What Does the Comparative Method Reveal About Adaptation?

What does the Comparative Method Reveal About Adaptation?

The Harvard community has made this article openly available. Please share how this access benefits you. Your story matters

Citation Leroi, A. M., M. R. Rose, and G. V. Lauder. 1994. “What Does the Comparative Method Reveal About Adaptation?” The American Naturalist 143 (3) (March): 381–402. doi:10.1086/285609.

Published Version doi:10.1086/285609

Citable link http://nrs.harvard.edu/urn-3:HUL.InstRepos:33948436

Terms of Use This article was downloaded from Harvard University’s DASH repository, and is made available under the terms and conditions applicable to Other Posted Material, as set forth at http:// nrs.harvard.edu/urn-3:HUL.InstRepos:dash.current.terms-of- use#LAA Vol. 143, No. 3 The American Naturalist March 1994

WHAT DOES THE COMPARATIVE METHOD REVEAL ABOUT ADAPTATION?

A. M. LEROI, M. R. ROSE, AND G. V. LAUDER* Departmentof Ecology and EvolutionaryBiology, Universityof California,Irvine, California92717

SubmittedJuly 6, 1992; Revised March 4, 1993; Accepted April27, 1993

Abstract.-It has been suggestedrecently that new quantitativemethods for analyzing comparative data permitthe identificationof evolutionaryprocesses. Specifically,it has been proposed that new comparative methods can distinguishthe direct effectsof natural selection on the distributionof a traitwithin a clade fromthe effectsof drift,indirect selection, genotype-by- environmentinteraction, and uncontrolledenvironmental variation. Such methodscan suppos- edly unravel t-herelative importanceof these factors by the phylogeneticanalysis of traits, performanceattributes, and habitats. We argue thatthey cannot. We show thatmany different evolutionarymechanisms can, in principle,account for any one interspecificpattern, and we illustrateour case using examples fromthe comparativeliterature. We argue that these con- founded mechanismscan only be unraveled if patternsof selection or genetic variationand covariationare directlymeasured in many species withina clade. Even thoughcomparative methodsare valuable for examiningthe evolutionaryhistory of traits,they will oftenmislead in the studyof adaptive processes.

The use ofcomparisons among extant species to identifyadaptations has many contemporarypractitioners, as well as a pedigreestretching back to the Origin of Species,if not to Paley's theisticbiology (Darwin 1859, 1871; Simpson 1953; Bock 1977; Clutton-Brockand Harvey 1979; Gould and Lewontin 1979; Trivers 1985; Coddington 1988; Baum and Larson 1991; Brooks and McLennan 1991; Ridley1992). More recently, the comparative method has also beenused to iden- tifytraits that are apparentlynot adaptive and theconstraints that cause themto be so (Coddington1988; Baum and Larson 1991;Brooks and McLennan1991). Thatis, comparativemethods are now beingused to revealthe mechanismsof evolution.This articleexamines the foundation of thisresearch strategy. First,we mustdefine the scope ofour inquiry. What is meantby "the mechanismsof evolution,"and whatkinds of studiesmake claimsabout them?By mechanismsof evolutionwe meanall thoseforces, such as geneticdrift, natural selection,mutation, or migration,that mold the evolutionof populationsor clades. We thinkthat claims about such mechanisms are mademore often than is readilyapparent, for instance, whenever a traitis identifiedas an adaptation. Thi-sposition may seem peculiarat first.After all, a distinctionis oftendrawn betweenadaptation as a processand adaptationsas traits(Futuyma 1986). But, as manydiscussions of adaptationand adaptationsshow, this distinction tends

* To whom correspondenceshould be addressed. E-mail: [email protected].

Am. Nat. 1994. Vol. 143, pp. 381-402. C 1994 by The Universityof Chicago. 0003-0147/94/4303-0002$02.00.All rightsreserved.

This content downloaded from 128.103.149.52 on Tue, 03 Mar 2015 16:41:08 UTC All use subject to JSTOR Terms and Conditions 382 THE AMERICAN NATURALIST to dissolve as these termsare more rigorouslydefined. Sober (1984, p. 208), for example, characterizesan adaptationas follows: "A is an adaptationfor task T in populationP ifand only ifA became prevalentin P because therewas selection for A, where the selective advantage of A was due to the fact that A helped performtask T." By Sober's definition,adaptations (as states) are, in part, de- finedby the evolutionaryprocess-direct selectionfor those traits-that gave rise to them. Some definitionsof adaptationdistinguish between the causes of the historical genesis of a traitand its currentmaintenance by selection: "we may designate as an adaptation any featurethat promotes fitness and was builtby selectionfor its currentrole" (Gould and Vrba, 1982, p. 6). This is a more historicalview of adaptation than Sober's. As such, it lends itselfnaturally to a comparativeor phylogeneticcontext, and as the analysis of interspecificpatterns has become more sophisticatedover recent years, it has been widely adopted. Coddington (1988, p. 3), forexample, definesan adaptationas an "apomorphicfunction pro- motedby naturalselection" and states elsewherethat "an evolutionarydefinition of adaptationmust have an historicalcomponent specifying selection as the evolu- tionaryagent responsible for the appearance of the feature" (p. 5). In theirdefini- tionof an adaptation,Baum and Larson (1991) add to thisby explicitlyincorporat- ing the idea of a trait'sperformance; stated in a way comparable to Sober's, it mightread as follows: A' is an adaptation for task T in clade C if and only if (1) A' is currentlymaintained by natural selection in C because of its superior performancea' at task T relativeto the ancestralcondition A and its performance a, and (2) A' originallybecame prevalentin C because of selectionfor its superior performanceax' at task T. Baum and Larson (1991) definemaladaptive and selec- tivelyneutral traits in an analogous fashion:if the performanceof the derivedtrait is lower thanthat of the ancestralcondition (at' < at), thenA' is a "disaptation"; if the derived performanceis equivalent in in-groupand out-grouptaxa (a-' = at), thenA' is a "nonaptation." The importantpoint is that, like Sober's, all these definitionsof adaptations are foundedon statementsabout an evolutionaryprocess: naturalselection. Given thatadaptations can be definedin principle,how should theybe identi- fiedin practice? Most biologistswould agree that,at a minimum,the functionof a putativeadaptation should be well understood.Many otherswould add thatthe phylogeneticdistribution of a traitshould also be known.Baum and Larson (1991) attemptto apply theirdefinition. They argue thatidentification of thephylogenetic distributionof a derived trait,its performancerelative to the ancestraltrait, and the "selective regime" thatputatively influences its evolutionare necessary and sufficientto distinguishbetween adaptive and nonadaptivetraits. This position, or one similarto it, is held by Greene (1986) and Coddington(1988), as well as Harvey and Pagel (1991, p. 11), who assert, "Stripped to the bone, . . . the evidence for adaptive evolution revealed by comparative studies is correlated evolution among charactersand environments."It is an attractiveview, for it offersthe hope thatclades can be inventoriedfor adaptations, nonaptations, and disaptationswithout having actually to measure natural selection operatingon traits.

This content downloaded from 128.103.149.52 on Tue, 03 Mar 2015 16:41:08 UTC All use subject to JSTOR Terms and Conditions COMPARATIVE METHODS AND ADAPTATION 383

In this article,we take issue withthis position. We will argue thatknowledge ofthe phylogeneticdistribution of traits,their performance, and selectiveregimes are not sufficientto identifyadaptations. This is not a completelynovel perspective; both Mishler(1988) and Brandon (1990) have statedthat a completeadaptive explanationrequires both a demonstrationof the relative fitnessof a putative adaptationand the traitit supplants,as well as genetic information.We agree. But we will go furtherthan these authorsand argue that,in the absence of such broader information,phylogenetic patterns will often suggest that a traitis an adaptationwhen in fact it is not and suggestthat it is not, when in fact it is. We begin our argumentwith a discussion of theoreticaland empiricalfindings frompopulation genetics and show how the effectsof selection, patternsof geneticcovariation, genetic background, environment, and the interactionof genotypes withtheir environment influence the observed distributionof traitsamong populations,species, and highertaxa. Given these factors,we reevaluate recent articles-both methodologicaland specificcase studies-that claim to inferadap- tationby the comparativemethod. Finally, we consider the general implications of our critique,particularly with respect to the circumstancesin which the comparativemethod can legitimatelybe used to analyze adaptationand evolutionary constraint.

POPULATION GENETICS OF MULTIPLE FITNESS-RELATED CHARACTERS It is our primaryintent in this section to show how any single comparative patterncan arise by means of very differentevolutionary genetic mechanisms. Formally,our basic argumentscan be developed in terms of a two-character, two-allele,one-locus model, thoughlater we consider otherloci and characters in the developmentof our argument.Say thatwe have two alleles at a locus, BI and B2, that can affecttwo distinctcharacters, Cl and C2, with any patternof dominance: Genotype B1B1 BIB, BIB, H +3, CharacterChaacerC1 Cl HIH2 +4PI+ YI HIH2 H2I + Y2

The Pi and -yispecify the allelic effectsand may have any sign; Hi specifiesthe phenotypiceffect of the genetic backgroundand may also have any sign. Some trivialpoints follow immediately.If the pi are both zero and the B locus is the only one affectingC1, then there will be no genetic variationfor C1 even when this locus is polymorphic.In quantitativegenetic notation, VG(C I) = 0 when all P = 0. Similarconclusions followfor C2. However, if at least one Pi3and at least one yi are differentfrom zero, then VG(CI) > 0 and VG(C2) > 0 when there is genetic polymorphism.Of greater interest,there will be a genetic correlation betweencharacters C1 and C2 when thereis segregatinggenetic variation affecting both at locus B. The particularvalues for the genetic variances and correlation will depend on the gene frequenciesof B, and B2 (since the total gene frequency

This content downloaded from 128.103.149.52 on Tue, 03 Mar 2015 16:41:08 UTC All use subject to JSTOR Terms and Conditions 384 THE AMERICAN NATURALIST mustequal one, thereis only one gene frequencyto keep trackof in a two-allele system). Let p designate the allele frequencyof B1, and assume that there is random matingwith respect to the genotypes at this locus. In this model, the values of the charactersC1 and C2 in any species are determinedby the values of Hi, i, yi, and p. The distributionof C1 and C2 across a set of populationsor species will, likewise, depend on the distributionsof these variables. We discuss the influenceof these variables on interspecificpatterns next. InterspecificPatterns Due to Pleiotropy,Epigenetics, and Correlated Selection Consider one of the simplestpossible patternsof interspecificassociation of C1 and C2: a positiveone. How mightthis be explained?At least threeelementary hypotheses,of a sort common in the comparativeliterature, present themselves; we call these thepleiotropy, the epigenetic,and the correlatedselection hypotheses. The pleiotropyhypothesis is so called because it invokes, on the basis of a presumedfunctional connection between C1 and C2, a positivegenetic covariance between these traits,arising from pleiotropy. In this scenario, selection acts on one of these charactersalone. In termsof the previouslycited model, the allelic effectsmight be t2 > P and Y2 > yi' and fitnessmight be a linear functionof C2 but have no dependence on C1. A range of species with varyingselection historieswould differin p: some would be fixedfor B1, some fixedfor B2, and some would be polymorphic.This conditionwould give rise to an interspecific association among C1, C2, and fitness. The epigenetic hypothesisis similar to the pleiotropyhypothesis in that the interspecificcorrelation between C1 and C2 is a consequence of selectionon one of these traits and the correlated evolution of the other. Here, however, the functionalconnection-which is a consequence of development-gives rise to a complex kind of pleiotropy:an epigenetic interaction(sensu Atchley and Hall 1991). An epigenetic interactionmight occur if, say, C1 were to influencethe developmentof C2, perhaps throughan inductiveevent, but C2 were to have no influenceon the developmentof C1. This means that most genes that affectC1 would also affectC2: i not equal to zero would necessarilyimply yinot equal to zero and vice versa. But forgenes thataffect C2, yinot equal to zero implies nothingabout i. This functionalasymmetry gives rise to an asymmetryin genetic correlations:C2 would be geneticallycorrelated with C1 when the latterhas genetic polymorphism,but the reverse would not be necessarilytrue (Atchleyand Hall 1991). This situation,in turn,could give rise to an asymmetryin selection responses depending on the genetic effectsand the charactersundergoing selection. The correlatedselection hypothesisin its simplestform postulates thatfitness is a linearfunction of traitsC1 and C2 in a given environment,but C1 and C2 are not associated by eitherphenotypic or geneticcorrelations. Here, locus B might influencecharacter C1 but not C2; thatis, the Pi are not equal to zero, but the -y are equal to zero, while alleles influencingonly C2 segregateat other loci. For instance, the activityof a glycolyticenzyme and a biomechanical traitsuch as stridelength, in both contributingto maximal runningspeed, mightbe selected

This content downloaded from 128.103.149.52 on Tue, 03 Mar 2015 16:41:08 UTC All use subject to JSTOR Terms and Conditions COMPARATIVEMETHODS AND ADAPTATION 385 upward togetherunder conditionsthat select for runningspeed. In such a case, selection will, over a range of species, also give rise to a positive interspecific correlationbetween C1 and C2 and fitness,despite an absence of genetic correlation. The pleiotropy,epigenetic, and correlatedselection hypothesesdiffer in their assumptionsabout the geneticand phenotypiccovariance structureof the evolv- ing species, as well as differingin theirassumptions about the historicalforces of selectionacting on each trait.Comparative data offerno hope of discriminating among these evolutionarygenetic hypothesesfor the simple reason thatnothing can be inferredabout patternsof phenotypicand geneticcorrelations within popu- lationsfrom population (or species) means. Functionalanalysis of the characters will also be generallyunhelpful, for a mechanical association between two traits may be reflectedin either or both of the genetic or the phenotypiccorrelation structureof the characters. Conversely, patternsof genetic and phenotypiccovariance need not be dependent on obvious mechanical associations. These hypothesesare but the most elementaryof an enormous range of evolutionary geneticmechanisms that can give rise to any given interspecificpattern of association. InterspecificPatterns Due to EnvironmentalCovariance We can extend this model to include phenotypicresponses to environmental variationof each character: Genotype BIB, BIB, BB2

Chacte C, HI + I1 + 8(E) HI A+ (E) HI + 2 + a(E) aracter C2 HI + YI + e(E) H2 + e(E) H, + Y2 + e(E) The 8 and E specifythe response of each characterto environmentalvariation (E). If the 8 and E termsare of the same sign over the range of environments, thenC1 and C2 respond to environmentalvariation in a similarway; thatis, they have a positive environmentalcovariance. In the absence of geneticcovariance (pi = yi = 0 at all loci), 8 and E will determinethe sign of the phenotypic covariance between C1 and CG. Environmentalcovariances, like geneticcovari- ances, may reflectfunctional connections between traits. The extent to which this is importantin the comparative patternof multipletraits depends on the extentto which environmentalvariation is allowed to obscure the geneticdiffer- ences among populations; if all populations compared are reared in a common environment,then this criticismhas no force. Comparative data are, however, oftenbased on organismsreared under very differentenvironmental conditions. When this is so, inferencesabout evolutionaryprocesses become dependenton physiologicalhomeostasis in a way thatis unpredictablein the absence of exten- sive knowledgeof the comparativephysiology of the species concerned. InterspecificPatterns Due to Driftand Linkage Even ifneither of the observed charactersis relatedto fitness(either by genetic effector by statisticalassociation withother allele frequenciesor fitnesscharac-

This content downloaded from 128.103.149.52 on Tue, 03 Mar 2015 16:41:08 UTC All use subject to JSTOR Terms and Conditions 386 THE AMERICAN NATURALIST ters), an interspecificcorrelation could still arise due to genetic drift.If C1 and C2 are the only charactersaffected by locus B, and alleles B1 and B2 are neutral, p will-be determinedby mutationpressure and genetic drift(see Lande 1979; Kimura 1983). Such purely neutralevolution is unlikelyto generate consistent comparativepatterns for any single characteror pair of characters.However, if many charactersare surveyed, then there is the possibilityof neutralevolution generatinga spuriouslysignificant comparative association, when proper allow- ance forfalse positives is not made (Felsenstein 1985). Of course, such characters mightstill be useful for reconstructingphylogenetic patterns among taxa. It may also be the case thatCl and C2 are not fitness-relatedcharacters but have accidental associations withother fitness-related characters, perhaps because Qf linkage disequilibrium,as reflectedin a transientgenetic correlationwith each otherand fitnesscomponents. In these cases, the values of p, Cl, and C2 will be determinedby accidents of samplingin the evolutionof each population.Unless locus B (and similarlyfor all otherloci affectingthe two characters)is extremely close to a locus under selectionor partof an inversioncontaining other loci under selection,then the evolutionof Cl and C2 over a rangeof disparatespecies should not follow any discerniblepattern. In the case of the Adh locus in Drosophila melanogaster,in whichpatterns of linkagehave been studiedintensively, linkage disequilibriumdecays rapidlywithin the locus (Kreitman 1983), which suggests that physical proximityof loci is not likely to be a factor in the evolution of charactersunrelated to fitness,at least forspecies witheffective population sizes comparable to those of D. melanogaster. However, evolutionaryunits such as mitochondria,inversions, and small chromosomes with suppressed recombina- tion all mightgenerate historicalassociations over groups of species in which these unitsare preserved.It is an open question how importantsuch associations are in the groups of species normallyused in comparativeresearch.

InterspecificPatterns Due to Evolution of the Genetic Background Species may vary in theirgenetic backgrounds because of a numberof causes and do so in such a way as to obscure or even alter the geneticrelations among traits.Since selection for eithercharacter will tend to alter the otheras a correlated response, it may be expected thatthe correlationbetween these two characters would be preserved among species. But this need not be so. Consider, for example, a case of antagonisticpleiotropy (see Rose 1982), in which P1 > 0 > 'Yi and Y2 > 0 > P2 Let fitnessbe an increasingfunction of C1 and C2, of some form(Rose 1982). With these genetic parameters,it will oftenbe the case that overdominancearises, and thus does protectedpolymorphism (Rose 1982, 1983). Moreover,at polymorphicequilibria, the additivegenetic correlation between the two characterswill be - 1. In otherwords, thisis a simplecase of an evolutionary trade-offbetween two fitness-relatedcharacters. One potentialproblem is otherloci forwhich the signs of the P3iand 'Yiare the same. Houle (1991) has considered cases of this kind,in which C1 and C2 are the finalproducts of a metabolic pathway that branchesjust before the characters. Thus, thereare allocative effectson the relativebalance between the two characters at the branchingpoint that could give rise to genetic effectslike those that

This content downloaded from 128.103.149.52 on Tue, 03 Mar 2015 16:41:08 UTC All use subject to JSTOR Terms and Conditions COMPARATIVE METHODS AND ADAPTATION 387 arise when the 3iand yiare of opposite signs,and acquisitionaleffects enhancing both characters,which would give rise to loci with geneticeffects that have the same sign. While the lattertype of locus would not have geneticpolymorphism sustained by selection, the fixationof differentbeneficial alleles of this type among differentspecies would result in genetic background parameters that would positivelyco-vary. If the magnitudeof the variance generatedbetween species by such substitutionsis sufficientlygreater than thatgenerated between species by differentvalues forp at allocative loci, then the existence of a functional trade-off,and an associated negative genetic correlationamong species, between the two characterscould be wholly obscured. Species may differin genetic background because of one of at least three causes. First,the amountand kind of geneticvariation may differamong species such that identical ecological pressures will give rise to differentevolutionary responses. Withrespect to the previouslymentioned case, it could be that,even thoughincreased acquisition is stronglyselected for in all species, only some species have the geneticvariation required for an evolutionaryresponse. Second, even thoughspecies may have identical classes of mutationsavailable to them, subtledifferences in ecological conditionscan cause gene frequenciesto converge on differentequilibria. Species selected just for varyingdegrees of acquisition will not show any trade-offin how they allocate theirresources. Third (though this is probably the least importantcause in nature), species may differin the degreeto which theyexperience inbreedingdepression. If the numberof acquisi- tiveloci is largerthan the numberof allocative loci, thenthe fixation of deleterious alleles withinbreeding should tendto resultin a positivecovariation for functional characters,in the same way that inbreedinglaboratory lines will (Rose 1984a). Variationin the overall degree of inbreedingdepression can be due to variation in historicaleffective population sizes (Ne) or, ifthese are similaramong species, the stochasticfixation of differentdeleterious alleles at differentloci in separate evolutionarylineages. This confoundinginfluence of geneticbackground on interspecificcomparisons is not restrictedto characters that are the end points of branched metabolic pathways. It arises whenever charactersare influencedby multipleloci, associated by complex patternsof pleiotropy,and undercomplex patternsof selection. We believe that these conditions will be met for many of the traits that are commonlythe subject of comparativeanalysis and thatit is merelythe absence of informationabout the geneticbasis of such traitsin most organismsthat permits comparativebiologists to ignore the vagaries of genetic structurein theirinfer- ences of evolutionarymechanism via comparativeanalyses. Much, however, is known about the geneticbasis of fitness-relatedtraits in at least one species, Drosophila melanogaster. Quantitativegenetic studies have revealed over the past decade thatpatterns of pleiotropyare pervasive and unpre- dictableamong charactersat all levels of organismalanalysis. For example, replicated strainsselected forlate-life fecundity in various laboratoriesexhibit correlated increases in longevityand flightendurance (Rose 1984b; Luckinbill et al. 1988) as well as stress resistance traitssuch as ethanol,desiccation, and starva- tion resistance (Service et al. 1985, 1988). These same strainsshow declines in

This content downloaded from 128.103.149.52 on Tue, 03 Mar 2015 16:41:08 UTC All use subject to JSTOR Terms and Conditions 388 THE AMERICAN NATURALIST early-lifefecundity (Rose 1984b), early-lifemale competitivemating ability (Ser- vice 1993), and early-lifemetabolic rate (Service 1987). The increases in stress resistanceand flightendurance are, in turn,associated withincreases in lipid and glycogenmetabolite pool sizes (Service 1987; Graves et al. 1992). If, as these studies suggest,pleiotropy is commonin large outbredpopulations, then the comparativeapproach is weakened as a methodfor inferringadaptive mechanisms.For not only does it become very difficultto attributeinterspecific evolutionin any given traitto selectionfor that trait (rather than for another with which it is geneticallycorrelated), but, as we argued for the case of a Y-shaped metabolic pathway, selection at one traitcan obscure the genetic relationships thatexist between other traits when viewed across species. Finally,the possibility exists that the genetic backgroundmight evolve in such a way so as not merely to obscure the genetic relationshipbetween traitswithin any given population but actuallyto alter them. Since genetic correlationsare dependenton allele frequenciesas well as patterns of allelic effects,selection (or drift)can, in principle,cause them to be highlyunstable in the course of evolution.The degree to whichthey are unstable has been the centerof theoreticaldisagreement for some time(Turelli 1988; Slat- kin and Frank 1990), and evidence bearing on their stabilityis scattered and weak. Genetic correlationshave occasionally been foundto evolve in laboratory populations.In the course of subjectingDrosophila populationsto two-traitselection, Sheridanand Barker (1974) demonstratedthat genetic correlations between coxal and sternopleuralbristles may diverge,though not necessarilyin a manner congruentwith theoretical expectations. Genetic correlationshave also been foundto respond to one-traitselection in Triboliumand Drosophila populations (see Wilkinsonet al. 1990). Much evidence forthe labilityof geneticcorrelations in naturalpopulations is ambiguous(see, e.g., Arnold 1981; Lofsvold 1986; Kohn and Atchley 1988; Cowley and Atchley 1990; other referencesin Barton and Turelli 1989), but there is at least one demonstrationof evolved-differencesin genetic covariance structure(Atchley et al. 1992). The intensityof directional selection required to change the signs and magnitudesof genetic correlationsin the laboratoryis not beyond the range of those found in nature (Endler 1986). The possibilityexists, then, that such correlationschange frequentlyin the course of the divergence of sister lineages, in which case a comparison of the mean phenotype of such lineages will reveal nothingof the genetic relations among traitsthat exist withineach particularlineage, at any one time. Genotype-by-EnvironmentInteraction and the Genetic Background The confoundingeffect of geneticbackground becomes even more acute when interactionsbetween specificalleles and geneticbackgrounds exist or when spe- cificgenetic backgroundsinteract with environment.The firstof these types of interaction,epistasis, is generallythought to be less pervasive than simple addi- tive backgroundeffects. Yet fromhis studies of gene action in guinea pig color- ation and morphology,Wright (1977) believed thatepistasis was importantin the evolutionof many traits,and it occupied a centralplace in his shiftingbalance theoryof evolution. Modern metabolic control theoryhas confirmedhis view

This content downloaded from 128.103.149.52 on Tue, 03 Mar 2015 16:41:08 UTC All use subject to JSTOR Terms and Conditions COMPARATIVEMETHODS AND ADAPTATION 389 thatepistasis is a systemicproperty of multistepmetabolic pathways (Keightley 1989). Epistatic effectsmay cause the effects(the Pi and y) of particularalleles to alter, taking,for instance, positive values on some genetic backgroundsbut negativevalues on others,and so to alter genetic correlations. While epistatic interactionsare commonlyfound among mutationsof large effecton arbitrarytraits (Wright1977) or in crosses of inbred lines (see, e.g., Jinkset al. 1973; Clark and Feldman 1981; Clark et al. 1981; Seager and Ayala 1982), far less is known about theirinfluence on fitness-relatedtraits in outbred populations(Crow 1987). In a few cases, however,epistatic modifiers have been shown to alter patternsof pleiotropyin laboratorypopulations. Clarke and Mc- Kenzie (1987) showed thatalleles thatconfer resistance to the insecticidediazinon in the blowfly,Lucilia cuprina, have the pleiotropiceffect of increasingdevelop- mentalinstability (as measured by fluctuatingasymmetry). These pleiotropicef- fectsare not, however,reflected in negativecorrelations between Diazinon resistance and these fitness-relatedtraits across populations;rather, Diazinon-resistant populations have compensated in some fashion for the deleterious pleiotropic effectsof resistance (McKenzie et al. 1982; Clarke and McKenzie 1987). This compensationwas shown to be due to the particularcomposition of the genetic backgroundof Diazinon-resistantflies-that is, due to evolved epistaticmodifi- ers. In a similarcase, Lenski (1988a, 1988b) showed thatmutations in Escherichia coli thatconfer resistance to the virus T4 have deleteriouspleiotropic effects on competitivefitness and thatthese effectscould be modifiedby subsequentevolu- tion in the absence of T4. Remarkably,the increase in fitnessin the resistant strainswas not attributableto a reversionat the resistance locus but ratherto the evolution of an allele elsewhere in the genome that epistaticallydiminished the cost of resistance. It is well-knownthat the expression of genetic relations(such as pleiotropy) can vary among environmentsbecause of interactionsbetween the environment and the collectionof genotypessegregating within populations (Service and Rose 1985; de Jong 1990; Stearns et al. 1991). Much the same phenomenoncan arise owing to interactionsbetween the environmentand the differentiatedgenetic backgroundsof a collection of species or populations. Indeed, many instances are knownof relatedpopulations having different patterns of environmentalsensi- tivityor "reaction norms" (see, e.g., Berven and Gill 1982; Schlichtingand Levin 1986). In this case, species thathave similarevolutionary genetic mechanisms at work in theirnormal environments could exhibitcomparative patterns in a common assay environmentthat do not reflectthese mechanisms.In particular,species can be expected to differwith respect to theirfunctional attributes under novel or stressfulconditions (Hoffmannand Parsons 1991) and so may be expected to exhibitpositive interspecific covariation for such traitsin some artificial or laboratoryenvironments. The Necessity of Skepticism Despite thislist of ways in whichvarious evolutionarygenetic mechanisms can be confoundedwith each other, there are, no doubt, cases in which the pattern of interspecificdifferentiation does reflecta simple evolutionarygenetic mecha-

This content downloaded from 128.103.149.52 on Tue, 03 Mar 2015 16:41:08 UTC All use subject to JSTOR Terms and Conditions 390 THE AMERICAN NATURALIST nism. For example, Arnold (1981) found that the chemoreceptiveresponse to slug and leech odor was positivelyassociated between an inland and a coastal populationof gartersnakes. His explanationof this association was of the sort thatwe have called the pleiotropyhypothesis: a geneticcorrelation (in this case positive) combined with divergentselection (forchemoreception of slug odor in the coastal population and against chemoreceptionof leech odor in the inland population). What makes Arnold's scenario more persuasive than most of this sortis the demonstrationthat the geneticcorrelation between these two traitsis stronglypositive and identical in each of the populations. Even here caution is required,however, for the population parameterswere estimatedin a common laboratoryenvironment and are confoundedwith differences in maternalenviron- ment.The influenceof this novel environmenton the populationmeans and geneticcorrelations is completelyunknown. Nevertheless, in providingthe information required to weaken the most elementary competing hypothesis (e.g., correlatedselection), this studysuggests a routeout of the thicketof confounding mechanisms. Very little evidence bears on the evolutionarygenetic mechanismsthat are responsiblefor distributions of traitsamong species, so it is difficultto say which of the mechanisms discussed earlier are most likely to be confoundedin any given study. We do not intendhere to make a claim for the importanceof any one patternof selection, allelic effects,or genetic structureover another. We claim only the necessityof a healthyskepticism when evolutionaryscenarios are offeredto explain comparativepatterns in the absence of directevidence about historicalselection forces and the geneticrelations among traits (see also Cracraft 1981).

DUBIOUS INFERENCES FROM COMPARATIVE DATA In this section we illustratethe precedingpopulation genetics arguments using examples of actual comparativestudies. We documentthree classes of erroneous inferencebased on comparativedata and the historicaldefinition of adaptation.

UnwarrantedInferences of Adaptationfrom Correlations between Characters and Environments Ecomorphologicalcorrelations have long been used as evidence forthe adaptive natureof organismalform (see, e.g., Cody and Diamond 1975; Bock 1977; Luke 1986; Brooks and McLennan 1991). Under the best possible situation,the repeated (phylogeneticallyindependent) occurrence of particularmorphological featurescongruently with a specificenvironmental factor may provide evidence of a causal relationshipbetween environmentand morphology-ifthe correlation is high and if the environmentalfactor is precisely identified.One such study (cited as "exemplary" in Mayr 1983) is Traub's (1980) demonstrationthat a par- ticularpattern of bristlemorphology in fleas (the "crown of thorns" phenotype) has evolved independentlyby differentdevelopmental routes in a number of lineages. This trait,thought to assist fleas in attachmentto the host, is usually foundin species thatparasitize hosts thatlive in habitatsespecially hazardous to

This content downloaded from 128.103.149.52 on Tue, 03 Mar 2015 16:41:08 UTC All use subject to JSTOR Terms and Conditions COMPARATIVE METHODS AND ADAPTATION 391 fleas and in which there is presumablya selective premiumfor stayingon the host. Anotherbroadly comparative study is that of Luke (1986), who showed that lizard toe fringeshave evolved at least 26 times in seven families. Fringe morphologyis generallycorrelated with the type of substrate,although several differenttypes of toe fringesmay all be found in species inhabitinga common environment.However, it is possible to argue that some type of toe projection maybe adaptive duringlocomotion in sand, withoutspecifying the exact morphology of the projection. In fact, few analyses in the literaturefit these best-case situations.Often the correlationthat is presentedis between a morphologicaltrait and a generalenvi- ronmentalcondition that has originatedbut once in phylogeny.The example of the plethodontidsalamanders discussed by Baum and Larson (1991) as the pri- maryexample of theirmethodology for analyzingadaptations (see also Larson et al. 1981; Larson 1984; Wake and Larson 1987) illustratesthese difficulties.The morphologicalnovelty of altered tarsal structurein the genus Aneides is considered to be an adaptation conferringenhanced climbingability relative to the ancestral condition and is correlated with the origin of an arboreal/scansorial habit. The correlationis, however, based on only two data points: both the morphologicalnovelty and change in habit arose once at the same internodeinterval on the phylogeny.This type of single historicalevent involvinga change in both morphologyand environmentis one thathas verylow power in historialanalysis (Lauder 1981): given a single event, a virtuallylimitless number of scenarios may be outlinedthat fit the observed datum. Fortunately,this problemhas been addressed in at least some of the recent comparative publications (see, e.g., Harvey and Pagel 1991). A second difficultywith such comparativeand historicalmethods of analyzing adaptationis thatoften the environmentis vaguelytreated. The notionof a selective regime(Baum and Larson 1991) or an "adaptive zone" (Simpson 1953; Van Valen 1971; Larson et al. 1981) is so general as to be useless in attemptingto identifyselection forces that have acted on morphology.Experimental population biology has repeatedlydemonstrated the sensitivityof phenotypicresponses to smallchanges in the natureof the imposed selectiveregime (see, e.g., King 1955). To use a selective regimesuch as "arboreal/scansorialhabit" as a characterfor the purpose of discerningadaptation combines so many distinctbiophysical fea- tures of the environmentinto a single label that comparativetests using such a characterare likelyto have low resolvingpower.

The Invocation of UndemonstratedConstraints on Adaptation in the Explanation of PhylogeneticConservatism Gould and Lewontin (1979) set comparative biology the task of elucidating "phyletic" and "developmental" constraintson adaptation,and not one of the recent expositions of the comparative method is withouta discussion of such constraints(Coddington 1988; McLennan et al. 1988; Baum and Larson 1991; Brooks and McLennan 1991; Harvey and Pagel 1991). Constraintsarguments are especially evidentin explanationsof phylogeneticconservatism. Coddington (1988, pp. 18-19), forexample, states that, "although several exactly specifiable

This content downloaded from 128.103.149.52 on Tue, 03 Mar 2015 16:41:08 UTC All use subject to JSTOR Terms and Conditions 392 THE AMERICAN NATURALIST convolutionsof the ejaculatory duct withinthe male secondary genitaliaof the theridiosomatidspiders definesthe family. . ., it is unlikelythat anyone would assert that this extraordinaryconservatism is caused by relentless stabilizing selection,on the route of the duct in all theriosomatidlineages." He continues, "The darwinianparadigm is best tested at low taxonomiclevels. At such levels, truenovelties may stillbe maintainedby stabilizingselection in all lineages. Their conversion into developmental 'constraint' and phylogenetic'inertia' may not have yet happened." In a similarfashion, Brooks and McLennan (1991, p. 166) suggestthat the presence of a male-territorial,polygynous, paternal-care mating systemin all species of gasterosteidfish is a "reflectionof tightphylogenetic constraintson mating-systemevolution within this family,"offering as evidence againstan adaptive explanationthe observationthat the Gasterosteidaelive in a wide range of habitats. Doubts have recentlybeen raised concerningthe coherence of notions such as phylogenetic,developmental, and genetic constraints(Antonovics and van Tienderen1991; but see Arnold 1992). Those who invokethem to explainphyloge- netic conservatismseem to believe one or more of the following.First, many traitsmay lack sufficientgenetic variation for selection,stabilizing or otherwise, to work on. This may, in turn,be due to the "canalizing" influenceof modifier loci (Waddington1975). Second, strong"coadapted" gene complexes or epistatic interactionsamong loci may ensure thatnearly all variationthat arises is severely deleterious.To use Wright'smetaphor, species are on peaks in a ruggedfitness landscape fromwhich it is difficultto shiftthem (Wright 1977). Third,if thereis variation,it is bound up in strongpleiotropic relations among traitsthat, cumula- tively,prevent selection on any one component of the organisms' phenotype (Coddington1990; Baum and Larson 1991; Harvey and Pagel 1991). How plausible is each of these explanations for phylogeneticconservatism? There are at least a few examples of phylogeneticallyconserved traitsfor which some evidence indicates that stabilizingselection does not play -a large role in theirmaintenance. Very few dipterans,for instance, show any formof directional asymmetry.Repeated laboratoryselection experiments on Drosophila, forexample, have failed to detect any heritablevariation for directional asymmetry (see, e.g., Tuinstraet al. 1990). On the otherhand, a possible example of a traitphylo- geneticallyconserved by stabilizingselection is providedby naturalexperiments in clades with cave-dwellingtaxa. The presence of eyes is uniformin the fish familyCharacidae, except forpopulations of the blindcave fishof Mexico, Astya- nax mexicanus, that were isolated fromtheir sighted surface dwelling relatives about 10,000yr ago. It is unclear whetherthis example of recessive evolutionis due to mutationpressure, direct selection for eye loss, or selection on some pleiotropiceffect (Culver 1982), but it is a strikingdemonstration of the presence of geneticvariation for a traitthat is ancient,conserved at hightaxonomic levels, and functional. Directional asymmetryin Drosophila is remarkablein possessing little se- lectable variation;few othertraits subjected to selectionin the laboratoryfail to respond (Maynard Smith 1989). Just as a lack of genetic variation cannot be generallyheld responsiblefor stasis in thefossil record (Charlesworth et al. 1982),

This content downloaded from 128.103.149.52 on Tue, 03 Mar 2015 16:41:08 UTC All use subject to JSTOR Terms and Conditions COMPARATIVEMETHODS AND ADAPTATION 393 so too it cannot explain stasis in phylogenies.We suspect that strikingcases of phylogeneticconservatism are the product of complex patternsof epistasis and pleiotropy,manifest as stronggenetic correlations among traits and complex patterns of correlated selection. In morphologicaltraits, such patternsarise as a consequence of development (Arnold 1992; Atchley et al. 1992). Sinervo and Licht (1991) experimentallydemonstrated complex patternsof selection in life- historytraits. They showed that the pelvic girdle dimensionsof a lizard, Uta stansburiana,impose an upper selective limitto egg size and, since clutch size and egg size trade offphenotypically, a lower selective limitto clutch size. Al- thoughthe question remainsopen as to what, in turn,limits pelvic girdledimen- sions, it is easy to see how selectionfor, say, locomotorperformance, could have ramifyingeffects on life-historytraits. Possibly such patternsof selection are responsiblefor strikingcases of conserved clutch sizes such as thatof the Gek- konid lizards, 600 species of which invariablylay two eggs (Stearns 1984), and those of the genus Anolis (anoles), which invariablylay but one (Sinervo and Licht 1991). The importantpoint is thatthere is no obvious way of distinguishing amongthe several notionsof "constraint" withoutmeasuring current and historical patternsof selection,genetic variation,and geneticcovariation. UnwarrantedInferences of Adaptation and Constraint from Correlationsamong Characters As discussed earlier,there are also at least threedistinct evolutionary genetic explanationsfor interspecific correlations among traits.We have called these the pleiotropy,epigenetics, and correlatedselection hypotheses.While the notionof "epigenetics" (sensu Atchleyand Hall 1991) is relativelynew and has not been used extensivelyin the comparativeliterature (but see Atchleyet al. 1992),pleiotropy and correlatedselection have long coexisted in the comparativeliterature as explanations for interspecificassociations among traits. Curiously,they are rarely encountered as alternativeexplanations for any single pattern. On the one hand, motivatedby particularnotions of the physiologicalor developmental relationsamong traits,comparative biologists have oftensearched the phylogenetic recordfor the influenceof pleiotropywrit large. Traits relatedto body size and lifehistory have oftenbeen subject to thissort of interpretation.On the other hand, a glance at any currenttext on behavioral or evolutionaryecology will reveal numerousadaptive hypothesesconcerning two or more traitsin which no mentionis made of the genetic relationsthat the traitsmay bear to each other. In this section we counterpose alternativeevolutionary explanations for two of thebest-documented and best-understoodcomparative patterns available, namely brain-bodyallometry and the effectsof costs of reproductionon lifehistory. Brain-bodyallometry.-The scaling of brain and body size over ordersof pla- cental mammalscan be described as a power function(Gould 1977; Lande 1979): log(Brain) = log(k) + b log(Body), where0.67 < b < 0.75. The allometriccoefficient, b, which is the slope of the linear relationshipbetween log-transformedbrain and body size, is sometimessmaller at lower taxonomic categories than at higher(e.g., the slope best describingthis relationship

This content downloaded from 128.103.149.52 on Tue, 03 Mar 2015 16:41:08 UTC All use subject to JSTOR Terms and Conditions 394 THE AMERICAN NATURALIST may be shallower among or withingenera, than among or withinorders; Lande 1979; Pagel and Harvey 1989). The interspecificrelationship between brain and body size and its variationamong taxa or taxonomic levels are unusual among such patternsin that somethingis known of the genetic relationshipsbetween these traits(Roderick et al. 1976; Atchleyet al. 1984). Furthermore,this pattern has been the subject of a numberof explicitquantitative genetic models-some comparable to those discussed earlier in this article (Lande 1979; Riska and Atchley 1985; Riska 1989)-that have attemptedto explain variationin the allometriccoefficient among taxa. Using quantitativegenetic data fromlaboratory mice, Lande (1979) arguedthat relativelysmall allometriccoefficients (b = 0.37) could be explained by selection forbody size alone. Larger allometriccoefficients (b = 0.67), on the otherhand, were closer to those that mightbe expected fromselection forbrain size alone. Riska and Atchley (1985), observingthat the genetic correlationbetween body size and adult brain size in mice changes over the course of ontogeny,modified Lande's hypothesis.They argued that evolutionarychanges in brain size could be explained by selection for body size alone and that differencesin allometric coefficientswere attributableto selection having acted on differentcomponents of body growth.Others have suggestedthat the evolutionof mammalianneonatal brain size of mammals is constrainedby maternalmetabolic rate (Martin 1981) or thatit has evolved withgestation length, litter size, and otherlife-history traits (Pagel and Harvey 1988). The historyof brain-bodyallometry studies illustratesat least three of the problems that we believe to be inherentto the comparative method. The first is that of confoundingselection pressures. Insofar as mammalianbody size is phenotypicallyand geneticallycorrelated with innumerable other traits-possibly includingfitness components such as fecundityand matingsuccess-the claim that selection on adult body size ratherthan brain size is responsible for the patternof allometrywithin a certaintaxon is tantamountsimply to choosing one of a very large numberof possibly equally correlatedvariables, and the same is truefor any claim concerningselection on brain size. The second problemlies in the confoundingof the causal influenceof selection withthat of genetic correlations.In the case of brain-bodyallometry, this error manifestsitself as the notionthat allometric coefficients can reflectthe degree to which one traitconstrains the evolution of another. For example, while Martin (1981) and Pagel and Harvey (1989) disagree as to whetherneonatal brain size and maternalbody weightscale isometrically,both apparentlyfeel thatthe value of b can be used as evidence for or against constrainthypotheses. This is a use of allometriccoefficients that ignores the salutarylesson of Lande's (1979) article, namelythat the constraintson evolution are not allometricrelations but rather the particulargenetic relations among traitsand the historicalselection forces thatgive rise to allometricrelations. Since neithergenetic relations nor historical selectionforces are recoverablefrom allometric relations without external knowl- edge of at least one or the other (Riska 1989), it followsthat the causal effectof the evolutionof one traiton anothercan be neitherconfirmed nor falsifiedby the findingof a particularallometric coefficient.

This content downloaded from 128.103.149.52 on Tue, 03 Mar 2015 16:41:08 UTC All use subject to JSTOR Terms and Conditions COMPARATIVEMETHODS AND ADAPTATION 395

Finally,we suspect thatthe empiricalquantitative genetic foundation of brain- body allometricmodels is less than firm.The estimates of genetic correlations betweenbrain and body size used by Lande (1979) and Riska and Atchley(1985) come frompopulations of laboratorymice. Althoughperhaps betterthan nothing at all, such data probablytell us littleabout the genetic structureof mammalian populationsin general. This observationis because genetic correlationschange withenvironment (Service and Rose 1985) and populationstructure, particularly since laboratorymouse populationsare at least somewhatinbred (see, e.g., Rod- erick et al. 1976) and inbreedingis known to alter genetic correlations(Rose 1984a). In addition,there is the problemof the phylogeneticgeneralization from mice to other species of mammal. We know of no evidence bearingon the con- stancy of genetic correlationsamong brain size, body size, and growthcompo- nents over highermammalian taxa. If any of these factorshave influencedthe patternsof genetic variance and covariance for brain and body size withinthe Mammalia,then there is virtuallyno limitto the combinationsof selectiveforces and geneticparameters that could account equally well forthe observed patterns (Riska 1989). Costs of reproductionand the evolution of life history.-Williams (1966a, 1966b) suggestedthat effortexpended on reproductionearly in life would cause a diminutionof residual reproductivevalue. In other words, early reproduction has a cost, one thatis paid out of subsequentreproduction (e.g., by the reduction of later survival). Williamsalso suggestedthat such a cost of reproductionmight be reflectedin a negativecorrelation over species between some index of reproductive effortand the residual reproductivevalue (of which survivalis the most easily measured component; Williams 1966a, 1966b). Many studies, using data fromnatural populations, have shown a negativerelationship across species between reproductiveeffort and survival(Tinkle 1969; Saether 1988; Harvey et al. 1990; Promislowand Harvey 1990). Are comparativestudies of thissort adequate tests of Williams's theory?If we understandthe theoryto be a statementabout evolutionarymechanism, specifically, the presence of negative genetic correlations among early and late reproductiveoutput, then, for at least threereasons, we believe theyare not. First,in commonwith comparative studies of brain-bodyallometry, these studies inevitablyconfound the effectsof correlatedselection on the distributionof traitsamong species withthose of geneticcorrelations. Are negativecorrelations amonglife-history traits across species the productof a truecost of reproduction (an antagonisticpleiotropy), or are they due to simultaneousdirect selection on separatelife-history traits? Short of the quantitativegenetic analysis of individual species, there seems to be no way to tell. Second, the effectsof environmentaldifferences among species on correlations among life-historytraits are oftenconfounded with those due to genetic differences (see, e.g., Harvey and Zammuto 1985; Saether 1988; Promislowand Har- vey 1990). This problemis especially acute in studies of life-historytraits, which are oftenvery sensitiveto environmentalvariation and oftenrespond in opposite ways to environmentalvariation (i.e., they have negativeenvironmental covariances). For example, if individualsadjust theirfecundity in response to mortality

This content downloaded from 128.103.149.52 on Tue, 03 Mar 2015 16:41:08 UTC All use subject to JSTOR Terms and Conditions 396 THE AMERICAN NATURALIST such thatpopulation sizes are roughlyconstant over time,then negative correlations between components of fecundityand survival among populations and highertaxa will be inevitableand will perniciouslymimic the effectsof a genetic trade-off(Sutherland et al. 1986; Gustafsson and Sutherland1988; Linden and M0ller 1989; Harvey and Keymer 1991; Partridgeand Sibly 1991). Ecological effectsmay be removed by comparinglaboratory or zoo populations(see, e.g., Schnebel and Grossfield1988; Harvey et al. 1990). However, if some species are preadapted to the laboratorywhile others are not, trade-offsthat obtain within each species may be obscured when they are compared. This resultis possibly why trade-offsbetween early fecundityand longevityhave not been foundwhen Drosophila species and semispecies are comparedin thelaboratory (Schnebel and Grossfield1988), even thoughsuch trade-offsdo exist withinoutbred laboratory populationsof Drosophila melanogaster (Rose and Charlesworth1981a, 1981b; Rose 1984b). Third, evolution of genetic backgroundmay obscure differencesin the costs of reproductionpaid by various species. For example, evolutionarychanges in body size should have widespread effectson organismalphysiology, including manyaspects of energeticacquisition. For thisreason, some workershave advocated testingfor interspecific costs of reproductiononly once the effectsof body size have been removed (Promislow and Harvey 1990; Blackburn 1991; Harvey and Pagel 1991). Typically, when the effectsof body size have been removed, negative interspecificcorrelations are found among some life-historytraits but not among others(see, e.g., Promislowand Harvey 1990). Such resultstell little, however, about which life-historytraits are influencedby costs of reproduction and which are not. Body size is but the most obvious of a multitudeof unknown physiologicaltraits on which selectioncould act so as to obscure trade-offs,other traitsthat cannot be easily measured or statistically"removed." The existence of such traitsmeans thatthe interspecificvariation that remains after the effects of body size have been removed are in no sense necessarilya truerepresentation of eitherthe functionalor geneticrelationships among traits. Taken together,the factorsthat we have discussed-correlated selection,co- variationof traits due to environmentvariation, differences among species in theirphenotypic responses to environmentalvariation, and the effectsof genetic background-can give rise to negative interspecificcorrelations among life- historytraits when costs of reproductionare absent and positive correlations when theyexist. Unlike Harvey and Keymer (1991), however,we findit difficult to see how the collectionof more comparativedata, or applicationof moresophis- ticated statisticalmethodology, could mitigatethese problems.

THE USES OF THE COMPARATIVE METHOD Comparison of the average characteristicsof species, theirperformances, and environmentsreveals littleabout adaptationand evolutionaryconstraint because the forcesthat mold evolution-selection, drift,mutation, and migration,and the substrateof genetic variationand covariation on which they act-are complex and hopelesslyconfounded in the way in which theyaffect the relativeattributes

This content downloaded from 128.103.149.52 on Tue, 03 Mar 2015 16:41:08 UTC All use subject to JSTOR Terms and Conditions COMPARATIVE METHODS AND ADAPTATION 397 of differenttaxa. Yet it should not be thoughtthat measuring genetic correlations and selectioncoefficients in any one populationwill reveal much about eitherits adaptedness to its environmentor lack thereof.A population is the productof its history:measurement of its currentparameters probably tells less about its past than its immediatefuture. However, we believe thatthe methodsof comparativebiology and evolutionary geneticsmight be usefullycombined. Specifically,patterns of selection,genetic variation,covariation, and interactionmight be measured for several species belongingto a singlemonophyletic group; the historicalinfluence of these various forces mightthen be inferredwith the aid of standardphylogenetic techniques such as optimization(see, e.g., Brooks and McLennan 1991). For example, to test the notionthat two traitshave been constrainedto evolve togetherin some clade, we suggest that theirgenetic correlationshould be measured in each of the terminaltaxa of thatclade. This approach is not an infalliblemethod; it is no betterthan the quantitativegenetic and phylogeneticmethods on which it is based. But we would be far surer of the historicalexistence of a constraintso inferredthan of one based only on a priorinotions of functionor development. The selectionhistory of a traitin a clade could be inferredin an analogous fashion instead of by the mere assertion that a clade has made a historicaltransition in lifestyleor habitat. Althoughthere are a few comparativestudies of genetic variance and covariance matrices (Arnold 1981; Lofsvold 1986; Atchley et at. 1992), phylogenetic analyses of the evolution of genetic parametershardly exist. One exception is that of Cohan and Hoffmann(1989), who compared the patternof correlated responses to selection forknock-down ethanol resistance in species belongingto the melanogaster subgroupand obscura subgroupof Drosophila. The most ele- gantexample thatwe know of a phylogeneticstudy of selectionis Basolo's (1990) study of female preferencefor male ornamentationin the genus Xiphophorus. The approach advocated here, althoughusing experimentaldata, remainspart of comparativemethodology and would take advantage of all of the recentconcep- tual and technicaladvances of thatfield. There are also situationsin which genuineexperimental manipulations can be done to reveal the basis of a comparativepattern. Reznick and colleagues (Rez- nick and Endler 1982; Reznick and Bryga 1987; Reznick 1989) have shown how among-populationdifferences in the life-historyattributes of Trinidadianguppies are attributableto varyingpatterns of predation. This conclusion was reached througha combinationof quantitativegenetic studies, replicated ecological com- parisons, and, most strikingly,experiments in which the selective pressures on naturalpopulations were altered and theirsubsequent evolutionobserved. Such experimentsmay well be the most powerfulway in which adaptation can be studied. Unlike Gould and Lewontin (1979), we do not hypothesizethat many traits are not adaptive. Rather,we are makingthe case thatthe adaptive (or nonadaptive) nature of traits cannot be determinedfrom most comparative data. If, as we suggest,historical study of the evolutionaryprocess is subjectto strictlimitations, thenthe various termsproposed to describe such processes or charactersare also

This content downloaded from 128.103.149.52 on Tue, 03 Mar 2015 16:41:08 UTC All use subject to JSTOR Terms and Conditions 398 THE AMERICAN NATURALIST subject to greatlyrestricted use. These include the termsadaptation, exaptation, disaptation,or aptation (Gould and Vrba 1982; Baum and Larson 1991), all of whichimply direct and specificknowledge of selectionforces in the past and the patternof genetic correlationamong characters. A furtherimplication of our thesis is that the promise of phylogeneticanalyses of congruenceamong traits, performancevariables, and environmentalcharacters to demonstrateadaptations will be unfulfilled. We do recognize the rathersevere limitationsthat our prescriptionsplace on retrospectivestudies of adaptationand evolutionaryconstraint. But we consider that such limitationsare betterseen clearly than obscured by false hopes about the abilityof comparativestudies to resolve the processes of evolution. Indeed, process is not all that is interestingin evolution; we can stilllearn a great deal about the natureof organismaldiversity by describingand analyzingpatterns of characterevolution alone (Eldredge and Cracraft1980; Lauder 1981; Wiley 1981; Lauder and Liem 1989; Brooks and McLennan 1991). Such analyses will be all the more robust for lacking unsupportedassumptions, unwarranted inferences, and untestablehypotheses about the historyof evolutionarymechanisms.

ACKNOWLEDGMENTS We thankA. F. Bennett,J. Cracraft,S. A. Frank, A. Gibbs, M. Matos, and L. D. Mueller for theirinsightful comments on the manuscript.This work was supportedin part by U.S. Public Health Service grantsAGO-6346 and AGO-9970 to M.R.R. and National Science Foundation grantIBN 91-19502to G.V.L.

LITERATURE CITED

Antonovics,J., and P. H. van Tienderen. 1991. Ontoecogenophyloconstraints?The chaos of con- straintsterminology. Trends in Ecology & Evolution 6:166. Arnold, S. J. 1981. Behavioral variationin naturalpopulations. I. Phenotypic,genetic and environ- mentalresponses to prey in the gartersnake, Thamnophiselegans. Evolution 35:489-509. 1992. Constraintson phenotypicevolution. AmericanNaturalist 140(suppl.):85S-107S. Atchley,W. R., and B. K. Hall. 1991. A model fordevelopment and evolutionof complexmorphologi- cal structures.Biological Reviews of the CambridgePhilosophical Society 66:101-157. Atchley, W. R., B. Riska, L. A. Kohn, A. A. Plummer,and J. J. Rutledge. 1984. A quantitative genetic analysis of brain and body size associations, theirorigin and ontogeny:data from mice. Evolution 38:1165-1179. Atchley,W. R., D. E. Cowley, C. Vogel, and T. McLellan. 1992. Evolutionarydivergence, shape change, and genetic correlationstructure in the rodent mandible. SystematicBiology 41: 196-221. Barton, N. H., and M. Turelli. 1989. Evolutionaryquantitative genetics: how littledo we know? Annual Reviews of Genetics 23:337-370. Basolo, A. 1990. Female preferencepredates the evolution of the sword in swordtailfish. Science (Washington,D.C.) 250:808-810. Baum, D. A., and A. Larson. 1991. Adaptationreviewed: a phylogeneticmethodology for studying charactermacroevolution. Systematic Zoology 40:1-18. Berven, K. A., and D. E. Gill. 1982. The geneticbasis of altitudinalvariation in the wood frog,Rana sylvatica. II. An experimentalanalysis of larval traits.Oecologia (Berlin) 52:360-369. Blackburn,T. M. 1991. The interspecificrelationship between egg size and clutch size in wildfowl. Auk 108:209-211.

This content downloaded from 128.103.149.52 on Tue, 03 Mar 2015 16:41:08 UTC All use subject to JSTOR Terms and Conditions COMPARATIVE METHODS AND ADAPTATION 399

Bock, W. 1977. Adaptationand the comparativemethod. Pages 57-82 in M. Hecht, ed. Major patterns in vertebrateevolution. Plenum, New York. Brandon,R. H. 1990. Adaptationand environment.Princeton University Press, Princeton,N.J. Brooks, D. R., and D. A. McLennan. 1991. Phylogeny,ecology and behavior. Universityof Chicago Press, Chicago. Charlesworth,B., R. Lande, and M. Slatkin. 1982. A neo-Darwiniancommentary on macroevolution. Evolution 36:474-498. Clark, A. G., and M. W. Feldman. 1981. The estimationof epistasis in componentsof fitnessin experimentalpopulations of Drosophila melanogaster.II. Assessmentof meioticdrive, via- bility,fecundity and sexual selection. Heredity46:347-377. Clark,A. G., M. W. Feldman, and F. B. Christiansen.1981. The estimationof epistasisin components of fitnessin experimentalpopulations of Drosophila melanogaster.I. A two stage maximum likelihoodmodel. Heredity46:321-346. Clarke,G. M., and J. A. McKenzie. 1987. Developmentalstability of insecticideresistant phenotypes in blowfly:a resultof canalizing naturalselection. Nature (London) 325:345-346. Clutton-Brock,T. H., and P. H. Harvey. 1979. Comparisonand adaptation.Proceedings of the Royal Society of London B, Biological Sciences 205:547-565. Coddington,J. A. 1988. Cladistic tests of adaptationalhypotheses. Cladistics 4:3-22. . 1990. Bridges between evolutionarypattern and process. Cladistics 6:379-386. Cody, M., and J. Diamond, eds. 1975. Ecology and evolutionof communities.Belknap, Cambridge, Mass. Cohan, F. M., and A. A. Hoffmann.1989. Uniformselection as a diversifyingforce in evolution: evidence fromDrosophila. American Naturalist134:613-637. Cowley, D. E., and W. R. Atchley. 1990. Developmental and quantitativegenetics of correlation structureamong body parts of Drosophila melanogaster.American Naturalist 135:242-268. Cracraft,J. 1981. The use of functionaland adaptive criteriain phylogeneticsystematics. American Zoologist 21:21-36. Crow, J. 1987. Population genetics history:a personal view. Annual Review of Genetics 21:1-22. Culver, D. C. 1982. Cave life. Harvard UniversityPress, Cambridge,Mass. Darwin, C. 1859. On the originof species. (Facsimile of the firstedition, 1967.) Atheneum,New York. 1871. The descent of man and selectionin relationto sex. Pt. 2. (Facsimile of the firstedition, 1989.) Pickering,London. de Jong,G. 1990. Genotype-by-environmentinteraction and the geneticcovariance betweenenviron- ments: multilocusgenetics. Genetica 81:171-177. Eldredge, N., and J. Cracraft. 1980. Phylogeneticpatterns and the evolutionaryprocess. Columbia UniversityPress, New York. Endler, J. 1986. Natural selection in the wild. PrincetonUniversity Press, Princeton,N.J. Felsenstein,J. 1985. Phylogeniesand the comparativemethod. American Naturalist 125:1-15. Futuyma,D. 1986. Evolutionarybiology. 2d ed. Sinauer, Sunderland,Mass. Gould, S. J. 1977. Ontogenyand phylogeny.Belknap, Cambridge,Mass. Gould, S. J., and R. C. Lewontin. 1979. The spandrelsof San Marco and the Panglossian paradigm: a critiqueof the adaptationistprogramme. Proceedings of the Royal Society of London B, Biological Sciences 205:581-598. Gould, S. J., and E. S. Vrba. 1982. Exaptation-a missingterm in the science of form.Paleobiology 8:4-15. Graves, J. L., E. C. Toolson, C. Jeong,L. N. Vu, and M. R. Rose. 1992. Desiccation, flight,glycogen, and postponed senescence in Drosophila melanogaster. PhysiologicalZoology 65:268-286. Greene, H. W. 1986. Diet and arborealityin the Emerald Monitor,Varanus prasinus, withcomments on the studyof adaptation. Fieldiana Zoology 31:1-12. Gustafsson, L., and W. J. Sutherland. 1988. The costs of reproductionin the collard flycatcher Ficedula albicollis. Nature (London) 335:813-815. Harvey, P. H., and A. E. Keymer. 1991. Comparinglife historiesusing phylogenies.Philosophical Transactionsof the Royal Society of London B, Biological Sciences 332:31-39. Harvey, P. H., and M. D. Pagel. 1991. The comparative method in evolutionarybiology. Oxford UniversityPress, New York.

This content downloaded from 128.103.149.52 on Tue, 03 Mar 2015 16:41:08 UTC All use subject to JSTOR Terms and Conditions 400 THE AMERICAN NATURALIST

Harvey, P. H., and T. M. Zammuto. 1985. Patternsof mortalityand age at firstreproduction in naturalpopulations of mammals. Nature (London) 315:319-320. Harvey, P. H., A. F. Read, and D. E. Promislow. 1990. Life historyvariation in placentalmammals: unifyingthe data withtheory. Oxford Surveys in EvolutionaryBiology 6:13-31. Hoffmann,A. A., and P. A. Parsons. 1991. Evolutionarygenetics and environmentalstress. Oxford UniversityPress, New York. Houle, D. 1991. Genetic covariance of fitnesscorrelates: what genetic correlationsare made of and why it matters.Evolution 45:630-645. Jinks,J. L., J. M. Perkins,and H. S. Pooni. 1973. The incidence of epistasis in normaland extreme environments.Heredity 31:263-269. Keightley,P. 1989. Models of quantitativevariation of flux in metabolic pathways. Genetics 121: 869-876. Kimura,M. 1983. The neutraltheory of molecularevolution. Cambridge University Press, Cambridge. King, J. C. 1955. Evidence for the integrationof the gene pool fromstudies of DDT resistance in Drosophila. Cold SpringHarbour Symposia on QuantitativeBiology 20:311-317. Kohn, L. A., and W. A. Atchley. 1988. How similarare genetic correlationstructures? data from rats and mice. Evolution 42:467-481. Kreitman,M. 1983. Nucleotide polymorphismat the alcohol dehydrogenaselocus of Drosophila melanogaster. Nature (London) 304:412-417. Lande, R. 1979. Quantitativegenetic analysis of multivariateevolution, applied to brain:body allome- try.Evolution 33:402-416. Larson, A. 1984. Neontological inferencesof evolutionarypattern and process in the salamander familyPlethodontidae. Evolutionary Biology 17:119-217. Larson, A., D. B. Wake, L. R. Maxson, and R. Highton. 1981. A molecularphylogenetic perspective on the originsof morphologicalnovelties in the salamandersof the tribePlethodontini (Am- phibia, Plethodontidae).Evolution 35:405-422. Lauder, G. V. 1981. Form and function:structural analysis in evolutionarymorphology. Paleobiology 7:430-442. Lauder, G. V., and K. F. Liem. 1989. The role of historicalfactors in the evolution of complex organismalfunctions. Pages 63-78 in D. B. Wake and G. Roth, eds. Complex organismal functions:integration and evolutionin vertebrates.Wiley, Chichester. Lenski, R. 1988a. Experimentalstudies of pleiotropyand epistasis in Escherichia coli. I. Variation in competitivefitness among mutantsresistant to virus T4. Evolution 42:425-432. 1988b. Experimentalstudies of pleiotropyand epistasis in Escherichia coli. II. Compen- sation for maladaptive effectsassociated with resistance to virus T4. Evolution 42:433- 440. Linden, M., and A. P. M0ller. 1989. Cost of reproductionand covariation of life-historytraits in birds. Trends in Ecology & Evolution 4:367-371. Lofsvold, D. 1986. Quantitativegenetics of morphologicaldifferentiation in Peromyscus. I. Tests of the homogeneityof genetic covariance structureamong species and subspecies. Evolution 40:559-573. Luckinbill, L. S., J. L. Graves, A. Tomkiw, and 0. Sowirka. 1988. A qualitative analysis of life historycharacters in Drosophila melanogaster. EvolutionaryEcology 3:31-39. Luke, C. 1986. Convergentevolution of lizard toe fringes.Biological Journalof the Linnean Society 27:1-16. Martin,R. D. 1981. Relative brain size and basal metabolic rate in terrestrialvertebrates. Nature (London) 293:57-59. Maynard Smith,J. 1989. Evolutionarygenetics. OxfordUniversity Press, New York. Mayr, E. 1983. How to carryout the adaptationistprogram? American Naturalist 121:324-334. McKenzie, J. A., M. J. Whitten,and M. A. Adena. 1982. The effectof genetic backgroundon the fitnessof diazinon resistance genotypes of the Australian sheep blowflyLucilia cuprina. Heredity49:1-9. McLennan, D. A., D. R. Brooks, and J. D. McPhail. 1988. The benefitof communicationbetween comparativeethology and phylogeneticsystematics: a case studyusing gasterosteidfishes. Canadian Journalof Zoology 66:2177-2190. Mishler, B. D. 1988. Reproductive ecology of bryophytes.Pages 285-306 in J. Lovett Doust and

This content downloaded from 128.103.149.52 on Tue, 03 Mar 2015 16:41:08 UTC All use subject to JSTOR Terms and Conditions COMPARATIVE METHODS AND ADAPTATION 401

L. Lovett Doust, eds. Plant reproductiveecology: patternsand strategies.Oxford University Press, New York. Pagel, M. D., and P. H. Harvey. 1988. How mammals produce large-brainedoffspring. Evolution 42:948-957. 1989. Taxonomic differencesin the scalingof brainon body weightamong mammals.Science (Washington,D.C.) 244:1589-1593. Partridge,L., and R. Sibly. 1991. Constraintsin the evolutionof lifehistories. Philosophical Transac- tions of the Royal Society of London B, Biological Sciences 332:3-13. Promislow,D. E. L., and P. H. Harvey. 1990. Living fast and dyingyoung: a comparativeanalysis of life-historyvariation among mammals. Journalof Zoology (London) 220:417-437. Reznick, D. N. 1989. Life-historyevolution in guppies. II. Repeatabilityof fieldobservations and the effectsof season on life histories.Evolution 43:1285-1297. Reznick, D. N., and H. Bryga. 1987. Life-historyevolution in guppies (Poecilia reticlulata).I. Pheno- typicand genetic changes in an introductionexperiment. Evolution 41:1370-1385. Reznick,D. N., and J. A. Endler. 1982. The impactof predationon lifehistory evolution in Trinidadian guppies (Poecilia reticiulata).Evolution 36:160-177. Ridley, M. 1992. Darwin sound on comparativemethod. Trends in Ecology & Evolution 7:37. Riska, B. 1989. Composite traits,selection response, and evolution. Evolution 43:1172-1191. Riska, B., and W. R. Atchley. 1985. Genetics of growthpredict patternsof brain-size evolution. Science (Washington,D.C.) 229:668-671. Roderick,T. H., R. E. Wimmer,and C. C. Wimmer.1976. Genetic manipulationof neuroanatomical traits. Pages 143-178 in L. Petrovich and J. L. McGaugh, eds. Knowing, thinking,and believing.Plenum, New York. Rose, M. R. 1982. Antagonisticpleiotropy, dominance, and geneticvariation. Heredity 48:63-78. 1983. Furthermodels of selection withantagonistic pleiotropy. Pages 47-53 in H. I. Freeman and C. Strobeck, eds. Population biology. Springer,Berlin. 1984a. Genetic covariationin Drosophila lifehistory: untangling the data. AmericanNaturalist 123:565-569. 1984b. Laboratoryevolution of postponed senescence in Drosophiilainelanogaster. Evolution 38:1004-1010. Rose, M. R., and B. Charlesworth.1981a. Genetics of life-historyin Drosophila inelanogaster.I. Sib analysis of adult females. Genetics 97:173-186. 1981b. Genetics of life-historyin Drosophila melanogaster. II. Exploratoryselection experiments. Genetics 97:187-196. Saether, B.-E. 1988. Pattern of covariation between life-historytraits of European birds. Nature (London) 331:616-617. Schlichting,C. D., and D. A. Levin. 1986. Phenotypicplasticity: an evolvingplant character. Biologi- cal Journalof the Linnean Society 29:37-47. Schnebel, E. M., and J. Grossfield.1988. Antagonisticpleiotropy: an interspecificDrosoplhila comparison. Evolution 42:306-311. Seager, R. D., and F. J. Ayala. 1982. Chromosome interactionsin Drosophila mnelanogaster.I. Viabilitystudies. Genetics 102:467-483. Service, P. M. 1987. Physiologicalmechanisms of increased stress resistancein Drosophila inelanio- gaster selected for postponed senescence. PhysiologicalZoology 60:321-326. 1993. Laboratory evolution of longevityand reproductivefitness components in male fruit flies: matingability. Evolution 47:387-399. Service, P. M., and R. M. Rose. 1985. Genetic co-variationamong life-history components: the effects of novel environments.Evolution 39:943-945. Service, P. M., E. W. Hutchinson,M. D. MacKinley, and M. R. Rose. 1985. Resistance to environmentalstress in Drosophlilamelanogaster selected forpostponed senescence. Physiological Zoology 58:380-389. Service, P. M., E. W. Hutchinson, and M. R. Rose. 1988. Multiple genetic mechanismsfor the evolutionof senescence in Dr-osophilamnelanogaster. Evolution 39:708-716. Sheridan,A. K., and J. S. F. Barker. 1974. Two-traitselection and geneticcorrelation. II. Changes in the geneticcorrelation during two-trait selection. Australian Journal of Biological Sciences 27:75-88.

This content downloaded from 128.103.149.52 on Tue, 03 Mar 2015 16:41:08 UTC All use subject to JSTOR Terms and Conditions 402 THE AMERICAN NATURALIST

Simpson, G. G. 1953. The major featuresof evolution. Columbia UniversityPress, New York. Sinervo, B., and P. Licht. 1991. Proximateconstraints on the evolutionof egg size, numberand total clutch mass in lizards. Science (Washington,D.C.) 252:1300-1302. Slatkin,M., and S. A. Frank. 1990. The quantitativegenetic consequences of pleiotropyunder stabilizing and directionalselection. Genetics 125:207-213. Sober, E. 1984. The nature of selection: evolutionarytheory in philosophical focus. MIT Press, Cambridge,Mass. Stearns, S. 1984. The tension between adaptation and constraintin the evolution of reproductive patterns.Advances in InvertebrateReproduction 3:387-398. Stearns, S. C., G. de Jong,and G. Newman. 1991. The effectsof phenotypicplasticity on genetic correlations.Trends in Ecology & Evolution 6:122-126. Sutherland,W. J., A. Grafen,and P. H. Harvey. 1986. Life historycorrelations and demography. Nature (London) 320:88. Tinkle,D. W. 1969. The concept of reproductiveeffort and its relationto the evolutionof lifehistories of lizards. American Naturalist102:501-516. Traub, R. 1980. Some adaptive modificationsin fleas. Pages 33-67 in R. Traub and H. Starcke, eds. Proceedingsof the InternationalConference on Fleas, Ashton Wold, U.K., June 1977. Balkema, Rotterdam. Trivers,R. 1985. Social evolution. Cummings,Menlo Park, Calif. Tuinstra,E. J.,G. de Jong,and W. Scharloo. 1990. Lack of response to familyselection for directional asymmetryin Drosophila melanogaster: leftand rightare not distinguishedduring development. Proceedings of the Royal Society of London B, Biological Sciences 241:146-152. Turelli,M. 1988. Phenotypicevolution, constant covariances, and the maintenanceof additivevaria- tion. Evolution 42:1342-1347. Van Valen, L. 1971. Adaptive zones and the orders of mammals. Evolution 25:420-428. Waddington,C. H. 1975. The evolution of an evolutionist.Cornell UniversityPress, Ithaca, N.Y. Wake, D. B., and A. Larson. 1987. Multidimensionalanalysis of an evolvinglineage. Science (Wash- ington,D.C.) 238:42-48. Wiley,E. 0. 1981. Phylogenetics:the theoryand practiceof phylogeneticsystematics. Wiley Intersci- ence, New York. Wilkinson,G., K. Fowler, and L. Partridge. 1990. Resistance of genetic correlationstructure to directionalselection in Drosophila melanogaster. Evolution 44:1990-2003. Williams,G. C. 1966a. Adaptationand naturalselection. PrincetonUniversity Press, Princeton,N.J. 1966b. Natural selection, the costs of reproductionand a refinementof Lack's principle. AmericanNaturalist 100:687-690. Wright,S. 1977. Evolution and the genetics of populations. Universityof Chicago Press, Chicago.

Associate Editor: BrentD. Mishler

This content downloaded from 128.103.149.52 on Tue, 03 Mar 2015 16:41:08 UTC All use subject to JSTOR Terms and Conditions