Phylogeny of Trichoptera (Caddisflies): … phylogenetic agreement has been reached about the...

Syst. Biol. 50(6):781–816, 2001

Phylogeny of Trichoptera (Caddis�ies): Characterization of Signaland Noise Within Multiple Datasets

KARL M. KJER,1 ROGER J. BLAHNIK,1,2 AND RALPH W. HOLZENTHAL2

1Department of Entomology, Rutgers University, Cook College, New Brunswick, New Jersey 08901, USA;E-mail: [email protected]

2Department of Entomology, University of Minnesota, St. Paul, Minnesota 55108, USA; E-mail: [email protected](current address), [email protected]

Abstract.—Trichoptera are holometabolous insects with aquatic larvae that, together with the Lep-idoptera, make up the Amphiesmenoptera. Despite extensive previous morphological work, lit-tle phylogenetic agreement has been reached about the relationship among the three suborders—Annulipalpia, Spicipalpia, and Integripalpia—or about the monophyly of Spicipalpia. In an effort toresolve this con�ict, we sequenced fragments of the large and small subunit nuclear ribosomal RNAs(1078 nt; D1, D3, V4-5), the nuclear elongation factor 1® gene (EF-1®; 1098 nt), and a fragment ofmitochondrial cytochrome oxidase I (COI; 411 nt). Seventy adult and larval morphological characterswere reanalyzed and added to molecular data in a combined analysis. We evaluated signal and ho-moplasy in each of the molecular datasets and attempted to rank the particular datasets accordingto how appropriate they were for inferring relationships among suborders. This evaluation includedtesting for con�ict among datasets, comparing tree lengths among alternative hypotheses, measuringthe left-skew of tree-length distributions from maximally divergent sets of taxa, evaluating the recov-ery of expected clades, visualizing whether or not substitutions were accumulating with time, andestimating nucleotide compositional bias.

Although all these measures cast doubt on the reliability of the deep-level signal coming from thenucleotides of the COI and EF-1® genes, these data could still be included in combined analyseswithout overturning the results from the most conservative marker, the rRNA. The different datasetswere found to be evolving under extremely different rates.

A site-speci�c likelihood method for dealing with combined data with nonoverlapping parameterswas proposed, and a similar weighting scheme under parsimony was evaluated.

Among our phylogenetic conclusions, we found Annulipalpia to be the most basal of the threesuborders, with Spicipalpia and Integripalpia forming a clade. Monophyly of Annulipalpia and In-tegripalpia was con�rmed, but the relationships among spicipalpians remain equivocal. [Annuli-palpia; Bayesian inference; dataset combination; homoplasy; Integripalpia; Spicipalpia; pseudorepli-cate reweighting.]

Trichoptera, or caddis�ies, are an order ofholometabolous insects with aquatic imma-ture stages. They are integral componentsof almost all freshwater communities (Reshand Rosenberg, 1984). With a fauna of some10,000 described extant species distributedamong 45 families (Morse, 1997a), the orderis diverse in terms of the microhabitats andtrophic niches the species occupy (Mackayand Wiggins, 1979). Ecological diversity andgeneral intolerance to pollution make the lar-vae excellent biological indicators of waterquality (Rosenberg and Resh, 1993). The net-spinning and case-making behaviors of thelarvae have long held the interest of biolo-gists (Williams et al., 1987). Despite this in-terest in Trichoptera, a comprehensive anal-ysis leading to a widely accepted phylogenyof higher-level relationships within the orderhas been elusive.

The purpose of this paper is, �rst andforemost, to provide a stable classi�cationfor the suborders of Trichoptera. Insights

into caddis�y evolution gained by thoroughphylogenetic analysis will provide a ro-bust framework for examining biological at-tributes of general interest within the order,such as case- and retreat-making behavior(Sukatsheva, 1980; Weaver and Morse, 1986;Frania and Wiggins, 1997), egg-laying andoviposition behavior (Weaver, 1983), evolu-tion of �lter feeding (Alstad, 1982; Thorp,1983; Thorp et al., 1986), pupation behavior(Wiggins and Wichard, 1989; Wichard andKlein, 1997), trophic relationships of larvae(Weaver and Morse, 1986), ancestral habi-tats (Shields, 1988; Kristensen, 1997), and bio-geography (Gall, 1997; Novokshovov andSukatcheva, 1997).

We collected sequence data from nu-clear ribosomal RNA (rRNA), elongationfactor 1alpha (EF-1®), and mitochondrial cy-tochrome oxidase I (COI), and we reeval-uated a morphological dataset from Franiaand Wiggins (1997). As in any study withdata from multiple sources, we had to make

781

at Pennsylvania State University on A

pril 24, 2014http://sysbio.oxfordjournals.org/

Dow

nloaded from

http://sysbio.oxfordjournals.org/

782 SYSTEMATIC BIOLOGY VOL. 50

decisions about how the combined datawould be treated. A secondary goal of thestudy was to evaluate the different sources ofdata so that we could perform an informedcombined analysis and conclude with a lim-ited set of hypotheses, from which the sup-port for any particular setof relationships canbe evaluated.

Dataset Combination

The debate over whether datasets shouldbe combined or analyzed separately (e.g.,Miyamoto, 1985; Kluge, 1989; Barrett et al.,1991; Bull et al., 1993; de Queiroz, 1993;Eernisse and Kluge, 1993; de Queirozet al., 1995; Huelsenbeck et al., 1996; Springeret al., 1999) has centered around either philo-sophical objections to excluding data or the-oretical explanations of why such exclusionswould be a good idea, with examples sup-porting both sides. Less attention has beenpaid to whether or not different partitionscould, in combination, answer questions indifferent parts of a tree (Sullivan, 1996;Wiens, 1998), or whether qualitative rank-ings of signal and noise could be used to�lter through alternative hypotheses. Yet sys-tematists who conduct preliminary analysesabout the level of information in candidategenes and then carefully select their genesto match the divergence times of their ques-tions implicitly expect different genes to beinformative at different hierarchical levels. Ifdifferent genes are informative at differentlevels, it is dif�cult to estimate relationshipsthat span 200 million years with a single geneor to categorize any single gene as “good” or“bad”. Here we evaluate datasets in amannerthat allows a ranking of datasets according tolocalized phylogenetic utility.

Taxonomic History

The sister group relationship between Tri-choptera and Lepidoptera, together com-prising the Amphiesmeniptera, has longbeen recognized and is among the moststrongly supported in entomology (Speyer,1870, and others [see Betten, 1934, for discus-sion]; Hennig, 1981; Kristensen, 1991, 1997;Pashley et al., 1993; Chalwatzis et al., 1996;Whiting et al., 1997). Within Trichoptera, itis now accepted that the order contains twomonophyletic suborders, Annulipalpia andIntegripalpia, with a third suborder, Spici-palpia, whose monophyly is equivocal. An-

nulipalpian larvae make �xed retreats, fromwhich they spin a silken net used to �lter�ne detritus or capture invertebrate prey. In-tegripalpian larvae make portable tube casesfrom which they feed in a variety of manners,most commonly as shredders or predatorsbut also as scrapers, �lterers, or herbivores,among others (Mackay and Wiggins, 1979).Spicipalpian larvae include the “free-living”predators (Rhyacophilidae and Hydrobiosi-dae); the “purse-case makers” (Hydroptili-dae), which feed by piercing algal cells orby gathering �ne detritus; and the “saddle-”or “tortoise-case makers” (Glossosomati-dae), which are specialized for scraping pe-riphyton off the upper surfaces of rocks. TheAnnulipalpia and Spicipalpia are primarilylotic, whereas the Integripalpia occur in bothlotic and lentic habitats (Wiggins, 1996).

Ross (1956, 1964, 1967) provided the�rst modern phylogenetic hypotheses ofsubordinal and superfamily relationships,but earlier workers also constructed gen-eral classi�cations for the order, includingUlmer (1912), Martynov (1924), and Milneand Milne (1939). Ross’ (1956, 1964, 1967)concept of Integripalpia contained twosuperfamilies, Limnephiloidea sensu latu(Integripalpia s.s.), and a paraphyletic “Rhy-acophiloidea” (Fig. 1a). His hypothesis of therelationships among Rhyacophilidae (thenincluding Hydrobiosinae), Glossosomati-dae, Hydroptilidae, and the Limnephiloideas.l. was based primarily on a presumedevolutionary transformation in larval case/pupal cocoon-making behavior. Recently,alternative morphologically based phylo-genies have been proposed challengingRoss’ view (summarized by Morse, 1997b).Weaver (1983, 1984, 1992a,b; Weaver andMorse, 1986) was the �rst to apply strictcladistic principles to caddis�y higher-levelclassi�cation and examined about twiceas many morphological characters as Ross(Fig. 1b). Wiggins and Wichard (1989; alsoWichard, 1991; Wiggins, 1992; Wichard et al.,1993; Wichard and Klein, 1997) suggestedthat the closed, semipermeable cocoonsof parchmentlike silk found in the spic-ipalpian families (limiting them to cold,well-oxygenated streams) represent thegroundplan condition of the order and thatthe cocoons of permeable silk with venti-lation openings found in the Annulipalpiaand Integripalpia are uniquely derived(Fig. 1c). Ivanov (1997) challenged Weaver’s



Dow

nloaded from


2001 KJER ET AL.—TRICHOPTERA PHYLOGENY 783

FIGURE 1. Five contemporary hypotheses of subor-dinal relationships of the Trichoptera. Equivalent taxo-nomic units are indicated by like symbols (e.g., Ross’Hydropsychoidea D Weaver’s Curvipalpia D Wigginsand Wichard’s Annulipalpia). (c) Representation of aphylogeny of pupation only (Wiggins, 1992). (e) Strictconsensus of �ve trees (Frania and Wiggins, 1997:Figs. 24 and 25). Spicipalpia includes the familiesRhyacophilidae, Hydrobiosidae, Glossosomatidae, andHydroptilidae.

hypothesis of Spicipalpia monophyly,providing evidence that each of Weaver’sfour spicipalpian apomorphies were ple-siomorphic or more generally distributedwithin Trichoptera (Fig. 1d). Most recently,Frania and Wiggins (1997) provided the�rst published analysis of Trichoptera re-lationships based on a computer- assistedsearch for most-parsimonious trees (usingHENNIG86). Their analysis of 70 adult andimmature characters resulted in �ve equallyparsimonious trees. Their strict consensussupported monophyly of Annulipalpia andIntegripalpia, but not of Spicipalpia (Fig. 1e).

In summary, at least �ve different hypothe-ses of basal Trichoptera relationships havebeen proposed or suggested (Fig. 1), differ-ing in the placement and monophyly of Spici-palpia and its included families. Phylogenieschallenging traditional classi�cations havealso been proposed for family relationshipswithin suborders (Fig. 2; see Morse, 1997b,for review).

Although the analysis of morphologicalcharacters has been extensive, consensusover relationships among suborders is at animpasse. However, molecular sequence datahave not been examined, and only one ofthe published phylogenetic studies utilizedautomated searches for most-parsimonioustrees (Frania and Wiggins, 1997). Additionaldata, rigorously analyzed, should provide afresh perspective to help stabilize caddis�yclassi�cation.

MATERIALS AND METHODS

Laboratory Protocols

The use of freshly frozen specimens waspreferred, but most taxa were available onlyas dried, pinned museum specimens. In bothcases, for DNA extraction, a single leg wastaken from larger specimens, or the headand thorax, including legs, from smallerspecimens. The remainder of the speci-men, including the wings and abdomen,the latter with its terminal genitalia, wasvouchered. Bar-coded and standardizedvoucher specimen labels were appliedto each specimen used in the study, andthis information was entered into a Biota(Colwell, 1996) database maintained at theUniversity of Minnesota Insect Collection,St. Paul (UMSP). Voucher specimens are de-posited in UMSP and other institutions (seeAppendix 2 at the Systematic Biology website,



Dow

nloaded from



FIGURE 2. Composite phylogeny of family-grouptaxa of Trichoptera and general global distribution. Re-lationships based on Weaver (1983, 1984) and Weaverand Malicky (1994) for families of Annulipalpia; Weaver(1983, 1984) for families of Spicipalpia; Gall (1994, 1997)for families of Plenitentoria; Weaver (1983) and Weaverand Morse (1986) for families of Leptoceroidea; and deMoor (1993) for families of Sericostomatoidea. AU DAustralasian, Cos D cosmopolitan, AF D Afrotropical,HO D Holarctic, NA D Nearctic, NT D Neotropical, ORD Oriental, PA D Palearctic, e D eastern, n D northern,s D southern, w D western.

www.utexas.edu/ftp/depts/systbiol). Spec-imens were identi�ed to species, except fora few female or larval specimens, for whichspecies determination was not possible.Samples were placed in labeled Eppendorftubes and ground over liquid nitrogen,using microtissue grinders (Phoenix Re-search). DNA was extracted with sodiumdodecyl sulfate, Proteinase-K, and phe-nol/chloroform, as described by Hillis andDavis (1986). Dried DNA pellets were resus-pended in 50–250 ¹l of Tris–EDTA buffer.Most of this material was separated as astock DNA collection and kept at ¡70±C. Therest was kept in a frostfree freezer for ampli-

�cation by polymerase chain reaction (PCR).Samples were ampli�ed on a thermal cyclerusing the reaction conditions described inSambrook et al. (1989). Initially, primers weredesigned from GenBank, and additionalprimers were then designed from our ownsequences. Primer sequences for the COIgene fragment were TAATTGGAGGATTTGGAAATG, paired with CCYGGTAAAATTAAAATATAAACTTC. The EF-1® fragmentwas ampli�ed in three overlapping pieces,starting with ATCGAGAAGTTCGAGAARGARGC paired with CCAYCCCTTRAACCANGGCAT; TTGCACGGRGAYAACATGTTRGA paired with either TTGAAACCAACGTTRTCRCC or GAAATYTGRCCAGGRTGRTT; and ACYACTGAAGTNAARTCNGT paired with GGGAAYTCCTGGAARGAYTC. The �rst variable region ofthe large subunit nuclear rRNA, D1, wasampli�ed with GGAGGAAAAGAAACTAACAAGGATT paired with CAACTTTCCCTTACGGTACT. The third variable region,D3, was ampli�ed with ACCCGTCTTGAAACACGGAC paired with either ATTCCCCTGACTTCGACCTGA or CTATCCTGAGGGAAACTTCGGA. The fourth through�fth variable region of the small subunit nu-clear rRNA, V4–5, was ampli�ed with eitherCAACTTTCCCTTACGGTACT or TGCG-GTTAAAAAGCTCGTAGT, paired with GC-CCTTCCGTCAATTCCTTTA. All primersare listed as 5’–3’. Some internal primers (notlisted) were designed for individual taxafrom the sequences we obtained by usingthe primers listed above. PCR conditionswere 94±C, 30 s; 50±C, 30 s; and 72±C, 45 s for35–45 cycles. Ampli�ed DNA was separatedon a 1.5% low-melting-point agarose gel(NuSieve 3:1; FMC Bioproducts). Bands ofDNA were cut from the agarose gel, puri�edwith GeneClean (Bio 101), and sequenced onan ABI 377 automated sequencer accordingto the manufacturer’s recommendations(Applied Biosystems), except that we usedone-half the recommended enzyme concen-tration in a reduced volume reaction. Se-quences were completed in both strands andedited manually with the help of SequenceNavigator (Applied Biosystems). Duringediting of each strand, nucleotides that werereadable but either showed irregular spacingbetween peaks or had some important com-peting background peak were coded withlowercase letters. These letters were con-verted to uppercase if the complementary



Dow

nloaded from



strand strongly con�rmed them, wereleft as lowercase when both strands werelowercase, or were expressed as N’s(nucleotide undetermined) when strandswere contradictory.

Taxon Sampling

Initially, we sought to obtain two repre-sentative taxa from each family, each as dis-tantly related as possible. In the absenceof published phylogenetic hypotheses, weused subfamily designation or geographi-cal distribution to select putative maximallydistant representative taxa. Later, when itbecame apparent that sequences from Inte-gripalpia were relatively similar to one an-other, whereas those from Annulipalpia andSpicipalpia were considerably more diver-gent, we increased representation from thelatter groups and decreased representationfrom Integripalpia (see Appendix 2, System-atic Biology website).

Our dataset includes outgroup represen-tatives from Mecoptera and Siphonaptera, aswell as representatives of every lepidopteransuborder: Zeugloptera (Micropterigidae),Aglossata (Agathiphagidae), Heterobath-mioidea (Heterobathmiidae), and Glossata(the vast majority of Lepidoptera). Ingrouptaxa include all trichopteran families ex-cept Limnocentropodidae (Leptoceroidea);Petrothrincidae, Hydrosalpingidae, Bar-barochthonidae, and Antipodoeciidae (Seri-costomatoidea); and Pisuliidae and Rossian-idae (Limnephiloidea). Missing taxa wereeither unavailable or we were unable to ob-tain PCR product from degraded specimens.

Genes Examined

We sequenced the �rst (D1: 334 nt) andthird (D3: 233 nt) variable regions of the largesubunit nuclear rRNA, the fourth throughthe �fth (V4–5: 511 nt) variable regions ofthe small subunit nuclear rRNA, and a frag-ment of the mitochondrial COI gene (441 nt)for most taxa. COI was sequenced to pro-vide additional resolution within suborders.Initially, nearly the entire gene for EF-1®(1098 nt) was sequenced to obtain a sec-ond conservative marker (using either itsnucleotides or its amino acids), indepen-dent of the rRNA genes. However, it be-came apparent from preliminary analysisthat the EF-1® gene was less than ideal forthe study of trichopteran suborders; third po-

sitions were excessively homoplastic, whilefew amino acids varied at all. Therefore,further sequencing of the EF-1® was lim-ited to Annulipalpia because the EF-1® se-quencing of this suborder was already nearlycomplete by the time of our preliminaryanalysis. Sequences were submitted to Gen-Bank under accession numbers AF436131-AF436645 (see Appendix 2, Systematic Biologywebsite).

Alignment

The COI gene was length invariant ex-cept for a single missing codon in Dipseu-dopsidae. The EF-1® gene lacked introns,did not vary in length, and alignment wastrivial. We used MacClade 4.0 (Maddisonand Maddison, 2000) to color-code sequencesby translated amino acids to check for stopcodons and proofread the edited sequences.When lowercase letters (see above section,Laboratory protocols) dictated amino acidchanges, we returned to the original gels tocon�rm the nucleotide.

The rRNA was aligned manually withreference to secondary structure, and itsnotation follows that of Kjer et al. (1994)and Kjer (1995). Alignments followedthe secondary structure models of Gutellet al. (1994), downloaded from the websitehttp://www.rna.icmb.utexas.edu, and weremodi�ed where compensatory substitu-tions con�rmed a customized secondarystructural model for amphiesmenopteranrRNA. Regions that could not be alignedwere excluded from the analysis. The cri-terion for data inclusion used secondarystructure to identify the boundaries oflength-heterogeneous regions. Nucleotidesin variable-length regions between �ankinghydrogen-bonded nucleotides were eitherexcluded (Kjer, 1997) or recoded (see below).Three regions that could be unambiguouslyaligned among all Trichoptera were notalignable across outgroups. In these regions,the outgroup sequences were replaced with“?” (coded as missing data).

Character Coding

Nucleotides were treated as unorderedcharacters with four alternative states. Thecoding of insertions and deletions is shownin Appendix 1. Typically, “indels,” when con-sidered at all, are treated by most inves-tigators as a single kind of character; we,



Dow

nloaded from



however, do not consider all “indels” to bethe same. Each length-heterogeneous regionthat contained phylogenetically informativeinsertions and/or deletions was evaluatedseparately and divided into one of threeclasses: insertions, deletions, or “indels”. In-sertions were de�ned as characters in whichsuccessive outgroups were all of identicallength and lacked a nucleotide present insome or all of the ingroup (Trichoptera). Thecharacter state of the missing nucleotideswas then de�ned as “ancestrally missing,”coded with an asterisk, and de�ned in the“symbols” option in PAUP¤. Using this sameoutgroup criterion, we de�ned deletions inthe ingroup as missing data, because al-though “ancestrally missing” is a de�nedstate, we could not de�ne the state of anucleotide that was lost. When we couldnot categorize a region as either an inser-tion or a deletion by outgroup comparison,we de�ned the region as an indel. Indel re-gions of variable length, even when these re-gions were not alignable across all taxa, of-ten contained phylogenetic signal in some (ifnot all) lineages. Therefore, we eliminatedthe nucleotide characters from the analy-sis but coded each unique combination ofnucleotides in these variable regions witha different symbol. PAUP¤ limits the num-ber of character states, and for some indelswe identi�ed as many as the maximum of32 states. Many of these states were synapo-morphic for only a few taxa; even so, we pre-ferred to retain some information rather thaneliminate these regions altogether. In somelength-heterogeneous regions, the number ofnucleotides in the region was coded. Thecoding of each individual indel shown inAppendix 1 is fully explained on the Sys-tematic Biology website. This method of cod-ing indels was similar to Wheeler (1999) andLutzoni et al. (2000), except that we did notuse a step matrix.

Data Analysis

As a starting point, phylogenetic anal-ysis was performed by way of character-based parsimony using PAUP¤ 4.0b3a(Swofford, 1999). Heuristic searches were im-plemented, with either 50 or 500 replicates ofrandom-taxa additions (depending on treeisland pro�les). Support at each node inthe cladogram was analyzed according tothe decay index (Bremer, 1988; Donoghue

et al., 1992) and nonparametric bootstrap-ping (Felsenstein, 1985). For bootstrapping,500 pseudoreplicates were performed, eachincluding 10 random addition searches. Ho-mogeneity of base frequencies across taxawas evaluated with a chi-square test usingPAUP¤. When signi�cant deviation from ho-mogeneity of nucleotide composition wasfound, we performed Log-Det minimumevolution analyses (Lockhart et al., 1994). Tofurther examine the possibility of nucleotidecompositional effects, we constructed a “GCtree” using a matrix of Euclidean distancesbetween nucleotide frequencies for each pairof taxa (Lockhart et al., 1994:formula 4), andcompared it with the parsimony tree withconsensus procedures.

We also used maximum likelihood as anoptimality criterion. Model and parameterswere selected after running MODELTEST3.04–3.06 (Posada and Crandall, 1998) andusing the Aikaike information criterion (AIC;Akaike, 1974). For all analyses, MODELTESTdictated use of a general time reversible(GTR) model with a gamma correction (Yang,1993, 1994a,b, 1996) for among-site rate vari-ation and invariant sites (Gu et al. 1995); nu-cleotide frequencies were estimated from thedata.

Bayesian inferences were used to estimatephylogeny and branch support under like-lihood, using the program MrBAYES 2.0(Huelsenbeck, 2000). For each analysis, twoMarkov chains were run, with 480,000 cyclesfor each chain. Treeswere saved to a �le every400 cycles, and the �rst 200 of these trees werediscarded. This left us with two tree �les,each containing 1,000 trees, which we thenpooled. A majority-rule consensus of these2,000 trees was then used to generate approx-imations of the posterior probability of eachclade. We used a starting tree generated froma neighbor-joining analysis, having selectedthe model and parameters by MODELTEST.

Hypothesis Testing

We used the constraint option in PAUP¤

to examine the differences among alter-native trees generated from other datapartitions and optimality criteria. Followingthe recommendations of Swofford et al.(1996) and Goldman et al. (2000) in re-stricting the use of the Templeton test, onlya priori alternative trees were comparedwith one another under parsimony with a



Dow

nloaded from



two-tailed Wilcoxon-ranked sum test(Templeton, 1983; Larson, 1994). The treesgenerated from the dataset under evaluationwere excluded from the test because thesetrees cannot be assumed to �t the nullhypothesis that they are no better than theother trees under comparison (Swoffordet al., 1996; Goldman et al., 2000). So thatcon�ict would be measured only amongsubordinal hypotheses and not amongfamily relationships within suborders, taxaconsidered in the constraints were con�nedto orders, the suborders Annulipalpia andIntegripalpia, and the families Hydroptili-dae, Glossosomatidae, Rhyacophilidae, andHydrobiosidae. Essentially, we treated theTrichoptera as a six-taxon group, resolvingthese six taxa as published by other authors,or as our analyses resolved them, but leavingthe relationships within Annulipalpia andIntegripalpia unconstrained. For example,even though Weaver and Morse (1986) pre-sented relationships among annulipalpianand integripalpian families, for our evalu-ation of con�ict, we were interested onlyin the relationships among the suborders.Accordingly, we de�ned our constraints asfollows: “Constraints Weaver and Morse D((Siphonaptera) (Mecoptera) ((Lepidoptera)(Integripalpia) ((((Hydrobiosidae) (Rhya-cophilidae)) ((Glossosomatidae) (Hydroptil-idae))) (Annulipalpia)))).” Our constraintscan be found at the bottom of our NEXUS�les, available on the Systematic Biologywebsite. In addition to constraints usedfor hypothesis testing, some of our phy-logenetic analyses included constraints.Constraints in the EF-1® data were imple-mented because one-third of the data wasmissing for Micropteryx, Pseudostenophylax,Monocosmoecus, Lype, Gumaga, and Marilia.Constraints in the amino acids–only analy-ses were necessary because unconstrainedanalyses required searching throughhundreds of thousands of equally parsimo-nious trees, preventing the completion ofsearches. Constraints are indicated on our�gures.

Data Combination and Signal Evaluation

We considered four independent datasets:the nuclear rRNAs, the nuclear EF-1®, the mi-tochondrial COI, and morphology, the latteras presented by Frania and Wiggins (1997).Each dataset was �rst analyzed separately

and then combined with the others. We eval-uated the potential for excessive homoplasyin three ways. First, we examined the leftskew of tree distributions from exhaustivesearches of �ve, six, seven, and eight taxa.Taxa were drawn according to a structuredrandom selection. One outgroup, Merope tu-ber (Mecoptera), was �xed in each set. A sec-ond outgroup for each set, either Catocolaor Agathiphaga (Lepidoptera), was selectedby coin toss. Finally, one species each fromAnnulipalpia, Spicipalpia, and Integripalpiawas selected by lot from a pool of taxa forwhich sequence data from all three molec-ular datasets were complete. Identical setsof taxa were used for both rRNA and COIdatasets, but for EF-1®, for which fewer In-tegripalpia with complete data were avail-able, we substituted taxa when necessary.Ten such sets of taxa were constructed. Sepa-rate exhaustive searches were completed oneach set of taxa for the rRNA, EF-1®, andCOI data, and the g1 statistic was recordedand compared to the signi�cance levels pre-sented by Hillis and Huelsenbeck (1992). Fora sixth taxon, an additional annulipalpianwas added and the analyses were repeated.If the �rst annulipalpian selected was in ei-ther the Philopotamidae or Stenopsychidaefamilies, then the second was selected fromanother family (to ensure that signal wasnot being con�ned to closely related taxa).This division corresponds to currently rec-ognized superfamily designations in the sub-order (families with and without ocelli). Theseventh taxon was an additional spicipalpianfrom a family different than the �rst. Finally,the eighth taxon was a second integripalpian.If the �rst integripalpian was from Pleni-tentoria, then a brevitentorian taxon was se-lected as the second, and vice versa. We werefortunate with Trichoptera, because both theingroup and the outgroups can be separatedinto well-de�ned groups, and we could berelatively certain that our sampling plan forthese groups provided maximally separatedsets of taxa.

Signal was also evaluated by assessingwhether or not trees from the g1 statisticsearches (above) recovered highly corrob-orated clades. Three clades, highly corrob-orated as monophyletic by morphologicaland molecular evidence, were identi�ed:Trichoptera, Annulipalpia, and Integri-palpia. Failure of the structured randomtaxon subsets to recover these clades renders



Dow

nloaded from



the data suspect, especially if excessivehomoplasy in the data is indicated by othermeans.

Finally, we attempted to visualize homo-plasy in the data with the construction ofa tree that spanned Trichoptera diversityand its outgroups and represented a reason-able estimate of phylogeny (Fig. 3). Uncor-rected mean percent nucleotide differenceswere calculated and placed on the nodesof this tree. Although a clocklike rate wasnot assumed, we did assume that ances-tors precede their descendants, and if true,then mean pairwise differences should in-crease with time unless the positions free tochange have already changed. The follow-ing tree was considered highly corroborated:(Mecoptera, Diptera (Lepidoptera ((Annuli-palpia) (Integripalpia)))). No Spicipalpiawere included because there is not consen-sus in the literature on their monophyly oron their placement relative to Annulipalpiaor Integripalpia, and because their inclusionwas not necessary for evaluating whether ornot divergence was increasing with time. Wethen plotted the observed mean uncorrecteddistances against the corrected distances es-timated by using models and parametersselected from the MODELTEST analysis,marking points on the nodes from the“highly corroborated tree” (above). Becausecharacter-based analyses may not be subjectto “saturation” as we are measuring it here,we evaluated patristic distances (total num-ber of changes separating taxa) on a com-bined data tree (including all taxa) for eachindividual dataset and traced the mean ofthese values through our “highly corrobo-rated tree.”

These analyses were designed to evaluatedatasets under parsimony. Even if we ques-tion the utility of a dataset for a distanceanalysis or an equally weighted parsimonyanalysis, some conservative signal could behidden within the noise of the quickly evolv-ing characters. For this reason, we evalu-ated “noisy” datasets under maximum likeli-hood, LogDet-corrected distances (Lockhartet al., 1994), and differentially weightedparsimony, because having the abilityto accommodate among-site rate-variation,branch-length heterogeneity, unequal nu-cleotide frequencies, and differences amongsubstitution frequencies, we might recover areasonable estimate of phylogeny from evenan apparently saturated dataset.

Morphology

The morphological data used in our anal-ysis was as presented by Frania and Wiggins(1997), with some exceptions. First, Franiaand Wiggins conducted only two heuristicsearches on the combined larval and adultcharacter sets, and they suspected the ex-istence of additional tree islands becausethey found additional trees in their sec-ond search. We conducted 50 random ad-dition searches of the morphological datawith no limit to the number of trees evalu-ated. Second, Frania and Wiggins assignedplesiomorphic states to outgroup taxa, evenwhen no homolog existed in the outgroup.We coded these characters in our dataset as“?”, rather than “0”. Finally, for the anal-ysis of the morphological data alone, weused Frania and Wiggins’ character polarityassignments because we felt that the differ-ences in our approach warranted a new anal-ysis that was as close as possible to the orig-inal dataset. In the combined data analysis,however, the morphological characters wereconsidered unordered because additionalmolecular characters should provide alterna-tive potential polarities. In our constrainedanalysis of alternative hypotheses, the “Fra-nia and Wiggins” hypothesis corresponds toour analysis of their characters (Fig. 7A, notFig. 1D). Composite taxa in the combinedanalysis were avoided whenever possible,but some taxa included in Frania and Wig-gins (1997) were not available to us, and taxathat we included as composites (one taxoncoded for morphology, combined with DNAsequences from another) are described in Ap-pendix 2 (see the Systematic Biology website,where they are marked with an asterisk inthe far right column). We also performedan unpolarized analysis, using differentiallyweighted parsimony (described below) onthe morphological data.

Combined Analysis

In determining potential combinationsused in a combined analysis, one criterionwe used to avoid dataset combination was re-ciprocal rejection according to the Wilcoxonsigned rank test (dataset A rejects the hypoth-esis from dataset B and vice versa). Anotherquestion we addressed was, “Does addingnoisy data bring us a more accurate hypoth-esis?” Of course, this is complicated withthe problem of how “accurate” is de�ned.



Dow

nloaded from



While we cannot predetermine an “accurate”hypothesis, we can evaluate signal in each ofthe partitions. An ideal combination wouldbe a slowly evolving dataset to provide sig-nal at the base of the tree while more rapidlyevolving genes resolve the tips of the tree. Ifdataset A were found to be more appropri-ate for the resolution of relationships amongsuborders, and dataset B were less appro-priate, then dataset B would be added to acombined analysis only if it did not over-turn results from dataset A. Similarly, if anuncorrected (equally weighted parsimony)combined analysis revealed deep phyloge-netic patterns shared with the partitioned“noisy” datasets, and a reasonable correctionof the combined analysis recovered a deephistory shared with the more conservativepartitioned dataset, then the “corrected” hy-pothesis would be favored. Similar to this ap-proach, others have constrained nodes thathave been estimated through characters orcombined analyses that have been deemedreliable (i.e., Moritz et al., 1992; Ballard et al.,1998).

The combined data included a subset oftaxa, starting with those used by Frania andWiggins (1997), and adding sequence datafor which the majority of sequences werecomplete. Characters in this combined anal-ysis included D1, D3, V4–5 rRNA, EF-1® nu-cleotides and amino acids, COI nucleotidesand amino acids, and morphology. Justi�-cation for including both nucleotides andamino acids in a combined analysis was dis-cussed by Agosti et al. (1996), Benabib et al.(1997), and Flores-Villela et al. (2000). Un-ordered data were both equally and differ-entially weighted in combined analyses.

In order to perform a combined analy-sis that would accommodate the extremedifferences in substitution rates among thedatasets, as well as among-site rate varia-tion within the datasets, we utilized “pseu-doreplicate reweighting” described by Kjeret al. (in press). Brie�y, this is a weightingscheme that generates 1,000 trees from a fastheuristic bootstrap analysis of the combineddata. Each character was then reweighted ac-cording to the highest rescaled consistencyindex from any of the 1,000 trees. Becauseeach of these trees comes from a differentpseudoreplicate of original data, and sincea strict consensus is an unresolved poly-tomy (only Trichoptera was recovered), wepredict that this weighting scheme will not

be as subject to the circularity imposed bythe original tree (Cunningham, 1997) as issuccessive weighting (Farris, 1969). How-ever, missing data will in�ate the weights(Archie, 1989), and smaller datasets willbe underrepresented in combined analyses.Generation of the 1,000 trees for the pseu-doreplicate reweighting of the morphologi-cal data included only the unpolarized mor-phological data rather than the combineddata.

We also performed a likelihood analysison the same set of taxa, using Bayesian infer-ences, on the combined nucleotide data. Bullet al. (1993) and Sullivan (1996) presentcontrasting conclusions about the combin-ability of datasets when different partitionsevolve under different models. We agreewith both studies, the differences betweenthem depending on whether the param-eters of separate datasets are suf�cientlyoverlapping or not (Sullivan, 1996). We mea-sured datasets with Bayesian estimates ofparameters, with 95% con�dence intervals,to see if they were closer to nonoverlapping(Bull et al., 1993) than complementary(Sullivan, 1996). Given excessive differencesin parameter estimates among genes, wedevised a method for assigning individualcharacters to site-speci�c rate classes fromthe pseudoreplicate reweighting schemedescribed above and have attempted to ac-commodate the criticism of site-speci�c ratemodels summarized by Buckley et al. (2001).We took the 1,000 bootstrap trees from thecombined data and reweighted them ac-cording to the “best consistency index,” witha base weight of 5. Each character was thenassigned an integer from 1 to 5 as a result,and the characters were then sorted intocharacter partitions according to their num-ber. Bayesian analysis was performed witha GTR model, with the �ve site-speci�c rate-classes. We hope the inaccurate assumptionthat all members of an a priori assigned class(such as third-codon positions or COI nu-cleotides) evolve at an identical rate is moreaccurately re�ected in this method, whereindividual rate classes were not assigneda priori.

RESULTS

Ranking Conserved Signal

All three of our tests for signal at thelevel of subordinal relationships indicated



Dow

nloaded from



that COI nucleotides and EF-1® nucleotideswere less appropriate for the estimation ofrelationships among suborders than was therRNA dataset at this level. As indicated inTable 1, the distribution of tree lengths forthe rRNA data has a signi�cant left skewin all analyses, whereas the COI nucleotideshave a signi�cant left skew in only 30% ofthe �ve-taxon (suborder) analyses. Similarly,EF-1® nucleotides failed to show signi�-cant left skew in all 10 of the �ve-taxondatasets. By examining the recovery of ex-pected clades, the rRNA data recovered amonophyletic Trichoptera in 100% of anal-yses, a monophyletic Annulipalpia in 100%of analyses, and a monophyletic Integri-palpia in 85% of analyses (Table 1). TheCOI nucleotides recovered the monophyly ofTrichoptera, Annulipalpia, and Integripalpiain 40%, 35%, and 20% of the analyses, re-spectively (Table 1). EF-1® nucleotides re-covered monophyletic Trichoptera, Annuli-palpia, and Integripalpia in 70%, 90%, and20% of the analyses, respectively (Table 1).

Figures 3 and 4 con�rm the left skew anal-ysis. These �gures convey similar informa-tion and are meant to be compared withone another. For example, we suggest select-ing one of the terminal nodes on Figure 3and tracing the values of a particular geneas you go “back in time.” Then �nd thesame starting point on Figure 4 and tracefrom nodes 1 to 6. Points should move upand to the right. At node 4 (Figs. 3 and4A), where pathways converge at the levelof subordinal relationships, the rRNA dataincreased in mean uncorrected differences,whereas neither the COI nucleotides nor theEF-1® nucleotides increased in divergencefrom node 4 to node 5. Patristic distancesaccumulated in all datasets up through Tri-choptera (Fig. 3, node 4), then decreased inboth EF-1® and COI between Trichopteraand Lepidoptera (node 5), and then increasedagain between Amphiesmenoptera and themost distant outgroups (node 6). None of thedatasets met the criterion of reciprocal con-�ict according to the Templeton test (Table 2;Templeton, 1983).

Summarizing the analyses of estimatedhomoplasy and signal strength, we found therRNA data to be most appropriate for esti-mating relationships among suborders, butthe signal from the rRNA within subordersis weak. We have little con�dence in COI orEF-1® for the recovery of the deepest nodes,

but increasing patristic distances (increasingat least within Trichoptera), combined withheavy taxon sampling and appropriate cor-rection, may still yield useful phylogeneticinformation from these noisy datasets. How-ever, the trees we present from individualpartitions are meant to be viewed as explo-rations of the data rather than competinghypotheses.

Phylogenetic Analyses

Results from our equally weighted par-simony analysis of the COI and EF-1® nu-cleotide datasets corroborated our analysisof signal because even with all taxa included,they do not show the monophyly of groupssuch as Lepidoptera or Trichoptera. Hy-potheses generated from equally weightedparsimony analyses of both COI andEF-1® nucleotide datasets were rejected un-der appropriately applied Templeton tests(Table 2). To save space in publication,equally weighted parsimony trees from thesedatasets, along with our other trees and ex-ecutable NEXUS �les, are included in theTREEBASE (herbaria.harvard.edu/treebase)websites but are not presented here. We con-centrated parsimony analyses on the rRNAand combined data, while attempting to es-timate phylogeny from the noisy datasetsthrough maximum likelihood.

The result of the equally weighted parsi-mony analysis of the rRNA data is shownin Figure 5. This analysis recovered a mono-phyletic Annulipalpia and Spicipalpia, thelatter allied with the Integripalpia. Plen-itentoria and Brevitentoria (LeptoceroideaC Sericostomatoidea) were also monophyl-etic (Weaver and Morse, 1986). However,Plenitentoria was not identical to the taxonde�ned by Weaver and Morse (1986) be-cause Kokiriidae was included within it inthe rRNA analysis; Kokiriidae (as Plectro-tarsidae) was included within Leptoceroideain the phylogeny presented by Weaver andMorse (1986).

The tree presented in Figure 5 representsthe �rst of our estimates of subordinal rela-tionships within Trichoptera. Bootstrap val-ues are given on Figure 5, but some nodesdiffered; the bootstrap analysis of the rRNAdata showed a paraphyletic Spicipalpia, withHydrobiosidae as the sister taxon to a mono-phyletic Integripalpia, and arctopsychines asthe sister taxon to the rest of the hydropsy-chids. Several well-established families do



Dow

nloaded from


TAB

LE

1.R

esu

lts

ofm

easu

rem

ents

ofle

ft-s

kew

intr

eed

istr

ibu

tion

s(H

illis

and

Hu

else

nbec

k,19

92).

The

top

row

labe

ls10

sep

arat

ean

alys

es,e

ach

mad

eu

pof

sets

of5,

6,7,

and

8ta

xa,w

hich

wer

esu

cces

sive

lyad

ded

.Lis

tsof

thes

eta

xa,i

ncl

ud

edin

each

seta

rein

dic

ated

belo

wth

eco

lum

nn

um

ber.

Inth

eco

lum

nat

the

far

left

,bel

owth

eli

stof

taxa

are

the

nu

mer

als

5,6,

7,an

d8

inth

eu

ppe

rle

ftco

rner

ofea

chbo

x;th

ese

ind

icat

eth

enu

mbe

rof

taxa

inth

ean

alys

esfo

llow

ing

toth

eri

ght(

see

text

for

des

crip

tion

ofho

wth

ese

taxa

wer

ese

lect

ed).

“#C

har.”

refe

rsto

the

nu

mbe

rof

info

rmat

ive

char

acte

rsin

the

dat

aset

,giv

enth

eta

xafo

rea

chan

alys

is(l

iste

dab

ove)

.“g1

”re

fers

toth

est

atis

tic

from

Hill

isan

dH

uel

sen

beck

(199

2)th

atd

escr

ibes

the

shap

eof

the

his

togr

amof

allt

ree

leng

ths.

Asi

ngl

eas

teri

skfo

llow

ing

the

g1st

atis

tic

ind

icat

essi

gni�

canc

eat

p<

0.05

.D

oubl

eas

teri

sks

ind

icat

esi

gni�

canc

eat

p<

0.01

.Ana

lyse

sd

eter

min

edto

cont

ain

sign

i�ca

ntle

ftsk

ewin

tree

dis

trib

uti

ons

are

inbo

ld.T

he

ques

tion

“Mon

ophy

leti

c?”

refe

rsto

wh

ethe

ror

not

each

anal

ysis

reco

vere

da

mon

ophy

leti

cte

stgr

oup

:YD

yes,

ND

no,S

Din

som

eof

the

mu

ltip

lesh

orte

sttr

ees.

The

ord

erof

thes

equ

esti

ons

isal

way

sth

esa

me.

The

test

grou

ps

are:

�rs

tle

tter

Dm

onop

hyle

tic

Tric

hop

tera

,sec

ond

lett

erD

mon

oph

ylet

icA

nnu

lipal

pia

,th

ird

lett

erm

onop

hyl

etic

Inte

grip

alp

ia.F

orth

ep

urp

oses

ofev

alu

atin

gsi

gnal

,we

cons

ider

edon

lyth

em

onop

hyly

ofTr

icho

pte

ra,A

nnu

lip

alp

ia,a

nd

Inte

grip

alp

iato

beex

pec

ted

.Th

ese

test

sin

dic

ated

that

the

CO

Iand

EF-

1®n

ucl

eoti

de

dat

ad

ono

tp

rovi

de

relia

ble

info

rmat

ion

wit

hre

spec

tthe

rela

tion

ship

sam

ong

subo

rder

s.

Ana

lysi

s1

23

45

67

Out

grou

pM

erop

eM

erop

eM

erop

eM

erop

eM

erop

eM

erop

eM

erop

eLe

pid

opte

raA

gath

apha

gaA

gath

apha

gaA

gath

apha

gaC

atoc

ola

Cat

ocol

aA

gath

apha

gaC

atoc

ola

Spic

ipal

ia#1

Oxe

thir

aA

gape

tus

Ato

psyc

heA

tops

yche

Oxe

thir

aR

hyac

ophi

laO

xeth

ira

Ann

ulip

alpi

a#1

Arc

tops

yche

Phy

loce

ntro

pus

Sten

opsy

che

Xip

hoce

ntro

nX

ipho

cent

ron

Ecn

omus

Dip

lect

rona

Inte

grip

alp

ia#1

Phr

ygan

opsy

che

Cae

nota

Lim

neph

ilus

Phr

ygan

eaP

sued

oste

noph

ylax

Tric

hove

spul

aLi

mne

philu

sA

nnu

lipal

pia

#2P

hylo

cent

ropu

sA

rcto

psyc

heX

ipho

cent

ron

Hyd

rops

yche

Dol

ophi

lode

sA

rcto

psyc

hePs

ycho

myi

aSp

icip

alpi

a#2

Pro

topt

ilaA

tops

yche

Oxe

thir

aR

hyac

ophi

laP

alae

agap

etus

Pal

aeag

apet

usG

loss

osom

aIn

tegr

ipal

pia

#2P

silo

tret

aP

hryg

anop

sych

eH

elic

opsy

che

Cae

nota

Con

tulm

aP

hryg

anea

Cae

nota

CO

I 5#

char

.16

122

716

916

218

215

917

8G

1¡

.232

¡.9

86¤¤

¡.3

99C

.897

¡.2

79¡

.806

¤¡

.166

Mon

ophy

leti

c?Y

YS

NN

YN

6#

Ch

ar.

230

232

181

171

197

166

187

G1

¡.4

24¡

.556

¡.4

83¡

.022

¡.1

70¡

.255

¡.7

35¤¤

Mon

ophy

leti

c?Y

YY

YY

NN

NN

NSN

NN

7#

Ch

ar.

243

244

198

185

199

177

194

G1

¡.5

42¤¤

¡.5

23¤¤

¡.3

83¡

.196

C.1

87¡

.401

¡.4

24¤

Mon

ophy

leti

c?S

SY

YSS

NN

NN

YY

NN

8#

Ch

ar.

249

246

209

195

206

190

206

G1

¡.4

83¤¤

¡.4

00¤¤

¡.2

85¡

.021

¡.1

27¡

.065

C.1

28M

onop

hyle

tic?

YY

NY

YN

NN

NN

NN

NN

YY

YY

NN

NR

NA 5#

Ch

ar.

314

293

296

300

310

308

307

G1

¡1.

401¤¤

¡1.

412¤¤

¡1.

430¤¤

¡1.

353¤¤

¡1.

412¤¤

¡1.

388¤¤

¡1.

361¤¤

Mon

ophy

leti

c?Y

YY

YY

YY

6#

Ch

ar.

317

307

309

320

317

316

312

G1

¡1.

735¤¤

¡1.

785¤¤

¡1.

702¤¤

¡1.

555¤¤

¡1.

538¤¤

¡1.

732¤¤

¡1.

742¤¤

Mon

ophy

leti

c?Y

YY

YY

YY

YY

YY

YY

Y

791



Dow

nloaded from


TAB

LE

1.(C

onti

nued

)R

esu

lts

ofm

easu

rem

ents

ofle

ft-s

kew

intr

eed

istr

ibu

tion

s(H

illis

and

Hu

else

nbe

ck,1

992)

.The

top

row

labe

ls10

sep

arat

ean

alys

es,e

ach

mad

eu

pof

sets

of5,

6,7,

and

8ta

xa,w

hic

hw

ere

succ

essi

vely

add

ed.L

ists

ofth

ese

taxa

,inc

lud

edin

each

set

are

ind

icat

edbe

low

the

colu

mn

nu

mbe

r.In

the

colu

mn

atth

efa

rle

ft,

belo

wth

elis

tof

taxa

are

the

num

eral

s5,

6,7,

and

8in

the

up

per

left

corn

erof

each

box;

thes

ein

dic

ate

the

num

ber

ofta

xain

the

anal

yses

follo

win

gto

the

righ

t(s

eete

xtfo

rd

escr

ipti

onof

how

thes

eta

xaw

ere

sele

cted

).“#

Cha

r.”re

fers

toth

en

um

ber

ofin

form

ativ

ech

arac

ters

inth

ed

atas

et,g

iven

the

taxa

for

each

anal

ysis

(lis

ted

abov

e).“

g1”

refe

rsto

the

stat

isti

cfr

omH

illis

and

Hu

else

nbec

k(1

992)

that

des

crib

esth

esh

ape

ofth

eh

isto

gram

ofal

ltr

eele

ngt

hs.A

sing

leas

teri

skfo

llow

ing

the

g1st

atis

tic

ind

icat

essi

gni�

can

ceat

p<

0.05

.Dou

ble

aste

risk

sin

dic

ate

sign

i�ca

nce

atp

<0.

01.A

nal

yses

det

erm

ined

toco

ntai

nsi

gni�

cant

left

skew

intr

eed

istr

ibu

tion

sar

ein

bold

.The

ques

tion

“Mon

oph

ylet

ic?”

refe

rsto

wh

eth

eror

note

ach

anal

ysis

reco

vere

da

mon

ophy

leti

cte

stgr

oup

:YD

yes,

ND

no,

SD

inso

me

ofth

em

ult

iple

shor

test

tree

s.T

heor

der

ofth

ese

ques

tion

sis

alw

ays

the

sam

e.T

hete

stgr

oup

sar

e:�

rstl

ette

rDm

onop

hyl

etic

Tric

hop

tera

,sec

ond

lett

erD

mon

ophy

leti

cA

nnu

lipal

pia

,thi

rdle

tter

mon

ophy

leti

cIn

tegr

ipal

pia

.Fo

rth

ep

urp

oses

ofev

alu

atin

gsi

gnal

,we

cons

ider

edon

lyth

em

onop

hyly

ofTr

icho

pte

ra,A

nnu

lip

alp

ia,a

ndIn

tegr

ipal

pia

tobe

exp

ecte

d.T

hese

test

sin

dic

ated

that

the

CO

Ian

dE

F-1®

nu

cleo

tid

ed

ata

do

not

pro

vid

ere

liabl

ein

form

atio

nw

ith

resp

ect

the

rela

tion

ship

sam

ong

subo

rder

s.

Ana

lysi

s1

23

45

67

7#

Ch

ar.

332

311

325

323

324

322

323

G1

¡1.

988¤¤

¡2.

023¤¤

¡1.

889¤¤

¡1.

631¤¤

¡1.

732¤¤

¡1.

819¤¤

¡1.

814¤

Mon

ophy

leti

c?Y

YY

YY

YY

YY

YY

YY

Y8

#C

har

.34

132

034

133

033

432

433

0G

1¡

1.96

9¤¤¡

1.94

8¡

1.93

5¤¤¡

1.56

4¤¤¡

1.87

8¤¤¡

1.88

1¤¤¡

1.79

0¤¤

Mon

ophy

leti

c?Y

YY

YY

YY

YN

YY

YY

YY

YY

YY

YY

EF-

1al

ph

aO

utgr

oup

Mer

ope

Mer

ope

Mer

ope

Mer

ope

Mer

ope

Mer

ope

Mer

ope

Lep

idop

tera

Cat

ocol

aC

atoc

ola

Cat

ocol

aC

atoc

ola

Cat

ocol

aC

atoc

ola

Cat

ocol

aSp

icip

alia

#1A

nchi

tric

hia

Aga

petu

sA

tops

yche

Ato

psyc

heB

ryso

pter

ixR

hyac

ophi

laA

nchi

tric

hiA

nnu

lipal

pia

#1A

rcto

psyc

heP

hylo

cent

ropu

sSt

enop

sych

eX

ipho

cent

ron

Xip

hoce

ntro

nE

cnom

usD

iple

ctro

naIn

tegr

ipal

pia

#1O

econ

esus

Olin

gaLi

mne

philu

sO

econ

esus

Lim

neph

ilus

Gum

aga

Lim

neph

iluA

nnu

lipal

pia

#2P

hylo

cent

ropu

sA

rcto

psyc

heX

ipho

cent

ron

Hyd

rops

yche

Dol

ophi

lode

sA

rcto

psyc

heP

sych

omyi

aSp

icip

alia

#2A

naga

petu

sA

tops

yche

Anc

hitr

ichi

aR

hyac

ophi

laR

hyac

ophi

laB

ryso

pter

ixG

loss

osom

aIn

tegr

ipal

pia

#2M

arill

iaO

econ

esus

Hel

icop

sych

eP

ycno

cent

rode

sC

hath

amiid

aeO

econ

esus

Olin

ga5

#C

har

.31

231

829

832

230

939

828

4G

1¡

.533

¡.4

86C

.012

C.0

59¡

.417

C.0

65¡

.894

Mon

ophy

leti

c?Y

NY

YY

NY

6#

Ch

ar.

346

339

328

348

323

352

298

G1

¡.5

05¡

.936

¤¡

.845

¤¡

.003

¡.3

34¡

.820

¤¡

.845

¤

Mon

ophy

leti

c?Y

NN

YY

YY

YY

NY

YY

Y7

#C

har

.35

936

134

438

136

736

731

1G

1¡

.591

¤¤¡

.744

¤¤¡

.765

¤¤¡

.237

¡.2

96¡

.754

¤¤¡

.532

¤¤

Mon

ophy

leti

c?Y

YN

YY

YY

YN

NY

YY

Y8

#ch

ar.

369

376

364

400

391

380

349

G1

¡.3

78¤

¡.6

77¤¤

¡.6

42¤¤

¡.3

02¡

.572

¤¤¡

.840

¤¤¡

.542

¤¤

Mon

ophy

leti

c?Y

NN

YY

YY

YN

YY

NY

YN

YY

NN

YN

792



Dow

nloaded from



FIGURE 3. Subset of taxa whose relationships are highly corroborated by multiple datasets. Numerals at the in-ternodes represent mean pairwise uncorrected percent differences and patristic distances. These values are spatiallyorganized as shown in the key at lower left. The large numerals inside the circles are for reference to pathwaysin Figure 4, the arrows indicating the direction the eye should follow. The x’s represent missing values for EF-1®,given the lack of data for Mecoptera and Arctopsyche.

not emerge as monophyletic in Figure 5,including the polyphyletic Philopotamidaeand Glossosomatidae and the paraphyleticHydrobiosidae. Similarly, Psuedoneureclipsisdid not group with other Polycentropodidae,but its placement within Polycentropodidaehas been questioned recently (Li et al., 2001).

The polyphyly of the Philopotamidae issurprising. The monophyly of the family, in-

cluding Wormaldia, is supported by severalmorphological synapomorphies, especiallythe uniquely modi�ed, T-shaped, membra-nous larval labrum (Wiggins, 1996). Fra-nia and Wiggins (1997) recovered a mono-phyletic Philopotamidae but did not includeWormaldia in their taxon set. When the rRNAsequence data analysis are constrained toa monophyletic Philopotamidae, the tree



Dow

nloaded from



FIGURE 4. Graphical representation of the accumulation of substitutions at successively deeper nodes in thetree presented in Figure 3. Points are plotted as mean uncorrected distances versus mean corrected distances, thediagonal line representing a 1:1 relationship. (a) rRNA, in circles; (b) EF-1®, in squares; (c) COI, in ovals. Small �lledsquares in (a) and (c) represent the points for EF-1®, indicating the differences in scale between panels. The small�lled square in (a) that is near the circled 5 represents the square marked 4 from (b) (the square marked 6 from [b] isnot shown in [a]). Similarly, the small �lled circle in (c) represents node 6 (the circled 6) from the rRNA for reference.



Dow

nloaded from


TAB

LE

2.R

esu

lts

ofth

eW

ilco

xon

rank

edsu

mte

stof

alte

rnat

ive

hyp

othe

ses

wit

hm

ult

iple

dat

aset

s.A

lter

nat

ive

hyp

othe

ses

are

liste

dac

ross

the

top

row

,by

dat

aset

from

this

pap

er,

orby

auth

orfr

omp

revi

ousl

yp

ubl

ishe

dhy

pot

hese

s(F

ig.1

).T

he�

tof

alte

rnat

ive

hyp

oth

eses

iste

sted

onth

ed

atas

ets

liste

din

the

far

left

colu

mn

.Val

ues

inp

aren

thes

esre

pre

sent

a-po

ster

iori

hyp

oth

eses

gene

rate

dfr

omth

ed

atas

etu

nder

eval

uat

ion,

and

wer

eth

eref

ore

not

incl

ud

edin

the

test

s.T

he

top

num

ber

liste

din

each

box

isth

ele

ngth

ofth

em

ostp

arsi

mon

iou

str

ee(s

),co

nstr

aine

dto

the

vari

ous

hyp

othe

ses.

The

bott

omnu

mbe

ris

the

hig

hest

pva

lue

give

nfo

rth

eW

ilcox

onra

nke

dsu

mte

stfo

ran

yof

the

a-pr

iori

con

stra

ined

tree

s.Si

gni�

canc

eis

ind

icat

edby

¤if

p<

0.05

and

¤¤if

p<

0.01

.Sho

rtes

tso

luti

ons

and

anal

yses

det

erm

ined

toco

ntai

nsi

gni�

can

tcon

�ic

tar

ein

bold

.

EF-

1aEF

-1a

CO

IW

eave

r&

Wig

gins

&Fr

ania

&H

ypot

hesi

sR

NA

RN

AM

LPa

rsim

ony

EF-

1aM

LL

ogD

etE

F-1a

AA

Par

sim

ony

CO

IML

CO

IAA

Ros

sM

orse

Wic

hard

Wig

gins

Ivan

ov

Dat

aset

(227

6)(2

285)

2463

2293

2293

2289

2317

2293

2291

2287

2293

2287

2297

2296

RN

A<

.000

1¤¤.3

00.1

23.5

42.0

01¤¤

.104

.452

.481

.176

1.00

0.0

67.1

70E

F-1a

4226

4225

(418

4)(4

217)

(421

3)42

2642

6642

4442

2242

2342

2542

2742

3242

23nu

cleo

tid

es.7

24.7

29.0

32¤

.248

1.00

0.8

12.7

95.7

17.4

61.9

89E

F-1a

amin

o33

834

0(3

39)

(342

)(3

39)

338

349

343

342

340

339

339

342

341

acid

s1.

000

.686

.066

.282

.400

.644

.822

.782

.449

.479

CO

I33

6533

9233

8433

8333

8033

86(3

354)

(337

7)(3

375)

3379

3386

3359

3382

3382

nucl

eoti

des

.812

.064

.249

.132

.153

.124

.164

.151

1.00

0.1

79.1

42C

OI

603

610

608

606

606

608

(610

)(6

10)

(600

)60

760

860

060

560

6am

ino

acid

s.8

16.2

22.2

52.4

91.6

10.2

15.5

64.2

441.

000

.628

.508

Mor

ph

olog

y35

335

636

036

535

835

036

435

435

8(3

44)

(357

)(3

49)

(342

)(3

58)

ord

ered

.433

.264

.071

.017

¤.1

061.

000

.036

¤.2

72.2

45M

orp

hol

ogy

301

301

308

309

306

298

313

309

307

(295

)(3

03)

(301

)(2

95)

(297

)u

nord

ered

.616

.616

.110

.118

.169

1.00

0.0

53.1

15.2

39

795



Dow

nloaded from



FIGURE 5. Results of an unconstrained, equally weighted parsimony analysis of rRNA data. This tree representsa strict consensus of 558 trees of 2,276 steps. Numerals above the internodes represent decay indices; those below theinternodes represent bootstrap proportions. Internodes without numbers represent decay indices of 1 or bootstrapproportions not recovered in the bootstrap analysis.

length is increased by a single step. Finally,other molecular datasets (see below) recovera monophyletic Philopotamidae in uncon-strained analyses. If Wormaldia is indeedmisplaced in Figure 5, its misplacement haspotentially severe consequences to the accu-racy of inferred relationships within the restof Annulipalpia. For example, if Wormaldia

is in reality a member of the Philopotamidae,where does Dipseudopsis then belong? IsDipseudopsis a member of a clade thatincludes Ecnomidae, or is it allied withPhilopotamidae? If close to Philopotamidae,would Dipseudopsis then drag the Ecnomi-dae with it, to an af�liation with the otherphilopotamids? A single misplaced taxon in



Dow

nloaded from



a tree potentially can affect the reconstructedancestral states at a large number of nodesand render the tree suspect in the neighbor-hood of the error.

The parsimony analysis of the rRNA(Fig. 5) has the Protoptilinae more closelyrelated to Hydrobiosidae than to otherGlossosomatidae. The polyphyly of Glosso-somatidae is also dif�cult to accept. Glosso-somatid monoplyly is supported by unam-biguous synapomorphies, including uniquelarval morphology and case-construction be-havior (Wiggins, 1996). The morphologicalanalysis indicates that the protoptiline glos-sosomatids ally with the other Glossosomati-dae, as they do in combined analyses. TherRNA sequences of the Protoptilinae, hererepresented by Protoptila and Culoptila, areamong the most autapomorphic in of allTrichoptera, and their placement in Figure 5could result from an accelerated substitutionrate for these taxa. In addition, if the protop-tilines are in the wrong place in Figure 5, thenthe paraphyly of the Hydrobiosidae may notbe supported.

The hypothesis generated from a maxi-mum likelihood analysis of the rRNA datasetis shown in Figure 6. Likelihood parametersare shown in Table 3. Empirical base frequen-cies of the rRNA data were 24%, 24%, 31%,and 21% A, C, G, and T, respectively. Al-though the likelihood-generated hypothesisagreed with the parsimony analysis in plac-ing the Spicipalpia in a clade with Integri-palpia, both groups emerged as polyphyletic

TABLE 3. Likelihood parameters, given as means, with standard deviation and 95% con�dence intervals, gener-ated by the Bayesian analysis. The third through the eighth columns (labeled A–C through G–T) indicate estimatedvalues from the general time reversible R-matrices, normalized to the RNA. “Inv” and “Alpha” refer to the estimatedproportion of invariable sites and the gamma shape parameter, respectively.

A–C A–G A–T C–G C–T G–T Inv Alpha

RNA mean 1.171 3.596 2.815 .768 7.003 1.000 .424 .326s.d. 0.046 0.215 0.268 0.044 1.731 0.000 .0002 .001295%ci .775– 2.619– 1.999– 0.563– 5.574– 1.000 .393– .291–

1.554 4.427 3.896 1.145 10.039 1.000 .449 .398EF-1® mean 3.728 11.405 3.739 5.803 19.655 1.553 .474 .8571

s.d. .336 3.600 .499 1.190 9.034 0.000 .0003 .003895%ci 2.292– 8.692– 2.636– 4.251– 15.07– 1.553– .4422– .742–

4.191 15.006 5.067 8.168 25.776 1.553 .5047 .985COI mean 1.386 11.455 .3262 9.168 71.548 2.462 .209 .2143

s.d. .1531 2.002 .0037 2.886 615.16 0.000 .0095 .003195%ci 0.4569– 7.649– .1553– 4.653– 24.59– 2.462– .075– .1546–

2.6119 16.250 .5219 14.697 134.7 2.462 .339 .2777Combined mean 3.362 8.085 9.133 3.8324 13.197 1.654

s.d. .0561 .3087 .3663 0.0733 .6499 0.00095%ci 2.850– 6.897– 7.890– 3.2359– 11.402– 1.654–

4.051 9.699 10.961 4.5676 15.490 1.654

in the likelihood analysis, with Culoptila(Glossosomatidae) nested within Integri-palpia. This placement is weakly supported,and Integripalpia monophyly is being foundin 6% of the Bayesian trees and protoptiline(Protoptila C Culoptila) monophyly in 16% ofthe Bayesian trees. One of the advantages ofa Bayesian analysis is that alternative cladescan be evaluated, and although 6% and 16%are not indications of strong support, thetrees that include a monophyletic Protoptili-nae and Integripalpia are among the set of“good” trees. Protoptilines are found to bemonophyletic in both the COI nucleotide-and amino acid-derived trees.

An unconstrained equally weightedparsimony tree (not shown) constructedonly from insertion, deletion, and “indel”characters that were excluded from thelikelihood analysis, resulted in a strict con-sensus tree that included (Outgroup ((Lep-idoptera) ((Annulipalpia) (“Spicipalpia”“Integripalpia”)))). Other features of thistree included a monophyletic Philopotami-dae. The protoptiline glossosomatids weremonophyletic but still separated from theother Glossosomatidae. Kokiriidae againgrouped inside a monophyletic Pleniten-toria, and Xiphocentronidae was the sistertaxon of a monophyletic Psychomyiidae.

Figure 7 represents the hypotheses gener-ated from the morphological characters ofFrania and Wiggins (1997), except that statesnot present in the outgroup were coded asmissing data (“?”) instead of plesiomorphic



Dow

nloaded from



FIGURE 6. Phylogram resulting from an unconstrained Bayesian maximum likelihood analysis of rRNA data.This is a majority-rule consensus of the �nal 1,000 of 1,200 trees from each of two Monte Carlo Markov chains. AGTR model with a gamma correction for among-site rate variation and invariant sites was used. The likelihood scorewas –11,327.0. Species are labeled as genera, with capital letter mnemonic abbreviations that refer to families, butthe taxa are identical to those used in Figure 5. Numerals on the nodes are percent recoveries from the consensusand can be translated as estimated posterior probabilities. In most cases, values are placed in the middle of theinternode; sometimes, for considerations of space, the value is placed above and to the right of the internode ordirectly to the left. The insert in the lower left is included to show the scale of the phylogram among Trichoptera andits outgroups; including outgroups with the ingroup at the same scale as used for the main �gure would require a�gure spanning �ve pages.



Dow

nloaded from



FIGURE 7. (a) Strict consensus of 18 trees (tree length D 354) from a reanalysis of the morphological characterspresented by Frania and Wiggins (1997). Capital letter mnemonic abbreviations refer to families. Values abovethe internodes are decay indices, and below the internodes outside the parentheses are the number of charactersthat support each node. Characters that support the node without homoplasy are placed in parentheses below thenode and numbered as in Frania and Wiggins (1997). (b) Hypothesis generated from the morphological data afterpseudoreplicate reweighting. Values above the nodes indicate bootstrap support, while the numerals below theinternodes are as in panel a. x indicates the node was not supported in a bootstrap analysis.

(“0”). The decision to code character statesthis way made a difference in the analysis(Frania and Wiggins, 1997). When we treatedthe morphological data exactly as Franiaand Wiggins did, we recovered 23 trees of358 steps, instead of 5 trees recovered byFrania and Wiggins (1997), but the topol-ogy of the strict consensus was virtuallythe same as theirs, differing only by a col-lapse in resolving Atriplectides with the koko-riid plus molannid clade and a slight dif-ference with the resolution of Lepidostoma(not shown). Our analysis, with theunknownoutgroup states coded as missing data, re-covered a monophyletic Annulipalpia andIntegripalpia, with a polyphyletic (Fig. 7A)or paraphyletic (Fig. 7B) “Spicipalpia.”The protoptiline glossosomatids groupedwith the other glossosomatid. Plenitento-

ria was monophyletic but excluded Kokiri-idae, as in Weaver and Morse (1986). TheLeptoceroidea and Sericostomatoidea (Fig. 7,bottom) showed little resolution, and whatresolution was obtained was only weaklysupported and is uncorroborated by ourdata. Within the Plenitentoria, Phryganei-dae and Phryganopsychidae were relativelybasal, in agreement with Gall (1994), whoused these taxa as outgroups in a study of theLimnephiloidea. Both the rRNA parsimonytree and the equally weighted morphologytree placed Phryganopsychidae as the mostbasal of the Plenitentoria. Our reanalysis ofFrania and Wiggin’s data supported theirconclusion that Limnocentropodidae is thesister taxon to the rest of Integripalpia, but wecould not con�rm this with an independentdataset because we were unable to obtain



Dow

nloaded from



DNA sequence data from Limnocentropodi-dae during our study. A subsequent anal-ysis did not support Frania and Wiggin’s(1997) placement of Limnocentropodidae(Kjer et al., in press). When subjected topseudoreplicate reweighting, using 1,000trees generated from a fast heuristic boot-strap analysis of unordered morphologicaldata, we �nd the relationships shown inFigure 7B. According to this analysis, Franiaand Wiggin’s larval character 34 (see alsoRoss, 1967) best de�nes higher groups with-out homoplasy in Trichoptera; the outgroupand Hydropsychoidea share two small scle-rites on abdominal tergum IX with at leastone long seta arising from each (Franiaand Wiggins, 1997). Tergum IX is membra-nous in the Philopotamoidea and lacks con-spicuous setae. “Spicipalpia,” Plenitentoria,Leptoceroidia, and Allocella have a singlelarge sclerite on tergum IX, with setae,whereas the nonhelicophid sericostomatoidshave a membranous ninth tergite but retainlong setae.

It is informative to examine the differ-ences between Figures 7A and 7B. Figure 7Bshows that when characters are differen-tially weighted, the topology changes withrespect to the relationship among suborders.Yet bootstrap analyses of the differentiallyweighted data do not support the new re-lationships but, in fact, support a mono-phyletic Rhyacophila and Atopsyche as a sis-ter taxon to Annulipalpia in 47% of the trees(not shown). This is not surprising, giventhat the weighting scheme favors the few(one) nonhomoplastic characters that maynot always be sampled in a bootstrap analy-sis. Interestingly, when the equally weighted,unpolarized morphological data are sub-ject to a bootstrap analysis, the topology inFigure 7B is supported! What this shows isthat the hypotheses of relationships amongsuborders from morphological data are notstable to analysis assumptions, changingamong many possibilities with every pertur-bation. This is important, because we arguethat noisy datasets do not pose con�ict withconservative datasets at the base of the tree,but the morphological dataset is not a noisydataset.

Three trees from the EF-1® data are shownin Figure 8. The trees are remarkably sim-ilar, especially considering the estimates ofhomoplasy and considering that the uncon-strained parsimony tree (TREEBASE web-

site) contained Micropteryx (Lepidoptera)inside the Annulipalpia as well as Diptera in-side the Integripalpia. Both nucleotide anal-yses recover a monophyletic Plenitentoria,Brevitentoria, and Sericostomatoidea, andBayesian estimates of posterior probabilitiesare generally high. Likelihood parametersare shown in Table 3. Empirical base fre-quencies of the EF-1® data were 25%, 30%,26%, and 19% A, C, G, and T, respectively.We consider nodes that are shared betweennucleotide and amino acid trees to indicaterelative con�dence that those nodes are rep-resentative of the “gene tree” (not to beconfused with independent congruence or“accuracy”). The amino acid tree recov-ers Brevitentoria and Sericostomatoidea.Dipseudopsidae and Stenopsychidae gr-oup together in all three analyses. Thegrouping of the predatory spicipalpians(Hydrobiosidae and Rhyacophilidae) andthe resolution within both Philopotamidaeand Integripalpia are shared between nu-cleotide trees and the amino acid tree. How-ever, we cannot overstate these results, bothbecause of the constraints we imposed andbecause of the alternative hypotheses withrespect to the placement of Hydroptilidae.

The chi-square tests showed that nu-cleotide composition for individual taxa,when compared with mean values, did notdeviate signi�cantly from expected valuesfor either the rRNA data or the COI databut did differ signi�cantly for the EF-1® se-quences. These differences remained whenthe outgroup was eliminated and each ofthe suborders was examined individually.This result dictated the analysis of EF-1®under a LogDet model (Lockhart et al.,1994), the only method we are aware ofthat accommodates among-taxon nucleotidecompositional heterogeneity (Fig. 8B). To ex-amine the possibility of nucleotide compo-sitional effects on the EF-1® data, we con-structed a “GC tree” (Lockhart et al., 1994).The GC tree was 544 steps longer than theparsimony tree (4,719 vs. 4,175), and a strictconsensus of both unconstrained analyseswas unresolved. However, after applicationof a few constraints, the tree length differ-ences between the GC tree and the parsi-mony tree dropped to 254 steps (4,466 vs.4,219), and some suspicious groupings wereshared between both analyses. A parenthet-ical strict consensus of the GC tree and theparsimony tree, with constraints shown in



Dow

nloaded from


FIG

UR

E8.

Phy

loge

neti

ch

ypot

hese

sre

sult

ing

from

anal

ysis

ofth

eE

F-1®

gen

e.C

rep

rese

nts

con

stra

ined

nod

es.(

a)H

ypot

hesi

sfr

oma

Bay

esia

nm

axim

um

likel

ihoo

dan

alys

isof

the

nucl

eoti

des

.T

helik

elih

ood

scor

ew

as¡

18,0

50.3

.Nu

mer

als

abov

eth

eno

des

are

esti

mat

edp

oste

rior

pro

babi

litie

s.(b

)R

esu

lts

from

am

inim

um

evol

uti

onhe

uri

stic

sear

chu

sing

the

Log

Det

corr

ecti

on.N

um

eral

sab

ove

the

nod

esar

ebo

otst

rap

valu

es.x

rep

rese

nts

lack

ofsu

pp

ort

bybo

otst

rap

.(c)

Hyp

oth

esis

from

aneq

ual

lyw

eigh

ted

par

sim

ony

boot

stra

pan

alys

isof

amin

oac

ids.

(In

this

anal

ysis

,no

mor

eth

an10

0tr

ees

wer

esa

ved

per

rep

lica

te.)

Ova

lsar

ound

boot

stra

pva

lues

rep

rese

ntu

ncon

stra

ined

nod

eson

the

amin

oac

idtr

eeth

atar

eal

sop

rese

ntin

one

orbo

thof

the

nu

cleo

tid

etr

ees.

The

boot

stra

pva

lues

sep

arat

edby

asl

ash

(e.g

.,C

/85)

rep

rese

ntbo

otst

rap

valu

esw

ith

alte

rnat

ive

cons

trai

nts

:Eit

her

Cer

acle

aw

asco

nst

rain

edw

ith

Mar

ilia,

orB

revi

ten

tori

aw

asco

nstr

ain

ed.

801



Dow

nloaded from



brackets, is [Merope[Diptera] [[Lepidoptera][Glossosomatidae Hydroptilidae Hydrob-iosidae (Rhyacophila) [[Limnephilidae] (Oe-conesus Gumaga Beraea Helicopsyche OlingaMarilia ((Pycnocentrodes Philanisus)Ceraclea))][Stenopsyche Polycentropus Cyrnellus Phy-locentropus [Cheumatopsyche Macrostemum((Diplectrona Homoplectra) Smicridea turrial-bana) (Hydropsyche Leptonema (Smicridea ta-lamanca Parapsyche))] [Philopotamidae] [Psy-chomyiidae] (Cyrnus Austrotinodes EcnomusXiphocentron)]]]]. Parentheses show the taxathat share similar nucleotide compositions.Phylogenetically related taxa that share nearidentical sequences will, of course, sharenearly identical nucleotide compositions, butmore distantly related taxa may also groupaccording to nucleotide compositional simi-larity. Suspicious groups shared between theGC tree and the parsimony tree involve theresolution within Hydropsychidae, and the

FIGURE 9. Phylogenetic hypotheses resulting from analysis of the COI gene fragment. C represents constrainednodes. (a) Majority rule consensus from a Bayesian maximum likelihood analysis of the nucleotides (GTR Cgamma C invariant sites). The likelihood score was ¡12,715.4. Numerals above the internodes represent poste-rior probabilities. (b) Hypothesis from an equally weighted parsimony bootstrap analysis of amino acids. Numeralsabove the internodes represent bootstrap proportions. Ovals around these values represent unconstrained nodeson the amino acid tree that are also present in the nucleotide tree.

grouping of Cyrnus with Xiphocentron andthe ecnomids, resulting in the paraphyly ofPolycentropodidae, and the grouping of Py-cnocentrodes with Philanisus, resulting in thepolyphyly of Conoesucidae (see Figs. 5 and6). The LogDet tree (Fig. 8B) did not al-ter these “suspicious” relationships, but theamino acid tree (Fig. 8C) contradicts themall except (Pycnocentrodes C Philanisus). AnAdams consensus (Adams, 1972; not shown)of the unconstrained GC tree and the uncon-strained parsimony tree includes both Mi-cropteryx (Lepidoptera) and Hydroptilidaenested within Annulipalpia.

Two trees generated from analyses of theCOI data are shown in Figure 9. Empiri-cal base frequencies of the COI data were30%, 19%, 13%, and 38% A, C, G, and T, re-spectively. Most of the agreement betweenthe COI nucleotide tree and the COI aminoacid tree is con�ned to the more apical



Dow

nloaded from



nodes, although both recover the mono-phyly of Annulipalpia and Integripalpia.Note the striking similarity of the two trees(Fig. 9A and B) concerning the higher-levelrelationships within Annulipalpia. Bothanalyses subdivide Annulipalpia into Hy-dropsychidae, Philopotamoidea (Stenopsy-chidae C Philopotamidae), and a thirdclade containing the other annulipalpianfamilies.

With the exception of basal placement ofthe Hydrobiosidae in both COI trees, for themost basal nodes, trees from nucleotides andamino acids were different from one anotherfor the most basal nodes for both the EF-1®gene (Fig. 8) and the COI gene fragment (Fig.9). Given that the amino acid and nucleotidedata from any gene must share a commonhistory, lack of agreement must represent aproblem with the data or the assumptionsbehind some or all analyses. With the substi-tution rate pro�les shown in Figures 3 and 4,we do not expect strong, concordant supportfor the most basal nodes, and we suspect thatthe lack of congruence re�ects homoplasy inboth genes (Table 1; Figs. 3 and 4).

We explored Wilcoxon signed rank testsa priori (Templeton, 1983) to evaluate the �tof a priori trees generated from other data inthis study as well as those from previouslypublished hypotheses oneach of our datasets(Table 2). In general, noisy datasets acceptalternative hypotheses without a large in-crease in tree length. Swofford et al. (1996)and Goldman et al. (2000) argued thatthe Kishino–Hasegawa test (Hasegawa andKishino, 1989; Kishino and Hasegawa, 1989)is inappropriate for comparison of phyloge-nies that are not de�ned a priori and statedthat this problem extends to the Temple-ton test as it is usually run: comparing thebest trees generated from a particular datasetwith other suboptimal trees. By using the“salvage” option described in Goldman et al.(2000) for inappropriately applied two-tailedTempleton tests, the EF-1® nucleotide datasetsigni�cantly rejects a tree generated from itsown amino acids (not shown); this datasetwould thus seem to be in con�ict with itself.Because alternative hypotheses constrainedon the EF-1® nucleotide data result intree-length additions of from 38 to 82 steps,we suspect that the homoplasy in the EF-1®nucleotides is not randomly distributed butrather has some pattern. This nonrandom ho-

moplasy could be coming from the among-taxon nucleotide compositional heterogene-ity observed in the data, and our “GC-tree”(Lockhart et al., 1994) lends some supportof this speculation. However, we cannotclaim that these �ndings demonstrate com-positional effects, because the “suspicious”groups were also found in the LogDet tree(Fig. 9B).

Complicating a combined likelihood anal-ysis, we �nd the datasets to be evolv-ing at excessively different rates; in mostcases, 95% con�dence intervals on parame-ter estimates are nonoverlapping (Table 3).The tree depicted in Figure 10 is a resultof the combined likelihood analysis of thenucleotide data, using a site-speci�c ratemethod designed to compensate for het-erogeneity among sites, both within andacross datasets. Figure 11 shows the re-sults of an equally weighted parsimonybootstrap analysis that includes all of thedata. Congruence among datasets is alsoshown in Figure 11. The differences betweenthe equally weighted and pseudoreplicatereweighted trees are minor enough to besummarized in the text rather than presentedin a separate �gure and are designatedwith x’s on Figure 11. The pseudorepli-cate reweighted tree placed Diplectrona atthe base of Hydropsychidae, with Arc-topsyche next, and Macrostemum and Hy-dropsyche together with a 97% bootstrap.Written parenthetically, with bootstrap val-ues included, the other differences in re-lationships within Annulipalpia from thepseudoreplicate reweighted tree were Phy-locentropus 39(89((Paduniella 51(Lype, Psy-chomyia)) 69((Xiphocentron 69(Nyctiophylax,Ecnomus))))). Within the Plenitentoria, Pan-gulia moved from its relatively apical po-sition to the sister taxon of the rest of theclade, supported by a bootstrap value of 69.All other differences were extremely weaklysupported; Pseudoeconesus was the next mostbasal taxon in the rest of Plenitentoria, af-ter which the Brachycentrus/Phryganopsycheclade was the sister taxon to the rest of thelimnephiloids. Goera switches place with theuenoids. In the Brevitentoria, Psilotreta andAtriplectides switched places, and Allocellaand Philanisus moved from their position inFigure 11 to the base of the sericostomatoids,with Allocella being most basal (see morphol-ogy, character 34; Fig. 7B).



Dow

nloaded from



FIGURE 10. Results of the Bayesian combined nucleotide data. The likelihood score was ¡29,954.3. Estimatedrelative rates for the �ve character partitions were 1, 4.823; 2, 3.072; 3, 1.546; 4, 1.324; and 5, 0.091, with 207, 330,188, 84, and 1,784 characters in each respective class. Numerals above the internodes represent estimated posteriorprobabilities.

DISCUSSION

Trichoptera Classi�cation

Summarizing our molecular and com-bined hypotheses (Figs. 10 and 11), we con-clude that Annulipalpia is the most basalsuborder, with Spicipalpia and Integripalpiaforming a clade, thus helping to resolve oneof the major disputes about evolutionary re-

lationships within the order Trichoptera. Al-though the support for this relationship ismodest under equally weighted parsimony,with a decay index of 1, it is consistent withthe rRNA data (parsimony, likelihood, andgap character analyses), the differentiallyweighted parsimony analysis of the mor-phology (Fig. 7B), and the EF-1® amino aciddata (Fig. 8C). Our analyses of homoplasy



Dow

nloaded from



FIGURE 11. Tree from the combined equally weighted parsimony bootstrap analysis. Characters in this analysisincluded D1, D3, and V4–5 rRNA; complete EF-1® amino acids; COI nucleotides; COI amino acids; and morphology.Tree length D 5,241. Numerals above the internodes are bootstrap values from the differentially weighted analysis,followed by bootstrap values from this analysis, followed by decay indices from this analysis. Nodes in con�ictbetween differentially weighted and equally weighted analyses are marked with an x. Numerals below the intern-odes refer to support by separate analyses of different datasets, according to the key (lower left), with parenthesesincluded to draw attention to the differences among taxa in the different analyses. Note that some decay indices D0 in the bootstrap tree, indicating nodes that collapsed in the strict consensus tree (the strict consensus tree can bereconstructed in this �gure by collapsing those nodes).



Dow

nloaded from



plays a qualitative role in evaluating sup-port, because the combined analyses agreewith the most appropriate of the molecu-lar datasets for inferring relationships amongbasal Trichoptera clades. Another qualitativevote of con�dence for a node in questionconcerns whether or not support increaseswith analyses designed to reduce the im-pact of homoplastic characters. Support forthe node linking the spicipalpians with theIntegripalpia increases both with likelihood(97% posterior probability) and with differ-ential weighting, where bootstrap supportincreases from 69 to 92 (Fig. 11). Alternativeanalyses support either a monophyletic ar-rangement or any of a variety of paraphyleticor polyphyletic arrangements of Spicipalpia,with no consistent pattern except that therelative relationships among the Spicipalpia,when considered unrooted, are remarkablyconsistent, with Hydrobiosidae and Rhya-cophilidae next to one another, as are Glos-sosomatidae and Hydroptilidae. Althoughmonophyly of Spicipalpia is not consistentlysupported, it does emerge from the anal-yses as a possibility (Fig. 5). We �nd thissuggestive enough to retain spicipalpianmonophyly as a suboptimal but viable hy-pothesis. Our hypothesis of relationships isnot in strong con�ict with the morphologicaldata of Frania and Wiggins (1997), and theinclusion of these data actually adds supportto our overall conclusions. In a general sense,the phylogeny lends support to one conclu-sion of Frania and Wiggins (1997), namely,that the “Spicipalpia” retain characters prim-itive for the order. According to our analy-sis, however, this results not from the basalposition of the suborder in the overall phy-logeny, but rather from its basal position inone of two major clades for the order, wheresome primitive characters might be expectedto be retained. If Spicipalpia is monophyletic,then equally parsimonious solutions exist forthe evolution of various key innovations,including cocoon-making and case-makingbehaviors. Both Annulipalpia and Integri-palpia are found to be monophyletic, withrelatively high bootstrap support, relativelyhigh decay indices, and support from mul-tiple independent datasets (Fig. 11). Clearly,characters shared between Annulipalpia andIntegripalpia cannot be invoked to have acommon origin, except with the possibilitythat the character state was primitive forthe order and lost in the Spicipalpia. Shared

primitive characters would be hard to recon-cile if Spicipalpia is paraphyletic.

At the level of superfamily, both equallyweighted morphology (Fig. 7A) and the com-bined analysis (Fig. 11) supported mono-phyly of both the Hydropsychoidea andthe Philopotamoidea within Annulipalpia.Support for most alternative arrangementsshown in Figures 5, 6, 7B, 8, and 9 is ex-tremely weak, but Figure 10 shows that noneof the Bayesian trees recover these superfam-ilies because of the placement of Stenopsy-che with Phylocentropus—a clade supportedstrongly but exclusively in the analysis ofthe EF-1® (Fig. 8). We could not placethe Pseudoneureclipsinae; rRNA places iteither with Xiphocentronidae (Fig. 5) orwith Wormaldia plus Dipseudopsidae (Fig.6), whereas COI places it (Antilopsyche)with the polycentropodids and the ecnomids(Fig. 9). We suspect that the grouping of Cyr-nus (Polycentropodidae) and the ecnomidswith Xiphocentronidae in the EF-1® analy-sis (Fig. 8) was due to nucleotide composi-tional similarity and that the EF-1® playeda relatively large role in the combined nu-cleotide analysis (Fig. 10); this suspicion iscorroborated by other data (see Figs. 2, 5, 6,8C, 9B, and 11, all of which place Xiphocen-tronidae with Psychomyiidae). Within theIntegripalpia, we �nd strong support frommultiple datasets for the division of thegroup into two infraorders: Plenitentoria andBrevitentoria, essentially as it has been di-vided, except that Pangullia (Kokiriidae) isnow placed in the Plenitentoria. From ourreview of the data, the inclusion of Pan-gullia (Kokiriidae) is probably not an error.We checked the voucher specimen for ac-curate identi�cation, and multiple molecu-lar markers support placing Pangullia withinPlenitentoria. However, the addition of an-other kokiriid taxon, particularly Kokiria, thenominate genus, would add strength to thisargument. Relationships within Plenitento-ria, however, are extremely weakly sup-ported, with almost no agreement amongalternative analyses, except for the group-ing of Limnephilidae sensu latu: Uenoidae,Goeridae, Apataniidae, and Limnephilidae,with Apataniidae as the sister taxon to amonophyletic Limnephilidae s.s. The mono-phyly of Leptoceroidea was not supported inmost analyses, but the nodes that contradictthe monophyly of this clade are extremelyweak, and our Bayesian likelihood (Fig. 11)



Dow

nloaded from



analysis does recover Leptoceroidea at 78%.Further analysis will be needed to evalu-ate relationships among trichopteran fami-lies, and we are working on expanding thedataset to address these questions.

Empirical Observations from the Analyses

“Saturation,” as generally discussed, isa distance-based concept, with saturationcurves derived from the pairwise compari-son of sequences. For character-based phy-logeny methods (e.g., parsimony or likeli-hood), “saturated” data may still providemeaningful phylogenetic information, evenwhen sequences no longer show an in-crease in distance with increasing diver-gence. This is because homoplasy, whenoccurring in different parts of the tree, canbe ef�ciently isolated with extensive taxonsampling (Swofford et al., 1996). Empiri-cal evidence (Hillis, 1998) and simulationstudies (Graybeal, 1998; Yang and Goldman,1997) show that adding taxa can greatly im-prove the accuracy of character-based meth-ods. Broughton et al. (2000) show that the re-moval of presumably noisy partitions (thirdpositions or transitions) is inadvisable be-cause even when homoplastic, these charac-tersmay not be problematic, given increasingpatristic distances, and these noisy partitionstend to contribute a large number of charac-ters to the analysis. Despite our utilizationof saturation curves in this paper, we agreewith results that indicate saturation can becompensated for with taxon sampling. How-ever, we caution against an overextention ofthese �ndings to the assumption that exten-sive taxon sampling will always lead to ac-curate phylogenetic hypotheses, especially atthe deepest nodes. Empirical examples can-not constitute proof that, for every set of taxa,sampling can always �x the problem. Havingunambiguous signal is preferable to relyingon the chance that the right taxa still exist tobe sampled and that homoplasy will be dis-tributed in the right places. The simulationsmentioned (e.g., Yang and Goldman, 1997;Graybeal, 1998) were done with modelparameters set constant across lineages,demonstrating that given the model, taxonsampling can help. However, the most se-rious problems in phylogenetic analyses ofbiological data occur when noise is not ho-mogeneously distributed across taxa, andsimulations may not address the effects

of nonrandom noise on biological datasets.With “good” data, we expect a left skew tothe tree distributions from �ve taxa; if thereis not, then at best, we have good reason tohope that taxon sampling will sort out the ho-moplasy. At worst, we are left with four con-�icting datasets with no way to rank them.

Some criterion had to be set for datasetcombinations of analysis and inclusion forpublication. A simple option would be thatof a single “total evidence” analysis (Kluge,1989), and we agree that this is prob-ably the most objective approach. How-ever, our goal was not to maximize objec-tivity at all costs, but rather to estimatephylogeny as objectively as we could with-out sacri�cing our ability to make deci-sions, perhaps even subjective ones, basedon observation. Although we combined alldata, it would be a misinterpretation to as-sume that our analyses of saturation sup-ports always combining data. Rather, we feltthat we could rank our hypotheses, evenif only qualitatively, and sort out con�ictfrom noise. Now that “tests” used to indi-cate whether or not data should be com-bined (e.g., Templeton, 1983; ILD tests, Far-ris et al., 1995) have come under criticism(Dolphin et al., 2000; Goldman et al., 2000;Yoder et al., 2001), we feel it is increasinglyimportant to examine data in a variety ofways and with respect for organismal exper-tise. We would have felt justi�ed avoiding theaddition of EF-1® or COI, simply because ourmeasures of excessive homoplasy in thesemarkers, but we were not forced to make thatdecision. In the end, we did combine all ofthe data, and the addition of even our noisi-est data resulted in the hypotheses of subor-dinal relationships that we consider our bestestimate (Figs. 10 and 11). We found we couldinclude all of our data in a combined analysiswithout overturning the results gained fromour most conservative marker, the rRNA.

The combination of approaches for eval-uating signal offered a consistent picture ofwhether or not signal was located primar-ily within suborders or among higher clades.Figures 3 and 4 can be examined with ref-erence to the numbers below the internodesin Figure 11, indicating which dataset wasmost responsible for the recovery of speci�cnodes. The g1 statistics and the recovery ofexpected clades in the �ve- to eight- taxonanalysis all gave the same answers, and Fig-ures 3 and 4 agree with the information from



Dow

nloaded from



Table 1. Since Hillis and Huelsenbeck (1992)showed that signi�cant signal can come fromany single clade in a dataset, even when sig-nal is randomized among all other taxa, we�nd the g1 statistic to be of little value inproving signal exists but meaningful in cast-ing suspicion on a dataset where signal mayhave broken down. This becomes importantwhen attempting to reconcile “con�icting”hypotheses from different datasets becausewithout some means of �ltering the data,one would often either have to collapse abest estimate to an unresolved strict consen-sus or allow saturated data to determine thetree.

That the COI seemed to have outper-formed the EF-1® within the Annulipalpia,as judged by the combined analysis, seemsincongruous with the results shown in Fig-ures 3 and 4 and Table 1. This outcome couldhave simply been due to greater taxon sam-pling in COI, but we also note that the ac-cumulation of differences in EF-1® aminoacids is surprisingly �at throughout the tree,perhaps indicating functional constraints.Amino acid sequences that are highly con-strained could be saturated at even very lowmean percent differences if the few posi-tions permitted to vary actually toggle freelyamong a few similar amino acids (Simonet al., 1994). If we accept that nonconstrainednucleotides become relatively rapidly homo-plastic in any sequence, then deeper-levelsignal in any protein-coding dataset is al-ways going to occur against a backgroundof inherently noisy silent sites (mostly third-codon positions; but see Broughton et al.,2000). The relevant issue then, in the searchfor appropriate protein-coding genes for re-covering ancient relationships, is the propor-tion of amino acids free to vary; EF-1® siteswere estimated to have 51% of its sites invari-ant. Perhaps counterintuitively, to be usefulin a phylogenetic analysis at the level weare considering, a protein would need aneven more rapid rate of amino acid diver-gence than that observed in the EF-1® gene,to give conservative, nonsilent substitutionsa chance to accumulate. In other words,a “conservative-looking” gene, with highlyconstrained �rst- and second-codon posi-tions, may be less appropriate for deep-levelphylogenetic analysis than a “noisy-lookinggene,” the underlying signal of which isfound in the variation occurring at nonsi-lent codon positions, given the likelihood

that the majority of third positions in bothgenes provide noise. Additionally, even theamino acid substitutions that do occur in atoo-conservative gene may contribute noiseif the positions free to vary alternate betweentwo or three structurally similar amino acidsfor which the precise character state is evo-lutionarily unconstrained, or nearly so.

We believe that the coding of insertionsand deletions is important. Swofford (1993)recommended that “indels” be treated asmissing data in the nucleotide sequenceblock and then coded in a presence/absencematrix at the end of the data �le; Kjer (1995),Crandall (1996), and others have followedthis advice. Wheeler (1999) and Lutzoni et al.(2000) proposed coding indels separately(even when they could not be aligned) asmultistate characters and then running thesecharacters through a step matrix. The ruleswe used to evaluate insertions and dele-tions shown in Appendix 1 are discussedat length in the Systematic Biology website.The widely used term “indel” may mistak-enly imply an inherent dif�culty in distin-guishing between insertions and deletionsin molecular data or at least indicate thata distinction between insertions and dele-tions is not important enough to use a sep-arate term for these different events. Wefound that the polarity for most insertionsand deletions could be unambiguously de-termined by the traditional examination ofsuccessive outgroups, as widely practiced bymorphological systematists. Our coding ofinsertions, deletions, and “indels” was ad-mittedly elaborate, but no more so than thecharacter coding performed by morpholo-gists. In our opinion, these characters shouldeach be treated individually and describedto the best ability of the investigator. Becauseof the lack of models for the evolution ofinsertions and deletions at the time of thisanalysis, these characters were left uncodedin the likelihood analysis (see Lewis, 2001).Because some of our analyses were con-ducted without the insertion/deletion char-acters, those uncomfortable with the com-plexity of the coding (or more comfortablewith an explicitly model-based analysis) will�nd likelihood trees that still place Integri-palpia and Spicipalpia together. Complexcoding of insertionsand deletionshas the dis-advantage that one cannot compare the per-formance of likelihood to that of parsimony,because the datasets are different. Although



Dow

nloaded from



we agree that likelihood remains a statis-tically superior criterion in many respects,we believe that parsimony’s advantage hasbeen its ability to combine data from multi-ple sources, such as morphology, insertions,deletions, and molecules evolving under ex-ceedingly different conditions.

Pseudoreplicate reweighting performedwell in resolving many more nodes thandid the equally weighted analysis (note thenodes with a decay index of zero in Fig. 11);moreover, enough nodes differed among theequally and differentially weighted analy-ses to reassure that the procedure was notentirely circular. However, at least withinAnnulipalpia, we prefer the results fromthe equally weighted analysis. With themorphology, pseudoreplicate reweightingyielded hypothesized relationships amongAnnulipalpia, Integripalpia, and the fourspicipalpian families that were identical toour combined analysis (Fig. 11); the charac-ters responsible for this change are apparenton Figure 7.

We present a rRNA dataset that seemedto be evolving at an appropriate rate for thesuborder study, a previously published mor-phological dataset that also resolved subor-ders but in an alternative way when datawere equally weighted, and two datasets,the COI and the EF-1®, that taken alone,were practically useless for estimating rela-tionships among suborders. Bootstrap val-ues and decay indices from individual uncor-rected analyses were weak, but it is expectedthat different sources of data to provide in-formation at different levels in the tree. Cor-roboration from independent sources comesonly from differentially weighted parsimonyof the morphological dataand from the EF-1®amino acids. However, these datasets, alongwith the rRNA, are the most slowly evolvingdatasets. We do not �nd “con�ict,” amongtrees, but rather �nd their differences to beeasily explained by the properties of the data.

We are not entirely satis�ed with any ofthe trees generated from single datasets. De-spite the problems with the COI and the EF-1®, no informed trichopterist would take therelationships among annulipalpian familiesgenerated from the rRNA data (Figs. 5 and6) seriously. The COI data provides a rea-sonable estimate of relationships within An-nulipalpia (see all the 4’s showing agreementwith the COI data in Fig. 11). And we haveconverged with some level of con�dence on

the relationships among taxa in the “back-bone” of the trees presented in Figures 10and 11. Although we have presented manytrees, all except Figures 7, 10, and 11 sim-ply represent explorations of the data. Thereis a danger in presenting such a wide va-riety of analyses, selecting to present someand ignore others. The alternatives wouldbe to perform and present either hundredsof arbitrary trees under every conceivablemethod, or to limit the analyses to a sin-gle optimality criterion, subjectively favoredon the basis of philosophy (e.g., parsimonyor likelihood). The guide in presentationmust be the data, rather than a favored hy-pothesis, although the data did lead us toa favored hypothesis. Subjectively, we �ndFigure 11 to represent our best current es-timate of phylogenetic relationships amongTrichoptera suborders and major family-level taxa. Although our preferred hypothe-sis comes from equally weighted parsimony,this statement should not be mis-cited as atriumph for parsimony, in view of its perfor-mance in estimating relationships with thenucleotides from COI and EF-1®. To eval-uate this performance, note that relativelyfew nodes in the combined analysis were re-covered from either of these noisy markers(Fig. 11; methods 3 and 4). Both parsimonyand likelihood analyses complemented oneanother. The agreement between Figures 10and 11 represents a more conservative esti-mate of phylogeny, with the differences be-tween them dependent on the relative con-tributions of the different datasets: Figure 11favors the morphological characters withrespect to the Spicipalpia and the Philopota-moidea, while Figure 10 (lacking morphol-ogy, gaps, and amino acids) looks morelike the RNA tree with respect to the spic-ipalpians, with the COI and EF-1® play-ing a larger role in dictating relationshipswithin Annulipalpia. We are currently in-volved in collecting additional data to esti-mate the phylogenetic relationships amongfamilies within Annulipalpia, Plenitentoria,Leptoceroidea, and Sericostomatoidea. It isour prediction that the nodes shared be-tween Figures 10 and 11 will not change withthe ongoing analyses. It was not our inten-tion to prove that “saturated” data shouldnever (or always) be used; rather, our anal-yses of homoplasy guided us in makinganalytical decisions that led to a crediblehypothesis.



Dow

nloaded from



ACKNOWLEDGMENTS

We are especially grateful to Dr. Oliver S. Flint,Jr. (Smithsonian Institution, Washington, D.C.), Dr.Alice Wells (Australian Biological Resources Study, Can-berra), Dr. Arturs Neboiss (Museum of Victoria, Abbots-ford, Victoria, Australia), Dr. Ferdy de Moor (AlbanyMuseum, Grahamstown, South Africa), Dr. AndrewNimmo (University of Alberta, Edmonton, Canada), andJames Bower for specimens used in the study. Lepi-dopteran DNA samples were provided byDr. Tim Fried-lander and Dr. David Wagner (University of Connecti-cut, Storrs). Thanks to Catherine Duckett (Universityof Puerto Rico); Jack Sites, Jr. (Brigham Young Univer-sity, Provo); Mike May, Frank Carle, and John LaPolla(Rutgers University); and Susan Weller (University ofMinnesota) for helpful comments on the manuscript.Richard Olmstead, Keith Crandall, and two anonymousreviewers made helpful comments that improved themanuscript. Thanks to Jim Archie (University of Cali-fornia, Long Beach) for helpful discussions. This workwas supported by National Science Foundation grantsDEB 9796097 (K.M.K., R.W.H.), DEB 9191091 (J. Sites,Jr., O. Flores, J. Archie), and DEB 9974081 (C. Duckett,K.M.K.) and by the New Jersey Agricultural ExperimentStation.

REFERENCES

ADAMS, E. N. 1972. Consensus techniques and the com-parison of taxonomic trees. Syst. Zool. 21:390–397.

AKAIKE, H. 1974. A new look at the statistical modelidenti�cation. IEEE Trans. Autom. Control 19:716–723.

AGOSTI, D., D. JACOBS, AND R. DESALLE. 1996. On com-bining protein sequences and nucleic acid sequencesin phylogenetic analysis: The homeobox protein case.Cladistics 12:65–82.

ALSTAD, D. N. 1982. Current speed and �ltration ratelink caddis �y phylogeny and distributional patternson a stream gradient. Science 216:533–534.

ARCHIE, J. W. 1989. Homoplasy excess ratios: New in-dices for measuring levels of homoplasy in phyloge-netic systematics and a critique of the consistency in-dex. Syst. Zool. 38:253–269.

BALLARD, J. W. O., M. K. THAYER, A. F. NEWTON, JR.AND E. R. GRISMER. 1998. Data sets, partitions, andcharacters: philosophies and procedures for analyzingmultiple datasets. Syst. Biol. 47:367–396.

BARRETT, M., M. J. DONAGHUE, AND E. SOGER. 1991.Against consensus. Syst. Zool. 40:486–493.

BENABIB, M., K. M. KJER, AND J. W. SITES , Jr. 1997. Mi-tochondrial DNA sequence-based phylogeny and theevolution of viviparity in the Sceloporus scalaris group.Evolution 51:1262–1275.

BETTEN, C. 1934. The caddis �ies or Trichoptera of NewYork State. N. Y. State Mus. Bull. 292:1–576.

BREMER , K. 1988. The limits of amino acid sequencedata in angiosperm phylogenetic reconstruction. Evo-lution 42:795–803.

BROUGHTON, R. E., S. E. STANLEY, AND R. T. DURRET.2000. Quanti�cations of Homoplasy for nucleotidetransitions and transversions and a reexamination ofassumptions in weighted phylogenetic analysis. Syst.Biol. 49:617–627.

BUCKLEY, T. R., C. SIMON, AND G. K. CHAMBERS. 2001.Exploring among-site rate variation models in a max-

imum likelihood framework using empirical data: Ef-fects of model assumptions on estimates of topol-ogy, branch lengths, and bootstrap support. Syst. Biol.50:67–86.

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.

CHALWATZIS, N., J. HAUF, Y. VAN DE PEER, R.KINZELBACH, AND F. K. ZIMMERMANN. 1996. 18Sribosomal RNA genes of insects: Primary structure ofthe genes and molecular phylogeny of Holometabola.Ann. Entomol. Soc. Am. 89:775–787.

COLWELL, R. K. 1996. Biota: The biodiversity data-base manager. Sinauer Associates, Sunderland,Massachusetts.

CRANDALL, K. A., AND J. F. FITZPATRICK, JR. 1996. Cray-�sh molecular systematics: Using a combination ofprocedures to estimate phylogeny. Syst. Biol. 45:1–26.

CUNNINGHAM , C. W. 1997. Is congruence between datapartitions a reliable predictor of phylogenetic accu-racy? Syst. Biol. 46:464–478.

DE MOOR, F. C. 1993. A cladistic analysis of charac-ter states in twelve families here considered as be-longing to the Sericostomatoidea. Ann. Cape Prov.Mus. Nat. Hist. 18:347–352. [In K. M. F. Scott. 1993.Three recently erected Trichoptera families fromSouth Africa, the Hydrosalpingidae, Petrothrincidaeand Barbarochthonidae (Integripalpia: Sericostoma-toidea). Ann. Cape Prov. Mus. Nat. Hist. 18:293–354.]

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

DE QUEIROZ, A., M. J. DONOGHUE, AND J. KIM. 1995. Sep-arate vs. combined analysis of phylogenetic evidence.Annu. Rev. Ecol. Syst. 26:657–681.

DOLPHIN, K., R. BELSHAW, C. D. L. ORME, AND D. L. J.QUICKE. 2000. Noise and incongruence: Interpretingresults of the incongruence length difference test. Mol.Phylogenet. Evol. 17:401–406.

DONOGHUE, M. J., R. G. OLMSTEAD, J. F. SMITH, ANDJ. D. PALMER. 1992. Phylogenetic relationships of Dip-sacales based on rbcl sequences. Ann. Mo. Bot. Gard.79:333–345.

EERNISSE, D. J., AND A. G. KLUGE. 1993. Taxonomiccongruence versus total evidence, and amniote phy-logeny inferred from fossils, molecules, and morphol-ogy. Mol. Biol. Evol. 10:1170–1195.

FARRIS, J. S. 1969. A successive approximations ap-proach to character weighting. Syst. Zool. 18:374–385.

FARRIS, J. S., M. KALLERSJO, A. G. KLUGE, AND C. BULT.1995. Testing signi�cance of incongruence. Cladistics10:315–319.

FELSENSTEIN, J. 1985. Con�dence limits on phylogenies:An approach using the bootstrap. Evolution 39:783–791.

FLORES-VILLELA, O., K. M. KJER, M. BENABIB, and J. W.SITES , JR. 2000. Multiple data sets, congruence, andhypothesis testing for the phylogeny of the basalgroups of the lizard genus Sceloporus (squamata,phrynosomatidae). Syst. Biol. 49:713–739.

FRANIA, H. E., AND G. B. WIGGINS . 1997. Analysisof morphological and behavioral evidence for thephylogeny and higher classi�cation of Trichoptera(Insecta). R. Ont. Mus. Life Sci. Contrib. 160:1–67.

GALL, W. K. 1994. Phylogenetic studies in the Lim-nephiloidea, with a revision of the world genera ofGoeridae (Trichoptera). Ph.D. Thesis. Univ. of Toronto,Toronto. 140 pp.



Dow

nloaded from



GALL, W. K. 1997. Biogeographic and ecologic relation-ships in the Plenitentoria (Trichoptera). Pages 109–116in Proceedings of the 8th International Symposium onTrichoptera (R. W. Holzenthal and O. S. Flint, Jr., eds.).Ohio Biological Survey, Columbus.

GOLDMAN, N., J. P. ANDERSON, AND A. G. RODRIGO.2000. Likelihood-based tests of topologies in phylo-genetics. Syst. Biol. 49:652–670.

GRAYBEAL, A. 1998. Is it better to add taxa or charactersto adif�cult phylogeneticproblem? Syst. Biol. 47:9–17.

GU, X., Y.-X. FU, AND W.-H. LI. 1995. Maximum likeli-hood estimation of the heterogeneity of substitutionrate among nucleotide sites. Mol. Biol. Evol. 12:546–557.

GUTELL, R. R., N. LARSEN, AND C. R. WOESE. 1994.Lessons from an evolving rRNA: 16S and 23S rRNAstructures from a comparative perspective. Microbiol.Rev. 58:10–26.

HASEGAWA, M., AND H. KISHINO . 1989. Con�dence lim-its on the maximum-likelihood estimate of the ho-minid tree from mitochondrial-DNA sequences. Evo-lution 43:672–677.

HENNIG , W. 1981. Insect phylogeny (translated by A. C.Pont and D. Schlee). Wiley, New York.

HILLIS , D. M. 1998. Taxonomic sampling, phylogeneticaccuracy, and investigator bias. Syst. Biol. 47:3–8.

HILLIS , D. M., AND S. K. DAVIS. 1986. Evolution of ribo-somal DNA: Fifty million years of recorded history inthe frog genus Rana. Evolution 40:1275–1288.

HILLIS , D. M., AND J. HUELSENBECK. 1992. Signal, noiseand reliability in molecular phylogenetic analysis.J. Hered. 83:189–195.

HUELSENBECK, J. P. 2000. MRBAYES: Bayesian inferenceof phylogeny. Distributed by the author. Departmentof Biology, Univ. of Rochester, New York.

HUELSENBECK, J. P., J. J. BULL, AND C. W. CUNNING HAM .1996. Combining data in phylogenetic analysis.Trends Ecol. Evol. 11:152–158.

IVANOV, V. D. 1997. Rhyacophiloidea, a paraphyletictaxon. Pages 189–193 in Proceedings of the 8th Inter-national Symposium on Trichoptera (R. W. Holzenthaland O. S. Flint, Jr., eds.). Ohio Biological Survey,Columbus.

KISHINO , H., AND M. HASAGAWA. 1989. Evaluation ofmaximum likelihood estimate of the evolutionary treetopologies from DNA sequence data, and the branch-ing order in the Hominoidea. J. Mol. Evol. 29:170–179.

KJER, K. M. 1995. Use of rRNA secondary structurein phylogenetic studies to identify homologous po-sitions: An example of alignment and data presen-tation from the frogs. Mol. Phylogenet. Evol. 4:314–330.

KJER, K. M. 1997. An alignment template for amphibian12S rRNA, domain III: Conserved primary and sec-ondary structural motifs. J. Herpetol. 31:599–604.

KJER, K. M., G. D. BALDRIDGE, AND A. M. FALLON. 1994.Mosquito large subunit ribosomal RNA: Simultane-ous alignment of primary and secondary structure.Biochim. Biophys. Acta 1217:147–155.

KJER, K. M., R. J. BLAHNIK , AND R. W. HOLZENTHAL.Phylogeny of caddis�ies (Insecta, Trichoptera). Zool.Scripta (in press).

KLUGE, A. G. 1989. A concern for evidence and a phylo-genetic hypothesis of relationships among Epicrates(Boidae, Serpentes). Syst. Zool. 38:7–25.

KRISTENSEN, N. P. 1991. Phylogeny of extant hexapods.Pages 125–140 in Insects of Australia, 2nd edition,

vol. 1. I. D. Nielsen, J. P. Spradberry, R. W. Taylor, M. J.Whitten, and M. J. Littlejohn, eds.

KRISTENSEN, N. P. 1997. Early evolution of the Lep-idoptera C Trichoptera lineage: Phylogeny and theevolutionary scenario. Mem. Mus. Natl. Hist. Nat. Ser.A 173:253–271.

LARSEN, N. 1992. Higher order interactions of 23S rRNA.Proc. Natl. Acad. Sci. USA 89:5044–5048.

LARSON, A. 1994. The comparison of morphological andmolecular data in phylogenetic systematics. Pages371–390 in Molecular approaches to ecology and evo-lution (B. Schierwater, B. Streit, G. P. Wagner, andR. De Salle, eds.). Birkhauser Verlag, Basel.

LEWIS , P. O. 2001. A likelihood approach to estimat-ing phylogeny from discrete morphological characterdata. Syst. Biol. 50:910–922.

LI, Y. J., J. C. MORSE, AND H.TACHET. 2001. Pseudoneure-clipsinae in Dipseudopsidae (Trichoptera: Hydropsy-choidea), with descriptions of two new species ofPseudoneureclipsis from East Asia. Aquat. Insects23:107–117.

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

LUTZONI, F., P. WAGNER, AND V. REEB. 2000. Integratingambiguously aligned regions of DNA sequences inphylogenetic analyses using unequivocal coding andoptimal character-state weighting. Syst. Biol. 49:628–651.

MACKAY, R. J., AND G. B. WIGGINS . 1979. Ecologicaldiversity in the Trichoptera. Annu. Rev. Entomol.24:185–208.

MADDISON, D. R., AND W. P. MADDISON. 2000.MacClade 4. computer program. Sinauer Associates,Sunderland, Massachusetts.

MARTYNOV, A. V. 1924. Rucheiniki (caddis�ies[Trichoptera]). Pages iv+384 in Prakticheskaya en-tomologiya, volume 5 (N. N. Bogdanova-Kat’kova,ed.). Gosudarstvennoe Izdatelstvo, Leningrad.

MILNE, M. J., AND L. J. MILNE. 1939. Evolutionary trendsin caddis worm cases construction. Ann. Entomol.Soc. Am. 32:533–542.

MIYAMOTO, M. M. 1985. Consensus cladograms andgeneral classi�cations. Cladistics 1:186–189.

MORITZ, C. C., J. SCHNEIDER, AND W. B. WAKE. 1992.Evolutionary relationships within the Ensatina esch-schultzii complex con�rm the ring species interpreta-tion. Syst. Biol. 41:273–291.

MORSE, J. C.1997a.Checklistof world Trichoptera. Pages339–342 in Proceedings of the 8th International Sym-posium on Trichoptera (R. W. Holzenthal and O. S.Flint, Jr., eds.). Ohio Biological Survey, Columbus.

MORSE, J. C. 1997b. Phylogeny of Trichoptera. Annu.Rev. Entomol. 42:427–450.

NOVOKSHOVOV, V., AND I. D. SUKATCHEVA. 1997. Earlyevolution of caddis�ies. Pages 95–99 in Proceedingsof the 7th International Symposium on Trichoptera(C. Otto, ed.). Backhuys, Leiden.

PASHLEY, D. P., B. A. MCPHERON, AND E. A. ZIMMER.1993. Systematics of holometabolous insect ordersbased on 18S ribosomal RNA. Mol. Phylogenet. Evol.2:132–142.

POSADA, D., AND K. A. CRANDALL. 1998. Modeltest:Testing the model of DNA substitution. Bioinformat-ics 14:817–818.

RESH, V. H., AND D. M. ROSENBERG (eds.). 1984. Theecology of aquatic insects. Praeger, New York.



Dow

nloaded from



ROSENBERG, D. M., AND V. H. RESH (eds.). 1993. Fresh-water biomonitoring and benthic macroinvertebrates.Chapman & Hall, New York.

ROSS, H. H. 1956. Evolution and classi�cation of themountain caddis�ies. Univ. of Illinois Press, Urbana.

ROSS, H. H. 1964. Evolution of caddis worm cases andnets. Am. Zool. 4:209–220.

ROSS, H. H. 1967. The evolution and past dispersal ofthe Trichoptera. Annu. Rev. Entomol. 12:169–206.

SAMBROOK, J., E. F. FRITSCH, AND T. MANIATUS. 1989.Molecular cloning: A laboratory manual, 2nd edition.Cold Spring Harbor Laboratory Press, Cold SpringHarbor, New York.

SHIELDS , O. 1988. Mesozoic history and neontology ofLepidoptera in relation to Trichoptera, Mecoptera,and Angiosperms. J. Paleontol. 62:251–258.

SIMON, C., F. FRATI, A. BECKENBACH, B. CRESPI, H. LIU,AND P. FLOOK. 1994. Evolution, weighting, and phy-logenetic utility of mitochondrial gene sequences anda compilation of conserved polymerase chain reactionprimers. Ann. Entomol. Soc. Am. 87:651–701.

SPEYER, A. 1870. Zur Genealogie der Schmetterlinge. En-tomol. Z. 1870:202–223.

SPRINGER, M. S., H. M. AMRINE, A. BURK, AND M. J.STANHOPE. 1999. Additional support for Afrotheriaand Paenungulata, the performance of mitochondrialversus nuclear genes, and the impact of data par-titions with heterogeneous base compositions. Syst.Biol. 48:65–75.

SUKATCHEVA, I. D. 1980. Evolution of the caddis�y (Tri-choptera) larval case construction. Zh. Obshch. Biol.41:457–469 [in Russian].

SULLIVAN, J. 1996. Combining data with different distri-butionsof among-site ratevariation. Syst. Biol. 45:375–380.

SWOFFORD, D. L. 1993. PAUP 3: Phylogenetic Analy-sis Using Parsimony. User manual. Univ. of Illinois,Urbana.

SWOFFORD, D. L. 1999. PAUP*: Phylogenetic Analy-sis Using Parsimony *and other methods, Version 4.Sinauer Associates, Sunderland, Massachusetts.

SWOFFORD, D. L., G. J. OLSEN, P. J. WADDELL, AND D. M.HILLIS . 1996. Phylogenetic inference. Pages 407–514in Molecular systematics (D. M. Hillis, C. Moritz, andB. K. Mable, eds.). Sinauer Associates, Sunderland,Massachusetts.

TEMPLETON, A. 1983. Phylogenetic inference from re-striction endonuclease cleavage site maps with par-ticular reference to the evolution of humans and apes.Evolution 37:221–244.

THORP, J. H. 1983. An evaluation of hypotheses onthe evolutionary differentiation of catchnets in net-spinning caddis�ies (Hydropsychidae). Oikos 40:308–312.

THORP, J. H., J. B. WALLACE, AND T. J. GEORGIAN, JR.1986. Untangling the web of caddis�y evolution anddistribution. Oikos 47:253–256.

ULMER, G. 1912. Die Trichopteren des Baltischen Bern-stein. Beitrage Naturkunde Preussens, volume 10.Leipzig.

VAN DE PEER, J., M. NEEFS, P. DERIJK, AND R.DEWACHTER. 1993. Reconstructing evolution fromeukaryotic small-subunit RNA sequences: Calibra-tion of the molecular clock. J. Mol. Evol. 37:221–232.

WEAVER, J. S., III. 1983. The evolution and classi�ca-tion of Trichoptera, with a revision of the Lepidos-tomatidae and a North American synopsis of this

family. Ph.D. Thesis. Clemson Univ., Clemson, SouthCarolina.

WEAVER, J. S., III. 1984. The evolution and classi�cationof Trichoptera, Part 1: the groundplan of Trichoptera.Pages 413–419 in Proceedings of the 4th InternationalSymposium on Trichoptera (J. C. Morse, ed.). Junk,The Hague.

WEAVER0, J. S., III. 1992a. Remarks on the evolution ofTrichoptera: A critique of Wiggins and Wichard’s clas-si�cation. Cladistics 8:171–180.

WEAVER, J. S., III. 1992b. Further remarks on the evo-lution of Trichoptera: A reply to Wiggins. Cladistics8:187–190.

WEAVER, J. S., III, AND H. MALICKY. 1994. The genusDipseudopsis Walker from Asia (Trichoptera: Dipseu-dopsidae). Tijdschr. Entomol. 137:95–142.

WEAVER, J. W., III, AND J. C. MORSE. 1986. Evolution offeeding and case-making behavior in Trichoptera. J.North Am. Benthol. Soc. 5:150–158.

WHEELER, W. 1999. Fixed character states and the op-timization of molecular sequence data. Cladistics15:379–385.

WHITING , M. F., J. C. CARPENTER, Q. D. WHEELER, ANDW. C. WHEELER. 1997. The Strepsiptera problem: Phy-logeny of the holometabolous insect orders inferredfrom 18S and 28S ribosomal DNA sequences and mor-phology. Syst. Biol. 46:1–68.

WICHARD, W. 1991. The evolutionary effect of over-coming osmosis in Trichoptera. Pages 337–342 in Pro-ceedings of the 6th International Symposium on Tri-choptera (C. Tomaszewski, ed.). Adam MickiewiczUniv. Press, Poznan.

WICHARD, W., AND H. P. KLEIN. 1997. Pupal cocoonof Amphiesmenoptera (Trichoptera C Lepidoptera)with evolutionary considerations of the Trichoptera.Pages 475–480 in Proceedings of the 8th Interna-tional Symposium on Trichoptera (R. W. Holzenthaland O. S. Flint, Jr., eds.). Ohio Biological Survey,Columbus.

WICHARD, W., H. H. SCHMIDT, AND R. WAGNER. 1993.The semipermeability of the pupal cocoon of Rhy-acophyla (Trichoptera: Spicipalpia) . Pages 25–27 inProceedings of the 7th International Symposium onTrichoptera (C. Otto, ed.). Backhuys Publishers,Leiden, the Netherlands.

WIENS , J. 1998. Combining data sets with dif-ferent phylogenetic histories. Syst. Biol. 47:568–581.

WIGGINS , G. B. 1992. Comments on the phylogeny of pu-pation behavior in Trichoptera: A response to Weaver.Cladistics 8:181–185.

WIGGINS , G. B. 1996. Larvae of the North Americancaddis�y genera (Trichoptera), 2nd edition. Univ. ofToronto Press, Toronto.

WIGGINS , G. B., AND W. WICHARD. 1989. Phylogeny ofpupation in Trichoptera, with proposals on the originand higher classi�cation of the order. J. North Am.Benthol. Soc. 8:260–276.

WILLIAMS , D. D., A. F. TAVARES, AND E. BRYANT. 1987.Respiratory device or camou�age? —A case for thecaddis�y. Oikos 50:42–52.

YANG, Z. 1993. Maximum likelihood estimation of phy-logeny from DNA sequences when substitution ratesdiffer over sites. Mol. Biol. Evol. 10:1396–1401.

YANG, Z. 1994a. Estimating the pattern of nucleotidesubstitution. J. Mol. Evol. 39:105–111.

YANG, Z. 1994b. Maximum likelihood phylogeneticestimation from DNA sequences with variable



Dow

nloaded from



rates over sites: Approximate methods. Syst. Biol.44:384–399.

YANG, Z. 1996. Among-site rate variation and its impacton phylogenetic analysis. TREE 11:367–372.

YANG, Z., AND N. GOLDMAN. 1997. Are big trees indeedeasy? TREE 12:357.

YODER, A. D., J. A. IRWIN, AND B. A. PAYSEUR. 2001.Failure of the ILD to determine data combinabilityfor slow loris phylogeny. Syst. Biol. 50:

Received 23 August 2000; accepted 7 November 2000Associate Editor:



Dow

nloaded from



APPENDIX 1. ALIGNMENT OF SELECTEDTAXA

Secondary structural symbols are presented as in Kjeret al. (1994). Nucleotides �anked by straight lines havebeen excluded from the analysis, as described in the “ex-

APPENDIX 1.

planation of Appendix 1” on the Systematic Biology web-site. Numerals above the blocks number the stems as inLarsen (1992) for the large subunit rRNA and in van dePeer et al. (1993) for the small subunit rRNA. Numeralsbelow the data blocks refer to alignment sites discussedon the Systematic Biology website.



Dow

nloaded from



APPENDIX 1. CONTINUED.



Dow

nloaded from



APPENDIX 1. CONTINUED.



Dow

nloaded from


Date post:	25-Apr-2018
Category:	Documents
Upload:	vandan
View:	220 times
Download:	2 times

Phylogeny of Trichoptera (Caddisflies): … phylogenetic agreement has been reached about the...

Documents