andanatomicaldiversity.None-
oforganismaldata. theless, most African amphibians remain poorly studied in
comparison to the diverse faunas of North and South America or
Southeast Asia and the Malay Archipelago. Within sub-Saharan
1 Present address: Natural History Museum and Biodiversity Research Center,
Africa, there are several regions of high anuran diversity (Stuart
UniversityofKansas,Lawrence,KS66045,USA.
E-mailaddress:[email protected] etal.,2004), a few ofwhichcorrespondtopreviouslyrecognized
1055-7903/$-seefrontmatter(cid:2)2008ElsevierInc.Allrightsreserved.
doi:10.1016/j.ympev.2008.08.015
D.C.Blackburn/MolecularPhylogeneticsandEvolution49(2008)806–826 807
biodiversity hotspots, including the Upper Guinean Forests, the characters.Malesofmostspeciesexhibitspinesliningthesurface
Cameroon Volcanic Line, and Eastern Arc Mountains (Brooks of the third and sometimes second and first fingers. In addition,
etal.,2002).Previousphylogeneticstudiesofanurangeneraspan- males of many species have unusually elongated third fingers,
ningregionsofsub-SaharanAfricahavefounddifferingpatternsof whichareuniqueamongvertebratesandmaybeasynapomorphy
relationshipamongregions(Wieczoreketal.,2000,2001;Vences unitingthesegenera(e.g.,Noble,1931).However,thefunctionof
et al., 2003a,b, 2004; Evans et al., 2004; Dawood and Uqubay, theselongfingersremainsenigmatic(Amiet,1989).Thediversity
2004;Idris,2004;Measeyetal.,2007).Otherstudieshavebeenre- ofmalesecondarysexualstructuresinthesegeneraareevolution-
strictedtosamplesfromparticulargeographicregionswithinAfri- ary labile and may be the product of changing sexual selection
ca(e.g.,RichardsandMoore,1996;Dawoodetal.,2002)andthus forces(Blackburn,inpress).
unabletoresolvethehistoricalrelationshipsbetweenregionswith The relationships among the smallest and largest Arthroleptis
high anuran diversity. Van der Meijden et al. (2004, 2005, 2007) specieshavebeencontentious.Historically,mostsmallArthroleptis
andBossuytetal.(2006)providedthefirstphylogeneticevidence specieswereplacedinthegenusSchoutedenella,whichwassynon-
fortworadiationsofranoidfrogsthatareendemictosub-Saharan ymizedwithArthroleptisbyFrostetal.(2006)basedontheresults
Africa. However, these studies did not focus on relationships be- ofapreliminaryphylogeneticanalysisofthisgroup.Someauthors
tween, and especially not within, the Africa genera. In general, (Loveridge, 1957;Schmidtand Inger,1959;Poynton,1964, 1976,
therehasbeenlimitedbiogeographicstudyonthescalenecessary 2003)havearguedthatthesmallspeciesareessentially‘‘smallver-
to understand the history of African anuran clades across the sions”ofthelargerArthroleptisandthusshouldbeincludedinthis
continent. genus. In contrast, study of morphological diversity in an evolu-
Severalfrogcladesfoundthroughoutsub-SaharanAfricaexhibit tionary context suggested to other authors (e.g., Laurent, 1940,
abroadrangeoforganismaldiversity.Squeakersandlong-fingered 1973,1981;LaurentandFabrezi,1985)thatthelarge(i.e.,Arthro-
frogs are one such clade, which currently comprises two genera, leptis) and small (i.e., Schoutedenella) species are not only each
Arthroleptis and Cardioglossa (Fig. 1; Frost et al., 2006). Together, monophyleticbutarealsonotsistertaxa.Clearly,thesearealter-
these two genera have an ecological, altitudinal, and geographic nativephylogenetichypothesesthatcannotbothbetrue.Thelatter
range spanning much of the habitat accessible to amphibians hypothesisimplieseithermorethanoneoriginofdirectdevelop-
acrosssub-SaharanAfrica(Fig.1).Ingeneral,theseareterrestrial ment in this clade, or that larvae re-evolved in Cardioglossa. To
leaf-litter frogs that feed on a diverse range of terrestrial arthro- date,therehasbeennoattempttoresolvethesediscrepanciesby
pods, including many ants and termites (Blackburn and Moreau, combining a broad taxon sampling with modern phylogenetic
2006). Arthroleptis and Cardioglossa exhibit a range of adult body methodology.
sizes(13–54mmsnout-ventlength;i.e.,morethanafourfolddif- ThisstudyofArthroleptisandCardioglossaaimstoestablish,as
ference),arecharacterizedbytwocontrastinglifehistories,andex- thoroughly as possible, an understanding of the evolution and
hibit the most reduced karyotype (2n=14) known in anurans diversity ofan anuranclade thatis restrictedto continentalsub-
(Bogart and Tandy, 1981). All but one species of Cardioglossa are SaharanAfrica.Therearethreeprincipalgoals.Thefirstistoinves-
believed to have the ancestral anuran life history characterized tigatethemonophyly,intergeneric,andinterspecificrelationships
byafree-living,feedingtadpole(Lamotte,1961;Blackburnetal., ofArthroleptisandCardioglossa.Aspartofthisanalysis,Ianalyzed
2008).CardioglossaaureoliandallArthroleptisarebelievedtohave theeffectofalignmentontheinferenceofphylogeneticrelation-
direct development in which metamorphosed froglets hatch di- ships.Second,givenanestimateofphylogeny,Ievaluatedtheevo-
rectlyfromterrestrialeggs(e.g.,GuibéandLamotte,1958;Lamotte lutionofbodysizeandlifehistory;thesecharactershaveplayeda
andPerret,1963;IUCN,2006).Lastly,malesofnearlyallArthrolep- centralroleinpreviousattemptstoorganizethetaxonomicdiver-
tis and Cardioglossa share a common suite of secondary sexual sity of these genera. Third, I evaluated the biogeographic history
Fig.1. (A)GeographicrangesofArthroleptis(red)andCardioglossa(blue);areasofoverlaparedepictedinpurple.Geographiczonesusedinthereconstructionofhistorical
biogeographyareindicated(W:WesternAfrica;C:CentralAfrica;A:AlbertineRiftMountains;E:EasternAfrica;S:SouthernAfrica).(B)Arthroleptis‘‘poecilonotus”inlife(MCZ
A-137982,subadult;Obang,NorthwestProvince,Cameroon).(C)Cardioglossamelanogasterinlife(MCZA-137905,adultfemale;Nsoung,SouthwestProvince,Cameroon).
808 D.C.Blackburn/MolecularPhylogeneticsandEvolution49(2008)806–826
withinArthroleptisandCardioglossatodeterminethespatialdiver- suprageneric taxonomies proposed in the era before molecular
sificationofthesegeneraacrossfivegeographiczonesinsub-Sah- phylogenetics (Laurent,1942, 1949,1951, 1961, 1973,1979; Du-
aran Africa. The results of the above analyses facilitate an bois, 1981, 1983, 1984, 1985, 1986, 1992; Frost, 1985) reflected
interpretationoftherelationship,ifany,thatexistsbetweenpat- some close relationship of these genera to the Astylosternidae
ternsofdiversityofbodysize,lifehistory,andbiogeography. (sensu Dubois, 1992; Astylosternus, Leptodactylodon, Nyctibates,
Scotobleps,andTrichobatrachus)andtheHyperoliidae(sensuLiem,
1970;Drewes,1984;e.g.,Afrixalus,Hyperolius,Kassina,andLeptop-
2.Materialsandmethods elis). Nearly all molecular phylogenetic studies have upheld the
closerelationshipof thesetaxa toa cladecomprisingArthroleptis
2.1.Taxonsampling (including Schoutedenella) and Cardioglossa (Vences et al., 2000,
2003a, 2003b; Biju and Bossuyt, 2003; van der Meijden et al.,
All species of Arthroleptis and Cardioglossa for which tissues 2004, 2007; Scott, 2005; Bossuyt et al., 2006; Frost et al., 2006).
wereavailableinmuseumcollectionswereincludedinthisanaly- Inaddition,mostmolecularphylogeneticstudieshavefoundthat
sis(Appendix1);theassignmentofspeciestogenerafollowsFrost the above taxa are part of a larger African ranoid clade that in-
(2007).Unfortunately,manyspeciesofArthroleptisandCardioglos- cludes the Brevicipitidae (sensu Frost et al., 2006; e.g., Breviceps
saarerepresentedonlybytypespecimens,whichcannotbedam- and Callulina) and the monotypic Hemisotidae (Hemisus). As this
aged and, in any event, tissues from these are likely unusable in studyfocusesontherelationshipsbetweenandwithinArthroleptis
molecular phylogenetic studies. Fieldwork in Malawi (in 2005) andCardioglossa,ratherthanthehigher-levelrelationshipsofAfri-
and Cameroon (in 2004 and 2006) significantly supplemented canranoidfrogs,representativespeciesofeachoftheabovegenera
the taxon sampling, including the discovery of new species (e.g., wereincludedtoserveonlyasoutgroupsinanalyses.Onespecies
Blackburn,2008;inthisstudy).Inrecognizingtaxaincludedinthis ofeachoutgroupgenuswasincluded(Appendix1)and,basedon
studyasnewspecies,Iadheretothegenerallineageconceptout- previous larger phylogenetic studies (e.g., Frost et al., 2006; Roe-
lined by de Queiroz (1998) and use the combination of phyloge- lants et al., 2007), the outgroup genera Hemisus, Breviceps, and
netic data and diagnosable morphological traits as criteria to Callulinawereusedtorootthetree.
recognizethesespecies.Inadditiontodatacollectedinthisstudy, To confirm or establish identification, voucher specimens for
DNAsequencedatawereobtainedfromGenBankfortwoingroup tissue samples were compared to the type specimens of nearly
species (Arthroleptis wahlbergii: FJ151052; Cardioglossa gratiosa: everyspeciesofArthroleptisandCardioglossaincludedinthisanal-
DQ283176).Thespeciesincludedinthisanalysisrepresentallspe- ysis(seeAppendix1). Theonlyvouchernotexaminedisa speci-
cies groups of Cardioglossa (Amiet, 1981; Blackburn, 2008) and men (MTSN 9178) of Arthroleptis nikeae that could not be
Arthroleptis species previously assigned to Schoutedenella (e.g., obtained on loan. Relevant type material was not examined for
Frost,2004),aswellasthesubgeneraAbroscaphus,Arthroleptulus, only three species: Arthroleptis stenodactylus, A. sylvaticus, and A.
and Coracodichus (Laurent, 1940, 1957, 1961); no tissue samples xenodactyloides.However,specimensfromthetypelocalitieswere
were available for Arthroleptis adolfifriederici, the type species of examined for both A. sylvaticus and A. xenodactyloides. The type
Abroscaphus(Laurent,1940,1957).Institutionalabbreviationsfol- materialofA.stenodactylusislostandpresumablydestroyeddur-
lowLevitonetal.(1985)withtwoadditions:AMCC,AmbroseMon- ing the Second World War (Frost, 1985); other A. stenodactylus
ell Cryo Collection (American Museum of Natural History, New wereexaminedfromnearthetypelocalityineasternTanzaniato
York);MM,MuseumsofMalawi,Blantyre. confirm identification. Because populations often referred to A.
IcollectedmitochondrialDNAsequencedata(seebelow)from poecilonotus likely represent multiple cryptic species (Rödel,
112specimensrepresenting26ingroupspeciesandsix outgroup 2000a; Rödel and Agyei, 2003; Rödel and Bangoura, 2004; Rödel
genera.Theingroupspeciesrepresentnearlyhalfofthedescribed et al., 2005), I refer to this taxon as A. ‘‘poecilonotus”. Even when
speciesdiversity(26of54species)ofArthroleptisandCardioglossa: checked against the holotype, specimens of Arthroleptis reichei
15of37speciesofArthroleptis,11and17speciesofCardioglossa, proved difficult to identify. There have been no previous sugges-
andatleastthreenewspeciesofArthroleptis.Notably,thesampling tionsthatthisisaspeciescomplex,butIcannotconsidertheiden-
of species previously assigned to Schoutedenella (Frost, 2004) is tificationofA.reicheispecimensinthisanalysisdefinitive.
verylow;onlyfourofthe19speciescouldbeincludedintheanal-
ysis(A.schubotzi,A.sylvaticus,A.xenodactylus,andA.xenodactylo-
ides).However,thisisparforthecoursewhenworkingonpoorly 2.2.Moleculardata
known African vertebrates. Much of the diversity of these small
speciesoccursinregions,suchaseasternDemocraticRepublicof GenomicDNAwasextractedfromtissuesamplesusingtheQia-
Congo, that have been essentially inaccessible to researchers for gen DNeasy Tissue Kit (Cat. No. 69506). A continuous stretch of
decades.SixspeciesofArthroleptis(A.bivittatus,A.brevipes,A.car- mitochondrial DNA ((cid:2)2400bp) comprising genes encoding 12S
quejai,A.krokosua,A.milletihorsini,andA.stridens)areknownonly and16SribosomalRNAandtheinterveninggeneencodingtransfer
fromtheholotypespecimen,andfullyelevenofthespeciesprevi- RNAforvaline,wasamplifiedusingeithertwoorfouroverlapping
ouslyassignedtoSchoutedenella(A.discodactylus,A.hematogaster, setsoftheprimerpairsofDarstandCannatella(2004)aswellas
A.milletihorsini,A.mossoensis,A.phrynoides,A.pyrrhoscelis,A.spina- standard PCR conditions similar to those used by those authors.
lis, A. stridens, A. troglodytes, A. vercammeni, and A. zimmeri) are In a few cases, taxonspecific primers were developed to amplify
known only from type specimens. In at least one case, the type regionsclosetopolynucleotideregionsthatinhibitedamplification
and only specimen is lost (A. milletihorsini; Ohler, pers. comm.), usingtheotherprimers(notshown).ThePCRconditionsusedwere
and most of the diversity of small species was unavailable even standard and the thermal cycle profile was as follows: 94(cid:3)C
for morphological study. However, as the species included span (3min); 35 cycles of 94(cid:3)C (30s), 48(cid:3)C (30s), 72(cid:3)C (1min);
the range of geography, elevation, ecology, life history, and body 72(cid:3)C(7min).Theannealingtemperaturewassometimesmodified
size found in Arthroleptis and Cardioglossa, this sampling is ade- (±3(cid:3)C) to improve the quality of the PCR product. PCR products
quateforaddressingbasicquestionsofphylogeneticrelationships, weresequencedusingcyclesequencingandtheABIPrismBigDye
characterevolution,andbiogeography. Terminatorkit(v.3.1;AppliedBiosystems).Thethermalcyclepro-
Overthepastseventyyears,thecloserelationshipofArthrolep- file for sequencing reactions was as follows: 25 cycles of 96(cid:3)C
tis and Cardioglossa has not been contested. Historically, most (10s), 50(cid:3)C (5s), 60(cid:3)C (4min). Sequencing reactions were puri-
D.C.Blackburn/MolecularPhylogeneticsandEvolution49(2008)806–826 809
fiedusingthePerformaDTRV396-WellPlatesfromEdgeBioSys- through WinClada v.1.00.08 (Nixon, 2002). Branches present in
tems(Cat.No.63887). P70%ofthebootstraptreeswereconsideredwell-supportedfol-
DNAsequencesofunequallengthfrom118terminaltaxawere lowingHillisandBull(1993;butseeWilcoxetal.,2002).
aligned in ClustalX v.1.83.1 (Thompson et al., 1997). This dataset The ML and Bayesian analyses were conducted using all four
included 12 exemplars of related outgroup genera, sequences of alignments (i.e., the three alignments generated using different
six ofwhichwereobtainedfromGenBank(Appendix1). Because gapcostsand theconsensusalignment).ModelTestv.3.7(Posada
a 30 fragment of the 12S rRNA gene of Cardioglossa aureoli (CAS and Crandall, 1998) was used to determine the best-fit model of
230187) could not be amplified, the data for this taxon were in- sequence evolution. A single partition was used for all analyses
cluded in the multiple alignment analysis in two fragments of these data. A ML search was conducted in GARLI V.0.95
(475bpof12Sand1447bpoftheremainingsequence).Following (Zwickl, 2006) using the model selected by ModelTest, with all
alignment, these two sequences were merged, using MacClade parameters estimated, and a neighbor-joining starting tree that
v.4.06(MaddisonandMaddison,2003),toformasinglesequence was generated in PAUP. This search was conducted three times
in the phylogenetic analyses with intervening bp considered as to make sure that GARLI was not stuck at a local optimum. Each
missingdata.Toexploretheeffectofalignmentonphylogenyesti- GARLIanalysiswasterminatedatleast3milliongenerationsafter
mation,threedifferentmultiplealignmentsweregenerated.These the last topological improvement. In each analysis, the last
alignmentsdifferinthevalueusedforthegapcost parameter in improvement occurred within the first 500,000 generations; the
multiplealignmentmodeinClustalX.Gapcostsof10,15,and20 topology was identical among replicates and (cid:3)lnL was nearly
were used. The gap cost value of 15 is the default in ClustalX identical among replicates. Three thousand bootstrap replicates
v.1.83.1; the default values were used for all other parameters. of ML searches were performed using PHYML v.2.4.4 (Guindon
Each alignment was checked manually for individual sequences and Gascuel, 2003) and a neighbor-joining tree as the starting
inwhichthealignmentwasobviouslyinerror(i.e.,largestretches topology. The PHYML bootstrap searches used the same model
ofdataconsistentlyonenucleotidepositionofffromtheposition of evolution as in GARLI in which all parameters were estimated
that would minimize the change across sites). The alignment of andsix ratecategorieswereused;asin theparsimonybootstrap
onlyonespecimen(A.stenodactylus,MCZA-137021)wasobviously analyses,topologiespresentinP70%ofbootstraptreeswerecon-
inerrorinallthreealignmentsdespitehavingsequenceessentially sideredwell-supported.Todeterminewhetherthesedataevolved
identical (<0.00% uncorrected pairwise divergence) to that of an- in a manner significantly different from a molecular clock (i.e.,
other specimen from the same locality (A. stenodactylus, MCZ A- constant rates), likelihood scores were calculated in PAUP for
137022); most nucleotide positions for MCZ A-137021 were one each of the phylogenies found from three replicate GARLI
base pair offfrom thoseofMCZ A-137022. The correction ofthis searches, for each of the four alignments, both with and without
alignmenterrorwas theonlymanual correctionmade.Thethree the molecular clock enforced and then evaluated using a likeli-
alignmentsweretrimmed relativeto asingle referencesequence hood ratio test (Felsenstein, 1981). The phylogenies from the
(A.‘‘poecilonotus”,MCZA-136818)suchthatmosttaxawererepre- three replicate GARLI analyses were each tested separately be-
sented for both the most 50 and 30 base pairs of the alignment, cause while the topology and (cid:3)lnL are essentially identical,
whichcorrespond,respectively,topositions2171–4486ofXenopus branch lengths vary slightly.
laevismitochondrialgenome(GenBankNC-001573). BayesianestimatesofphylogenywereobtainedusingMrBayes
Subsequently, a consensus alignment was generated in which v.3.1.1 (Huelsenbeck and Ronquist, 2001; Ronquist and Huelsen-
all nucleotide positions differing between the three alignments beck, 2003) and the model of evolution selected by ModelTest.
wereexcluded.Thisfollowstheexclusionalignmentmethodadvo- Bayesian analysis was run for five million generations, sampled
catedbyGatesyetal.(1993)thathasbeenemployedrecentlyin every 1000 generations, using four chains, a temperature of 0.2,
studies using mitochondrial data (e.g., Leaché and McGuire, and default priors. Because MCMC searches within this Bayesian
2006;Wiensetal.,2007).Inthisway,Iattemptedtoexcludechar- frameworkmaysometimesbetrappedatlocaloptima,threeaddi-
actersforwhichthehomologystatementsareuncertain(e.g.,Gate- tionalreplicateanalyseswereperformed.StationarityoftheBayes-
sy et al., 1993). The results of phylogenetic analyses using the iansearcheswasdeterminedbyplotting(cid:3)lnLofeachgeneration
consensus alignment were then compared to those of the align- against generation time. The phylogeny, posterior probabilities,
ments generated using different gap costs. This comparison en- andparametervalueswerethenestimatedfromthepostburn-in
abled me to determine what information, if any, is lost through treesfromallfourreplicates.Topologieswithposteriorprobabili-
theuseofaconsensusalignmentinphylogenyestimation. ties P95% were considered well-supported (Wilcox et al., 2002;
Mean genetic distances within and between Arthroleptis and butseeLeachéandReeder,2002).
CarioglossawerecalculatedinMEGAv.4.0.1(Tamuraetal.,2007) The effect of alignment on phylogeny estimation was deter-
usingtheuncorrectedpairwisemethodandthealignmentgener- minedinseveralways.Topologiesresultingfromaparticularana-
ated using a gap cost of 15. Standard error was estimated using lyticalmethod,butusingdifferentalignments, werequalitatively
500bootstrapreplicates. comparedtocontrastspecificdifferencesbetweentrees(i.e.,taxon
AissistertotaxonBinthistreebutnotthattree).Thedistances
2.3.Phylogeneticanalysis betweenparsimony(fromPAUP)andML(fromGARLI)topologies
resultingfromanalysisofdifferentalignmentswerequantitatively
Estimates of phylogeny were obtained using parsimony and comparedusingthepartitionmetricofPennyandHendy(1985)as
maximum likelihood (ML) criteria as well as within a Bayesian implementedinPAUP(‘‘symmetricdifferencemethod”).Lastly,the
framework. Parsimony searches, using each of the three align- measures of nonparametric bootstrap support and Bayesian pos-
mentsgeneratedwithdifferentgapcosts,wereperformedinPAUP* teriorprobabilitieswerecomparedforanalysesofeachalignment.
v.4.0b10(Swofford,2003)usingTBRbranchswappinginaheuris- Onemightexpectthatthecharactersexcludedfromanalysisina
ticsearchwith500randomadditionsequencereplicatesandgaps consensus alignment would be those with the most homoplasy.
treatedasmissingdata.Asimilaranalysiswasconductedforthe Thus,excludingthesefromanalysiswouldproducerelativelyhigh-
consensusalignment.Supportforbranchesintheparsimonytopol- ersupportvalues.Alternatively,aconsensusalignmentmayresult
ogieswasevaluatedusingnonparametricbootstrappingwith1000 in the exclusion of characters with autapomorphic states, espe-
bootstrapreplicates,usingTBRand10randomadditionsequences ciallyiftheseareindels,whichmaycauseuncertaintyinthealign-
per replicate, as implemented in Nona v.2.0 (Goloboff, 1993) mentofneighboringnucleotides.
810 D.C.Blackburn/MolecularPhylogeneticsandEvolution49(2008)806–826
2.4.Hypothesistesting 2007).Dataonthelifehistoryofeachingroupspeciesweregath-
ered from the literature(Appendix 1). Thereare surprisingly few
Support for alternative hypotheses of phylogenetic relation- reliablepublished accountsoflife history of Arthroleptisand Car-
shipswasevaluatedintwoways.Alltopologicalconstraintswere dioglossa.Inaddition,ongoingresearchusingDNA-barcodingiden-
generatedusingMacClade4.06and,forthesakeofsimplicity,only tified tadpoles of several Cardioglossa species (Blackburn and
theconsensusalignmentwasusedtotestalternativehypotheses. Sterne,unpublisheddata).Lifehistorydataforoutgrouptaxawere
First, tree lengths were obtained from parsimony analyses using takenfromLamotteandZuber-Vogeli(1954),Amiet(1970,1971,
unconstrainedandconstrainedtreesearchesconductedinPAUP*. 1989), Schiøtz (1999), Müller et al. (2007) and my own unpub-
These alternative phylogenetic hypotheses were evaluated using lisheddata.WiththeexceptionofLeptopelis,thelifehistoryofeach
atwo-tailedTempletontest(amodificationoftheWilcoxonsigns outgroupgenusisbelievedtobeinvariable.Becausetheonlypos-
ranked test; Templeton, 1983). Because both constrained and sible direct-developing Leptopelis (L. brevirostris) is nested within
unconstrained parsimony tree searches produced multiple most therestofLeptopelisdiversity(Idris,2004),Iconsidertheancestral
parsimonious trees, all trees were evaluated to determine if the lifehistoryofLeptopelistohavebeenlikelycharacterizedbyatad-
constrained trees were significantly different. The P-values that pole(e.g.,Schiøtz,1999).
summarize these results are the most conservative possible. Sec- Theevolutionofbodysize,acontinuouscharacter,wasrecon-
ond,theShimodaira–Hasegawa(SH)testwasusedtotestsupport structed using weighted squared-change parsimony in Mesquite.
for alternative phylogenetic hypotheses in a ML framework (Shi- Snout-ventlengthwasmeasured(±0.1mm)frommuseumspeci-
modaira and Hasegawa, 1999; Goldman et al., 2000). Using con- mensusingdigitalcalipers.Maximumbodysizeforeachingroup
strained tree searches in GARLI, maximum likelihood topologies species was estimated by using the largest snout-vent length
foralternativephylogenetichypothesesweregenerated,witheach (SVL)measuredforaspecies(Appendix2),regardlessofwhether
analysisterminatingthreemilliongenerationsafterthelasttopo- thatspecimenismaleorfemale.Thisapproachhastwoadvanta-
logical improvement. These topologies were then tested against ges:(1)itapproximatestheupperboundsofSVLforeachspecies,
theMLtopologyobtainedfromunconstrainedsearchesusingthe and(2)itenablestheinclusionofbothmaleandfemalespecimens.
SH test and 10,000 RELL (resampling estimate of log-likelihood) Because of error associated with different sampling intensity for
replicates in PAUP. A significant result (a=0.05) from either the eachspecies, these values servesimply as a heuristicmeasure of
TempletontestortheSHtestindicatesthatthealternativehypoth- maximumbody size. Two species in this analysis,Arthroleptissp.
esisisrejectedinfavorofthetopologyfoundbyparsimonyorML nov. 2 and Arthroleptis sp. nov. 3 are known only from juvenile
analyses,respectively. specimens,andthusbodysizedataarecodedasmissingforthese
species. Ancestral states at nodes immediately ancestral to these
2.5.Biogeographichistory twospeciesarenotreconstructed.Thetopologiesusedforances-
tral state reconstruction are the same ML phylogenies used in
Historicalbiogeographywasinvestigatedusingdispersal–vicar- theabovebiogeographicanalyses.
ianceanalysisasimplementedinDIVAv.1.1(Ronquist,1997).The Body size of outgroup taxa was estimated as follows. For all
topologiesusedforthisanalysisaretheMLphylogeniesgenerated genera except Nyctibates, the maximum snout-vent length was
byGARLIforeachalignment;thesetreesweresimplifiedsuchthat obtained for each species treated in key taxonomic literature
allintraspecificrelationships,exceptforthosewithinA.‘‘poecilon- (Afrixalus: Schiøtz, 1999; Astylosternus: Amiet, 1977; Breviceps:
otus”,wereexcluded.Becausetheanalysesoftheconsensusalign- Channing, 2001, B. fichus: Channing and Minter, 2004; Callulina:
mentfailedtoresolvethemitochondrialgenelineagesofC.oreas deSáetal.,2004;Hemisus:Laurent,1972b,H.barotseensis:Chan-
andC.manengoubaasmonophyletic(seeSection3),thesetwospe- ning, 2001; Hyperolius: Schiøtz, 1999; Kassina: Schiøtz, 1999;
cieswereeachgivenabranchlengthtwoordersofmagnitudeless Leptodactylodon: Amiet, 1980, L. blanci: Ohler, 1999, L. stevarti:
thanotherbranchesinthistree(i.e.,0.00001)toapproximatethe Rödel and Pauwels, 2003; Leptopelis: Schiøtz, 1999; Scotobleps:
unresolvedrelationshipbetweenthesespecies;branchlengthsof Perret, 1966; Trichobatrachus: Perret, 1966). Then using these
0.0mayinterferewithancestralstatereconstructionsofcontinu- maximumsnout-ventlengthvalues,themedianvaluewascalcu-
ouscharacters(Revell,pers.comm.). lated for each genus (Appendix 1); results are very similar if
Thegeographicdistributionofeachspecieswasobtainedfrom mean values of maximum snout-vent lengths are used instead
therecentGlobalAmphibianAssessment(IUCN,2006).Thesedis- (data not shown). Because the phylogenetic relationships within
tributions were then coded as corresponding to five geographic each of these genera remain largely unclear and were not esti-
zones(Fig.1)basedlooselyonbioticregionsrecognizedbyBurgess matedinthisanalysis,usingthemediansnout-ventlengthvalues
etal.(2004).Thesezonesare(1)WesternAfrica(westof,butnot isaconservativeapproachthatestimatesbodysizeinthesegen-
including, Nigeria; thus the Upper Guinean Forest Zone is in- era without skewing ancestral state reconstructions in outgroup
cluded); (2) Central Africa (including the Lower Guinean Forest taxa to extreme values. Maximum snout-vent length for Nycti-
Zone,CameroonVolcanicLine,andlowlandforestsofCongoRiver bates corrugatus was measured directly from three MCZ speci-
Basin);(3)AlbertineRiftMountains(includingeasternDemocratic mens (MCZ A-136788–89, A-136939).
RepublicofCongo,Burundi,Rwanda,andUganda);(4)EasternAfri-
ca (including the Eastern Arc Mountains); (5) Southern Africa 3.Results
(countries south of, and including, Malawi and Mozambique).
The geographic ranges assigned to outgroup taxa summarize the 3.1.Parsimonyanalyses:differentgapcostsvs.consensusalignment
entirerangeofthatgenus.Thisispredicatedontwoassumptions.
First, each outgroup genus is assumed to be monophyletic, and, Because of differences in the numbers of insertions and dele-
second,thegeographicrangeofextanttaxapreservessomerepre- tions inferred through multiple alignment using different gap
sentationoftheirhistoricaldistribution. costs, the final data matrices for the three alignments generated
vary slightly in length (gap cost 10 [gc10]: 2548bp; gap cost 15
2.6.Ancestralstatereconstructions [gc15]: 2543bp; gap cost 20 [gc20]: 2533bp). Homology state-
mentsdifferformorethan700charactersbetweenthethreealign-
Ancestralstatesforlifehistorywerereconstructedusingunor- ments; when these characters are removed, the resulting
dered parsimony in Mesquite v.2.0 (Maddison and Maddison, consensusalignmentis1800bpinlength.Theresultsofheuristic
D.C.Blackburn/MolecularPhylogeneticsandEvolution49(2008)806–826 811
Table1
Summaryofparsimony(Pars.;PAUP),ML(GARLI),andBayesiananalyses
Alignment TotalChar. #Constant #Pars.inform. Treelength #Pars.trees ML(cid:3)lnL Meanbayes(cid:3)lnL±standarddeviation
Gapcost=10 2548 936 1366 9601 22445 43229.73a 43387.17±13.05
Gapcost=15 2543 906 1371 9731 7219 43677.72 43835.84±13.11
Gapcost=20 2533 900 1382 9875 2900 44150.50 44307.33±12.98
Consensus 1800 812 799 4743 624 23532.96 23709.79±14.25
a Mean(cid:3)lnLofthreeGARLIruns:43229.44,43229.88,43229.86.
parsimony searches using these four alignments are summarized In four instances, parsimony analysis of gc10, gc15, and gc20
inTable1. alignments found a particular interspecific topology that differed
MP analysis of different alignments produced largely similar fromthatoftheconsensusalignment.Intwocases,thereishigh
patterns of phylogenetic relationships among ingroup taxa bootstrap support from the gc10, gc15, and gc20 analyses. These
(Fig. 2). The primary differences in topologies between analyses topologiesandtheirrespectivesupportfromtheselatteranalyses
pertaintotherelativeplacementofArthroleptiswahlbergiiandCar- are as follows: (C. gracilis (C. melanogaster, C. schioetzi)), 94/83/
dioglossaaureoli.Inallparsimonyanalyses,thereisstrongsupport 93; ((C. leucomystax, C. occidentalis), (C. gratiosa (C. pulchra (C.
for A. wahlbergii being more closely related to large Arthroleptis manengouba, C. oreas)))), 84/84/89; (C. gratiosa (C. pulchra (C.
than to C. aureoli, A. sylvaticus, or A. xenodactyloides. However, in manengouba,C.oreas))),61/63/65;(Arthroleptissp.nov.3(A.xeno-
theparsimonyanalysesofthegc10andgc15alignments,A.wahl- dactylus(A.schubotzi,A.xenodactyloides))),63/64/73.
bergii is the sister taxon to the remaining large Arthroleptis, MPanalysisofallfouralignmentsresolvedScotoblepsasthesis-
whereasintheanalysesofthegc20andconsensusalignments,it tertaxontoArthroleptisandCardioglossa.However,thesupportfor
is the sister species of A. francei. Similarly, in analyses of gc10, this topology is low. Indeed, only the parsimony analysis of the
gc15,orconsensusalignments,C.aureoliiseitherthesistertaxon consensus alignment produced bootstrap support >50%, but it
toornestedwithinArthroleptis,whereasanalysisofthegc20align- was still low in this case (64%). All parsimony analyses found
mentplacesitasthesistertaxontoArthroleptisandotherCardio- strongsupport(>70%bootstrapsupport)forthreeoutgroupclades:
glossa.Thetopologyfoundinanalysisoftheconsensusalignment (1)Hemisus(Breviceps,Callulina);(2)Kassina(Afrixalus,Hyperolius);
also differs within Cardioglossa. In contrast to the analyses of and (3) ((Astylosternus, Trichobatrachus) (Leptodactylodon,
gc10, gc15, gc20 alignments in which the most parsimonious Nyctibates)).
topology is (C. elegans (C. leucomytax, C. occidentalis) (C. gratiosa
((C. manengouba, C. oreas) C. pulchra))), the most parsimonious 3.2.MLandBayesiananalyses
topology resulting from the consensus alignment is (C. gratiosa
(C. elegans (C. leucomystax, C. occidentalis)), C. pulchra (C. manen- BothhierarchicallikelihoodratiotestsandtheAkaikeinforma-
goba, C. oreas)). The gc10, gc15, and gc20 parsimony analyses all tioncriterion,asimplementedinModelTest,selectedtheGTR+I+C
produced reciprocal monophyly of C. manengouba and C. oreas, modelasthebest-fittingmodelforallfouralignments.Thismodel
but the analysis of the consensus alignment produced paraphyly of sequence evolution was used in ML analyses using GARLI and
of C. oreas with respect to C. manengouba. Likewise, analysis of PHYMLaswellasinBayesiananalyses.Estimatedbasefrequencies
thegc10,gc15,andgc20alignmentsresolveC.gracilisasthesister weresimilaramongallfouralignments,buttheestimatedtransi-
taxonto(C.melanogaster,C.schioetzi),whereasanalysisofthecon- tion and transversion rates were consistently higher for the con-
sensusalignmentlocatedC.gracilisasthesistertaxontoallother sensusalignment;parametervaluesestimatedbyModelTestare
Cardioglossa (except C. aureoli). Lastly, parsimony analysis of the reported in Table 2. For each alignment, the hypothesis that the
consensusalignmentfoundArthroleptissp.nov.3tobethesister dataareultrametricisstronglyrejected(P<0.0001).
species of C. aureoli, whereas analysisof the otheralignmentsall Each Bayesian analysis reached stability by 500,000 genera-
produceatopologyinwhichArthroleptissp.nov.3isthesistertax- tions. Thus, to be conservative, the first one million generations
on to (A. xenodactylus (A. schubotzi, A. xenodactyloides)). Several (i.e.,1000trees)fromeachanalysiswereexcludedfromcomputa-
nodes found in the parsimony analyses of the gc10, gc15, gc20 tionofposteriorprobabilities.Themean(cid:3)lnLandstandarddevia-
alignmentsdidnothaveP50%bootstrapsupportintheparsimony tionforpostburn-intreesarereportedinTable1.
analysisoftheconsensusalignment(Fig.2). Thespecies-leveltopologiesproducedfromBayesianmajority-
Nonparametricbootstrappingprovidessimilarlevelsofhighor ruleconsensusandMLanalysesfrombothGARLIandPHYMLare
low support among analyses of the three alignments generated allsimilar.IntheMLtopologyfromthegc10,gc15,andgc20GARLI
usingdifferentgapcosts(Fig.2).Thereareonlytwoingroupnodes analyses,everyspeciesismonophyletic(Fig.3).Thisisnottruefor
that have <70% bootstrap support in all three of these analyses the ML topology produced from the analysis of the consensus
(Fig.2).Moreinterestingly,therearetwootheringroupnodesfor alignment in whichone species,C. oreas, isparaphyleticwithre-
whichthesupportis<70%inoneortwoanalysesbut>70%inthe specttoC.manengouba(seeSection4);thisissimilartotheresults
remaininganalyses(grayovalsinFig.2).Inoneofthesecases,the from the parsimony analyses. In general, the Bayesian posterior
supportforthenodeunitingArthroleptissp.nov.1,Arthroleptissp. probabilitiesandMLbootstrapvaluesaresimilarinthatmanyof
nov.2,andA.variabilisrangesfrom67%to90%.Inalmosteverycase, thesamenodeshaveeitherloworhighsupportinbothanalyses.
bootstrap support from analysis of the consensus alignment is TherearethreemajorpointsofdifferencebetweentheMLspe-
eitherthesameorlower,sometimesmuchlower,thanthesupport cies-level topology and the consensus summary of post burn-in
forthesametopologiesfoundinanalysisoftheotherthreealign- Bayesian trees produced from analyses of the four alignments.
ments.Thereareonlythreecasesinwhichthesupportforthecon- First,whilethereisstrongsupportforacladecomprisingA.sylvat-
sensus alignment exceeds that for the other alignments; in two icus,A.xenodactylus,andA.xenodactyloides,therelationshipamong
cases,itisonlymarginallyhigher(consensusvs.gc10/gc15/gc20: thesethreespeciesarenotresolveddefinitively(Fig.3).Second,a
90vs.89/88/81;82vs.75/78/77),whereasitissubstantiallyhigher sister species relationship between A. wahlbergii and A. francei
foronenode(90vs.72/75/78;seeFig.2formoredetails). exhibitshighsupportfromBayesianand MLanalysisofthegc20
812 D.C.Blackburn/MolecularPhylogeneticsandEvolution49(2008)806–826
Fig.2. Strictconsensustopologyofinterspecificrelationshipsbasedonparsimonyanalysis.Intraspecificrelationshipshavebeencollapsedandonlythenearestoutgroup
taxonisshown.Numbersabovebranchesarenonparametricbootstrapproportionsforparsimonyanalysesinwhichgapcostsequal10,15,and20,respectively;numbers
belowbranchesarenonparametricbootstrapproportionsforparsimonyanalysisoftheconsensusalignment.Grayovalsindicatenodesforwhichnonparametricbootstrap
*
proportionsfromdifferentanalysesspan70%support.SpeciespreviouslyassignedtoSchoutedenellaareindicated.100%bootstrapsupportisindicatedby‘‘”;lessthan50%
bootstrapsupportisindicatedby‘‘-‘‘.
Table2
(cid:3)Lnlikelihood,AIC,andparametervaluesfortheGTR+I+CmodelselectedbyModelTestv.3.7
Alignment (cid:3)LnL AIC f(A) f(C) f(G) f(T) R[A-C] R[A-G] R[A-T] R[C-G] R[C-T] I a
Gapcost=10 43276.54 86573.08 0.408 0.211 0.140 0.241 4.426 9.296 4.669 1.216 32.786 0.192 0.684
Gapcost=15 43711.96 87443.92 0.412 0.211 0.136 0.241 4.076 8.968 4.382 1.345 30.500 0.177 0.664
Gapcost=20 44193.21 88406.41 0.417 0.208 0.134 0.241 3.821 8.195 4.031 1.378 28.146 0.184 0.676
Consensus 23560.77 47141.54 0.376 0.218 0.167 0.240 5.354 15.418 6.622 1.672 47.905 0.214 0.557
andconsensusalignmentsbutnotthegc10andgc15alignments. EasternArcArthroleptisisalsonotwellresolved.Severalanalyses
TherelationshipofthesetwospeciestoA.tanneriandotherlarge (ML:gc10,gc15,gc20;Bayes:gc10,gc15)findthatallofthesespe-
D.C.Blackburn/MolecularPhylogeneticsandEvolution49(2008)806–826 813
Fig.3. Maximumlikelihoodphylogramsfromanalysisofgapcost10(gc10;A),gc15(B),gc20(C),andconsensus(D)alignments.Brancheswithin‘‘dottedline”boxare
magnifiedby100(cid:4)relativetootherbranches.NumbersabovebranchesarenonparametricbootstrapproportionsfromPHYMLanalysis;numbersbelowbranchesare
Bayesianposteriorprobabilities.100%bootstrapsupportand1.00posteriorprobabilityareindicatedby‘‘(cid:5)”;lessthan50%bootstrapsupportor0.50posteriorprobabilityare
indicatedby‘‘(cid:3)”.SpeciespreviouslyassignedtoSchoutedenellaareindicated.
814 D.C.Blackburn/MolecularPhylogeneticsandEvolution49(2008)806–826
Fig.3(continued)
D.C.Blackburn/MolecularPhylogeneticsandEvolution49(2008)806–826 815
Fig.3(continued)