Embed Size (px)
Citation preview
2003 POINTS OF VIEW 853
Syst. Biol. 52(6):853–864, 2003Copyright c© Society of Systematic BiologistsISSN: 1063-5157 print / 1076-836X onlineDOI: 10.1080/10635150390252279
Are the Dental Data Really at Odds with the Molecular Data? Morphological Evidencefor Whale Phylogeny (Re)Reexamined
MAUREEN A. O’LEARY,1 JOHN GATESY,2 AND MICHAEL J. NOVACEK3
1Department of Anatomical Sciences, Stony Brook University, Stony Brook, New York 11794-8081, USA; E-mail: [email protected] of Biology, Spieth Hall, University of California, Riverside, California 92521, USA; E-mail: [email protected]
3Division of Paleontology, American Museum of Natural History, 79th Street at Central Park West, New York, New York 10024-5192, USA;E-mail: [email protected]
Paraxonia, the clade that includes cetaceans (whales)and artiodactylans (even-toed hoofed mammals), is ex-ceptionally well documented for diverse character sys-tems (reviewed by Gatesy and O’Leary, 2001). Given thiswealth of phylogenetic data, the group has been utilizedas a model system to examine a variety of critical issuesin modern systematics, including taxon sampling (e.g.,Philippe and Douzery, 1994), the importance of fossils(e.g., O’Leary, 2001), supertree analysis (e.g., Liu et al.,2001), supermatrix analysis (e.g., Matthee et al., 2001),and the relative quality of different data sets (e.g., Gatesy,2002). In this paper, we explore this last issue further.
Specifically, we critique an interpretation of the patternof evolution of dental characters presented in a recent re-analysis of morphological data for paraxonians (Naylorand Adams, 2001). Within the context of a broader re-analysis of morphological and molecular data, we arguethat (1) a variety of logical and empirical mistakes weremade by Naylor and Adams (2001), (2) the conclusionsof that study are therefore not supported by the data, and(3) more extensive total evidence/simultaneous analyses(Kluge, 1989; Nixon and Carpenter, 1996a) of morpho-logical, behavioral, and molecular data will be requiredto choose decisively among competing hypotheses ofwhale origins.
NAYLOR AND ADAMS (2001)The matrix of O’Leary and Geisler (1999) included
gross anatomical data from multiple extant families ofparaxonians and enabled direct comparisons of morpho-logical characters with previously published molecularevidence. Simultaneous analyses (Nixon and Carpenter,1996a) of the combined molecular and morphologicaldata yielded poorly resolved topologies, and the sys-tematic position of the wholly extinct clade Mesonychia,traditionally considered among the closest phylogeneticrelatives of Cetacea, was highly unstable (O’Leary, 1999,2001). Gatesy and O’Leary (2001) suggested that addi-tional characters and taxa would be required to sort outthis difficult phylogenetic problem.
In contrast to this total evidence approach, Naylorand Adams (2001) examined the phylogenetic signal inO’Leary and Geisler’s (1999) data set and attempted to re-solve paraxonian phylogeny by the exclusion of differentclasses of systematic data. In an appendix, O’Leary
and Geisler (1999) had arbitrarily divided their morpho-logical matrix into five sets of characters (dental, cranial,basicranial, postcranial, and soft tissue) and warned that“these partitions are . . . simply constructs to organizeinformation and are not meant at present to convey anyknowledge of biologically based character-relatedness”(1999:466). Naylor and Adams, however, used the book-keeping structure of O’Leary and Geisler’s (1999) ap-pendix as classes of phylogenetic evidence and assesseddisagreements among these classes of data using theincongruence length difference (ILD) test of Farris et al.(1995). The dental, basicranial, and postcranial charac-ters showed significant conflicts, but there was no cleardistinction between anomalous and consistent phyloge-netic signals in the morphological matrix. Naylor andAdams (2001) then applied taxonomic congruence; theyanalyzed partitions separately, analyzed the morpholog-ical data set with different partitions excluded, and com-pared the topological results of these individual analysesto those of previously published molecular trees.
Based on this series of analyses, Naylor and Adamsstated that “the dental characters emerged as the mostdifferent from the other partitions” (2001:447) and that“analysis of all morphological data excluding dentalcharacters results in a topology entirely consistent withthat of the molecular data” (2001:448). They then ar-gued that the dental characters offered an “anomaloushierarchical signal in one character partition of the data”(2001:448) and exhibited “nonindependence” (2001:450).They cited descriptive research on tooth development inmammalian model organisms (e.g., rodents) and gener-alizations about adaptive evolution to support their ar-gument that developmental constraints and directionalselection might predispose teeth to evolutionary conver-gence. Naylor and Adams concluded that “—the most-parsimonious reconciliation [of the morphological andmolecular evidence for Paraxonia] is that the dentalsignal is suspect” (2001:453).
We contend that the procedure used by Naylor andAdams (2001) to identify conflicting morphologicaldata was an amalgam of taxonomic congruence (e.g.,Miyamoto and Fitch, 1995), conditional combination(e.g., Bull et al., 1993), and evolutionary taxonomy (e.g.,Mayr, 1969), and we would not advocate its general ap-plication. The conclusions reached by Naylor and Adams(2001) regarding the inconsistency of dental evidence
854 SYSTEMATIC BIOLOGY VOL. 52
were quickly incorporated, without critical reexamina-tion, into subsequent studies (Gingerich et al., 2001;Thewissen et al., 2001; Harris et al., 2003). This acceptanceis problematic because many of Naylor and Adams’s(2001) conclusions are not supported by the data.
DEBATED PHYLOGENETIC POSITIONS FOR MESONYCHIA
Initially taking a taxonomic congruence approach, acomparison of topologies supported by different datasets (see Miyamoto and Fitch, 1995), Naylor and Adams(2001) did not state clearly the spectrum of possible out-comes to the debate about cetacean phylogeny. In the firstparagraph of their paper, they outlined a dichotomy ofsupposedly competing hypotheses (2001:444):
The traditional paleontological view is that an extinct order ofmammals, the Mesonychia, is the sister taxon to Cetacea.. . . Themolecular evidence, by contrast, supports a phylogenetic hypothesisin which Cetacea is nested deeply within the Artiodactyla, implyingthat Artiodactyla is paraphyletic with respect to Cetacea.
In the context of combining data from extinct andextant organisms, these hypotheses, however, shouldnot be viewed as mutually exclusive (Figs. 1a, 1b).Mesonychia could group closest to Cetacea regardless ofwhether Artiodactyla is paraphyletic or monophyletic(Fig. 1c; also see Gatesy et al., 1996; O’Leary, 2001).The real topological incongruence among previousmorphological and molecular analyses of Paraxoniacan be observed very clearly by ignoring the extinctclade Mesonychia and directly contrasting relation-ships among extant artiodactylan families and Cetacea(Figs. 1d, 1e). Morphological data support a mono-phyletic Artiodactyla with a sister group relationshipbetween suine artiodactylans (pigs) and hippopotamidartiodactylans (hippos), but molecular data generallyfavor a paraphyletic Artiodactyla, with a close associ-ation between whales, hippos, and ruminant artiodacty-lans (sheep, mouse deer, and close relatives). Becausemolecular data have not been collected from mesonychi-ans (these animals have been extinct for approximately45 million years) any placement of Mesonychia relativeto extant paraxonian taxa is equally parsimonious formolecular data.
Put another way, disproving a sister taxon relationshipbetween Mesonychia and Cetacea is not tantamount toachieving congruence between molecular and morpho-logical data partitions. This is because the hypothesisthat Mesonychia is the sister taxon of Cetacea is not syn-onymous with the hypothesis that Artiodactyla is mono-phyletic. Naylor and Adams (2001), however, treatedthese two hypotheses somewhat interchangeably (Fig. 1),which has contributed to a variety of misinterpretationsin the remainder of their analysis.
INCONCLUSIVE ILD TESTS
Naylor and Adams (2001) also took a conditionalcombination approach by using statistical tests to ex-pose incongruent classes of data (see Bull et al., 1993;de Queiroz, 1993). They “examined the homogeneity of
FIGURE 1. Some possible phylogenetic relationships among parax-onians. (a) A sister group relationship between Cetacea and Mesony-chia. (b) Paraphyly of Artiodactyla with Cetacea nested within. (c) Atopology consistent with those of tree a and tree b. (d) Relationshipsamong extant paraxonians supported by morphological data fromO’Leary and Geisler (1999; also see Geisler, 2001). Red circles markclades supported by morphological data that conflict with molecu-lar studies (see Gatesy, 1998; Gatesy and O’Leary, 2001, and referencestherein). (e) Relationships among extant paraxonians supported by ap-proximately 1 million base pairs of DNA sequence and insertions oftransposons (Gatesy et al., 2002). Asterisks signify that a taxon is en-tirely extinct. Lineages that connect extant taxa are dark gray; all othertaxa are extinct.
the phylogenetic signal across the five different mor-phological data partitions presented in [O’Leary andGeisler’s (1999)] paper”—(2001:446). As noted above,O’Leary and Geisler (1999) previously had warned thatthese partitions (basicranial, cranial, dental, postcranial,and soft tissue) were just arbitrary bookkeeping con-structs, one way of organizing a list of cladistic char-acters according to anatomical proximity. For example,
2003 POINTS OF VIEW 855
the postcranial partition could just as easily have beendivided by O’Leary and Geisler (1999) into four parti-tions (e.g., forelimb, hind limb, vertebral, and pelvic),and any character in the dental or basicranial partitionscould have been considered part of the cranial partition.If partitions were to be used for anything other thanbookkeeping, it was incumbent upon Naylor and Adams(2001) to provide a phylogenetic argument for their cho-sen partition boundaries, because no such rationale wasused by O’Leary and Geisler (1999). As Siddall (1997),Kluge (1998), and others have noted, there are as manyarbitrary character partitions as the imagination allows.Unlike character cliques that are homogeneous in termsof phylogenetic signal (Estabrook et al., 1976), partitionsdefined solely by anatomical proximity or morphologi-cal criteria often contain mixed phylogenetic signals, asdid those of O’Leary and Geisler (1999).
Pairwise ILD tests among the five arbitrary charac-ter sets showed that three comparisons were signifi-cant: basicranial versus dental (P = 0.001), basicranialversus postcranial (P = 0.004), and postcranial versusdental (P = 0.001). In ILD tests between individual par-titions and the remaining data, two comparisons weresignificant: basicranial versus other morphological traits(P = 0.001) and dental versus other morphological traits(P = 0.001). According to Naylor and Adams (2001),the ILD test results indicated that “the dental charac-ters emerged as the most different from the other parti-tions, followed by the basicranial characters” (2001:447).The postcranial partition, which had two significant ILDscores, was not considered problematic in the remainderof Naylor and Adams’s (2001) study.
Given these ILD analyses, and the standard con-ditional combination argument that “if data setsare demonstrably heterogeneous they should not becombined in an analysis that assumes character homo-geneity” (Bull et al., 1993:385), there would appear to beno defensible way in this example to sort out a singlediscrepant partition. The test results showed that therewere multiple conflicting interactions among three char-acter sets (basicranial, postcranial, and dental); removalof any one of these incongruent partitions leaves at leasttwo significantly conflicting partitions in the residuum.Therefore, Naylor and Adams’s (2001) statement that thenondental character sets represented “data partitions forwhich the hierarchical signals were mutually compati-ble” (2001:446) is inconsistent with their ILD test resultsbecause the nondental partition included the basicranialand postcranial character sets that were significantly in-congruent. The five morphological partitions could bemerged into minimally three data sets if significant inter-nal conflicts among dental, basicranial, and postcranialcharacters were to be avoided.
These specifics underscore a more general point. Theoriginal authors of the ILD test (Farris et al., 1995) neversuggested that it should be implemented as a criterionfor excluding data, and numerous authors recently havecriticized this test as a basis for separation of charactersets (e.g., Barker and Lutzoni, 2002; Darlu and Lecointre,2002; Dowton and Austin, 2002).
INTERPRETATION OF ILD TEST RESULTSBY TAXONOMIC CONGRUENCE
Based on the ILD results, which showed no clear dis-tinction between anomalous and consistent phylogeneticsignals in the morphological matrix, Naylor and Adams(2001) again implemented a taxonomic congruence ap-proach (Miyamoto and Fitch, 1995). They comparedtopologies supported by separate analyses of differ-ent morphological partitions with previously reportedmolecular topologies (Fig. 1e) “to determine the natureof the phylogenetic signal differences among data parti-tions found to be distinct by the ILD tests”—(2001:447).
An analysis of all morphological characters, excludingthe dental data, yielded a 50% majority rule consensustree (Naylor and Adams, 2001: fig. 2) that was inconsis-tent with O’Leary and Geisler’s (1999) results (Fig. 2a).Naylor and Adams (2001:447) suggested that
the resulting phylogeny was similar to that of the molecular dataonly. Specifically, Artiodactyla was paraphyletic with respect toCetacea, and the Hippopotamidae were found to be the closest livingrelatives to Cetacea. . . .
When the basicranial partition was excluded [from the com-bined morphological data set], the relationships among the majorclades were nearly identical to those found by O’Leary and Geisler(1999). Artiodactyla, Cetacea, Mesonychia, and Perissodactyla wereall found to be monophyletic, and Mesonychia was the sister-groupto Cetacea. This implies that the basicranial partition has little influ-ence on the topology obtained by O’Leary and Geisler (1999).
When both the dental and basicranial partitions were ex-cluded, Artiodactyla was found to be paraphyletic with respect toCetacea. . . .Although the remaining signal supports Artiodactyl [sic]paraphyly, it does so only weakly.
The tests described above identify the dental partition as the oneexerting the most influence on topology for O’Leary and Geisler’s(1999) data set. . .. When the dental data are removed, the remainingmorphological characters (extant and fossil combined) yield a treecorresponding to that obtained from an analysis of the moleculardata alone. In fact, a phylogenetic analysis of the dental data alonesupports a close relationship between Cetacea and Mesonychia, con-firming this finding.
Because the ILD test results were ambiguous, the taxo-nomic congruence arguments in the passage above out-line Naylor and Adams’s (2001) primary evidence forpinpointing the dental partition as the source of “mis-leading” (2001:450) signal in O’Leary and Geisler’s (1999)data set. However, this majority of statements in thisquoted passage are either problematic or refutable forthe following five reasons.
1. Naylor and Adams (2001) reported 46,209 most par-simonious trees from their analysis of the morphologicalmatrix with dental characters excluded. A grouping ofhippos and whales was resolved in the 50% majority ruleconsensus of these topologies but was not supported bya strict consensus tree. Several authors have noted thatgroups resolved in a 50% majority rule consensus treebut not in the strict consensus tree are of dubious empiri-cal value because conflicts among minimum length treesfor a data set are disregarded (e.g., Nixon and Carpenter,1996b). Sharkey and Leathers (2001:283) explicitly notedthat “using majority rule consensus as a criterion forselecting trees equates reliability with ambiguity.” Thesecriticisms notwithstanding, Naylor and Adams (2001)employed 50% majority rule consensus trees throughout
856 SYSTEMATIC BIOLOGY VOL. 52
FIGURE 2.
2003 POINTS OF VIEW 857
their reanalysis of O’Leary and Geisler’s (1999) data setand apparently (based on our attempts to repeat theirwork) did not complete branch swapping in many oftheir searches. In our analysis of the nondental data set(Fig. 2b), we found 93,578 most-parsimonious trees ofthe same length as those found by Naylor and Adams(2001).
2. Naylor and Adams stated that “when the dental dataare removed, the remaining morphological characters(extant and fossil combined) yield a tree correspondingto that obtained from an analysis of the molecular dataalone” (2001:447). Later in their paper, they added thatthe nondental data set “results in a topology entirelyconsistent with that of the molecular data” (2001:448)and that molecular data “yield a phylogenetic signalidentical to that found from an analysis of fossil andextant morphological data from which only the dentalcharacters have been excluded” (2001:452). These state-ments are false. The strict consensus tree for the non-dental data set exhibited numerous discrepancies withpreviously published molecular results, which do notgroup camels, sheep, and mouse deer to the exclusion ofhippos and whales nor do they show camels to be moreclosely related to sheep than to the mouse deer (Fig. 1e).The nondental data also grouped the horse (Equus)(Fig. 2b) closer to artiodactylans (e.g., sheep, camels) thanto Eocene fossil stem horses and tapirs (Hyracotherium,Heptodon); paraphyly of Perissodactyla is a result in fullconflict with prior analyses of mammalian phylogeny(e.g., Holbrook, 1999; Murphy et al., 2001). The onlyapparent consistency with previous molecular topolo-gies was the hippo + whale grouping that was not evenresolved in the strict consensus of most-parsimonioustrees. Overall, six groupings of extant taxa were resolvedby analysis of the nondental data (Fig. 2b), and three ofthese are contradicted by molecular data (Fig. 1e; see alsoGatesy et al., 1999, 2002).
The topological incongruence with molecular data cre-ated by Naylor and Adams’s (2001) elimination of dentalcharacters was extensive (Fig. 2b). The WHIPPO-2 ma-trix (Gatesy et al., 1999), a large molecular data set ofparaxonian DNA sequences published 2 years prior to
←FIGURE 2. Strict consensus trees derived from minimum length topologies for different partitions of O’Leary and Geisler’s (1999) morpholog-
ical matrix (see that paper for support values). (a) All characters. (b) Nondental characters. (c) Nonbasicranial characters. (d) Dental characters.(e) Nonpostcranial characters. (f) Non-hind limb characters or nonfeet characters. Red circles mark groupings of extant taxa contradicted bymolecular data (Fig. 1e). The orange circle indicates a node that is inconsistent with perissodactylan monophyly. Black circles designate cladessupported by the nonbasicranial data set that conflicted with the complete morphological matrix. Green circle marks the only improvement intopological congruence with molecular results after the deletion of different morphological data partitions. For each data set, minimum treelength and the number of extra steps required to resolve artiodactylan monophyly are shown. Blue = cetaceans; purple = mesonychians; yellow= artiodactylans; brown = perissodactylans; and gray = other taxa. Common names for extant taxa and an extinct tapiroid perissodactylan areshown to the right of generic names. Lineages that connect extant taxa are colored dark gray; all other taxa are extinct. Phylogenetic analyses inPAUP∗ generally were heuristic with 1,000 random taxon addition replicates, tree bisection–reconnection (TBR) branch swapping, tree lengthscalculated using all characters, and the “amb-” option in effect (Swofford, 1998). The dental partition was an exception. Because of the number ofminimum-length trees for this data set and computer memory limitations, branch swapping could not be completed in any single random taxonaddition replicate. For the dental partition, 453,448 minimum-length trees were saved using multiple heuristic searches in PAUP 3.1.1 (Swofford,1993), and the number of extra steps required to recover artiodactylan monophyly was estimated by a heuristic PAUP∗ search with 1,000 randomtaxon addition replicates and maximally 1,000 trees retained per replicate, with the “amb-” option in effect. For the nondental partition, phyloge-netic analyses in PAUP∗, with and without the constraint of artiodactylan monophyly, were heuristic with 1,000 random taxon addition replicates,TBR branch swapping, and the “amb-” option in effect. Additional searches of the nondental data set were done in PAUP 3.1.1 as a comparisonto the analysis in Naylor and Adams (2001). These searches yielded 93,578 minimum-length trees. Because of the limited number of informativecharacters, extensive missing data, and character conflicts, bootstrap/jackknife analyses were not executed for the partitioned data sets.
the study of Naylor and Adams (2001), required an ad-ditional 119 character steps beyond minimum length toaccommodate the conflicting relationships strictly sup-ported by Naylor and Adams’s (2001) analysis of non-dental morphological data; 188 extra steps are required ifperissodactylan paraphyly also is enforced. Each of theseconstrained topologies is a very poor fit to the moleculardata relative to minimum-length molecular topologies(P = 0.0001 for Wilcoxon signed rank test of charactersupport for a priori comparisons, Templeton, 1983). Forcomparison, conflicting groups in the original morpho-logical topology of O’Leary and Geisler (1999) required140 extra steps from the molecular matrix (Fig. 2a). Ifanything, removal of the “misleading” (2001:450) dentaldata increased topological incongruence with previousmolecular results.
The extraction of any arbitrarily defined morphologi-cal partition from an analysis risks the removal of highlyconsistent characters with the deletion of any supposedinconsistent ones. The dental partition was very hetero-geneous in terms of phylogenetic signal, as evidencedby extensive homoplasy within this partition when ana-lyzed separately (consistency index of 0.4444; Kluge andFarris, 1969). The deletion of all dental traits eliminatedmany characters that were congruent with indepen-dent DNA data and resulted in a topology that, contraNaylor and Adams (2001:448), was inconsistent with pre-vious molecular results (Fig. 2b), not “a topology entirelyconsistent with that of the molecular data” (2001:448).All large published molecular matrices for Paraxoniaseverely contradict the nondental tree (see Fig. 1).
3. Naylor and Adams made two points regardingthe strength of support for artiodactylan paraphyly intheir partitioned analyses. Surprisingly these statementswere not based on quantitative indices of support. Theystated that “when both the dental and basicranial parti-tions were excluded, Artiodactyla was found to be para-phyletic with respect to Cetacea. . .. Although the remain-ing signal supports Artiodactyl [sic] paraphyly, it doesso only weakly” (2001:447). In contrast, the nondentaldata “strongly supported a phylogeny identical to thatfound for the molecular data, namely, that Artiodactyla
858 SYSTEMATIC BIOLOGY VOL. 52
is paraphyletic with respect to Cetacea” (2001:451). Theirfirst statement is true; with both dental and basicranialcharacters removed, only three extra character steps wererequired to resolve artiodactylan monophyly. However,their second statement is false. For the nondental parti-tion, artiodactylan monophyly also demanded only threeextra character steps beyond minimum-length (Fig. 2b;P > 0.400 in Wilcoxon signed rank test for many com-parisons of minimum-length trees with and without ar-tiodactylan monophyly). Therefore, we contend that thenondental data did not “strongly” support artiodactylanparaphyly.
4. In the ILD analyses, the basicranial partition, likethe dental partition, was significantly incongruent withthree other morphological partitions. Naylor and Adams(2001) however stated that dental characters, not basi-cranial characters, were problematic, based on their tax-onomic congruence results (2001:447):
when the basicranial partition was excluded, the relationshipsamong the major clades were nearly identical to those found byO’Leary and Geisler (1999). Artiodactyla, Cetacea, Mesonychia, andPerissodactyla were all found to be monophyletic, and Mesonychiawas the sister-group to Cetacea. This implies that the basicranial par-tition has little influence on the topology obtained by O’Leary andGeisler (1999).
These statements are false; 22 of 31 clades favored bythe nonbasicranial partition conflicted with the tree ofO’Leary and Geisler (1999) (Figs. 2a, 2c). The non-basicranial data set strictly supported a paraphyleticArtiodactyla, with extant cetaceans nested within, di-phyly of Cetacea, and paraphyly of Mesonychia (Fig. 2c).Of the five groups that Naylor and Adams (2001) listedas being supported by this data set, only one, Peris-sodactyla, was recovered in our parsimony reanalysis.Removal of the basicranial characters radically rear-ranged relationships supported by O’Leary and Geisler’s(1999) data set (Fig. 2a) and contradicted Naylor andAdams’s primary thesis.
5. Naylor and Adams’s (2001) argued that “a phyloge-netic analysis of the dental data alone supports a close re-lationship between Cetacea and Mesonychia” (2001:447)and that this finding points to the dental partition asthe “anomalous hierarchical signal” (2001:448) in themorphological data set. Again, we could not replicatethis result. Phylogenetic analysis of the dental partitionproduced hundreds of thousands of minimum-lengthtopologies, and we retained 453,448 trees found in multi-ple heuristic searches initiated by different random taxonaddition sequences (Naylor and Adams [2001] truncatedtheir search at 88,737 trees; D. Adams, pers. com.). Thestrict consensus of our 453,448 trees had little structureand did not group Cetacea with Mesonychia (Fig. 2d). A50% majority rule consensus tree also did not supportthe quoted statement above from Naylor and Adams(2001) and instead resolved a paraphyletic Artiodactyla,a monophyletic Perissodactyla, and a large unresolvedpolytomy that included Leptictis, Asioryctes, three lin-eages of cetaceans, Eoconodon, Arctocyon, eight lineages ofmesonychians, and the Eocene artiodactylan Diacodexis.
Even if analysis of the dental data alone did support“a close relationship between Cetacea and Mesonychia”(Naylor and Adams, 2001:447), this result would be con-gruent with all published molecular topologies, becauseany placement of the fossil clade Mesonychia (whichhas not been scored for molecular characters) is consis-tent with molecular results (e.g., Fig. 1c). Artiodactylaand its traditional subclades (Fig. 1d), which representreal topological discrepancies with molecular trees (Fig.1e), were not strictly supported by analysis of the den-tal partition alone (Fig. 2d). Artiodactylan monophylydemanded three extra character steps beyond minimum-length for this data set, the same number of steps requiredfrom the nondental partition that “strongly supported”(2001:451) artiodactylan paraphyly.
WHAT ABOUT THE INCONGRUENTPOSTCRANIAL PARTITION?
After showing that the postcranial characters weresignificantly incongruent with two other morphologicalpartitions, Naylor and Adams (2001) did not commentagain on the postcranial partition as a potentially anoma-lous phylogenetic signal. Our analysis of the morpholog-ical matrix, with postcranial characters deleted, yieldeda strict consensus that supported Cetacea, Mesonychia,Perissodactyla, and a nonmonophyletic Artiodactyla(Fig. 2e). All minimum-length trees favored a Cetacea+ Mesonychia clade, grouped hippopotamid artiodacty-lans closer to Cetacea than to other extant artiodactylans,and therefore supported both of Naylor and Adams’s(2001) “conflicting” phylogenetic hypotheses (Fig. 1c).
In further exploratory analyses of the postcranial data,we noticed that removal of all characters relevant tothe hind limb or removal of all foot characters yieldeda topology that was quite congruent with Naylor andAdams’s (2001) nondental tree (Figs. 2b, 2f). Naylor andAdams (2001) characterized the dental data as the pri-mary aberrant signal in the morphological data set be-cause removal of this partition (45 characters, 37% of thetotal matrix) resulted in a 50% majority rule consensustree with groups that they suggested perfectly matchedthe “molecular” results: whales grouping closest to hip-pos, a paraphyletic Artiodactyla, and mesonychians dis-tantly related to whales. However, removal of 17 hindlimb characters, only 14% of the total matrix, yieldedall of these groups in a well-resolved strict consensus ofsix minimum-length trees (Fig. 2f). Deletion of anotherarbitrary postcranial subpartition, forefoot and hindfootcharacters (16 traits), supported the same set of trees (Fig.2f). Thus, morphological analyses that included the en-tire “suspect” (2001:453) dental partition supported all ofthe relationships that Naylor and Adams (2001) arguedwere inhibited by this “anomalous hierarchical signal”(2001:448) (Fig. 2f). Four unequivocally optimized den-tal synapomorphies grouped hippos and whales to theexclusion of other extant paraxonians in these trees.
Clearly the patterns of character support, conflict, andmissing data in O’Leary and Geisler’s (1999) matrix aremuch more complicated than Naylor and Adams (2001)
2003 POINTS OF VIEW 859
FIGURE 3. Putative serial homologues mapped onto the strict consensus of minimum-length trees for O’Leary and Geisler’s (1999) com-plete morphological matrix. Parsimony character optimizations for four sets of characters selected by Naylor and Adams (2001) are shown(68/70/72, 69/71/73, 74/75/76/52, and 80/81; character numbers from O’Leary and Geisler, 1999). Solid bars at internodes show unequivocaloptimizations, and open bars at internodes are equivocal optimizations (only one position for each equivocal optimization is shown). None ofthese characters supported groups that conflicted with molecular trees (red circles). When the morphological data set was analyzed withoutthe putative serial homologues, subgroups of Mesonychia (the clade formed by the common ancestor of Harpagolestes and Hapalodectes) wereunresolved, but no improvement in congruence occurred between molecular and morphological results. Nodes that conflict with moleculartopologies are supported by unequivocal synapomorphies that are primarily nondental. The number of unequivocal dental synapomorphies isshown above each conflicting internode, and the number of unequivocal nondental synapomorphies is indicated below. Colors and labels are as inFigure 2.
reported. Even the removal of small character partitionshad profound and unpredictable effects on systematicresults. This situation illustrates the potential liability ofremoving large heterogeneous data partitions from anal-ysis. An examination of unequivocal synapomorphies
from the O’Leary and Geisler (1999) data set showed thatcharacters from multiple partitions, not primarily dentalcharacters, supported each of the clades that conflictedwith molecular topologies (Fig. 3). In fact, the majority ofthese character transformations (80%) were in nondental
860 SYSTEMATIC BIOLOGY VOL. 52
traits, a result that again contradicts the primary thesisof Naylor and Adams (2001).
PROBLEMATIC EXPLANATIONSFOR DENTAL INCONGRUENCE
In the final section of their paper, Naylor and Adams(2001) expanded arguments about the hypothesized in-congruent nature of dental characters. Specifically, theyused descriptive work from developmental geneticsand arguments about convergent selection pressures toexplain “why dental characters might be misleading”(2001:450). As discussed above, Naylor and Adams’s(2001) ILD tests and taxonomic congruence comparisonsdid not single out the dental partition as an “anoma-lous” phylogenetic signal. They nonetheless developedthree reasons to question the phylogenetic utility of den-tal traits: (1) developmental dependence, (2) functionaldependence, and (3) excessive homoplasy.
On the issue of developmental dependence, Naylorand Adams said (2001:450), “Because the same under-lying genetic architecture generates teeth in a particulartooth group, similar structures on different teeth (e.g.,the hypocone) are de facto serially homologous.” Theyargued, furthermore, that identification of different char-acters as serial homologues implied that these characterswere phylogenetically redundant.
Identification of serial homologues (particularly infossils) is complicated by a number of factors. Hall(1995:10–11) defined serial homologues as “serial rep-etition of features . . . based on common developmen-tal processes.” Developmental papers cited by Naylorand Adams (2001) showed (1) that in ontogenetic stud-ies conducted primarily in the mouse (some also inthe vole), similar developmental pathways are sharedamong groups of teeth (Jernvall and Thesleff, 2000) and(2) that in ontogenetic studies of the mouse using knock-out experiments, impeding one gene affects the develop-ment of groups of teeth (Tucker et al., 1998; Tucker andSharpe, 1999). It is unclear therefore how Naylor andAdams (2001:450) had evidence that “similar structureson different teeth” in extinct whales or mesonychians”are de facto serially homologous.” The studies Naylorand Adams (2001) cited did not describe the develop-ment of teeth from extant artiodactylans, cetaceans, orother ungulates and said nothing about developmen-tal pathways in extinct taxa, which comprise ∼85% ofthe genera most relevant to whale origins (O’Leary andGeisler, 1999). For their argument Naylor and Adams(2001) had to assume that the ontogeny of a rodent wasidentical to the ontogenies of modern and extinct parax-onian species. This could be a gross oversimplification.
For example, empirical work on eye lens inductionhas shown that even intrageneric species can possessdramatically different ontogenetic trajectories to achievehomologous structures (see review by Hall, 1995). Suchresults warn against assuming that similar develop-mental pathways automatically underlie similar pheno-types. Wray and Lowe (2000) also emphasized the op-posite, that similar ontogenetic pathways can produce
extremely different end products. Hall (1995:29) stressedthat “many homologous features share common devel-opment but many do not.” Thus, applying Hall’s (1995)definition of serial homology (also see Wagner, 1989),Naylor and Adams (2001) did not establish that certaindental characters from O’Leary and Geisler (1999) are“de facto serially homologous” (2001:450) because theydid not present comparative developmental data for theappropriate taxa.
Also very relevant to this discussion is the follow-ing question: what is the specific implication for phy-logeny reconstruction of labeling certain characters “se-rial homologues”? The developmental studies cited byNaylor and Adams (2001) were important descriptivecontributions to the study of ontogeny, and their devel-opmental discoveries could be incorporated into phy-logenetic analysis directly as characters (see Janies andDeSalle, 1999). These studies were not, however, fun-damentally comparative or phylogenetic. Naylor andAdams (2001) implied that simply calling characters “se-rial homologues” means that the investigator has knowl-edge that these characters amount to phylogeneticallyredundant observations. That different tooth charactersare serially homologous is, however, just an additionalassumption made by Naylor and Adams (2001) in theirphylogenetic reanalysis, and their recognition of serialhomologues does not necessarily lead directly to the con-clusion that these traits are phylogenetically correlated.
Even though they lacked specific developmental data,Naylor and Adams (2001) still argued that many dentalcharacters from O’Leary and Geisler (1999) were seriallyhomologous and “nonindependent” (2001:450). Naylorand Adams “identified a priori six sets of characters mea-sured on multiple teeth that were presumed to exhibitsuch redundancy” (2001:450). When characters withinany set covaried in a principal coordinates analysis, thatset was considered “serially homologous” and “redun-dant” (2001:450). They argued that four of six sets covar-ied appropriately and concluded (2001:450), “11 char-acters in the dental partition exhibit nonindependenceand pseudoreplication reflecting multiple coding of thesame features.” Naylor and Adams (2001:450) conceded,however, that “shared-derived history can cause char-acter states to covary over a tree or a principal coordi-nates ordination in the same way as developmental non-independence.” Simply mapping these characters ontoO’Leary and Geisler’s (1999) topology (Fig. 3) demon-strated that none of the 11 putative serial homologuessupported artiodactylan monophyly or any other mor-phological clades that conflicted with molecular evi-dence. Removing or downweighting these “serial ho-mologues” therefore does not change the phylogenetichypothesis of O’Leary and Geisler, except to decrease res-olution within Mesonychia (Fig. 3). A minority of thesecharacters equivocally supports a grouping of Cetaceawith Mesonychia (Fig. 3), but again this grouping iscompatible with both morphological and molecular trees(Fig. 1c).
Problems with the serial homology argumentsnotwithstanding, Naylor and Adams (2001) next
2003 POINTS OF VIEW 861
discussed functional correlation as another potential rea-son dental data confounded phylogenetic results. Theirprincipal coordinate analysis indicated that, “the meta-cone on the upper molar M2 fell close to the metaconidcluster” (2001:450) and that this possibly was an “an ad-ditional instance of functional and developmental non-independence” (2001:450). They presented, however, noargument for why these traits should be functionallycorrelated. Furthermore, on this subject Farris (1969:374)noted that “even if we could find some way to measure,for example, the ‘functional importance’ of a character, itwould still remain to be demonstrated that the measureof functional importance was correlated with the util-ity of the character for purposes of cladistic inference.”Even if Naylor and Adams (2001) had presented an argu-ment about the functional relatedness of these characters,it would be an oversimplification to equate functionalnonindependence with phylogenetic nonindependence.In any case, variations in M2 metacone shape (character52 in Fig. 3) did not support artiodactylan monophyly orany other clades that conflicted with molecular evidence(Fig. 3).
Naylor and Adams (2001) introduced a third line ofargumentation as to why dental data are problematic forphylogeny reconstruction. They hypothesized (2001:452)that “mammalian teeth are more evolutionarily plas-tic than was originally believed, and that any phylo-genetic signal initially present in the dental data hasbeen eroded because of convergent evolution.” In sup-port of this statement Naylor and Adams (2001) citedJernvall’s (2000:2641) assertion that “small cusps maybe unreliable as phylogenetic signals,” an inference thatemerged primarily from descriptions of mouse develop-ment (Jernvall et al., 1998; Jernvall and Thesleff, 2000).The generality of this assertion is, however, wholly spec-ulative. Jernvall’s (2000) observation of ontogenetic vari-ation in a character in an individual species is not ev-idence of the phylogenetic variability of the characteracross other living and extinct species. Extrapolating di-rectly from ontogeny to phylogeny has little relation-ship to phylogenetic hypothesis testing and greatly over-simplifies the phylogenetic role of developmental data(Janies and DeSalle, 1999).
In this line of argument, Naylor and Adams (2001)cited a recent systematic study of Cretaceous mammals(Luo et al., 2001) as an example in which dental traits havebeen shown to be evolutionarily labile and prone to con-vergence. Luo et al. (2001) suggested that tribosphenicmolars evolved multiple times in early mammals, andNaylor and Adams (2001) argued that this result indi-cated that homoplasy in teeth has generally been un-derestimated in mammalian systematics. Naylor andAdams (2001) failed to point out, however, that Luoet al.’s (2001) tree was derived from a data matrix thatcontained ∼35% dental characters and that each of thesedental characters was treated as an independent pieceof phylogenetic evidence. It was illogical for Naylor andAdams (2001) to endorse the final result of Luo et al.(2001), where dental data served as synapomorphies atcritical nodes, and to argue simultaneously that the fun-
damental phylogenetic usefulness of dental data shouldbe questioned. In the face of this inconsistency, it is hardto understand how Naylor and Adams concluded that“there is every reason to suspect that . . . dietary selec-tion pressures . . . would override any phylogenetic sig-nal in an evolutionarily malleable tooth morphology”(2001:452).
Naylor and Adams’s (2001) extrapolations from on-togeny to phylogeny and from one phylogenetic analy-sis to another were not justified. In separate analyses, thedental data in O’Leary and Geisler’s (1999) matrix hada higher ensemble consistency index (0.4444; Kluge andFarris, 1969) and retention index (0.7727; Farris, 1989)than the remaining morphological characters (consis-tency index = 0.4024; retention index = 0.7112). Thevalues for these two partitions are actually very simi-lar but higher for the dental data. Dental convergencemight have occurred in Cretaceous mammals, but thisconvergence is not necessarily a predictor of the evolu-tionary behavior of dental characters across Mammalia.In Paraxonia, the clade of interest, dental homoplasy hasnot been shown to be excessive relative to homoplasy inother morphological traits.
HISTORICAL PRECEDENCE FOR CHARACTER DELETION
Arguments such as those presented by Naylor andAdams (2001) share features in common with the phi-losophy underlying schemes of relationship and classi-fication proposed much earlier in the history of system-atics by scientists such as Cuvier (1800) and Blainville(1816). These 19th century authorities argued that theyhad particular insights for judging the relative systematicimportance of anatomical traits. In the 20th century, thesearguments were embraced by evolutionary taxonomistswho used perceived generalities of evolutionary processto determine which characters were reliable and whichwere not (Simpson, 1945; Mayr, 1969; see review by Cain,1959). With the development of cladistics, many raisedquestions about the testability (and therefore the scien-tific value) of such descriptions of evolutionary history(e.g., Farris et al., 1982).
These challenges also apply to the arguments ofNaylor and Adams (2001) because their criteria for distin-guishing phylogenetically misleading from highly infor-mative groups of characters are deficient for many of thesame reasons. Without insights into some yet undiscov-ered law of nature, there is no particular reason to thinkthat a functional, developmental, or ecological explana-tion for homoplasy is a better explanation of covariationthan is synapomporphy. Simply proposing such general-ities does not condemn characters to being phylogeneti-cally uninformative.
The analysis of Naylor and Adams (2001) is the mostrecent manifestation of a disturbing trend towards dataselectivity in paraxonian systematics. The phylogeneticdatabase for this group is becoming large, but in most re-cent studies only small subsets of the available data havebeen examined, in isolation from the majority of poten-tially conflicting evidence. For example, in a study of the
862 SYSTEMATIC BIOLOGY VOL. 52
digestive tract, Langer (2001:87) noted that morpholog-ical characters, which happen to contradict the hypoth-esis he accepts, “can be seen as the result of convergentadaptations to an aquatic or amphibious lifestyle andhave to be dismissed as phylogenetic evidence.” Langer(2001) did not actually include these characters or anymolecular data in his matrix, and his analysis yielded avariety of unconventional relationships. In an analysisof mitochondrial cytochrome b and selected morpholog-ical characters, Luckett and Hong (1998:154) contendedthat “even though 1140 nucleotide bases or characterscomprise the cytochrome b gene of most eutherians, themajority of these traits appear to be of minimal valuefor assessing higher-level relationships.” Citing Naylorand Brown (1998), Luckett and Hong (1998) argued that>95% of the nucleotide sites they examined were toovariable to be informative and were eliminated from phy-logenetic analysis.
Including certain data and excluding other publisheddata when conducting phylogenetic analyses can resultin well-resolved trees, but such trees are weakly testedeven when they have high standard support measures(Kluge, 1997). As a result, data exclusion has question-able relevance to empirical phylogenetics. The basis forsuch evaluations of characters is often the credibility ofa particular null model (Maddison, 1990; Wollenbergand Atchley, 2000; O’Keefe and Wagner, 2001), a per-ceived evolutionary generality, or an arbitrary scheme ofcharacter partitioning. Characters suspected of covary-ing for developmental or functional reasons, however,actually might record phylogenetic congruence ratherthan convergence, and different characters assigned toan arbitrary class can contain mixed phylogenetic sig-nals. Determinations of character reliability do not needto be based on arbitrarily chosen models or data par-titions but instead can be decided by the combinedcongruence of all characters. Unjustified partitioningand character exclusion should not be used to rejectanalyses in which these additional assumptions wereavoided.
CONCLUSIONS
None of Naylor and Adams’s ILD tests and taxo-nomic congruence comparisons singled out the den-tal partition as a “signal markedly different from therest of the morphological data” (2001:451), and many ofNaylor and Adams’s (2001) conclusions were not sup-ported by reanalysis of O’Leary and Geisler’s (1999) data.There was, therefore, no empirical reason for Naylorand Adams (2001) to explain why dental traits wereso incongruent with all other morphological characters.Our (re)reanalysis of O’Leary and Geisler’s (1999) ma-trix showed that (1) the interactions among arbitrarycharacter partitions in phylogenetic analysis were com-plex, (2) the dental partition was very heterogeneous interms of phylogenetic signal, (3) dental characters, sin-gled out by Naylor and Adams (2001) as serial homo-logues, did not support morphological clades that con-flicted with molecular results (these conflicting clades
primarily were supported by nondental morphologicalevidence), and (4) homoplasy of dental characters inParaxonia was not extreme relative to other morpholog-ical partitions.
Naylor and Adams ended their paper by stating that“over-reliance on any form of data should be viewedwith caution” (2001:452). We agree with this generalstatement; however, we disagree that to put this ideainto practice requires deleting entire partitions of data,as done by Naylor and Adams (2001) and certain otherrecent analyses of paraxonian phylogeny. Such practicesincrease rather than decrease reliance on particular sub-sets of data. Inclusive simultaneous analysis of all avail-able data minimizes overreliance on any one form of data(Nixon and Carpenter, 1996a). When conflicts remain,new data should arbitrate the incongruence, rather thanad hoc ideas about the existence of arbitrary data parti-tions and how some of those partitions might be corrupt.
In our previous work, we have used partitions forheuristic exploration of phylogenetic data (e.g., Novacek,1994; Gatesy, 1998; O’Leary, 1999) and to assess the sta-bility of systematics results to data removal (e.g., Gatesyet al.,1999; Gatesy, 2002). A hallmark of these studies isthat they each presented a combined analysis and recog-nized that it best explained the data. Heuristic data ex-ploration should not be confused with the approach ofNaylor and Adams (2001), in which selected partitionsare discredited and a phylogeny based on incompletedata is preferred over the combined result.
The recent explosion of published phylogenetic anal-yses of Paraxonia includes contributions from such his-torically disparate fields as histology, paleontology, andmolecular biology and challenges systematists to absorbdata collected outside their area of specialization. We donot argue here in favor of a particular phylogenetic resultor that no dental characters are convergent (doubtlessmany are). Instead, we simply suggest that more exten-sive systematic analyses will be required to sort out therelationships among living whales, artiodactylans, andtheir extinct relatives. A combined analysis of all poten-tial homologues provides the greatest explanatory powerbecause it casts phylogenetic analysis as an accretionarysynthesis of detailed comparative work across all phe-notypic and genotypic systems and in all taxa.
ACKNOWLEDGMENTS
For comments on earlier versions of the manuscript, we thank R.Baker, J. Carpenter, A. de Queiroz, R. DeSalle, C. Hayashi, D. Janies, M.Kearney, M. Siddall, and J. Wilson. This work was supported by NSFgrants DEB-9903964 and DEB-9985847.
REFERENCES
BARKER, F. K., AND F. M. LUTZONI. 2002. The utility of the incongruencelength distance test. Syst. Biol. 51:625–637.
BLAINVILLE, H. M. D. DE. 1816. Prodrome d’une nouvelle distri-bution systematique du regne animal. Bull. Soc. Philom., 1816:105.
BULL, J. J., J. P. HUELSENBECK, C. W. CUNNINGHAM, D. L. SWOFFORD,AND P. J. WADDELL. 1993. Partitioning and combining data in phy-logenetic analysis. Syst. Biol. 42:384–397.
CAIN, A. J. 1959. Function and taxonomic importance. Oxford Univ.Press, New York.
2003 POINTS OF VIEW 863
CUVIER, G. 1800. Lecons anatomie comparee. Baudouin, Imp. del’Institut national des sciences et des arts, puis Genets, an VIII, Paris.
DARLU, P., AND G. LECOINTRE. 2002. When does the incongruencelength difference test fail? Mol. Biol. Evol. 19:432–437.
DE QUEIROZ, A. 1993. For consensus (sometimes). Syst. Biol. 42:368–372.
DOWTON, M., AND A. D. AUSTIN. 2002. Increased congruence does notnecessarily indicate increased accuracy—The behavior of the incon-gruence length difference test in mixed-model analysis. Syst. Biol.51:19–31.
ESTABROOK, G. F., C. S. JOHNSON, AND F. R. MCMORRIS. 1976. A math-ematical foundation for the analysis of cladistic character compati-bility. Math. Biosci. 29:181–187.
FARRIS, J. S. 1969. A successive approximations approach to characterweighting. Syst. Zool. 18:374–385.
FARRIS. J. S. 1989. The retention index and the rescaled consistencyindex. Cladistics 5:417–419.
FARRIS, J. S., M. KALLERSJO, A. G. KLUGE, AND C. BULT. 1995. Con-structing a significance test for incongruence. Syst. Biol. 44:570–572.
FARRIS, J. S., A. G. KLUGE, AND M. F. MICKEVICH. 1982. Immunologicaldistance and the phylogenetic relationships of the Rana boylii speciesgroup. Syst. Zool. 31:479–491.
GATESY, J. 1998. Molecular evidence for the phylogenetic affinities ofCetacea. Pages 63–111 in The emergence of whales (J. Thewissen,ed.). Plenum, New York.
GATESY, J. 2002. Relative quality of different systematic data sets forcetartiodactyl mammals: Assessments within a combined analysisframework. Pages 45–68 in Molecular systematics and evolution:Theory and practice (R. DeSalle, W. Wheeler, and G. Giribet, eds.).Birkhauser Verlag, Basel.
GATESY, J., C. HAYASHI, M. A. CRONIN, AND P. ARCTANDER. 1996.Evidence from milk casein genes that cetaceans are close relatives ofhippopotamid artiodactyls. Mol. Biol. Evol. 14:537–543.
GATESY, J., C. MATTHEE, R. DESALLE, AND C. HAYASHI. 2002. Res-olution of a supertree/supermatrix paradox. Syst. Biol. 51:652–664.
GATESY, J., M. MILINKOVITCH, V. WADDELL, AND M. STANHOPE. 1999.Stability of cladistic relationhips between Cetacea and higher-levelartiodactyl taxa. Syst. Biol. 48:6–20.
GATESY, J., AND M. A. O’LEARY. 2001. Deciphering whale origins withmolecules and fossils. Trends Ecol. Evol. 16:562–570.
GEISLER, J. 2001. New morphological evidence for the phylogeny ofArtiodactyla, Cetacea, and Mesonychidae. Am. Mus. Novit. 3344:1–53.
GINGERICH, P. D., M. UL HAQ, I. S. ZALMOUT, I. H. KHAN, AND M. S.MALKANI. 2001. Origin of whales from early artiodactyls: Hands andfeet of Eocene Protocetidae from Pakistan. Science 293:2239–2242.
HALL, B. K. 1995. Homology and embryonic development. Evol. Biol.28:1–37.
HARRIS, S. R., D. J. GOWER, AND M. WILKINSON. 2003. Intraorganismalhomology, character construction, and the phylogeny of aetosaurianarchosaurs (Reptilia, Diapsida). Syst. Biol. 52:239–252.
HOLBROOK, L. 1999. The phylogeny and classification of tapiromorphperissodactyls (Mammalia). Cladistics 15:331–350.
JANIES, D., AND R. DESALLE. 1999. Development, evolution andcorroboration. New Anat. 257:6–14.
JERNVALL, J. 2000. Linking development with generation of novelty inmammalian teeth. Proc. Natl. Acad. Sci. USA 97:2641–2645.
JERNVALL, J., T. ABERG, P. KETTUNEN, S. KERANEN, AND I. THESLEFF.1998. The life history of an embryonic signaling center: BMP-4 in-duces p21 and is associated with apoptosis in the mouse tooth enamelknot. Development 125:161–169.
JERNVALL, J. A., AND I. THESLEFF. 2000. Reiterative signaling and pat-terning during mammalian tooth morphogenesis. Mech. Dev. 92:19–29.
KLUGE, A. G. 1989. A concern for evidence and a phylogenetic hypoth-esis of relationships among Epicrates (Boidae, Serpentes). Syst. Zool.38:7–25.
KLUGE, A. G. 1997. Testability and the refutation and corroboration ofcladistics hypotheses. Cladistics 13:81–96.
KLUGE, A. G. 1998. Total evidence or taxonomic congruence: Cladisticsor consensus classification. Cladistics 14:151–158.
KLUGE, A. G., AND J. S. FARRIS. 1969. Quantitative phyletics and theevolution of anurans. Syst. Zool. 18:1–32.
LANGER, P. 2001. Evidence from the digestive tract on phylogeneticrelationships in ungulates and whales. J. Zool. Syst. Evol. Res. 39:77–90.
LIU, F., M. MIYAMOTO, N. FREIRE, P. ONG, M. TENNANT, T. YOUNG,AND K. GUGEL. 2001. Molecular and morphological supertrees foreutherian (placental) mammals. Science 291:1786–1789.
LUCKETT, W. P., AND N. HONG. 1998. Phylogenetic relationships be-tween the orders Artiodactyla and Cetacea: A combined assessmentof morphological and molecular evidence. J. Mammal. Evol. 5:127–182.
LUO, Z.-X., R. L. CIFELLI, AND Z. KIELAN-JAWOROWSKA. 2001. Dualorigin of tribosphenic mammals. Nature 409:53–57.
MADDISON, W. 1990. A method for testing the correlated evolution oftwo binary characters: Are gains or losses concentrated on certainbranches of a phylogenetic tree? Evolution 44:539–557.
MATTHEE, C. A., J. D. BURZLAFF, J. F. TAYLOR, AND S. K. DAVIS. 2001.Mining the mammalian genome for artiodactyl systematics. Syst.Biol. 50:367–390.
MAYR, E. 1969. Principles of systematic zoology. McGraw-Hill,New York.
MIYAMOTO, M. M., AND W. M. FITCH. 1995. Testing species phyloge-nies and phylogenetic methods with congruence. Syst. Biol. 44:64–76.
MURPHY, W., E. EIZRIK, W. JOHNSON, Y. ZHANG, O. RYDER, ANDS. O’BRIEN. 2001. Molecular phylogenetics and the origins ofplacental mammals. Nature 409:614–618.
NAYLOR, G. J. P. AND D. C. ADAMS. 2001. Are the fossil data really atodds with the molecular data? Morphological evidence for Cetartio-dactyla phylogeny reexamined. Syst. Biol. 50:444–453.
NAYLOR, G. J. P., AND W. M. BROWN. 1998. Amphioxus mitochondrialDNA, chordate phylogeny, and the limits of inference based on com-parisons of sequences. Syst. Biol. 47:61–76.
NIXON, K. C., AND J. M. CARPENTER. 1996a. On simultaneous analysis.Cladistics 12:221–241.
NIXON, K. C., AND J. M. CARPENTER. 1996b. On consensus, collapsibil-ity, and clade concordance. Cladistics 12:305–321.
NOVACEK, M. J. 1994. Morphological and molecular inroads to phy-logeny. Pages 85–131 in Interpreting the hierarchy of life: From sys-tematic patterns to evolutionary process theories (L. Grande and O.Rieppel, eds.). Academic Press, San Diego.
O’KEEFE, F. R., AND P. J. WAGNER. 2001. Inferring and testing hypothe-ses of cladistic character dependence by using character compatibil-ity. Syst. Biol. 50:657–675.
O’LEARY, M. A. 1999. Parsimony analysis of total evidence from extinctand extant taxa and the cetacean–artiodactyl question (Mammalia,Ungulata). Cladistics 15:315–330.
O’LEARY, M. A. 2001. The phylogenetic position of cetaceans: Fur-ther combined data analyses, comparisons with the stratigraphicrecord and a discussion of character optimization. Am. Zool. 41:487–506.
O’LEARY, M. A., AND J. H. GEISLER. 1999. The position of Cetacea withinMammalia: Phylogenetic analysis of morphological data from extinctand extant taxa. Syst. Biol. 48:455–490.
PHILIPPE, H., AND E. DOUZERY. 1994. The pitfalls of molecu-lar phylogeny based on four species, as illustrated by theCetacea/Artiodactyla relationships. J. Mammal. Evol. 2:133–152.
SHARKEY, M. J., AND J. W. LEATHERS. 2001. Majority does not rule:The trouble with majority-rule consensus trees. Cladistics 17:282–284.
SIDDALL, M. E. 1997. Prior agreement: Arbitration or arbitrary. Syst.Biol. 46:766–770.
SIMPSON, G. G. 1945. The principles of classification and a classificationof mammals. Bull. Am. Mus. Nat. Hist. 85:1–350.
SWOFFORD, D. L. 1993. PAUP: Phylogenetic analysis using parsimony,version 3.1.1. Illinois Natural History Survey, Champaign.
SWOFFORD, D. L. 1998. PAUP*: Phylogenetic analysis using par-simony (*and other methods), version 4. Sinauer, Sunderland,Massachusetts.
TEMPLETON, A. 1983. Phylogenetic inference from restriction endonu-clease cleavage site maps with particular reference to the evolutionof humans and the apes. Evolution 37:221–244.
864 SYSTEMATIC BIOLOGY VOL. 52
THEWISSEN, J. G. M., E. M. WILLIAMS, L. J. ROE, AND S. T. HUSSAIN.2001. Skeletons of terrestrial cetaceans and the relationships ofwhales to artiodactyls. Nature 413:277–281.
TUCKER, A. S., K. L. MATHEWS, AND P. T. SHARPE. 1998 Transforma-tion of tooth type induced by inhibition of BMP signaling. Science282:1136–1138.
TUCKER, A. S., AND P. T. SHARPE. 1999. Molecular genetics of toothmorphogenesis and patterning: The right shape in the right place. J.Dent. Res. 78:826–834.
WAGNER, G. P. 1989. The origin of morphological characters and thebiological basis of homology. Evolution 43:1157–1171.
Syst. Biol. 52(6):864–865, 2003Copyright c© Society of Systematic BiologistsISSN: 1063-5157 print / 1076-836X onlineDOI: 10.1080/10635150390252288
Total Evidence Versus Relevant Evidence: A Response to O’Leary et al. (2003)
GAVIN J. P. NAYLOR AND DEAN C. ADAMS
Department of Ecology, Evolution, and Organismal Biology, Iowa State University, Ames, Iowa 50010, USA; E-mail: [email protected] (G.J.P.N.)
We commend O’Leary et al. (2003) on their reanalysisof their own data. We could take issue with their list ofcriticisms of our reanalysis point by point. For example,we disagree with their assertion that a heuristic searchyielding 93,578 most-parsimonious trees (MPTs) is neces-sarily more “rigorous” than a heuristic search that yields46,209 MPTs. There simply is no way of knowing if theirset of 93,578 heuristically derived MPTs represents a bet-ter sampling of the landscape of MPTs without conduct-ing an exhaustive search of the tree space, which couldpotentially yield hundreds of millions of MPTs. Simi-larly, whether or not one should use a strict or majorityrule consensus as a summary of a set of MPTs dependscritically on how representative the subsample of heuris-tically derived MPTs is of the hierarchical patterns con-tained in the complete set of MPTs. If the subset weretruly representative of the complete set, a majority ruleconsensus would give a better estimate of the overall hi-erarchical tendency than would a strict consensus. Whenthe subset of heuristically derived MPTs is not represen-tative of the complete set, both strict and majority ruleconsensuses are poor summaries of hierarchical signal.However, what is at stake is much more fundamentalthan the list of technical issues raised by O’Leary et al.For us, the argument is about how best to estimate phy-logeny. The conflicting views expressed by Naylor andAdams (2001) and O’Leary et al. (2003) are a reexpressionof the debate between the total evidence versus the whatwe term herein “relevant evidence” schools of thought.
Both schools are faced with character variation acrosstaxa. One school proposes that any measurable trait hasthe potential to be phylogenetically informative, whereasthe other asserts that some traits are collectively better in-dicators of phylogeny than others. The first school (totalevidence) is skeptical of invoking any reliance on pro-cess to differentially weight characters and asserts thatthe best way to estimate phylogeny is to “let the dataspeak for themselves.” There is an implicit assumption
WOLLENBERG, K. R., AND W. R. ATCHLEY. 2000. Separation of phy-logenetic and functional associations in biological sequences byusing the parametric bootstrap. Proc. Natl. Acad. Sci. USA 97:3288–3291.
WRAY, G. A., AND C. J. LOWE. 2000. Developmental regulatory genesand echinoderm evolution. Syst. Biol. 49:28–51.
First submitted 26 November 2002; reviews returned 27 April 2003;final acceptance 19 June 2003
Associate Editor: Pam Soltis
that the distribution of measurable traits can be re-lied upon to tell us the truth about phylogeny; thatany historical signal present in the data will increasewith the number of characters scored, and that thissignal will override signals in the data present fromother forces. In other words, this approach assumesthat character covariance generated from forces otherthan history will cancel themselves out, allowing thehistorical signal to be identified. The second schoolis skeptical of this assumption. Because of this, itasserts that careful choice of characters is of pri-mary importance. In choosing characters, advocatesof the second school consider assumptions about de-velopmental processes, morphological integration, andfunctional morphology and use these factors as a guidein choosing characters that are more likely to contain his-torical signal (or more commonly to eliminate suites ofcharacters predisposed to contain covarying signals dueto forces other than evolutionary history).
In addressing the question “what is a character?”, pro-ponents of both schools would agree that a morpholog-ical character constitutes a heritable feature that can bereliably identified across taxa. They disagree on the im-portance ascribed to the idea that a character should rep-resent a tightly integrated module whose presence or absenceis functionally and developmentally dissociated from the pres-ence or absence of other such characters (the independentand identically distributed assumption). The total evi-dence school is less concerned about character indepen-dence but rightfully wary of the lure of selecting charac-ters that favor one hypothesis over another. The relevantevidence school is rightfully wary of character covari-ances due to forces other than history but perhaps lessconcerned about the lure of favoring certain hypothesesover others.
Resolution of these issues will best be addressedthrough empirical developmental biology and carefulfunctional anatomy. Two key questions remain: (1) Are
