combined molecular phylogenetic analysis of the orthoptera...

233

Syst. Biol. 48(2):233–253, 1999

Combined Molecular Phylogenetic Analysis of the Orthoptera(Arthropoda, Insecta) and Implications for Their Higher Systematics

P. K. FLOOK,1 S. KLEE, AND C. H. F. ROWELLZoology Institute, University of Basel, 4051-Basel, Switzerland

Abstract.— A phylogenetic analysis of mitochondrial and nuclear rDNA sequences from species ofall the superfamilies of the insect order Orthoptera (grasshoppers, crickets, and relatives) con�rmedthat although mitochondrial sequences provided good resolution of the youngest superfamilies,nuclear rDNA sequences were necessary to separate the basal groups. To try to reconcile thesedata sets into a single, fully resolved orthopteran phylogeny, we adopted consensus and combineddata strategies. The consensus analysis produced a partially resolved tree that lacked several well-supported features of the individual analyses. However, this lack of resolution was explained byan examination of resampled data sets, which identi�ed the likely source of error as the relativelyshort length of the individual mitochondrial data partitions. In a subsequent comparison in whichthe mitochondrial sequences were initially combined, we observed less con�ict. We then used twoapproaches to examine the validity of combining all of the data in a single analysis: comparativeanalysis of trees recovered from resampled data sets, and the application of a randomization test. Be-cause the results did not point to signi�cant levels of heterogeneity in phylogenetic signal betweenthe mitochondrial and nuclear data sets, we therefore proceeded with a combined analysis. Recon-structing phylogenies under the minimum evolution and maximum likelihood optimality criteria,we examined monophyly of the major orthopteran groups, using nonparametric and parametricbootstrap analysis and Kishino–Hasegawa tests. Our analysis suggests that phylogeny reconstruc-tion under the maximum likelihood criteria is the most discriminating approach for the combinedsequences. The results indicate, moreover, that the caeliferan Pneumoroidea and Pamphagoidea,as previously suggested, are polyphyletic. The Acridoidea is rede�ned to include all pamphagoidfamilies other than the Pyrgomorphidae, which we propose should be accorded superfamily status.[Combined analysis; insect phylogeny; molecular evolution; Orthoptera; ribosomal DNA.]

We have used phylogenies reconstructedfrom nucleotide sequences to examine theevolutionary history of the insect order Or-thoptera (grasshoppers, crickets, and rel-atives) (Flook and Rowell, 1997a, 1997b,1998). Of particular interest are relationshipsof several of the higher taxa, and we haveattempted to relate our results to existingsystematic disputes. The molecular phylo-genies are well suited to this task and havethe potential to resolve several outstandingproblems concerning the ecology, characterevolution, and biogeographic distribution ofmember taxa. However, the success of theseanalyses has been limited since, on theirown, the different sequences haveproven in-adequate for reconstructing phylogeny overthe whole range of orthopteran evolution.Because of this, we have so far been unableto reduce the �ndings of our work to a singleschema.

The failure of these analyses to generatefully resolved phylogenies may stem fromthe antiquity of the Orthoptera. Divergencedates of the major groups are estimated torange over a period of ~ 200 million years,with fossils of the oldest groups appearingin the Permian (Carpenter and Burnham,1985). Consequently, although the relativelyrapidly-evolving mitochondrial sequenceshave proven valuable for examining com-paratively recent events (Flook and Row-ell, 1997b), the basal branching patterns ofthe Orthoptera are resolved only by themore slowly evolving nuclear ribosomalRNA gene sequences (Flook and Rowell,1998). In contrast, these nuclear sequencesare almost invariant in the more recentlyevolved groups, and wewere previously un-able to detect signi�cant phylogenetic signalamong the four youngest caeliferan super-families. On the basis of these results, we ex-pect that the data sets contain enough phy-logenetic signal between them to determinemost of the major features of orthopteranphylogeny in a single analysis. The purpose

1Address correspondence to Dr. P. K. Flook, Zool-ogisches Institut, Rheinsprung 9, 4051-Basel, Switzer-land. E-mail: �[email protected].

234 SYSTEMATIC BIOLOGY VOL. 48

of the work reported here was to identifyan effective strategy for resolving the or-thopteran phylogeny from the three datasets.

Two fundamentally different approachesfor treating multiple phylogenetic data setsare commonly applied: consensus, and com-bined analyses. Both of these strategies havebeen criticized (Bull et al., 1993; de Queiroz,1993), and selection of one over the other iscomplicated. An argument in favor of com-bining data is that, because most reconstruc-tion methods are consistent (for most un-derlying tree shapes), an analysis is morelikely to recover the correct phylogeny asdata are added. On the other hand, phylo-genetic reconstruction can become compli-cated when data that evolve at very differ-ent rates (e.g., mitochondrial and nuclearDNA) are combined in a single analysis. Ashas been demonstrated, when heterogene-ity exists in phylogenetic signal from dif-ferent data partitions, the overall signal issometimes diminished after pooling of data;in such cases, data combination should beavoided (Bull et al., 1993; de Queiroz, 1993).Bull et al. (1993), discussing the alternativesfor analyzing multiple data sets, maintainthat a combining of data should be precededby a search for any con�ict between the in-dividual data sets.

Here we estimate relationships amongorthopterans from new and previously pub-lished molecular data and, in doing so,examine several important aspects of phy-logeny reconstruction. First, we consider thechoice of reconstruction method in situa-tions where contrasting phylogenetic lev-els are examined simultaneously. Second,we compare the use of both consensus anddata combination approaches for analyz-ing the mitochondrial and nuclear riboso-mal DNA (rDNA) sequences. We suggestthat consensus methods are tooconservativefor the treatment of the orthopteran data.We demonstrate that combining nuclear andmitochondrial DNA (mtDNA) sequences ina single analysis is legitimate, in spite of thefact that the sequences are evolving at dif-ferent rates. Reconstructing trees from thecombined data set under the criteria of min-imum evolution (ME) and maximum likeli-

hood (ML) optimality, we obtain a phyloge-netic scheme in which the majority of nodesare resolved.

MATERIALS AND METHODS

Samples

Samples from all the orthopteran super-families were included in this study. In thesuborder Caelifera (shorthorned grasshop-pers) seven superfamilies are commonlyrecognized (see Dirsh, 1975; Rentz, 1991):Tridactyloidea (false mole crickets, sandgropers); Tetrigoidea (pygmy grasshoppers,grouse locusts); Eumastacoidea (monkeygrasshoppers, false stick insects); Pneu-moroidea (�ying gooseberries, desert long-horned grasshoppers, razor-back bush-hoppers); Pamphagoidea (rugged earthhoppers, true bush-hoppers); Acridoidea(grasshoppers, locusts); and Trigonoptery-goidea. In the other orthopteran suborder,Ensifera (long-horned grasshoppers, katy-dids, crickets), we sampled all four super-families (following the higher classi�cationof Gorochov, 1995b): Tettigonioidea (bush-crickets, katydids); Hagloidea (hump-winged crickets); Stenopelmatoidea (cavecrickets, Jerusalem crickets, wetas); andGrylloidea (true crickets, mole crickets). Inaddition we used outgroup taxa from threerelated orders: Phasmida (walking sticks);Blattodea (cockroaches); and Grylloblat-todea (ice-crawlers). A full list of the mate-rial used in this study is given in Table 1. Theseven taxa in which sequences were deter-mined for the �rst time were Rhainopommamontanum (Caelifera, Acridoidea, Lentu-lidae), an unidenti�ed species of Systellafrom Borneo (Caelifera, Trigonopterygoid-ea, Trigonopterygidae), Tanaocerus koebeli(Caelifera, Pneumoroidea, Tanaoceridae),Xyronotus aztecus (Caelifera, Pneumoroidea,Xyrolnotidae), Physemacris variolosa (Caeli-fera, Pneumoroidea, Pneumoridae), Comi-cus campestris (Ensifera, Stenopelmatoidea,Stenopelmatidae), and Ceuthophilus carlsbad-ensis (Ensifera, Stenopelmatoidea, Rhaphid-ophoridae). Procedures for collection andstorage of material and for DNA isolation

1999 FLOOK ET AL.—PHYLOGENETIC ANALYSIS OF THE ORTHOPTERA 235

TAB

LE

1.Ta

xono

mic

sam

ple

exam

ined

inth

isst

udy.

The

high

erta

xono

mic

grou

psof

the

Cae

lifer

aan

dE

nsif

era

iden

ti�ed

here

corr

espo

ndto

thos

elis

ted

byD

irsh

(197

5)an

dG

oroc

hov

(199

5b),

resp

ectiv

ely.

The

Acr

idom

orph

oid

ease

nsu

Dir

sh(1

975)

corr

espo

nds

toth

esu

perf

amili

esA

crid

oide

a,Pa

mph

agoi

dea

,Pn

eum

oroi

dea,

and

Trig

onop

tery

goid

ea.

EM

BL

acce

ssio

nno

s.O

rder

Subo

rder

Supe

rfam

ilyFa

mily

Gen

usSp

ecie

s12

S16

S18

S

Ort

hopt

era

Cae

lifer

aA

crid

oid

eaA

crid

idae

Gom

phoc

erip

pus

rufu

sZ

9324

7Z

9328

5Z

9759

1A

crid

atu

rrit

aZ

9759

6aZ

9761

2aZ

9756

0O

edip

oda

coer

ules

cens

Z93

255

Z93

293

Z97

573

Lent

ulid

aeR

hain

opom

ma

mon

tanu

mZ

9760

1aZ

9761

8aZ

9758

1a

Pam

phag

oid

eaPa

mph

agid

aeG

laui

ate

rrea

Z93

258

Z93

296

Z97

568

Bat

rach

otet

rix

sp.

Z93

259

Z93

297

Z97

590

Pyrg

omor

phid

aeP

rosp

hena

scud

deri

Z93

261

Z93

299

Z97

579

Pyr

gom

orph

aco

nica

Z97

600a

Z97

616a

Z97

580

Pneu

mor

oide

aPn

eum

orid

aeB

ulla

cris

mem

brac

ioid

esZ

9326

4Z

9330

2Z

9756

2P

neum

ora

inan

isZ

9326

5Z

9330

3Z

9757

7P

hyse

mac

ris

vari

olos

aZ

9759

9aZ

9761

5aZ

9757

6a

Tana

ocer

idae

Tana

ocer

usko

ebel

iZ

9760

5aZ

9762

1aZ

9758

6a

Xyr

onot

idae

Xyr

onot

usaz

tecu

sZ

9760

7aZ

9762

3aZ

9758

8a

Trig

onop

tery

goid

eaTr

igon

opte

rygi

dae

Syst

ella

raf�

esi

Z97

604a

Z97

620a

Z97

585

Syst

ella

sp.

Z97

603a

Z97

619a

Z97

584a

Eum

asta

coid

eaPr

osco

piid

aeP

rosa

rthr

iate

retr

iros

tris

Z93

266

Z93

304

Z97

578

Stip

hra

robu

sta

Z93

267

Z93

305

Z97

583

Eusc

hmid

tiid

aeE

usch

mid

tia

cruc

iform

isZ

9760

9aZ

9762

5aZ

9756

7Eu

mas

taci

dae

Hom

eom

asta

xde

ntat

aZ

9759

8aZ

9761

4aZ

9757

1Te

trig

oid

eaBa

trac

hid

eida

eIn

det

sp.

Z93

274

Z93

312

Z97

631

Tetr

igid

aeP

arat

etti

xcu

culla

tus

Z93

273

Z93

311

Z97

574

Trid

acty

loid

eaTr

idac

tylid

aeN

eotr

idac

tylu

sau

stra

lisZ

9327

5Z

9331

3Z

9757

2C

ylin

drac

hetid

aeC

ylin

drau

stra

liako

chii

Z93

277

Z93

315

Z97

565

Ensi

fera

Tetti

goni

oid

eaTe

ttig

oniid

aeTe

ttig

onia

viri

diss

ima

Z97

606a

Z97

622a

Z97

574

Rus

polia

niti

dula

Z97

602a

Z97

618a

Z97

582

Sten

opel

mat

oid

eaSt

enop

elm

atid

aeC

omic

usca

mpe

stri

sZ

9769

8aZ

9762

4aZ

9756

4a

Mim

nerm

idae

Hem

idei

nacr

assi

dens

Z93

278

Z93

316

Z97

570

Rha

phid

opho

rid

aeC

euth

ophi

lus

carl

sbad

ensi

sZ

9759

7aZ

9761

3aZ

9756

3a

Hag

loid

eaH

aglid

aeC

ypho

derr

ism

onst

rosu

sZ

9327

9Z

9331

7Z

9756

6G

ryllo

idea

Gry

llida

eG

ryllu

sca

mpe

stri

sZ

9328

0Z

9331

8—

b

Ach

eta

dom

esti

cus

Z97

611a

Z97

627a

—b

Phas

mid

aPh

yllid

aeP

hylli

umbi

ocul

atum

Z93

281

Z93

319

Z97

575

Pseu

dop

hasm

atid

aeA

gath

emer

acr

assa

Z93

282

Z93

320

Z97

561

Gry

llobl

atto

dea

Gry

llobl

attid

aeG

ryllo

blat

taro

thi

Z93

283

Z93

321

Z97

569

Bla

ttod

eaPa

nchl

orid

aeG

rom

phad

orhi

napo

rten

tosa

Z97

610a

Z97

626a

Z97

592

aPr

evio

usly

unp

ublis

hed

sequ

ence

sob

tain

edfo

rth

isst

ud

y.b

Sequ

ence

sno

tinc

lud

edin

this

stud

y.


have been published previously (Flook andRowell, 1997a, 1998).

Sequence Data

We obtained 3,177 bp (total alignmentlength) of sequence from three genes: 393bp of the 3 0 half of the mitochondrial small-subunit rRNA gene (12S); 558 bp from the3 0 half of the mitochondrial large-subunitrRNA gene (16S); and the complete se-quence (2,226 bp) of the nuclear small-subunit rRNA gene (18S). For details ofthe PCR ampli�cation of the mitochon-drial gene sequences, see Flook and Rowell(1997a) . Nuclear 18S sequences were ampli-�ed, cloned, and sequenced as described inFlook and Rowell (1998). All sequences havebeen deposited in the EMBL database; a listof the corresponding accession numbers isgiven in Table 1.

Raw data were analyzed with the Li-CorImage-Analysis software (version 2.3) andcontig assembly was performed by usingthe AssemblyLign package (Oxford Molec-ular Group). New sequences were added tothe existing sequence alignments and editedmanually with use of the SeqApp program(D. Gilbert, Univ. Indiana). We attempted toresolve dif�cult regions of the DNA align-ment by referring to the secondary structureof the three genes (Flook and Rowell, 1997b,and unpubl. data); in several regions, how-ever, the alignment remained uncertain, andwe omitted these regions from subsequentanalyses. A computer �le containing the se-quence data in NEXUS format is availablefrom P.F.K. on request.

Analyses

Substitution patterns.—Preliminary anal-yses indicated that sequences differ fromeach other with respect to base composi-tion (determined by comparison of uncor-rected and LogDet nucleotide substitutiondistances: Lockhart et al., 1994), variability,and distribution of among-site rate varia-tion (through estimation of gamma distribu-tion shape parameters). To examine the un-derlying patterns of nucleotide substitutionin more detail, we adopted a ML approach(e.g., see Huelsenbeck and Crandall, 1997).Tocompare the�tof substitution models, we

used likelihood ratio tests (LRTs) to assessthe statistical signi�cance of differences be-tween likelihood scores obtained for a givenphylogeny under various assumptions. Thephylogeny in question was estimated byusing equally weighted parsimony. Likeli-hoods were calculated with the computerprogram PAUP* 4d61 (test version providedby D. Swofford, Smithsonian Institution).We calculated the test statistic as 2(ln L0– lnL1) = –2 ln L, where L0 and L1 are the likeli-hoods under the null and alternate hypothe-ses, respectively. The signi�cance of the LRTstatistic was tested by using a x 2 distributionwith n degrees of freedom, where n is thenumber of parameters that differ betweensubstitution models.

Phylogenies were reconstructed and an-alyzed by using PAUP * . Heuristic searcheswere conducted under the maximum parsi-mony (MP), ME, and ML optimality criteria.For MP we performed equally weighted andcharacter state–weighted heuristic searches.Under the ME optimality criterion, after theresults of the preliminary likelihood analy-ses searches, we used the generalized nu-cleotide substitution model of Hasegawaet al. (1985), HKY85, with and without agamma correction for among-site rate vari-ation. We calculated the shape parameter byusing the methods of Sullivan et al. (1995)and Yang and Kumar (1996) as implementedin PAUP * . To calculate phylogenies underthe ML optimality criterion, we used an it-erative strategy (Swofford et al., 1996). Ini-tially, we used a MP tree as a starting tree forthe analysis. We then performed a heuris-tic search, using the optimal substitutionmodel identi�ed in the preliminary like-lihood analysis (see above), and simulta-neously estimating model parameters. Weused the resulting phylogeny as the startingpoint for another search and repeated thisprocess until the analysis converged on asingle phylogeny. For ME and MP trees, weused nonparametric bootstrap analysis toassess the con�dence that could be attachedto the individual nodes.

Effect of sequence length on phylogeny estima-tion.—We adopted a resampling approachto simulate the effect of increasing sequencelength and examined its in�uence on phylo-


genetic reconstruction. We sampled the orig-inal data sets with replacement to generatenew samples of various sequence lengths.For each sequence length we generated 50new samples. New samples of between 100and 2,000 bp in size were generated in 100-bp increments; new samples of between2,500 and 5,000 bp were generated in 500-bp increments. Resampled data sets weregenerated with use of a program written byP. K. F. (DOS program available from P.K.F.on request). Phylogenies were reconstructedby two methods: heuristic searches underthe MP criterion, and heuristic searches un-der ME criterion (HKY85 distance). We didnot perform analyses under the ML opti-mality criterion because computation timeswere prohibitive. In cases when multipletrees were obtained for data sets, strict con-sensus trees were retained for analysis. Theresults of searches were then assessed bycalculating the average symmetric differ-ence (Penny and Hendy, 1985) between alltrees in different size samples (by using theTREEDIST command in PAUP * ). Using thisapproach, we were then able toobtain a mea-sure of convergence with resampled datasets upon an unspeci�ed tree.

Randomization test of phylogenetic hetero-geneity among data partitions.—The random-ization test of Rodrigo et al. (1993) was usedto investigate the legitimacy of combiningthe data by examining the hypothesis thatdifferences between phylogenies estimatedfrom different data sets were due to sam-pling error. The test has the advantage inthe present situation of being applicable un-der any optimality criterion; we used it toexamine data combination under ME. The�rst stage of performing the test involvedgenerating two sets of 500 nonparametricbootstrap samples from each of the individ-ual data partitions and from the 12S + 16Sdata partition. After estimating trees fromthe bootstrap replicates by using ME, we cal-culated symmetric differences for each of the500 pairs of trees, using the compute pro-gram Component (Page, 1993). The result-ing tree-to-tree distances were then plottedas frequency histograms, which should cor-respond to the null distribution of tree-to-tree distances when two data sets estimate

the same historical events. To perform thetest, we then compared the observed tree-to-tree distance for phylogenies reconstructedfrom two data sets against the correspond-ing null distributions. This has three pos-sible outcomes. First, if the observed tree-to-tree distance is less than the 95% valueof both null distributions, we do not rejectthe hypothesis that the observed differencesare due to sampling error. Second, if the ob-served tree-to-tree distance is greater thanthe 95% value in only one of the compar-isons, this indicates that the error associatedwith one of the data sets is greater than thatwith the other but would not be suf�cientgrounds to reject combination. However, thethird possibility, that the observed tree-to-tree distance is greater than the 95% level inboth comparisons, is explicable only for datasets that estimate signi�cantly different treesand so combination of the data would not beappropriate.

Parametric bootstrap analysis of ME trees.—We used parametric bootstrapping to as-sess the possibility that the presence of agiven grouping in a tree might be the re-sult of cumulative phylogenetic error incomponent branches, rather than a signi�-cant phylogenetic signal. Although this ap-proach has been previously applied in a like-lihood framework (e.g., Huelsenbeck andRannala, 1997), as indicated above, the com-putation times for calculating ML trees wereprohibitive. Instead we implemented the ap-proach under the criterion of ME optimality.To generate parametric bootstrap samples,we initially reconstructed ME trees with spe-ci�c constraints enforced (i.e., monophyly ofa given group) and used HKY85 distancesto estimate sequence divergence. Using theresulting topologies and branch lengths, acomputer program simulated 100 new datasets of the same length as the original dataset. The program included C source codefrom the Siminator program (J. Huelsen-beck, Univ. California-Berkeley) to gener-ate the transition probabilities under theHKY85 model. Empirical estimates of basefrequencies were used to generate ancestralsequences in the simulation, and among-site rate variation was incorporated by usinga gamma shape parameter estimated from


the original data. Phylogenies were recon-structed from the simulated data under theME criterion and the resulting trees sum-marized in a majority rule consensus tree.We then calculated the parametric bootstrapsupport (PBS) for the prespeci�ed node asthe percentage of times it was recovered inthe consensus. We interpreted the PBS valueas indicative of whether a departure fromthe given hypothesis (i.e., nonmonophyly ofa group) might be explicable by phyloge-netic error in the data set. Thus, if we areinterested in the absence of a speci�c group-ing in our reconstruction, a very low PBSvalue would signify high levels of randomerror in the data for that hypothesis, and lit-tle importance should be attached to the ob-servation of polyphyly in the original anal-ysis. Conversely, a high PBS value wouldbe indicative of low levels of phylogeneticnoise associated with an hypothesis, and itsobservation would be therefore signi�cant.

Kishino–Hasegawa Tests.—To comparecompeting phylogenetic hypotheses un-der the ML optimality criterion, we usedthe Kishino–Hasegawa test (KHT; Kishinoand Hasegawa, 1989) as implemented inPAUP * . To test speci�c hypotheses of mono-phyly, we reconstructed phylogenies underthe constraint of monophyly for the givengroup, using the CONSTRAINTS option ofPAUP * . We estimated ML parameters andtrees by using the iterative approach de-scribed above. In this way we were able torecover the best tree compatible with a givenhypothesis of monophyly.

RESULTS AND DISCUSSION

Sequence Data

In addition to data collected in the previ-ous studies, we determined new sequencesin a total of 15 species. Both mitochondrialand nuclear sequences were obtained for the�rst time in seven of those taxa, and in theother eight we sequenced those fragmentsnot included in the earlier work. One impor-tant omission from the data reported hereis the 18S sequences from members of theensiferan Grylloidea (crickets, mole crick-ets, and related insects). We obtained se-quences from several members of this su-

perfamily (including Gryllus and Acheta),but analysis of the 18S sequences indicatedthat they were extremely diverged from theother orthopterans. A preliminary analysisof the pattern of variation in these genesindicated that any ensiferan phylogeny in-cluding the grylloid 18S gene sequences isunlikely to correspond to the species phy-logeny (unpubl. results). However, becausethe Grylloidea represent an important en-siferan group, we have attempted to recon-cile them into this analysis by performing aseparate analysis of the mtDNA.

The main properties of the DNA se-quences are summarized in Table 2. Thealignment lengths of 12S, 16S, and 18S se-quences were 393, 558, and 2,226 bp, respec-tively. After removal of ambiguous align-mentpositions, the lengths of the threegeneswere reduced to 316, 448, and 1,773 bp, re-spectively, giving a total of 2,537 bp. Theproporion of variable sites in the mtDNA(67.1%) was three times greater than that inthe 18S gene (21.6%). The nucleotide com-positions of the sequences also contrast. Ta-ble 2 emphasizes the high A + T content inthe mitochondrial sequences (69.5%) com-pared with the 18S sequences, which haveeffectively no bias (52.7%). Variation in basecomposition, particularly in the mtDNA, isalso present between different taxonomicgroups (see Table 3), although the phy-logenetic component of this variation isminimal. For example, the highest averagemtDNA A + T content (72.4%) is recorded inthe Trigonopterygoidea, whereas the secondlowest mtDNA A + T content is recordedin the Pamphagoidea (67.5%), a relativelyclosely related group. Because such vari-ation can be confounding in phylogenyreconstruction in the mtDNA sequences(Flook and Rowell, 1997b), we also usedthe LogDet transformation (Lockhart et al.,1994), which corrects for distortions in phy-logenetic signal that are due to composi-tional bias. Although trees based on uncor-rected distances (p-distance, Jukes–Cantordistance) differed from LogDet trees, wedetected no such differences between treesthat were based on distances corrected forunequal base composition (e.g., HKY85).Therefore we rejected the condition of non-


TABLE 2. Descriptive statistics for separate and combined data partitions.

Data partition 12S 16S 18S 12S + 16S 12S + 16S + 18S

Initial length (bp) 393 558 2226 951 3177Final length (bp)a 316 448 1773 764 2537No. parsimony-informative sites 175 225 176 400 576No. variable sites (% of �nal) 217 (68.7) 296 (66.1) 383 (21.6) 513 (67.1) 896 (35.3)A 0.31 0.32 0.25 0.32 0.27C 0.11 0.12 0.23 0.11 0.20G 0.17 0.20 0.28 0.19 0.25T 0.41 0.36 0.24 0.38 0.28MP tree length (n)b 1,086 (16 ) 1,378 ( 4 ) 735 (24 ) 2,516 ( 3 ) 3,274 ( 1 )CIc 0.35 0.38 0.67 0.36 0.43RId 0.41 0.38 0.66 0.37 0.43Gammae

MP, YK 0.733 0.676 0.166 0.670 0.204MP, S 0.802 0.799 0.253 0.777 0.246ME, YK 0.714 0.656 0.146 0.669 0.203ME,S 0.784 0.782 0.205 0.771 0.244

a Length of sequence after removal of ambiguous alignment sites.b n = number of trees recovered.c Consistency index for MP tree.d Retention index for MP tree.e Estimate of gamma shape parameter from MP and ME trees by methods of Yang and Kumar (1996) and Sullivan et al. (1995).

stationarity of base composition bias as re-quiring any special attention in this analysis.

Properties of Parsimony Trees

We initially reconstructed phylogenies byequally weighted parsimony. The resultingtrees are not shown here but their lengthsand corresponding statistics are summa-rized in Table 2. Values for consistency

indices (CIs) were relatively low in themtDNA sequences (average = 0.36), re�ect-ing the decreased levels of character congru-ence at the base of the orthopteran molecu-lar phylogeny. The CI for the 18S data washigher (0.67), although the number of infor-mative sites (210) was approximately halfthat in the mtDNA sequences (402). Non-parametric bootstrap support (NBS) for the

TABLE 3. Summary of phylogenetic analyses of orthopteran superfamilies and phasmids.

A + T%

Group 12S + 16S 18S 12S + 16S + 18S Autapomorphies NBSa

Caelifera 0.6978 0.4953 0.5523 0 89Acridoidea 0.6939 0.4947 0.5509 0 0Pamphagoidea 0.6746 0.4963 0.5466 0 0Pneumoroidea 0.6892 0.4940 0.5496 0 0Trigonopterygoidea 0.7243 0.4931 0.5606 9 100Eumastacoidea 0.7099 0.4971 0.5562 0 11Tetrigoidea 0.7127 0.4993 0.5588 7 100Tridactyloidea 0.7133 0.4924 0.5525 7 100

Ensifera 0.6773 0.4960 0.5471 0 97Tettigonioidea 0.6709 0.4954 0.5450 2 100Stenopelmatoidea 0.6827 0.4971 0.5498 1 52

Orthoptera 0.6934 0.4955 0.5512 0 87

Phasmida 0.7148 0.4950 0.5543 19 100aNonparametric bootstrap support. Values correspond to the number of times a particular group was recovered as monophyletic

in 1000 replicates.


TABLE 4. Maximum likelihood analysis of hierarchical substitution models for the combined sequence dataset. The log likelihood of a maximum parsimony phylogeny was estimated under different models of nucleotidesubstitution. Likelihoods were assessed with likelihood ratio tests as described in Materials and Methods. Modelscompared: JC69, Jukes and Cantor (1969); F81, Felsenstein (1981); HKY, Hasegawa, Kishino, and Yang (1985);GTR, general time-reversible model (Lanave et al., 1984); GTR + G , general time-reversible model with gammadistribution correction for among-site variation.

H0 vs. H1 ln L0 ln L1 –2 ln L df P

JC69 vs. F81 –19621.64 –19518.29 206.7 3 < < 0.001F81 vs. HKY85 –19518.29 –18912.96 1210.66 1 < < 0.001HKY85 vs. GTR –18912.96 –18864.14 97.64 4 < < 0.001GTR vs. GTR + G –18864.14 –18323.86 1080.56 1 < < 0.001

unweighted parsimony trees was generallylow, particularly among the higher caelifer-ans. We also estimated character transitionmatrices from these trees and reconstructedphylogenies by using weighted parsimony.For character-state weighting we performedseveral searches, using a range of transi-tion:transversion weightings varying from1:2 to 10:1. As in the equally weighted par-simony, several nodes were poorly sup-ported in nonparametric bootstrap analy-sis. We concluded that the low overall levelof character congruence in the individualgenes at speci�c phylogenetic levels pre-vents the parsimony method from effec-tively recovering phylogeny for the entireorder. Character-state weighting is similarlyineffective because of the contrasting pat-terns of substitution at the different phylo-genetic levels in the data sets, particularlyin the mtDNA. For this reason we did notpursue parsimony methods further.

Analysis of Substitution Methods

A more effective way of analyzing thedata than by parsimony is to use model-based approaches, namely, ME and ML.However, to apply these optimality crite-ria effectively, one must identify an appro-priate substitution model. For this purposewe used a likelihood approach to examinea hierarchy of different models. We �rstexamined the assumption of equal basefrequencies, then the assumption of equaltransition and transversion rates, then thegeneral time-reversible (GTR) model (whereall substitution probabilities are assumed tobe independent), and �nally the assumptionof equal rates among sites. The results, sum-

marized in Table 4, indicate that as moreparameters are added to the substitutionmodel, the �t of the model to the combinedsequences is signi�cantly improved as as-sessed by likelihood ratio tests. Therefore,for subsequent analyses with ML, we usedthe GTR model with a gamma correctionfor among-site rate variation. However, foranalyses under theME criteria wedecided touse the transformation based on the HKY85model. This model is still a good approxi-mation to the underlying pattern of substi-tution but has a smaller variance than theGTR distance because of the lower numberof parameters,—a desirable property for re-construction of phylogenies from distancedata (Kumar et al., 1993).

Reconstruction and Consensus Analysis of METrees

Phylogenies were reconstructed underthe ME optimality criteria from HKY85 dis-tance matrices as described above (Fig. 1a).The shapes of the mtDNA trees are very sim-ilar, with short branches between the basalgroups, whereas the 18S phylogeny showsthe reverse pattern. The symmetric differ-ence, a measureof the amountof congruencebetween two trees (Penny and Hendy, 1985),between the 12S and 16S trees was 44, andthe average distance between the mtDNAtrees and the 18S tree was 38. The three treeswere compared in a strict consensus, andthe results (Fig. 1b) con�rm that the levelof congruence between data sets is low, theonly signi�cant feature of orthopteran phy-logeny recovered in this analysis being themonophyly of the Caelifera.


FIGURE 1. Consensus analyses of minimum evolution (ME) trees. Combined data partitions are indicated by“+” symbol (e.g., 12S + 16S); different data partitions included in consensus tree are indicated by “&” symbol(e.g., 12S & 16S). (a) HKY85 ME phylograms for 12S, 16S, combined (12S + 16S), and 18S data partitions. (b) Strictconsensus for 12 S & 16S & 18S trees. (c) Strict consensus for (12S + 16S) & 18S trees. (d) Adams consensus for (12S +16S) & 18S trees. (e) Agreement subtree for (12S + 16S) & 18S trees. Abbreviations for taxa: Gom = Gomphocerippus;Acr = Acrida; Oed = Oedipoda ; Rha = Rhainopomma; Gla = Glauia; Bat = Batrachotetrix; Prosph = Prosphena; Pyr= Pyrogomorpha; Bul = Bullacris; Pne = Pneumora; Phys = Physemacris; Tan = Tanaocerus; Xyr = Xyronotus; SysR =Systella raf�esi; SysB = Systella sp.; Pros = Prosarthria; Sti =Stiphra; Eus = Euschmidtia; Hom = Homeomastax; Tetr =Batrachideidae sp.; Par =Paratettix; Neo = Neotridactylus; Cyl = Cylindraustralia; Tetti = Tettigonia; Rus = Ruspolia;Com = Comicus; Hem = Hemideina; Ceu = Ceuthophilus; Cyp = Cyphoderris; Phyl = Phyllium; Aga = Agathemera; Gry= Grylloblatta; Gro = Gromphadorhina.

Previous analysis suggests that the in-dividual mtDNA sequences are too shortto provide resolution at deep levels, andthat combining sequences greatly improvesthe effectiveness of reconstructions (Flookand Rowell, 1997b). We therefore performeda ME search for the combined 12S + 16Sdata (Fig. 1a) and used the resulting treein a second strict consensus analysis withthe 18S data. The consensus tree obtained(Fig. 1c) contains several more bifurca-tions than does the tree recovered from theconsensus of the 12S, 16S, and 18S datapartitions, although the overall congruencebetween the trees is still low (symmetric dif-

ference = 30). The source of disagreementbetween the different trees is clari�ed byadopting a different comparison method. Inan Adams (1986) consensus, where nestingswithin the individual trees are preserved,many of the more recently evolved caelif-eran taxa are represented in the �nal tree(Fig. 1d); however, several unlikely group-ings are also contained in this tree (e.g.Eumastacidae + Tetrigoidea). A third com-parison method, agreement subtree anal-ysis (Finden and Gordon, 1985), excludesthe con�icting clusters from the �nal phy-logeny (Fig. 1e). This reveals that the incon-sistencies are precisely those relationships


previously demonstrated to be the resultof reconstruction failure attributable tosystematic error (Flook and Rowell, 1997a,1998): long branch attraction between eu-mastacoid and tridactyloid mtDNA se-quences, very short internodes between thepneumorid and pyrgomorph lineages inthe mtDNA phylogeny, and poor resolutionbetween tetrigoid and eumastacid 18S se-quences. Thus, the low level of congruencebetween the individual phylogenies re�ectsa lack of resolution of the individual datasets rather than serious con�ict in phyloge-netic signal.

Effect of Sequence Length on PhylogenyReconstruction

To investigate our concerns about the con-tribution of the length of sequences tophylo-genetic recovery of these data, we examinedtheeffect of increasing sequence length of se-quences from the different data partitions.Having found no “correct” phylogeny forthe whole 33-taxon data set, we used agree-ment among the recovered phylogenies asthe criterion by which to judge the success ofthe searches on different sequence lengths.Generating random samples from 12S, 16S,and 18S sequences, we reconstructed MEphylogenies for different sequence lengthsand measured the degree of congruence bycalculating the average symmetric differ-ence between trees. The results (Fig. 2) showthat curves obtained for the different se-quences are very similar. This indicates thatafter compensating for sequence length, theoverall ability of the sequences to resolve thephylogeny is approximately equal. None ofthe lines converge on an average symmetricdifference of 0, but this is explained by thevarious de�ciencies of the individual datasets discussed above. In addition, when wegenerated samples from the combined 12S+ 16S + 18S data (Fig. 2), the curve �tted tothese data was very similar to that obtainedfor the individual data partitions.

We also used this simulation strat-egy to examine our reservations aboutthe use of parsimony for reconstructingphylogenies from the data. We gener-ated random samples for a subsample ofthe taxa (Fig. 3) for which there was a

FIGURE 2. Analysis of the effect of sequence lengthon the effectiveness of phylogenetic reconstruction fordifferent sequences. Each data point represents the av-erage symmetric difference between 50 phylogenies re-constructed from random data samples of each indi-cated length. Random samples were obtained by re-sampling with replacement from the alignment of theindividual genes. u= 12S; n = 16S;s= 18S. Thick line(comb.) represents a power function �tted for the datapoints estimated for the combined data set.

strong phylogenetic expectation and forwhich the individual data provided veryclear resolution. We then reconstructed treesby using equally weighted parsimony andME. For the latter method we used equal-rate HKY85 distances. The results (Fig. 3)show that the ME analysis converged ona relatively small number of closely re-lated trees (i.e., average symmetric differ-ence close to 0). These trees matched theexpectation for the 12 taxa (see legend forFig. 3). In contrast, the parsimony analysesclearly failed to converge on a single tree.

Combination of Data Sets

The results of the above analyses indicatethat the limited range over which the dif-ferent sequences resolve orthopteran phy-logeny is due more to the short length of the


FIGURE 3. Analysis of the effect of sequence lengthon the effectiveness of phylogenetic reconstructionfor different methods and optimality criteria. Com-bined data sets from sequences from the follow-ing taxa: two acridoids (Acrida, Gomphocerippus), thepamphagoids, the tetrigoids, the tridactyloids, andtwo outgroups (the phasmids). The expected phyloge-netic relationships between these taxas are (Phasmida(Tridactyloid (Tetrigoidea (Pyrgomorphidae (Pam-phagidae, Acridoidea))))). Each data point representsthe average symmetric difference between 50 phyloge-nies reconstructed from random data samples of eachindicated length. Random samples were obtained byresampling with replacement from the alignment of thecombined sequences. Results are shown for searchesperformed with MP (n) and ME (u).

individual sequences than to con�ict or ab-sence of phylogenetic signal. We have al-ready examined the issue of data combi-

nation for the mitochondrial data, usingincongruency indices, and demonstratedthat combining the 12S and 16S sequences isjusti�ed (Flook and Rowell, 1997b). Further-more, the convergence in the phylogeneticreconstructions observed in the analysis ofrandomly sampled sequences provides oneline of support that legitimizes the com-bination of nuclear and mitochondrial se-quences. A �nal source of support comesfrom applying the test of Rodrigo et al.(1993) for heterogeneity in phylogenetic sig-nal. Comparison of the observed symmetricdifferences (Table 5) with the estimated nulldistributions (Fig. 4) shows that in none ofthe comparisons do the observed tree-to-tree distances exceed the 95% level. In otherwords, we do not reject the null hypothesisthat the observed differences between phy-logenies reconstructed from different datasets are due to sampling error. However, itis notable that the outcome of the test forthe 12S versus 18S comparison is very closeto the 95% level. We attribute this to thelimitations of the symmetric difference in-dex, in which rearrangements of only one ortwo taxa can result in maximal differences(Penny and Hendy, 1985). We also note thatmany suboptimal trees were very close tothe ME tree in reconstructions from themito-chondrial data. Comparing the �rst 10 sub-optimal trees for the two data sets, we foundthat 20% of the comparisons produced crit-ical values that would not be rejected bythe estimated null distributions (i.e., < 46).This suggests that the symmetric differencemight be an inadequate metric; therefore,we recalculated the null distributions, us-ing an alternative and potentially morediscriminating comparison measure: the

TABLE 5. Results of randomization tests. Numbers correspond to the percentage of the estimated null distri-bution (see Fig. 4) lower than the calculated symmetric differences between the ME trees for the compared datapartitions.

Sym.diff.(AS)a 12S 16S 12S + 16S 18S

12S vs. 16S 44 (16) 88.3 (82.2) 89.26 ( 0.8) — —12S vs. 18S 40 (15) 47.0 (70.2) — — 93.8 (86.4)16S vs. 18S 32 (13 — 10.0 (49.9) — 49.9 (62.6)12S + 16S vs. 18S 26 (12) — — 16.4 (50.8) 43.6 (40.8)

aFigures in parentheses correspond to results when the tree comparisons are made using the agreement subtree (AS) distanceprovided in Component (Page, 1993).


FIGURE 4. Frequency distributions for symmetricdifferences between trees calculated from pairwisecomparisons of bootstrapped data samples. Solid barscorrespond to thedistribution of treedistances based onagreement subtree differences; white bars correspondto tree distances calculated by using symmetric differ-ences.

agreement subtree distance (Finden andGordon, 1985). The results (Table 5) demon-strate that in all combinations the null hy-pothesis of the differences being due to sam-pling error is clearly not rejected.

ME and ML Analysis of the Combined Data Set

Taken together, we interpret the resultsof the randomization tests and of the sim-ulation analyses as a suf�cient basis forcombining the data in the subsequentexamination. We therefore proceeded tothe phylogenetic reconstruction of the com-bined 12S + 16S + 18S sequences. Initially,we performed heuristic searches under the

ME criterion and used nonparametric boot-strapping toassess the con�dence that couldbe associated with individual nodes. Recon-structions based on equal-rate and gamma-corrected HKY85 distances recovered verysimilar trees and levels of bootstrap sup-port. The equal-rate tree (Fig. 5) is equidis-tant from the 12S + 16S and the 18S ME trees(symmetric differences of 20 and 16, respec-tively). We also reconstructed trees underthe ML optimality criteria, using the GTRmodel with a gamma distribution correc-tion for among-site rate variation. The MLtree (Fig. 6) is very similar to the ME tree(symmetric difference = 16). The most im-portant difference is that the Tetrigoidea arerecovered in a clade with the Proscopiidae.Another interesting difference is that the Ha-gloidea are recovered as the primitive en-siferan lineage.

Assessment of Individual Nodes

Because the different approaches used inour analysis recovered similar trees, a re-maining task was to assess the con�dencethat could be attached to those relationships.We chose three strategies for this. First,we use nonparametric bootstrapping (Efron,1982; Felsenstein, 1985) to examine the con-�dence that could be attached to speci�cnodes in ME trees. Second, we performedparametric bootstrapping to examine thesigni�cance of prespeci�ed groups in theME trees. This latter approach is particularlyattractive by offering �exibility in examina-tion of phylogenetic hypotheses (Huelsen-beck et al., 1996; Huelsenbeck and Rannala,1997). Third, we used the KHT to compareML trees. We applied the �rst method, non-parametric bootstrapping, to examine thedifferent nodes suggested in the full or-thopteran phylogeny (Fig. 5) and the othertwo methods to investigate the followingthree controversial relationships suggestedby the analysis.

First, we examined the apparent poly-phyly of the Pamphagoidea. In the nonpara-metric bootstrap analysis, a monophyleticPamphagoidea was not recovered from anyof the replicates (Table 6). For the paramet-ric bootstrap analysis we simulated the dataas described, enforcing the single constraint


FIGURE 5. Phylogram of ME tree reconstructed from HKY85 distance matrix for combined 12S + 16S + 18Sdata set. Numbers above nodes indicate NBS (based on 1,000 replicates) for equal-rate/gamma-corrected HKY85distances. Values are indicated only for nodes that received > 50% in at least one of the analyses. The phylogenyshown was reconstructed from equal-rate distances. The gamma-corrected tree differs only in its placing of theTetrigoidea, which are recovered inside the eumastacids.

of a monophyletic Pamphagoidea. The pam-phagoid clade was recovered in 31% of thereplicates, suggesting that the data were not

informative enough topositively rejectpam-phagoid monophyly under the ME optimal-ity criteria. This inference was supported

TABLE 6. Summary of results of bootstrap analyses and Kishino–Hasegawa tests (KHTs) of speci�c hypothesesof monophyly.

Hypothesis NBS PBS ML(ln L)a KHTb

(Xyronotidae, Trigonopterygoidea) 60 81 –18387.93 NDc

(Xyronotidae, Tanaoceridae) 12 30 –18405.10 1.461 (0.144)(Acridoidea, Pamphagidae, Xyronotidae, 34 89 –18400.04 1.976 (0.048)

Trigonopterygoidea, Pneumoridae)(Pamphagidae, Acridoidea) 74 78 –18387.93 NDc

(Trigonopterygoidea, Pyrgomorphidae) 1 17 –18400.14 1.337 (0.182)(Pamphagoidea) 0 31 –18401.38 0.855 (0.393)(Pneumoroidea) 0 10 –18419.08 1.928 (0.054)(Eumastacoidea) 11 32 –18392.34 0.817 (0.414)

a Log likelihood score for optimal tree under given constraints. Phylogenies were compared with the ML phylogeny recoveredfrom an unconstrained search (ln L = –18387.93) .

b Number indicates T value for different comparisons. Number in parentheses is the probability of getting a more extreme valueof T under the null hypothesis (of no difference between trees).

c ND = not done; KHT not performed because constrained tree identical to unconstrained tree.


FIGURE 6. Phylogram of ML tree reconstructed for combined 12S + 16S + 18S data set. The phylogeny wasreconstructed under the GTR model of substitution. The log likelihood of the phylogeny = –18387.93.

by the KHT, which failed to detect signif-icant differences between the best uncon-strained and pamphagoid-constrained MLtrees. A slightly different approach to exam-ining this problem was to examine the puta-tive grouping of the Pamphagidae with theAcridoidea. Here, the relatively high NBSfor this clade (73%) also corresponded toa relatively high PBS value (78%), indicat-ing that this observation was unlikely to bedue to cumulative random error. However,the KHT was not particularly informativebecause the acridoid–pamphagid groupingwas recovered in all of the unconstrainedML searches. Thus, although we are unableto reject monophyly of the Pamphagoideaoutright, it obtains no support from ouranalyses.

The second relationship investigated wasamong the four eumastacoid taxa. Previ-ous analyses with the mitochondrial datahad suggested a sister group relationship ofthe Eumastacidae and Proscopiidae (Flook

and Rowell, 1997a), but the results of thepresent analyses and those of the 18S analy-sis suggested that the Eumastacidae mightbe basal. This initially seemed to be sup-ported by the nonparametric bootstrap anal-ysis, where a monophyletic Eumastacoideawas recovered from only 11% of replicates.However, the results are complicated by thefact that the proscopiids were grouped withthe tetrigoids in some of the analyses (e.g.,in the ML tree in Fig. 6). No such groupinghas previously been suggested in the liter-ature, and the two groups are morphologi-cally very distinct. As a result of this discrep-ancy, no clear alternative hypothesis is sug-gested by our analyses of the combined data,and we can only examine to what extentthe data reject monophyly. Using the samestrategy as outlined for the Pamphagoidea,we performed a search enforcing the con-straint of a monophyletic Eumastacoidea,but the PBS value of only 32% indicatedthere was relatively little informative phy-


logenetic signal relating to the monophylyof the Eumastacoidea in the ME analysis.Furthermore, in the KHT, the difference be-tween the unconstrained ML tree and theeumastacoid tree was not signi�cant. There-fore, we conclude that the sequences do notcontain enough phylogenetic signal to settlethe issue of eumastacoid monophyly.

Finally, we tested the monophyly of thePneumoroidea. The interrelationships ofthe three pneumoroid families (Table 6)have been the subject of much specula-tion (e.g., Kevan, 1982) and we initiallytested the hypothesis of the monophylyof this superfamily. No evidence for amonophyletic Pneumoroidea was obtainedfrom the nonparametric bootstrap analy-sis, but a low PBS value (10%) indicatedthat the observation of pneumoroid poly-phyly might be explained by random er-ror in the data. However, large differenceswere detected between the unconstrainedtree and the Pneumoroidea-constrained MLtree. We tested several other hypotheses in-volving members of the Pneumoroidea (Ta-ble 6) and obtained high PBS values for aXyronotidae + Trigonopterygidae grouping(81%) and for placing the Tanaoceridae andPyrgomorphidae outside of the (Acridoidea,Pamphagidae, Pneumoridae, Xyronotidae,Trigonopterygidae)(89%); these results sug-gest that the occurrence of these groupingsin the ME tree may be well founded. Wetherefore reject the monophyly of the Pneu-moroidea on the basis for the analysis of thecombined data.

Phylogenetic Position of the Grylloidea

In terms of taxonomic sampling, the mainde�ciency of this analysis was the absenceof any grylloid taxa. To rectify this, we at-tempted a further combined mtDNA anal-ysis, this time including sequences for twogryllids, Gryllus campestris and Acheta do-mesticus. The aims here were to establishthe monophyly of the Ensifera and to de-termine the position of the Grylloidea. Us-ing the mtDNA sequences, wereconstructeda phylogeny from several subsamples oforthopteran taxa and their outgroups. Weomitted most of the caeliferan taxa becausetheir position as a sister group of the En-

sifera is unquestioned and because exclu-sion of these sequences greatly reduced theamount of computation. The ME tree forone of the samples (Fig. 7) illustrates thatthe mtDNA sequences are unable to resolvethe ensiferan phylogeny when the grylloidtaxa are included. A low NBS value is ob-tained for a grylloid/tettigonioid grouping(54%), but the signi�cance of this value ismade unclear by the fact that when dif-ferent combinations of outgroups are used(e.g., no phasmids; addition of dipterans),the grylloid taxa are placed outside of theother ensiferans. However, in all the recon-structions, the branch connecting the gryl-loid sequences to the other ensiferans (Fig. 7)is the longest internal branch, exceedingthe lengths of branches connecting the out-groups. Combined with their highly un-usual 18S sequences (which will be dis-cussed elsewhere), this suggests to us thatthe Grylloidea probably represent a sistergroup to the rest of the extant Ensifera.

Discussion of Phylogenetic Findings

The main phylogenetic results are sum-marized in Figure 8. This phylogeny in-cludes only those internal branches that re-ceived high support from the nonparametricbootstrap analyses or for which we believephylogenetic support obtained in this anal-ysis is compelling. This results in a loss ofresolution compared with the tree shown inFigure 5, for example, but allows us to ad-dress con�dently several speci�c points.

Caelifera.—The consensus tree shows theCaelifera as the sister group of the En-sifera (node O), but with the tridactyloidsas the sister group of all other caelifer-ans (node C1) and widely separated fromthem. Our previous analyses had showna con�ict with respect to the relationshipof the Tridactyloidea to the rest of theCaelifera. The mtDNA analysis recovereda tetrigoid/tridactyloid grouping, but thismay have been due to inadequate taxonomicsampling with respect to outgroups for theorthopteran taxa. In the present analysis ourresults �rmly reinforce our previous inter-pretation of the 18S tree: The tridactyloidsare a single clade; there therefore appears tobe no justi�cation for splitting off the cylin-


FIGURE 7. Unrooted phylogram of ME tree reconstructed from HKY85 distance matrix for combined (12S + 16S)data set, including sequences from two grylloid species. Thick lines indicate branches for which > 89% bootstrapsupport was obtained.

drachetids as a separate superfamily, as pro-posed by Kevan (1982).

Of the remaining caeliferans, we placethe Tetrigoidea as the most basal (nodeC2), closely followed (node C3) by the Eu-mastacidae sensu lato and the Proscopiidae.All these groups are monophyletic (and re-main so in other analyses with much largertaxon samples; data not shown). The sepa-ration of the Tetrigoidea and Tridactyloideais very good and clari�es earlier resultsbased on mitochondrial sequences (Flookand Rowell, 1997a). The separation of theTetrigoidea from the eumastacoids is notentirely clear in the present analysis be-cause of their occasional grouping with theproscopiids. However, the Tetrigoidea werevery clearly identi�ed as basal to the eumas-tacoid taxa in the mtDNA analyses (Flookand Rowell, 1997a). At node C3, our datado not allow us to decide whether the Eu-mastacidae s.l. and Proscopiidae are sistergroups within a monophyletic clade (the Eu-mastacoidea sensu Dirsh, 1996) or whetherthe latter indeed branch separately fromthe main trunk of the tree. The mitochon-drial and nuclear data con�ict on this point,the former rejecting a sister group rela-tionship. However, because mitochondrial

genes are more likely to resolve short intern-odes than are nuclear genes (Moore, 1995),we can speculate that the results donot favora monophyletic Eumastacoidea. This inter-pretation favors recognizing the proscopi-ids as a superfamily, as proposed by (e.g.)Descamps (1973). The remaining taxa (i.e.,the Acridomorphoidea sensu Dirsh [1975]plus the Pneumoroidea and Trigonoptery-goidea) constitute a fourth clade (node C4),within which the branching order is not cer-tainly resolved in our analyses.

The various resolved lineages within thislast clade, however, are of considerable sys-tematic interest.

1. The Sonoran Tanaoceridae (though un-fortunately represented in our analysisby only a single genus) form a solitarybranch of their own (L1), and contraryto Kevan (1982) do not cluster with theSoutheast Mexican Xyronotidae. We cancon�dently exclude the possibility thatthe Tanaoceridae are related to the eu-mastacids, as proposed by Rehn (1948).These results seem to con�rm Dirsh’s(1955) decision to remove them fromthe latter group and give them indepen-dent family status. As noted above, the


FIGURE 8. Summary of the phylogenetic conclusions of this study, indicating proposed revisions of orthopteranclassi�cation. O corresponds to the ancestral orthopteran node. Nodes E1 and E2 correspond to the two resolvedensiferan nodes. Nodes C1–C4 indicate the nodes clearly resolved within the Caelifera on the basis of our molec-ular analyses. Although the remaining caeliferan lineages are separated in individual analyses (e.g., Fig. 5), wedo not consider their branching order to have received suf�cient support from the bootstrapping and likelihoodanalyses.

parametric bootstrap analysis (Table 5)gives support for the proposition that theTanaoceridae, followed by the Pyrgomor-phidae, are the most primitive branchesof the �fth clade.

2. The worldwide Pyrgomorphidae (L2)form a monophyletic grouping clearlydistinct from all the others. There is nosupport for the occasional suggestion of aclose relationship of the Pyrgomorphidae

with the Lentulidae (e.g., Amedegnato,1993; Dirsh, 1956).

3. The African Pneumoridae are resolvedas a monophyletic group (L3) with noclose relationship to any other taxa.The Tanaoceridae and the Xyronotidae,grouped by Dirsh (1966) and Kevan(1982) with the Pneumoridae in a su-perfamily Pneumoroidea on the basisof similar stridulatory mechanisms, are


excluded from this clade. We also per-formed a KHT to examine the hypoth-esis that the Pneumoridae were primi-tive to the other higher caeliferans exceptfor the Tanaoceridae (i.e., Trigonoptery-goidea, Pamphagoidea, Acridoidea, Xy-ronotidae). The difference between thebest tree consistent with this hypothesisand the unconstrained ML tree was sig-ni�cant (P < 0.05), and we thus reject thehypothesis.

4. The Xyronotidae, also represented in ouranalysis by only a single genus, do notcluster with any Pneumoroidea sensuDirsh, but instead are grouped withthe Southeast Asian Trigonopterygidaein a further distinct clade (lineage L4).These latter are de�nitely not associ-ated with the eumastacids or proscopi-ids, contrary to the suggestion of Dirsh(1975). The hypothesis of Bolõ var (1909)and Kirby (1910) , according to which theTrigonopterygidae are allied to the Pyr-gomorphidae, also received no supportin our analysis.

5. The last lineage (lineage L5) includesthe Old World Pamphagidae, the Cen-tral African Lentulidae, and the world-wide Acrididae (for details, see Fig. 5).The unity of this assemblage is sup-ported with high NBS (73%) and PBSvalues (78%), indicating that this ob-servation is unlikely to be accountedfor by random error alone. In otheranalyses (data not shown) this cladealso includes the New World familiesPauliniidae, Tristiridae, Ommexechidae,and Romaleidae. On the basis of theirmorphology we would expect the OldWorld Lathiceridae and Charilaidae tobe included as well, but we have nosequence data for these two families.

The identi�cation of these lineages, andmore signi�cantly the rejections of severalpreviously proposed groupings (Tables 3, 5),have several important consequences forcurrent interpretations of their relation-ships. We will discuss further the interrela-tionships of the Pneumoroidea sensu Dirshin another paper, but our data clearly donot support the retention of this superfamily.

The failure of the Pamphagidae and the Pyr-gomorphidae to cluster together in any anal-ysis using any of the three sequenced partsof the genome has been discussed in ourprevious papers. The combined sequencesclearly link the pamphagids with the acri-dids rather than with the pyrgomorphidsand indicate that the Pamphagoidea sensuDirsh (1975) or Kevan (1982) are a poly-phyletic grouping. As the Pamphagidae reg-ularly emerge from the analyses embeddedwithin an assemblage of acridid subfami-lies and acridoid families, the Acridoidea inthe usual modern sense (i.e., excluding thePamphagidae) form a paraphyletic group-ing. Also, the Oedipodinae are invariablyseparated from the clade containing bothAcrida and Gomphocerippus, in accordancewith their rather basal and isolated posi-tion within the Acrididae, already noted inour analyses of the mitochondrial sequences(Flook and Rowell, 1997a).

Ensifera.—Our analyses split the Ensiferainto three groups. The grylloids may bethe most basal (node E1) and are also veryhighly diverged, rather similar to but farexceeding the situation of the tridactyloidswithin the Caelifera. In the ME tree (Fig. 5),the remaining ensiferans fall into two sis-ter groups: a tettigonioid clade (representedin our sample by Haglidae + Tettigoniidae)and a stenopelmatoid clade (in our sam-ple, Mimnermidae + (Stenopelmatidae +Rhaphidophoridae)). This agrees with theconclusions of Gorochov (1995a) , based onmorphology, but disagrees markedly withthe hypothesis of Gwynne (1995), which de-rives from a cladistic analysis of the mor-phological data of Ander (1939) and otherauthors. However, in the ML analysis (Fig. 6)the hagloid is recovered as the most primi-tivegroup. Therefore, because of this con�ictand the low NBS and PBS values obtainedfor the ensiferan taxa (Fig. 5) our interpre-tation in Figure 8 (node E2) is conservativeand tentative.

Implications for the Systematic Nomenclatureof the Orthoptera

Many modern writers have recom-mended a node-based or a lineage-basedclassi�cation (see de Querioz and Gauthier,


1994), and if the necessary phylogeneticknowledge is available, the advantages in-deed seem compelling. As our analysis hasnot completely resolved branching orders, anode-based system is still impracticable, butthe less revolutionary lineage-based systemmay well be explored. What are the implica-tions of the above �ndings for orthopteranclassi�cation?

We have not investigated the Ensiferawith the same depth of taxon samplingas the Caelifera. However, reconstructionsbased on different taxonomic samples ofmtDNA sequences consistently support amonophyletic Ensifera. Furthermore, al-though direct comparison of all the ensiferansuperfamilies is dif�cult for the combineddata, the long branch length connecting thegrylloid taxa to the other ensiferans, com-bined with their highly diverged 18S se-quences (unpubl. data), suggests to us thatthe Grylloidea represent the basal ensiferanlineage. Thus in the Ensifera we identifyonly two nodes in our summary: E1 and E2.

Within the Caelifera, many of the resolvedclades already bear classical names. Basallineages at nodes C1 to C3 in Figure 8 cor-respond to the Tridactyloidea, Tetrigoidea,and Eumastacoidea, respectively. The �-nal clade (C4) is practically identical withthe Acridomorphoidea sensu Dirsh (1975),except that it additionally encompassesthe Trigonopterygoidea, which Dirsh erro-neously placed in the Eumastacoidea. Thelineages we distinguish within this cladeare mostly currently recognized as fami-lies (Tanaoceridae, Pyrgomorphidae, Pneu-moridae, Lentulidae, Pamphagidae, andAcrididae). The proposed grouping of theTrigonopterygidae and the Xyronotidae cor-responds to Kevan’s original (1952) placingof Xyronotus, though he abandoned this inlater publications.

How can the existing names be recon-ciled with the phylogenetic relations heredescribed? We propose that the lineages de-�ned by well-resolved nodes on the back-bone of the tree be allotted superfamily rank.Strict application of this strategy to our cur-rent data would divide the Caelifera intofour superfamilies, rather than the existingseven: (1) Tridactyloidea (2) Tetrigoidea, (3)

Eumasticoidea, and (4) an unnamed cladeof higher Caelifera, containing �ve resolvedlineages of uncertain branching order. How-ever, we hope that further data or improvedanalysis (see below) will eventually succeedin resolving the branching order of theselineages, and we would then have max-imally (4) Tanaoceroidea new superfamily,(5) Pyrgomorphoidea new superfamily (seealso Vickery [1997], who, however, givesno reasons for this grouping), (6) Pneumor-oidea, (7) Trigonopterygoidea sensu novo,and (8) Acridoidea sensu novo. Our Pneu-moroidea would, however, include only onefamily, not the three conceived by Dirsh(1975); the Trigonopterygoidea would be en-larged from two to three families; and ourAcridoidea would include all the currentPamphagoidea except the Pyrgomorphidae.

It seems logical to treat resolved lineageswithin these superfamilies as families. Weare preparing more-detailed analyses of re-lationships within both the Eumastacoideaand the Acridoidea (both used in the senseproposed above), using much larger num-bers of taxa. When these are completed, wehope to be able to suggest further re�ne-ments to our classi�cation. However, thepresent restricted data set supports the re-tention of at least the Lentulidae, Pamphagi-dae, and Acrididae as separate families withthe Acridoidea.

Prospects for a Fully Resolved MolecularPhylogeny of the Orthoptera

Although the resolution afforded by ourcombined analysis compares favorably withother higher-level studies of insect molecu-lar phylogeny (e.g., Campbell et al., 1995;Dowton and Austin, 1994; Kambhampati,1995), outstanding dif�culties remain. Inparticular, the basal relationships of theAcridomorphoidea are still unclear, and thelow PBS values associated with several ofthe putative groupings suggest that we areunable to distinguish suf�cient phyloge-netic signal to recover the precise branchingpatterns. Furthermore, simulations (Fig. 2)indicate that the accumulation of longersequences cannot be guaranteed to im-prove our understanding of this part ofthe phylogeny. Comparing the paleontolog-


ical data (for a summary, see Flook andRowell, 1997a) with the evolutionary dis-tances estimated here (Fig. 5), we estimatethat the lineages corresponding to nodes L1to L5 probably appeared in the late Juras-sic/early Cretaceous. This is signi�cant be-cause it parallels the situation in amniotephylogeny, where resolution of the basal lin-eages has not yet been reached, althoughfar more molecular data are currently avail-able (e.g., Huelsenbeck and Bull, 1996). Be-cause the accumulation of data from alter-native genes carries additional dif�cultiesof interpretation (e.g., identi�cation of ho-mology), we suggest that the solution toour problem does not necessarily lie in thesequencing of more genes. One alternativewe have explored is the use of other formsof molecular data such as gene rearrange-ments (e.g., Boore, et al., 1995; Flook et al.,1995), but in our surveys of orthopteran mi-tochondrial genomes we so far have iden-ti�ed only one informative rearrangement(unpublished data). Another possibility ismore-intensive taxonomic sampling. How-ever, in several of the lineages treated above(e.g., Xyronotidae and Tanaoceridae), it isdif�cult to suggest unsampled taxa thatare likely to be phylogenetic intermediates.Thus we conclude that an increased under-standing of orthopteran evolution throughusing the methods of molecular phylogenet-ics depends as much on improved methodsof hypothesis testing as on additional datacollection.

ACKNOWLEDGMENTS

Weacknowledge thehelpof the following colleaguesin obtaining material used in this study: B. Aerne, G.Bruyea, C. S. Carbonell, J. S. Edwards, A. Floren, M.Gentili, G. Halffter, B. Helfert, O. von Helversen, M.Impekhoven, J. Irish, A. Mason, N. M. Mungai, D. E.Perez, M. Rahier, D. R. Rentz, K. Richards, M. Richards,K. Riede, T. Romero, and M. Van Staaden. Several ofthese persons also provided taxonomic determinationsof their material. We also thank the authorities of theCarlsbad Caverns National Park for supplying us witha specimen of Ceuthophilus carlsbadensis and Mr. U.Stiefel for technical assistance. Finally, we are gratefulto Jack Sullivan, Chris Simon, Dick Olmstead, and ananonymous reviewer for suggesting valuable improve-ments to an earlier draft of the paper. This study was�nanced in part by a grant to P. K. F and C. H. F. R. fromthe Swiss National Science Foundation.

REFERENCES

ADAMS, E. N. 1986. N-trees as nestings: Complexity,similarity, and consensus. J. Classif. 3:299–317.

AMEDEGNATO, C. 1993. African-American relation-ships in the acridians (Insecta, Orthoptera). Pages59–75 in The Africa-South America connexion (W.George and R. Lavocat, eds.). Oxford Univ. Press,Oxford, England.

ANDER, K. J. 1939. Vergleichend anatomische undphylogenetische Studien uber die Ensifera (Saltato-ria). Opuscula Entomologica, Lund, Sweden.

BOLIVAR, I. 1909. Family Acridiae, subfamily Pyrgo-morphinae. Genera Insectorum, Bruxelles.

BOORE, J. L., T. M. COLLINS, D. STANTON, L. L. DAEHLER,AND W. M. BROWN. 1995. Deducing the pattern ofarthropod phylogeny from mitochondrial DNA re-arrangements. Nature 376:163–165.

BULL, J. J., J. P. HUELSENBECK , C. W. CUNNINGHAM, D.L. SWOFFORD, AND P. J. WADDELL. 1993. Partitioningand combining data in phylogenetic analysis. Syst.Biol. 42:384–397.

CAMPBELL, B. C., J. D. STEFFEN-CAMPBELL, J. T. SORENSEN,AND R. J. GILL. 1995. Paraphyly of Homopteraand Auchenorrhyncha inferred from 18S rDNA nu-cleotide sequences. Syst. Entomol. 20:175–194.

CARPENTER, F. M., AND L. BURNHAM. 1985. The geolog-ical record of insects. Annu. Rev. Earth Planet. Sci.13:297–314.

DE QUEIROZ, A. 1993. For consensus (sometimes).Syst. Biol. 42:368–372.

DE QUERIOZ, K., AND J. GAUTHIER. 1994. Toward a phy-logenetic system of biological nomenclature. TrendsEcol. Evol. 9:27–30.

DESCAMPS, M. 1973. Notes preliminaires sur les geni-talia de Proscopoidea (Orthoptera, Acridomorpha).Acrida 2:77–95.

DIRSH, V. M. 1955. Tanaoceridae and Xyronotidae:Two new families of Acridoidea (Orthoptera). Ann.Mag. Nat. Hist. 8:255– 288.

DIRSH, V. M. 1956. The phallic complex in Acridoidea(Orthoptera) in relation to taxonomy. Trans. R. Ento-mol. Soc. Lond. 108:223–356.

DIRSH, V. M. 1966. Acridoidea of Angola I, II. Publ.Cult. Cie. Diamantes Angola 74:11–305; 307–527.

DIRSH, V. M. 1975. Classi�cation of the acridomor-phoid insects. Classey, Farringdon, Oxford, England.

DOWTON, M., AND A. D. AUSTIN. 1994. Molecular phy-logeny of the insect order Hymenoptera: Apocritanrelationships. Proc. Natl. Acad. Sci. USA 91:9911–9915.

EFRON, B. 1982. The jackknife, thebootstrap, and otherresampling plans. SIAM, Philadelphia.

FELSENSTEIN , J. 1981. Evolutionary trees from DNA se-quences: A maximum likelihood approach. J. Mol.Evol. 17:368–376.

FELSENSTEIN , J. 1985. Con�dence limits on phyloge-nies: An approach using the bootstrap. Evolution39:783–791.

FINDEN, C. R., AND A. D. GORDON. 1985. Obtainingcommon pruned trees. J. Classif. 2:255–276.

FLOOK, P. K., AND C. H. F. ROWELL. 1997a. The phy-logeny of the Caelifera (Insecta, Orthoptera) as de-


duced from mt rRNA gene sequences. Mol. Phylo-genet. Evol. 8:89–103.

FLOOK, P. K., AND C. H. F. ROWELL. 1997b. The effec-tiveness of mitochondrial rRNA gene sequences forthe reconstruction of the phylogeny of an insect or-der, Orthoptera. Mol. Phylogenet. Evol. 8:177–192.

FLOOK, P. K., AND C. H. F. ROWELL. 1998. Inferencesabout orthopteroid phylogeny and molecular evo-lution from small subunit nuclear ribosomal RNAsequences. Insect Mol. Biol. 7:163–178.

FLOOK, P. K., C. H. F. ROWELL, AND G. GELLISSEN .1995. The sequence, organisation and evolution ofthe Locusta migratoria mitochondrial genome. J. Mol.Evol. 41:928–941.

GOROCHOV, A. V. 1995a. Contribution to the systemand evolution of the order Orthoptera. Zool. Zh.74:39–45.

GOROCHOV, A. V. 1995b. [Systematics and evolution ofthe suborder Ensifera (Orthoptera)]. In [System andevolution of the suborder Ensifera (Orthoptera)] (A.F. Emeljanov, ed.). Two Parts. 305 pp. Proceedingsof the Zoological Institute, Russian Academy of Sci-ences 260; 224 pp, 212 pp.

GWYNNE, D. T. 1995. Phylogeny of the Ensifera (Or-thoptera): A hypothesis supporting multiple originsof acoustical signalling, complex spermatophoresand maternal care in crickets, katydids, and weta. J.Orthopt. Res. 4:203–218.

HASEGAWA, M., H. KISHINO, AND T. YANO. 1985. Datingof the human–ape splitting by a molecular clock ofmitochondrial DNA. J. Mol. Evol. 22:160–174.

HUELSENBECK , J., AND K. CRANDALL. 1997. Phylogenyestimation and hypothesis testing using maximumlikelihood. Annu. Rev. Ecol. Syst. 28:437–466.

HUELSENBECK , J. P., AND J. J. BULL. 1996. A likelihoodratio test to detect con�icting phylogenetic signal.Syst. Biol. 45:92–98.

HUELSENBECK , J. P., D. M. HILLIS, AND R. JONES. 1996.Parametric bootstrapping in molecular phylogenet-ics: Applications and performance. Pages 19–45 inMolecular zoology: Advances, strategies, and proto-cols (J. D. Ferraris and S.R. Palumbi, eds.). Wiley-Liss,New York.

HUELSENBECK , J. P., AND B. RANNALA. 1997. Phyloge-netic methods come of age: Testing hypotheses in anevolutionary context. Science 276:227–232.

JUKES, T. H., AND C. R. CANTOR. 1969. Evolution ofprotein molecules. Pages 21–132 in Mammalian pro-tein metabolism (H. N. Munro, ed.). Academic Press,New York.

KAMBHAMPATI, S. 1995. A phylogeny of cockroachesand related insects based on DNA sequence of mi-tochondrial ribosomal RNA genes. Proc. Natl. Acad.Sci. USA 92:2017–2020.

KEVAN, D. K. M. 1952. On the systematic position oftwo anomalous genera previously placed in the sub-family Pyrgomorphinae (Orth., Acrididae). Entomol-ogists Monthly Mag 88:265–272.

KEVAN, D. K. M. 1982. Orthoptera. Pages 352–379 inSynopsis and classi�cation of living organisms, Vol-ume 2 (S. P. Parker, ed.). McGraw-Hill, New York.

KIRBY, W. F. 1910. A synonymic catalogue of Or-thoptera. Volume 3. Orthoptera Saltatoria. Part 2 (Lo-custidae vel Acrididae). British Museum, London.

KISHINO, H., AND M. HASEGAWA. 1989. Evaluation ofthe maximum likelihood estimate of the evolution-ary tree topologies from DNA sequence data, and thebranching order in Hominoidea. J. Mol. Evol. 29:170–179.

KUMAR, S., K. TAMURA, AND M. NEI. 1993. MEGA:Molecular Evolutionary Genetics Analysis. ThePennsylvania State Univ., University Park.

LANAVE, C., G. PREPARATA, C. SACCONE, AND G.SERIO. 1984. A new method for calculating evolu-tionary substitution rates. J. Mol. Evol. 20:86–93.

LOCKHART, P. J., M. A. STEEL, M. D. HENDY, AND D.PENNY. 1994. Recovering evolutionary trees undera more realistic model of sequence evolution. Mol.Biol. Evol. 11:605–612.

MOORE, W. 1995. Inferring phylogenies from mtDNAvariation: Mitochondrial-gene trees versus nuclear-gene trees. Evolution 49:718–726.

PAGE, R. D. M. 1993. Component. Natural HistoryMuseum, London.

PENNY, D., AND M. D. HENDY. 1985. The use of treecomparison metrics. Syst. Zool. 34:75–82.

REHN, J. A. G. 1948. The acridoid family Eumastaci-dae (Orthoptera). A review of our knowledge of itscomponents, features and systematics, with a sug-gested new classi�cation of its major groups. Proc.Acad. Nat. Sci. Phila. 100:77–139.

RENTZ, D. 1991. Orthoptera. Pages 290–296 in The in-sects of Australia (I. D. Naumann, ed.). MelbourneUniversity Press, Victoria.

RODRIGO, A. G., M. KELLY-BORGES, P. R. BERQUIST, AND P.L. BERQUIST. 1993. A randomisation test of the nullhypothesis that two cladograms are sample estimatesof aparametric phylogenetic tree. N.Z. J. Bot. 311:257–268.

SULLIVAN, J., K. E. HOLSINGER, AND C. SIMON. 1995.Among-site rate variation and phylogenetic analy-sis of 12S rRNA in sigmodontine rodents. Mol. Biol.Evol. 12:988–1001.

SWOFFORD, D. L., G. J. OLSEN, P. J. WADDELL, AND D.M. HILLIS. 1996. Phylogenetic inference. pages 407–514 in Molecular systematics (D. M. Hillis, C. Moritz,and B. K. Mable, eds.). Sinauer, Sunderland, Mas-sachusetts.

VICKERY, V. R. 1997. Classi�cation of the Orthoptera(sensu stricto) or Caelifera. Pages 5–40 in The bio-nomics of grasshoppers, katydids, and their kin (S. K.Gangwere, M. C. Muralirangan, and M. Muraliran-gan, eds.). CAB International, Wallingford, England.

YANG, Z., AND S. KUMAR. 1996. Approximate methodsfor estimating the pattern of nucleotide substitutionand the variation of substitution rates among sites.Mol. Biol. Evol. 13:650–659.

Received 1 February 1998; accepted 3 July 1998Associate Editor: C. Simon

combined molecular phylogenetic analysis of the orthoptera...

Documents