Influence of Generation Time on the Rate of Response to Selection

Vol. 137, No. 4 The American Naturalist April 1991

INFLUENCE OF GENERATION TIME ON THE RATE OF RESPONSE TO SELECTION

JAY A. ROSENHEIM* AND BRUCE E. TABASHNIK Departmentof Entomology,3050 Maile Way, Room 310, Universityof Hawaii, Honolulu, Hawaii 96822

SuibmittedSeptember 11, 1989; Revised Januat-y30, 1990; Accepted March 13, 1990

Abstract.-We examined the influenceof generationtime on the rate of evolution of a trait underintense natural selection: pesticide resistancein arthropodpests. Previous empiricaland theoreticalanalyses supporteda positivelinear relationship between the numberof generations per year and the rate of evolutionof pesticideresistance. To testthis relationship, we assembled a data base that integratedinformation on resistance evolution, generationtime, and other biological parametersfor 682 NorthAmerican arthropod pests. The data did not supporta linear relationshipbetween generationsper year and the evolutionof resistance,revealing instead a nonlinearand highlyvariable relationship,with peak rates of resistance evolution for species withintermediate generation times. This resultwas independentof the differencebetween introduced and native species and of differencesamong the major arthropodtaxonomic orders in abilityto evolve resistance.A reevaluationof evidence fromanalytical and computer-simulation models of resistanceevolution suggests that the linearrelationship between generations per year and resistanceevolution is also withoutfoundation in theory.An extensionof a simpleanalytical model of resistanceevolution suggests instead that the rate of resistanceevolution is independent of generationtime. We also findlittle support for the suggestionthat species withmany generations per year are regularlysubject to elevated levels of selection for pesticide resistance. Per-generationfitness values for genotypes conferringresistance to pesticides or genotypes conferringincreased fitnessin response to any density-independentselective agent are related exponentiallyto generationtime, resulting in the independenceof generationtime and the rate of response to selection in the simplest-casemodel. Generationtime can influencethe rate of resistanceevolution; however, ratherthan actingin a simple,uniform manner, generation time interactswith a varietyof genetic,ecological, and operationalfactors to produce a multitudeof effects.

The molecular-clockhypothesis has spurreda continuingcontroversy regarding the rate of evolutionarychange in lineages withdifferent cohort generation times. Most studies attemptingto resolve thisissue have been conductedat the molecular level and have examinedgenetic changes thatare arguablyselectively neutral (Sarich and Wilson 1973; Wu and Li 1985; Easteal 1988; Graur et al. 1989; see also Palumbi 1989). The degreeto whichthe rateof molecularevolution of neutral traitsis constantis importantbecause rate constancymust be assumed to recon- structevolutionary trees with correct topologies and to inferthe phylogenetic divergencetimes of taxa of known molecularsimilarity. The influenceof generationtime on the evolutionof nonneutraltraits is also a

* Presentaddress: Departmentof Entomology,University of California,Davis, California95616.

Am. Nat. 1991. Vol. 137, pp. 527-541. ? 1991 by The Universityof Chicago. 0003-0147/91/3704-0005$02.00.All rightsreserved.

This content downloaded from 128.120.194.195 on Tue, 23 Sep 2014 01:11:58 AM All use subject to JSTOR Terms and Conditions 528 THE AMERICAN NATURALIST theoreticallyimportant question. Variation in generationtime is one of several factorsthat could cause the rateof adaptiveevolution to varyin differentlineages. If the fitnessvalues of given alleles are independentof generationtime, then lineages witha largernumber of generations per year (GPY) exhibitgreater annual changes in allele frequencies.Rate constancyof evolutionfor selectivelyimpor- tant traits would argue against the claim that a constant rate of evolution is consistentonly with a neutraltheory of evolution (Hartl and Dykhuizen 1979; Gillespie 1986; Zuckerkandl 1987). The effectof generationtime on the rate of adaptationof populationscould also influencethe evolutionof life-historystrate- gies. If a largerGPY increases evolutionaryplasticity, shorter generation times could be favored over longer generationtimes. Shortergeneration times might thereforebe favored evolutionarilyfor one of the same primaryreasons that sexual reproductionhas been hypothesizedto be favoredover asexual reproduction (Maynard Smith 1984; Nunney 1989). Nevertheless,few studies have attemptedto assess the influenceof generationtime on the rate of adaptive evolution. In a seminal study,Hartl and Dykhuizen(1979; Dykhuizenand Hartl 1981) found that adaptation of Escherichia coli to a novel habitat was a functionof absolute timerather than the numberof elapsed generations.These authorswere able to change generationtime by manipulatingthe rate of nutrientflow into bacterial cultures in a chemostat. Similar experimentaltests of the influenceof generation time on adaptive evolution have not been performedwith higher eucaryotes. The evolution of pesticide resistancein arthropodpests provides an opportu- nity to test the influenceof generationtime on the evolution of a selectively importanttrait. The economic importanceof pesticide resistancehas resultedin the documentationof resistance in a large numberof arthropodspecies (Geor- ghiou 1981); these data reveal the results of a large-scale natural experiment in organic evolution,observed at the organismalrather than the molecularlevel. Generationtime has been identifiedas one of the few factorsthat appears to influenceresistance evolutionin a strongand consistentmanner. Empirical and theoreticalanalyses have uniformlysupported a positive linear relationshipbetween GPY and the rate of resistanceevolution (Comins 1979; Georghiou 1980; Tabashnik and Croft 1982, 1985; May and Dobson 1986; Hartl 1988; Tabashnik 1990). Here we present an empiricalanalysis of resistance evolution in North American arthropodpests thatchallenges this view. We reevaluate existingthe- oryin lightof our resultsand show thatit failsto supporta consistentrelationship between generationtime and the rate of resistanceevolution. Finally, we extend our conclusions to a broad array of traitsinfluenced by selective forces that operate in a density-independentmanner. Elsewhere we present the results of computer simulationselucidating the various influencesof generationtime on resistanceevolution (Rosenheim and Tabashnik 1990).

METHODS Data-Base Compilation Our analysis is based on a data base thatintegrates information on biology and resistanceevolution for North American arthropod pests. We used Davidson and

This content downloaded from 128.120.194.195 on Tue, 23 Sep 2014 01:11:58 AM All use subject to JSTOR Terms and Conditions GENERATION TIME AND RESPONSE TO SELECTION 529

Lyon's (1987) extensive list of pest species as the foundationfor the data base. All pests of agriculture,stored products,and human or animal health were included. Pests of ornamentalshrubs and trees, rangelands,and foresttrees were included only if subject to pesticidal control. For each pest species, we then sought five pieces of information:(1) a measure of abilityto evolve resistance, (2) an estimateof GPY, (3) a measure of pest severity,(4) whetherthe species was introducedor native, and (5) the species' taxonomicorder. An index of each species' abilityto evolve resistancewas generatedby count- ing the number of insecticide/acaricideclasses to which at least some North American field populations had been reported as resistant(Georghiou 1981). Thirty-sixpest species reportedas resistantby Georghiou(1981) but not present in Davidson and Lyon (1987) were added to the data base. Pesticides were grouped into six classes followingGeorghiou (1981): (1) DDT and analogues, (2) benzene hexachloride and cyclodienes, (3) organophosphates,(4) carbamates, (5) pyrethroids,and (6) other miscellaneous compounds, includingbinapacryl, chlordimeform,ovex, propargite,quinomethionate, sulfenone, and tetradifon. Resistances to inorganic,botanical, and microbialinsecticides, which are rela- tivelyuncommon, were excluded. Althoughwe feel that our measure of ability to evolve resistance is unbiased with respect to species GPY, the index is not withoutlimitations. Most important,the exposure of differentpests to pesticide selection pressures probably varied withinand between differentmanaged eco- systems. We have attemptedto accommodate variable selection intensityby including in the analyses a measure of pest severity.We analyzed the data base firstas a singleunit and thengrouped species by pest typeand crop attacked (see below). Estimates of each species' GPY were compiled fromthe literature.For cases in which a range of GPY associated with latitudewas cited, or when different studies produced differentGPY estimates,the arithmeticmean of the observa- tions was used. No attemptswere made to estimateGPY fromlaboratory studies or egg-to-adultdevelopment rates. We estimatedGPY for 90 species for which literaturereferences could not be found by using the mean value reportedfor congeners. This approximationwas not applied to mosquitoes, which demon- stratedlarge within-genusvariation in GPY. In thisarticle, we use the termsGPY and generationtime to referto the same biological parameter;however, because many arthropodsundergo diapause forpart of the year, the inverse relationship between GPY and generationtime will not be exact. This distinction,although potentiallysignificant in some cases (Rosenheim and Tabashnik 1990), is not importantto the currentdiscussion. It should be noted, however, that our data base included values forGPY and not generationtime per se. The remainingthree parameters were includedto explain variationin resistance evolutionnot relatedto GPY. The thirdparameter, an index of pest severity,was generatedby summingthe numberof timesa species was cited in the Review of Applied Enitomology,1950-1953 (Hall 1950-1953a, 1950-1953b). The severity index was includedbecause more importantpests may be subject to more intense selection pressure from pesticide applications and, further,because all other thingsbeing equal, resistancein more importantpests may be more likelyto be documented by researchers. Only North American field studies or laboratory

This content downloaded from 128.120.194.195 on Tue, 23 Sep 2014 01:11:58 AM All use subject to JSTOR Terms and Conditions 530 THE AMERICAN NATURALIST studies publishedin North Americanjournals were tallied. A time period before the major onset of resistance (1950-1953) was chosen to minimizethe extentto which a species' abilityto evolve resistancewould inflateits severityscore. To cross-referencecurrent taxonomic names withthose used in the 1950-1953 vol- umes of the Review of Applied Entomology,all names were amended to their 1950 formsusing reviews of economic entomologyfrom that period and taxonomic works providinghistorical data on name synonymies.Thirty species not clearly definedtaxonomically in 1950 were excluded. Population bottlenecksthat occur when exotic species are introducedto new regionscan have profoundeffects on a population's geneticstructure. We therefore included a fourthparameter that grouped native and introducedspecies. This categorizationwas obtainedfrom the NorthAmerican Introduced Arthropod Database (NAIAD, an unpublisheddata base compiled by R. I. Sailer and the United States Departmentof Agriculture;Sailer 1978, 1983) and fromnFurniss and Carolin (1977), Clausen (1978), and Davidson and Lyon (1987). Species listed in NAIAD as havinginvaded by a continuousrange expansion were categorized as native because population bottleneckswere unlikelyto have occurred and gene flow with indigenous populations was unlikelyto have been obstructed. Twelve pests introducedafter the advent of syntheticorganic pesticides in 1945 were excluded. Finally,to assess the possibilitythat different arthropod taxonomic orders have differentabilities to evolve resistance,species were grouped by order following the systematicscheme of Borror et al. (1981). The finaldata set included 888 species; GPY estimates were obtained for 682 species. The portionof the data base describingresistance evolution for pests of cottonis presentedas an example in the Appendix; the fulldata base withcomplete literature citations is available fromthe firstauthor. Statistical Analysis Resistance scores were transformedas the log(resistancescore + 1.0) before regressionanalyses to satisfythe assumptionof homoscedasticity.Analyses of transformedand untransformeddata yieldedidentical results; therefore, only the results fromthe transformeddata analyses are presented. To test for a linear relationshipbetween resistanceevolution and GPY, we performedmultiple linear regressionsand partialregression analyses of resistancescores (dependentvariable) on GPY and pest severity(independent variables). The data base was analyzed as a single unit and by groupingspecies by pest type (i.e., key pests, operationallydefined as species withseverity score -8, and pests of agriculture, stored products,and humanor animal health) or by crop attacked. To test for a nonlinearrelationship between resistance and GPY, we performedpolynomial regressionsof the residualsfrom a linearregression of resistanceon pest severity (dependentvariable) on GPY (independentvariable). Residuals were analyzed as a means of removingthe influenceof confoundingindependent variables (in this case, pest severity).Effects of taxonomicorder were investigatedby averaging across all species withineach orderthe residualvalue froma multipleregression of resistance score (dependent variable) on pest severityand GPY, with GPY

This content downloaded from 128.120.194.195 on Tue, 23 Sep 2014 01:11:58 AM All use subject to JSTOR Terms and Conditions GENERATION TIME AND RESPONSE TO SELECTION 531

TABLE 1 MULTIPLE-REGRESSIONANALYSIS OF THEINFLUENCE OF GENERATIONSPER YEAR (GPY) ONRESISTANCE EVOLUTION

Pest Group Type and Regression IndependentVariables Coefficient SE t p 2*

All species (682): Pest severity .0063 .0006 10.22 <.001 .14 GPY .0017 .0012 1.50 .13 .01 Agriculture(586): Pest severity .0087 .0009 9.95 <.001 .15 GPY .0011 .0011 .93 .35 .01 Stored products(37): Pest severity .0087 .0064 1.36 .18 .06 GPY .0162 .0165 .99 .33 .02 Human and animal health(59): Pest severity .0039 .0011 3.70 <.001 .22 GPY .0044 .0059 .75 .46 .04 Key pests (70): Pest severity .0039 .0012 3.13 <.01 .13 GPY .0015 .0043 .36 .72 .01 Cotton (36): Pest severity .0085 .0019 4.60 <.001 .39 GPY - .0021 .0045 - .48 .64 .00 Grains and corn (81): Pest severity .0078 .0019 4.12 <.001 .18 GPY - .0020 .0032 -.62 .54 .00 Pome fruits(apple and pear) (56): Pest severity .0071 .0019 3.81 <.001 .22 GPY .0114 .0049 2.32 .02 .09 Solanaceous crops (primarilypotato, tomato,tobacco) (37): Pest severity .0081 .0028 2.84 .01 .20 GPY .0012 .0076 .15 .88 .01

NOTE.-Number of species in parentheses. * r' values are given for both significant(P < .05) and nonsignificantvariables. For regressions withone or no significantindependent variables, r2 values reflectthe contributionof each independent variable considered alone (i.e., simple bivariatelinear regression);note that in these cases t and P values continueto referto the resultsof the multipleregression analysis. For the regressionwith two significantindependent variables (pests of pome fruits),partial correlation coefficients are reported. included as a cubic polynomial(independent variables). All analyses were con- ducted using the BMDP computerstatistics package (Dixon 1985).

RESULTS AND DISCUSSION

Data-Base Analysis Multiple linear regressionof resistance score on pest severityand GPY revealed a consistentpositive effectof pest severityon resistancescore (table 1). However, except forone case, pests of pome fruits,GPY had no significanteffect on resistancescore. The two independentvariables, pest severityand GPY, were themselves almost completelyuncorrelated (for all species, r,2 = 0.026); thus, species with rapid generationturnover were not, in general, more severe pests,

This content downloaded from 128.120.194.195 on Tue, 23 Sep 2014 01:11:58 AM All use subject to JSTOR Terms and Conditions 532 THE AMERICAN NATURALIST

0.15

0.10

0.05~=-.5 ~ < .0 -0.05 23 -0.10 y =-0.063 + O.O31GPY- 0.0017GPY + 0.000022GPY 2 r = 0.057 P << 0.001

0 1 0 20 30 GPY FIG. 1.-Mean residuals (?SE) froma regressionof log-transformedresistance score on pest severityfor 682 arthropodpests. Circles, mean residual values for pests grouped by GPY value. Classes are formedwith upper boundariesat generations-per-yearvalues of: 0.5, n = 32; 1.5,n = 276; 2.5, n = 129;3.5, n = 90; 4.5, n = 37; 6.5, n = 40; 10.5,n = 34; 15.5, n = 13; 25.5, n = 9; >25.5, n = 16. Dashed line, fittedthird-order polynomial regression curve. and the pest severityvariable could not hide an importantcontribution by GPY. For pests of pome fruits,GPY explained only 9% of the observed variance in resistancescores. Given thatnine regressionswere computed,a singlesignificant result is not compellingevidence; if we maintainoverall ox = 0.05 by applying Bonferroni'sinequality (Dixon 1985) to the data in the table, the criticalP value becomes .05/9 = .004, and the resultfor pome fruits(P = .02) is not significant. Thus, the data do not supporta linear relationshipbetween resistanceevolution and generationtime. An examinationof the residuals froma regressionof resistance score on pest severityfor all species combined revealed a patternof positive residuals for species withintermediate GPY values (i.e., GPY = 3.5-10.5; fig. 1). The magni- tude of the effect(largest mean residual was 0.097 for6.5 ? GPY ? 10.5; fig. 1) was similar to that of pest severity: key pests with severityscores of 10-73 yielded predicted increases in resistance scores of 0.063-0.441 (from table 1, regressionfor all species combined). A polynomialregression of residuals on GPY yielded a curvilinearrelationship, which although highly significant(P < .001 for the linear, quadratic, and cubic terms) explained only 5.7% of the variationin residuals. Thus, althoughgeneration time does influenceresistance evolution,the effectappears to be nonlinearand highlyvariable. The influencesof two variables that mightcomplicate the demonstrationof a significanteffect for generation time were investigated.First, the native species/ introducedspecies dichotomyproved to be highlysignificant, with introduced species exhibitinga decreased ability to evolve resistance (J. A. Rosenheim, M. W. Johnson,R. F. L. Mau, B. E. Tabashnik, and S. C. Welter,unpublished data). However, no significantlinear effect for GPY was detectedfor either native

This content downloaded from 128.120.194.195 on Tue, 23 Sep 2014 01:11:58 AM All use subject to JSTOR Terms and Conditions GENERATION TIME AND RESPONSE TO SELECTION 533

0.09 . 38

0.06

- -0.03 . ..

0) 0.00

c~-0.03 3 0)~ ~ ~ ~ ~~~~~~3 33 -0.06 -130 20

12 Aca Dip Col Lep Hem Ort Thy Hom Hym FIG. 2.-Mean residuals (?SE) froma regressionof log-transformedresistance score on pest severityand a third-orderpolynomial in generationsper year fornine major arthropod taxonomic orders. The numberabove or below each bar is the sample size. Aca, Acarina; Dip, Diptera; Col, Coleoptera; Lep, Lepidoptera; Hem, Hemiptera; Ort, Orthoptera;Thy, Thysanoptera;Horn, Homoptera; Hym, Hymenoptera.Only the mean residualsfor the Ho- mopteraand Hymenopteradiffer significantly from zero (t = 3.93, P < .01, and t = 9.38, P < .001, respectively). or introducedpests alone, and the subtlecurvilinear relationships between resistance score and GPY were not significantlydifferent for the two species groups. Thus, the differentabilities of nativeand introducedspecies to evolve resistance did not mask an importantrole forgeneration time. Second, differentarthropod taxonomic orders exhibited similar abilities to evolve resistance (fig. 2). Of the nine orders with sample sizes greaterthan 10, only two, the Homoptera and Hymenoptera,showed mean residuals fromthe regressionof resistance score on severityand a third-orderpolynomial in GPY thatdiffered significantly from zero. The retardedresistance evolution observed in the Homoptera was not due to a higherproportional representation by introduced species (G = 0.09, P > .5). However, 12 of 20 hymenopteranspecies were introduced,a significantlygreater proportion of introducedspecies thanobserved forother orders (otherorders, 149 of 662 species were introduced;G = 12.48, P < .001). Thus, the negativemean residualobserved forthe Hymenopteramay in part reflectthe decreased abilityof introducedspecies to evolve resistance. The magnitudeof the taxonomicorder effect was small comparedto the effectof GPY (largestabsolute value for a mean residualfor a taxonomicorder was 0.046 [fig. 2], halfas large as the maximummean residualfor GPY [fig.1]). Thus, variation in resistanceevolution capacities among differenttaxonomic orders also did not appear sufficientto mask possible underlyinglinear effectsfor GPY. In summary,a surveyof fielddata on resistanceevolution in 682 NorthAmeri- can arthropodpests did not supporta significantlinear relationship between GPY

This content downloaded from 128.120.194.195 on Tue, 23 Sep 2014 01:11:58 AM All use subject to JSTOR Terms and Conditions 534 THE AMERICAN NATURALIST and resistanceevolution. A modestcurvilinear relationship was identified;species with intermediateGPY values showed increased abilities to evolve resistance. This relationshipaccounted foronly 5.7% of the observed variationin residuals, however,suggesting that the effectwas highlyvariable. Thus, available empirical evidence does not supportprevious theoreticalanalyses, which predicta linear relationshipbetween GPY and the rate of resistanceevolution. Reappraisal of ExistingEvidence Empirics.-Previous empiricalsupport for a linear relationshipbetween GPY and resistanceevolution is of limitedscope. May and Dobson (1986) argued that the numberof generationsrequired for resistance to evolve in arthropods,avian coccidia, and nematodes was less variable than the correspondingabsolute time requirements.Their data set, however,included only three data pointsfor arthropod species for which the numberof generationsrequired to develop fieldresis- tance was known. Georghiou(1980) describeda linearrelationship between GPY and the time required for resistance evolution to aldrin/dieldrinin seven soil- dwellinginsect species occurringin differentcrop systemsin differentcountries. The time required to manifestresistance to azinphosmethylamong 12 pest and 12 beneficialarthropods in NorthAmerican apple orchardsalso supporteda direct influenceof generationtime (Tabashnik and Croft 1985). This latterexample is of special interest;despite being derived froma differentmeasure of resistance evolution and a differentsample of arthropodspecies, the data in table 1 also identifiedpests of pome fruitsas a group significantlyinfluenced by GPY. Al- thoughwe cannot fullyexplain why such an effectshould exist forapple arthropods other than as a chance event, the agreementbetween our study and the earlier one (Tabashnik and Croft 1985) is suggestive.Generation time interacts with various genetic, ecological, and operationalfactors to produce an array of effectson resistance evolution (Rosenheim and Tabashnik 1990); it is therefore possible thatconditions in apple orchardsresult in a positiverelationship between GPY and resistance evolution. Because of the predominantlycool climates in which apples and pears are grown,none of the 56 pome fruitpests had a GPY value greaterthan 12; of the remaining626 species in the data base, 37 had GPY values greater than 12. The positive role of GPY observed for pests of pome fruitsmay thereforesimply reflect the increasingtrend in resistance score for species withlow to moderateGPY (fig. 1). Analytical models.-The dominantview regardingthe importanceof generation time has been most lucidly expressed by May and Dobson (1986), who derived a basic formularelating generation time, Tg,to the absolute time required for resistance to appear, TR. Because of the crucial role that this theory has played in shapingthought on resistanceevolution, we review May and Dobson's (1986) analysis. We then demonstratehow a simpleextension of theirmodel can explain the observed absence of a linearrelationship between GPY and resistance evolution. In theirsimplest-case model, May and Dobson (1986) analyzed resistancecon- ferredby one locus withtwo alleles, a resistantallele R and a susceptible allele S, existingat frequenciesPt and qt, respectively,in generationt. The population

This content downloaded from 128.120.194.195 on Tue, 23 Sep 2014 01:11:58 AM All use subject to JSTOR Terms and Conditions GENERATION TIME AND RESPONSE TO SELECTION 535 was assumed to be diploid and closed to gene flow.The per-generationfitnesses, 1tV,of the three genotypes RR, RS, and SS in the presence of pesticide were assumed to satisfythe conditionwRR ? ?VRS "wSs.Standard population-genetics theory(Hedrick 1983) then relates the R allele frequencyin successive generations as

Pt+ (t'RRP? + " RSPtqt)1( P + 2' RSPtqt+ t (1) Two approximationswill be generallyvalid duringthe early stages of resistance: first,pt << qt, and, second, qt 1.0. Equation (1) can thereforebe reduced to

Pt+ i/Pt WRS/tR SS . (2) Equation (2) can be applied to successive generationsto obtain a relationship describingthe numberof generations,n, requiredfor the R allele frequencyto increase fromits initialfrequency, po, to a finalfrequency at which resistance may be considered to have evolved, pf:

Pf/po- (w'RSA/wss * (3)

Notingthat n = TR/Tg,equation (3) may be rearrangedto yield

TR Tgln(pf/po)/ln(11vRS/WSS) . (4) May and Dobson (1986) completed theirderivation with this equation and concluded thereforethat a linear relationshipexists between time to evolve resistance, TR,and generationtime, Tg. They furthermoreconcluded thatthe influence of generationtime should be strongcompared to the influencesof othervariables presentin equation (4), which are relatedonly logarithmicallyto TR. These conclusions rely,however, on the assumptionthat the ratio of fitnesses (lt'RS/vt'ss) remainsconstant for all valuesof Tg,or, in otherwords, that the per- generationselection intensityis the same for species with differentgeneration times. This mightbe true for laboratoryselection experiments,but it does not generallyhold in the field.An example will demonstratethis. Assume thatin the absence of pesticides "t'RS = W55 and thata singlepesticide applicationkills 50% of SS individualsand 0% of RS individuals;for each applicationof pesticide per generation,the ratio of per-generationfitnesses, W'RS/WSS, will thereforeincrease by a factor of two. Consider three pests in an agro-ecosystemthat is sprayed twice per year at 6-mo intervals. Species A has two generationsper year (Tg = 0.5 yr);thus, each generationis sprayedonce, Wi)RS/WSS = 2, and equation(4) can provide an expression forthe timeto resistancefor species A, TR(A), TR(A) 0.5 ln(pf/p0)/ln2 0.72In(pf/po). For the purposes of this discussion, let ln(pf/po)= k, some constant, for all species. Equation (5) may thenbe rewrittenas TR 0.72k. Consider now species B in the same agro-ecosystem,with Tg = 1 yr. Each generationof species B will experience both of the annual sprays; VRS/WSS is thereforeequal to (2)2 or 4. From equation (4), TR(B) lk/ln(4) = 0.72k, exactly the same result as for

This content downloaded from 128.120.194.195 on Tue, 23 Sep 2014 01:11:58 AM All use subject to JSTOR Terms and Conditions 536 THE AMERICAN NATURALIST species A. A thirdspecies, species C, with Tg = 4 yr, experiences eight sprays = per generation;WRS/WSS is then(2)8 _256, and TR(C) 4k/ln(256) 0.72k. The timeto resistanceis clearlyindependent of generationtime. Althoughthis example presentsthe case of regularlyspaced, discretesprays, our argumentis independentof the temporalpattern of pesticide-inducedmortality. The crucial ele- ment is that species with longer generationtimes will, on the average, suffer greaterper-generation pesticide-induced mortality than species withshorter generationtimes. We can generalizethis result within the analyticalframework of May and Dob- son (1986). Let b denote the numberof pesticide applications per year, and a denote the (fractionalsurvival of RS individualsper pesticide application)/(frac- tional survival of SS individualsper pesticide application). Assume that these parametersare constantacross species withinan agro-ecosystem.Then the ratio of per-generationfitnesses can be expressed as

WRSWSS= a (Tgb) (6) and substitutinginto equation (4)

TR Tgln(pf/p0)/ln(a(Tgb)) Tgln(pf/po)/(Tgblna) (7) ln(pf/p0)!(b ln a). Observe that Tg has dropped out of equation (7). This simplest-casemodel pre- dicts, therefore,that the time to resistance is independentof generationtime. The theoreticalanalysis of Comins (1979) also suggestedthat the rate of selec- tionfor resistance could be independentof generationtime. Comins (1979), however, wenton to assume thatspecies withrapid generation turnover require more intensepesticidal suppression, because of theirpopulation's high intrinsic growth rate, and thereforeexhibit rapid selection for resistance. Expressed in termsof the parametersin equation (7), Comins's argumentis thatb, the numberof pesticide applicationsper year, is a species-specificparameter, inversely proportional to each species' generationtime. AlthoughComins's (1979) argumentmay be applicable to some pest species (see, e.g., Longstaff1988), we do not believe it is generallyapplicable for two reasons. First,the typicalcrop is attackedby a largenumber of arthropodspecies, only a handfulof which become "key" pests, that is, pests requiringpesticidal suppression.The relativelylarge numberof secondarypests continues,however, to experience the pesticide applicationsdirected at the key pests. Althoughnot all secondary pests are susceptibleto all pesticide applications(Rosenheim and Hoy 1986), secondarypest populationsare commonlysuppressed by at least some pesticide applications. The timingof these pesticide applications is, however, completelyindependent of life-historyparameters of secondarypests, including theirgeneration times. Thus, thereis no causal linkbetween generation time and pesticide-applicationfrequency for the large majorityof pests that are not the primarytargets of pesticide applicationsand thereforedo not influencethe deci- sion of when to apply pesticides.

This content downloaded from 128.120.194.195 on Tue, 23 Sep 2014 01:11:58 AM All use subject to JSTOR Terms and Conditions GENERATION TIME AND RESPONSE TO SELECTION 537

Second, even for key pests, the relationshipbetween generationtime and pesticide-applicationfrequency may be weak. Generationtime has a stronginflu- ence on the intrinsicrate of population growth(Price 1984). However, many factorsmay cause the realized rate of populationgrowth to fallbelow the intrinsic growthrate, and it is the realized populationgrowth rate, not the intrinsicgrowth rate or the generationtime per se, whichmay influencethe frequencyof pesticide applications. Realized populationgrowth rate is also shaped by the entirerange of biotic and abiotic factors that influencedevelopment rates and age-specific survivaland reproduction,including population interactions with host-plant con- dition, weather, the farmer'sagronomic practices, predators,parasitoids, and pathogens.In addition,there are many instancesin whichthe rate of population rebound followinga pesticide applicationwill not affectpest-management deci- sions. Many crops have relativelynarrow temporal windows of susceptibilityto pest damage. A singlepesticide application made at the onset of plantsusceptibil- itymay be sufficientto decrease populationsto levels low enough thatadditional applications are not required untilthe next crop cycle. The rate of population rebound outside the window of susceptibilitydoes not influencethe pesticide- application frequency.Thus, for many pest species, not all generationsexperi- ence pesticide applications. The frequencyof pesticide applicationsmay also be influencedby otherfactors not relatedto populationdynamics, including the type of damage induced by the targetpest and the monetaryvalue of the crop. In summary,Comins's (1979) argumentlinking pesticide-application frequency to pest-species generationtime does not apply to secondarypests and applies to key pests only under the restrictiveconditions that (1) generationtime is the primarydeterminant of realized populationgrowth rate and (2) realized population growthrate is the primarydeterminant of pesticide-applicationfrequency. Perhaps the strongestevidence arguingagainst the generalityof Comins's thesis is our failureto observe a close relationshipbetween GPY and eitherpest severity or resistance score (table 1). Even forkey pests, comprisingthe 70 species with pest severityscores of 8 or more, there was no significantrelationship between GPY and severityscore (r2 = 0.027, P > .1) or GPY and resistance score. If pesticide-applicationfrequency is generallyindependent of generationtime, our extensionof May and Dobson's (1986) analyticalmodel (eq. [7]) suggeststhat the rate of resistanceevolution is independentof generationtime. This simplest-case model, althoughheuristic, does ignoremany less directinfluences of generation time on resistance evolution, which are explored elsewhere (Rosenheim and Tabashnik 1990). Computer-simulationmodels. -The only computer-simulationstudy explicitly testingthe effectof GPY (Tabashnik and Croft 1982) concluded that increasing GPY consistentlyaccelerated resistanceevolution under a varietyof conditions (with and withoutimmigration, with high or low doses of pesticide). However, in theirsimulations, increasing GPY resultedin a correspondingincrease in the numberof pesticide applicationsper year, and it is clear thatincreased pesticide applicationsdo accelerate resistance.As discussed above withregard to Comins's (1979) analysis, the assumptionof increasingpesticide-application frequency with increasingGPY is not generallyvalid. Thus, computer-simulationmodels cur- rentlyprovide no supportfor a role of generationtime in resistance evolution.

This content downloaded from 128.120.194.195 on Tue, 23 Sep 2014 01:11:58 AM All use subject to JSTOR Terms and Conditions 538 THE AMERICAN NATURALIST

GenerationTime and Fitness Values Our conclusion thatthe evolutionof pesticideresistance depends on the absolute time elapsed ratherthan on the numberof generationselapsed parallels that of Hartl and Dykhuizen (1979; Dykhuizenand Hartl 1981) studyingadaptation of Escherichia coli to a novel environment.Because of the well-definednature of pesticide-inducedselection pressures,we were able to demonstrateanalytically thatper-generation fitness values forresistant genotypes are relatedexponentially to species' generationtimes (eq. [6]). In studies of E. coli strainscompeting for limitingnutrients, Dykhuizen (1978) foundthat the per-generationselection rate was linearlyrelated to generationtime over a wide range of generationtimes. If we ignorepossible effectsof generationtime on mutationrates, the question of whetherthe rate of adaptive evolution is influencedby generationtime becomes equivalentto the questionof whethergeneration time influences the fitness values of given genetic variants. The distinctionbetween traits influencedby density-dependentand density-independentselective forces appears to be crucial in this regard. We do not know whetherthere is a general relationshipbetween generation timeand the relativefitnesses of differentgenetic variants for traits influenced by selective forceswhose intensityis linkedto populationdensity. It seems possible thatfitness values associated withsome such traits,like competitiveability, might be influencedby generationtime, whereas fitnessvalues for other such traits, perhaps includingthose influencedby selection acting on shortparts of the life cycle, the durationof which may be only loosely relatedto total generationtime (e.g., gameticselection or sexual selection),might be less likelyto show such an influence.For traitsshaped by density-dependentselection, the question of the influenceof generationtime on fitnessvalues of givengenetic variants may remain primarilyempirical. We suggest,however, that the relationshipbetween generation time and fitness values thatwe have demonstratedfor pesticide resistance (eq. [6]) is generalizable to traitsmolded by selection that is decoupled fromthe species' life historyand population dynamics,that is, density-independentselection. This type of selection is generatedby manyabiotic factors,including harsh weather (e.g., extreme temperatureor humidityfluctuations, drought, heavy rain,hail) and various envi- ronmentaldisturbances (e.g., fires,floods, volcanic emissions). Generalistpreda- tors and parasites may also act in a density-independentmanner if the prey or host species in question consistentlyforms a small fractionof the total prey or host pool. Finally, interspecificcompetition may be independentof densityif a resource is exploited by anotherspecies, the populationdynamics of which are unaffectedby the species in question. Regardless of the specificselective agent, the per-generationintensity of selectiongenerated by density-independentfactors is relatedexponentially to generationtime. For example, in a habitatthat experiences periodic freezes, a long-livedplant may, on the average, experience a large numberof freezes (selection bouts) per generation,whereas an annual or a short-livedperennial might experience a singlefreeze or escape selectionentirely; in general the numberof freezes experiencedper generationis directlypropor-

This content downloaded from 128.120.194.195 on Tue, 23 Sep 2014 01:11:58 AM All use subject to JSTOR Terms and Conditions GENERATION TIME AND RESPONSE TO SELECTION 539 tional to generationtime. The argumentdeveloped above (eqq. [1]-[7]) for the evolution of pesticide resistance is thereforerelevant to the evolution of traits contributingto cold tolerance. In fact, for many traitslike cold tolerance, our argumentis actuallysimplified somewhat because thereis a greatlyreduced likeli- hood of the species' populationdynamics' influencingthe frequencyor intensity of selection bouts. We conclude that in our simplest-casescenario the rate of evolutionaryresponse to density-independentselection is independentof generation time.

ACKNOWLEDGMENTS We thankR. Carlson for providingNAIAD, and M. A. Caprio, L. A. Freed, S. Gilboa, M. A. Hoy, R. T. Roush, and C. M. Simon fortheir critical reviews of the manuscript.This work was supportedby U.S. Departmentof Agriculture (USDA) grant HAW00947H, Western Regional IntegratedPest Management grant 8902553, USDA/CSRS (Cooperative State Research Service) Special Grants in Tropical and Subtropical Agriculture8902558, and a Fujio Matsuda Scholar Award (to B.E.T.) fromthe Universityof Hawaii Foundation. This is paper 3441 in the Hawaii Instituteof Tropical Agricultureand Human Resources JournalSeries.

LITERATURE CITED

Borror, D. J., D. M. DeLong, and C. A. Triplehorn.1981. An introductionto the studyof insects. 5th ed. Saunders, Philadelphia. Clausen, C. P., ed. 1978. Introduced parasites and predatorsof arthropodpests and weeds. U.S. Departmentof AgricultureHandbook no. 480. Comins, H. N. 1979. The managementof pesticide resistance: models. Pages 55-69 in M. A. Hoy and J. J. McKelvey, Jr.,eds. Geneticsin relationto insectmanagement. Rockefeller Founda- tion, New York. Davidson, R. H., and W. F. Lyon. 1987. Insect pests of farm,garden, and orchard. 8th ed. Wiley, New York. Dixon, W. J., ed. 1985. BMDP statisticalsoftware. University of California,Berkeley. Dykhuizen, D. 1978. Selection for tryptophanauxotrophs of Escherichia coli in glucose-limitedche- mostatsas a test of the energyconservation hypothesis of evolution. Evolution 32:125-150. Dykhuizen, D., and D. Hartl. 1981. Evolution of competitiveability in Escherichia coli. Evolution 35:581-594. Easteal, S. 1988. Rate constancyof globingene evolutionin placental mammals. Proceedingsof the National Academy of Sciences of the USA 85:7622-7626. Furniss, R. L., and V. M. Carolin. 1977. Westernforest insects. U.S. Departmentof Agriculture Forest Service Miscellaneous Publicationno. 1339. Georghiou, G. P. 1980. Insecticide resistance and prospectsfor its management.Residue Reviews 76:131-145. 1981. The occurrenceof resistanceto pesticidesin arthropods.Food and AgricultureOrgani- zation of the United Nations, Rome. Gillespie, J. H. 1986. Natural selection and the molecular clock. Molecular Biology and Evolution 3:138-155. Graur, D., Y. Shuali, and W.-H. Li. 1989. Deletions in processed pseudogenes accumulate fasterin rodentsthan in humans. Journalof Molecular Evolution 28:279-285. Hall, W. J., ed. 1950-1953a. Review of Applied Entomology,Ser. A: Agricultural.Vols. 38-41. CommonwealthInstitute of Entomology,London.

This content downloaded from 128.120.194.195 on Tue, 23 Sep 2014 01:11:58 AM All use subject to JSTOR Terms and Conditions 540 THE AMERICAN NATURALIST

. 1950-1953b. Review of Applied Entomology,Ser. B: Medical and Veterinary.Vols. 38-41. CommonwealthInstitute of Entomology,London. Hartl, D. L. 1988. A primerof populationgenetics. 2d ed. Sinauer, Sunderland,Mass. Hartl, D., and D. Dykhuizen. 1979. A selectively driven molecular clock. Nature (London) 281:230-231. Hedrick, P. W. 1983. Genetics of populations.Science Books International,Portola Valley, Calif. Longstaff,B. C. 1988. Temperaturemanipulation and the managementof insecticide resistance in stored grain pests: a simulationstudy for the rice weevil, Sitophilus oryzae. Ecological Modelling43:303-313. May, R. M., and A. P. Dobson. 1986. Population dynamicsand the rate of evolution of pesticide resistance. Pages 170-193 in Pesticide resistance: strategiesand tactics for management. National Academy of Sciences, Washington,D.C. Maynard Smith,J. 1984. The ecology of sex. Pages 201-221 in J. R. Krebs and N. B. Davies, eds. Behavioural ecology. 2d ed. Sinauer, Sunderland,Mass. Nunney, L. 1989. The maintenanceof sex by group selection. Evolution 43:245-257. Palumbi, S. R. 1989. Rates of molecular evolution and the fractionof nucleotide positions free to vary. Journalof Molecular Evolution 29:180-187. Price, P. W. 1984. Insect ecology. 2d ed. Wiley, New York. Rosenheim, J. A., and M. A. Hoy. 1986. Intraspecificvariation in levels of pesticide resistance in fieldpopulations of a parasitoid,Aphytis melinus (Hymenoptera: Aphelinidae): the role of past selection pressures. Journalof Economic Entomology79:1161-1173. Rosenheim,J. A., and B. E. Tabashnik. 1990. Evolutionof pesticideresistance: interactions between generationtime and genetic,ecological, and operationalfactors. Journal of Economic Ento- mology 83:1184-1193. Sailer, R. I. 1978. Our immigrantinsect fauna. Bulletin of the Entomological Society of America 24:3-11. 1983. Historyof insect introductions.Pages 15-39 in C. L. Wilson and C. L. Graham, eds. Exotic plant pests and North Americanagriculture. Academic Press, New York. Sarich, V. M., and A. C. Wilson. 1973. Generationtime and genomicevolution in primates.Science (Washington,D.C.) 179:1144-1147. Tabashnik, B. E. 1990. Modeling and evaluationof resistancemanagement tactics. Pages 153-182 in R. T. Roush and B. E. Tabashnik,eds. Pesticide resistancein arthropods.Chapman & Hall, New York. Tabashnik, B. E., and B. A. Croft.1982. Managingpesticide resistance in crop-arthropodcomplexes: interactions between biological and operational factors. Environmental Entomology 11:1137-1144. 1985. Evolution of pesticideresistance in apple pests and theirnatural enemies. Entomophaga 30:37-49. Wu, C.-I., and W.-H. Li. 1985. Evidence for higherrates of nucleotide substitutionin rodentsthan in man. Proceedings of the National Academy of Sciences of the USA 82:1741-1745. Zuckerkandl,E. 1987. On the molecularevolutionary clock. Journalof Molecular Evolution26:34-46.

This content downloaded from 128.120.194.195 on Tue, 23 Sep 2014 01:11:58 AM All use subject to JSTOR Terms and Conditions APPENDIX

AN EXAMPLE OF THE DATA USED TO INVESTIGATE THE INFLUENCE OF THE NUMBER OF GENERATIONS PER YEAR (GPY) ON THE EVOLUTION OF PESTICIDE RESISTANCE: NORTH AMERICAN ARTHROPOD PESTS OF COTTON

Introduced (i) or Pest Resistance Taxonomic ScientificName Native (n) GPY* Severityt Scoret Order?

Heliothis zea (Boddie) n 4 70 4 Lep Anuraphisinaidiradicis (Forbes) n 12 0 0 Hom Petrobia latens (Muller) n 18.5 4 0 Aca Lygiuslineolaris (Palisot de Beauvois) n 4 20 1 Hem Anthoniomusgrandis grandis Boheman n 5 47 2 Col Aphis gossypiiGlover n 31.5 41 1 Hom Aphis craccivora Koch i 30.211 2 0 Hom Tetranychuscinnabarinus (Boisduval) n 20 1 Aca T. turkestani(Ugarov & Nikolskii) n 12 4 0 Aca T. urticae Koch n 15.5 73 5 Aca T. pacificus McGregor n 15 15 2 Aca T. desertorlumBanks n 15 0 0 Aca T. tiumidusBanks n 18.5 0 1 Aca T. schoenei McGregor n 9 2 2 Aca T. canadensis (McGregor) n 18.5 0 1 Aca Pseudatomoscelis seriatus (Reuter) n 7 6 1 Hem Lygus elisus Van Duzee n 4 4 1 Hem L. hesperus Knight n 4 4 3 Hem Adelphocorisrapidus (Say) n 211 4 0 Hem A. superbus (Uhler) n 211 0 0 Hem Alabamnaargillacea (Hubner) i 5 11 2 Lep Pectiniophotagossypiella (Saunders) i 5 4 1 Lep Estigmeneacrea (Drury) n 2.5 15 2 Lep Strymnonmelinus (Hubner) n 2.5 0 0 Lep Frankliniellafusca (Hinds) n 8 3 0 Thy F. exigua Hood n 8 1 0 Thy F. gossypiana Hood n 8 0 Thy F. occidentalis (Pergande) n 6 2 1 Thy F. tritici(Fitch) n 12.5 3 1 Thy Caliothripsfasciatus (Pergande) n 6 0 0 Thy Thripstabaci Lindeman i 7.5 17 2 Thy Sericothripsvariabilis (Beach) n 0 0 Thy Homalodisca triquetra(F.) n 1 0 Hom Aulacizes irrorata(F.) n 0 0 Hom Oncomnetopiaundata (F.) n 1 0 Hom Cuerna costalis (F.) n 1 0 Hom Anthonomusgrandis thurberiaePierce n 3 0 0 Col Chlorochroaligata (Say) n 1 0 0 Hem Acrosternumhilare (Say) n 2 0 0 Hem Euschistus impictiventris(Stal) n 1 0 0 Hem Nezara viridula(L.) i 2.5 2 0 Hem Chlorochroasayi Stal n 2 0 0 Hem Dysdercus suturellus(Herrich-Schaffer) n 0 0 Hem D. mimulusHussey n 0 0 Hem Bucculatrixthurberiella Busck n 5 0 4 Lep

* Missing values are species for which we were unable to locate estimatesof GPY. Full literature citationsfor GPY estimatesare available fromthe authors. t Species with missingvalues were not clearly definedtaxonomically in 1950. t The resistance score is the numberof insecticide/acaricideclasses to which at least some North Americanfield populations have been reportedas resistant(Georghiou 1981). ? Lep, Lepidoptera; Hom, Homoptera; Aca, Acarina; Hem, Hemiptera; Col, Coleoptera; Thy, Thysanoptera. 11Values were estimatedby using the mean value reportedfor congeners.

This content downloaded from 128.120.194.195 on Tue, 23 Sep 2014 01:11:58 AM All use subject to JSTOR Terms and Conditions