Received 28January 2003 Accepted 14 May 2003 Publishedonline 25 July 2003

Reproductive isolationdriven bythe combinedeffects of ecologicaladaptation and reinforce ment P. Nosil1* ,B.J.Crespi 1 and C.P.Sandoval 2 1Departmentof Biosciences, BehavioralEcology Research Group, 8888 University Drive,Simon FraserUniversity, Burnaby, BritishColumbia V5A 1S6, Canada 2MarineScience Institute, University ofCalifornia, Santa Barbara, CA 93206,USA Recentyears have seena resurgenceof interestin theprocess of speciationbut few studies have elucidated themechanisms either driving or constraining theevolution of reproductive isolation. In theory, thedirect effectsof reinforcing selectionfor increasedmating discrimination whereinterbreeding produceshybrid offspring with lowfitness and the indirect effectsof adaptation todifferent environments can both promote speciation.Conversely, high levels ofhomogenizing geneflow can counteract the forces of selection. We demonstratethe opposing effectsof reinforcing selectionand gene flow in Timema cristinae walking-stick insects.The magnitude offemale mating discrimination against males from other populations is greatest whenmigration rates betweenpopulations adaptedto alternate hostplants are high enoughto allow the evolutionof reinforcement, but low enough to prevent gene flow from eroding adaptive divergencein mate choice.Moreover, reproductive isolation is strongestunder the combined effects of reinforcement andadaptation toalternate hostplants. Our findingsdemonstrate the joint effects of reinforcement, eco- logical adaptation andgene flow on progress towardsspeciation in thewild. Keywords: geneflow; hybridization; natural selection;speciation; Timema;walking-stick insects

1. INTRODUCTION distributedthroughout westernNorth America (Crespi& Sandoval 2000). Timema cristinae exhibits twogenetically Speciation via natural selectioncan occur as an indirect determinedcolour-pattern morphs (Sandoval 1993), with by-productof adaptive divergence(Mayr 1963; Jiggins et anunstriped morph being more commonon thehost plant al. 2001) andcan also involve directselection for repro- Ceonothusspinosus anda stripedmorph being more com- ductiveisolation in geographical regions wherehybridiz- mon on Adenostomafasciculatum (Sandoval 1994 a). Pre- ation is maladaptive (i.e.reinforcement) (Dobzhansky dation on T.cristinae by birds andlizards is strong and 1951; Howard1993; Butlin 1995; Noor 1999). When eachmorph ismost cryptic onthe plant onwhich it is speciationoccurs as a by-productof adaptive divergence, more common(Sandoval 1994 a,b).Populations usingdif- ecologically divergent populations exhibit greater repro- ferenthost plants have also diverged in asuiteof other ductiveisolation than ecologically similar populations of morphological andbehavioural traits, including bodysize similar age (reviewedin Funk1998; Rundle et al. 2000; andshape, host preference and cryptic resting behaviour, Schluter2000; Funk et al. 2002). The key predictionof wherehost preference refers to a preferenceto rest on a thereinforcement hypothesis is that non-allopatric particular host,given thechoice (Nosil et al. 2002; Nosil & (geographically contiguousor overlapping) populations Crespi2003). will exhibit greater mating discrimination than allopatric Levelsof sexual isolation are higher betweenpairs of T. (geographically separated)populations. Previous empirical cristinae populations usingdifferent host plants ( n = 15 studiesof reinforcementhave provided evidencefor sucha pairs) than betweensimilar-aged pairs usingthe same host pattern (Noor 1995; Saetre et al. 1997; Rundle& Schluter plant (n = 13 pairs) (Nosil et al. 2002). Becauseinsect col- 1998; Higgie et al. 2000) butthere are fewdata onhow ourmorph (stripedversus unstriped) does not influence ecological adaptation andreinforcement interact during between-populationmate choice,the sexual isolation that thespeciation process (Schluter 2000). Furthermore, hasevolved betweenpopulations adaptedto alternate although theoretical modelspredict that theoutcome of hostsis independentof colour pattern (although colour reinforcementreflects a balance betweenthe strength of pattern might influencewithin-population mate choice; reinforcing selectionand the ability ofhomogenizing gene Nosil et al. 2002). In this study,we expand previous work flowbetween populations tocounteract selection by consideringthe effects of reinforcing selectionand gene (Sanderson1989; Servedio& Kirkpatrick 1997; Cain et flowon between-population mating preferences. al. 1999; Servedio2000), this prediction has never,to our Timemacristinae satisfiesall ofthe preconditions for knowledge,been tested using natural populations. reinforcementto contribute to the reproductive isolation In this paper, weuse Timema walking-stick insectsto observedbetween populations. First, interbreeding and analyse thejoint roles of ecological adaptation, reinforce- geneflow between adjacent populations usingdifferent mentand gene flow in theevolution of reproductive iso- hostplants (i.e.‘ hybridization’between the host-adapted lation. Timema are winglessphytophagous insects forms)is strongly suggestedby theobservations that indi- viduals from adjacentpopulations ondifferent hosts are well within oneper-generation dispersal distanceof each *Authorfor correspondence ([email protected]). other (Sandoval 1993, 1994 a)andthat between-morph

Proc.R. Soc.Lond. B (2003) 270, 1911–1918 1911 Ó 2003 TheRoyal Society DOI10.1098/ rspb.2003.2457 1912P. Nosil andothers Natural selection andspeciation matings are commonin naturewhere populations using differenthosts are in contact(Nosil et al. 2002). More- over,geographical variation in morphology isindicative of parapatric 1 abalance betweenhost-specific selection and gene flow population A C betweenadjacent patches (Sandoval 1994 a;Nosil &Cre- C spi 2003), andthe populations are conspecific.In this allopatric study,we use mitochondrial DNA(mtDNA) sequence C variation toprovide further evidenceof interbreeding and population 2 geneflow between adjacent populations of T.cristinae . Second,the evolution of reinforcement traditionally parapatric C A 3 requiresreduced hybrid fitness.In T.cristinae, between- population hostmigrants are more likely tobe of the locally non- C matching colourmorph than are residents(Sandoval allopatric A 1994a)andthe non-matching morph is at alarge selective population 4 disadvantage owing todifferential predation (therelative fitnessof the less-cryptic morph is 0.30; Sandoval Figure 1. Representative examplesof thefour ‘types’ of 1994a,b).Thusoffspring derivedfrom between-hostmat- between-population crosses (C, Ceonothus population; A, Adenostoma population). Eachsquare represents a ing (i.e. ‘hybridization ’)will tendto exhibit reducedfitness homogeneous patchof asingle host plant, whichmay or relative tooffspring derivedfrom within-population mat- maynot havea neighbouring population using the ing becausefemales who mate with locally less-cryptic alternative host plant (parapatric and allopatric populations, males producea higher frequencyof offspring that are the respectively). ‘Study’ populations used in mating trials are locally non-matching colourmorph or intermediate in col- represented byunfilled boxes, and blackboxes represent ourpattern (i.e.bear afaint stripe) than femalesthat mate populations thatare adjacent toastudy population but were with cryptic males (experimental geneticcrosses show that not used in mating trials. In thefigure, malesfrom allopatric Ceonothus populations are used asan illustrative example; in only 5% ofwithin-morph crossesproduce F 1 offspring of thealternate morph or intermediates,whereas 80% of themating trials malesfrom eachof the12 study between-morphcrosses produce both morphs or inter- populations were used (i.e. malesfrom allopatric and parapatric populations on eachhost). 1, femaleparapatric, mediates;Sandoval 1993). Although previous workdem- male from thesame host; 2, femaleallopatric, male from the onstratesthat femalesproducing offspring ofthe samehost; 3, femaleparapatric, male from thealternative non-matching colour-patternmorph will beat aselective host; 4, femaleallopatric, male from thealternative host. disadvantage (Sandoval 1994 a,b),thecost of producing offspring that are intermediate in colourpattern is lesswell known.In thecurrent study, we test whether selection against theintermediate colour-patternmorph also are geographically separatedfrom all other hostpatches imposesreinforcing selection,predicting that if theinter- by regions lacking suitable hosts.We define a ‘population’ mediate morph exhibits lowfitness then its frequencywill ofwalking-stick insectsas all theinsects collected within declinethrough time (i.e.between sequential time periods ahomogeneouspatch ofa single hostplant. ‘Parapatric’ andage classes;Endler 1986). Becausepopulations using insectpopulations are in contactwith apopulation of differenthosts have diverged in anumberof traits other insectsadapted to the alternate host,while ‘allopatric’ than colourpattern (Nosil et al. 2002; Nosil &Crespi populations are separatedfrom all populations adaptedto 2003), selectionagainst ‘hybrids’ may extendbeyond that thealternative hostby distancesof more than 50 timesthe basedupon colour pattern alone. 12 mper-generation gene-flowdistance (Sandoval 1993). Finally, wenote that recenttheoretical workindicates Reinforcementwould be supported by higher between- that directselection against hybrid mating canpromote population copulation frequenciesin mating trials involv- reinforcement,even without reduced hybrid fitness ing allopatric femalesthan in trials involving parapatric (Servedio2001). In T.cristinae ,themale rideson the back females. ofthe female during themating period andthus females Reinforcing selectionis predictedto be frequency that mate with locally less-crypticmales are likely tosuffer dependent,with increasing rarity ofapopulation (relative reducedindividual survival during mating, favouring mat- tothe population with whichit co-occurs)raising the ing discrimination against migrants from thealternate probability ofindividuals from theco-occurring popu- host. lation mating with individuals ofthe wrong population Giventhat thepreconditions for reinforcementare met, andthus the opportunity for reinforcing selection wetestedfor its presenceby assessinglevels offemale mat- (Howard1993; Noor 1995). However,increased opport- ing discrimination against foreign males (males that were unity for between-populationmatings also increasesthe collectedfrom adifferentpopulation from thefemales) for potential for geneflow between populations, which can walking-stick insectscollected from eight parapatric popu- retard or preventreinforcement. Thus, the actual magni- lations (reinforcementpossible) and from fourallopatric tudeof female mating discrimination that evolvesis populations (noopportunity for reinforcement).Mating expectedto reflect a balance betweenthe opposing forces trials wereconducted for all 66 possiblepairwise combi- ofreinforcing selectionand gene flow, with theeffects of nationsof the 12 studypopulations (figure 1). Most reinforcementbeing greatest whenpopulation sizesare patchesof these two host-plant species are distributedin similar (Sanderson1989; Servedio& Kirkpatrick 1997; parapatric patchesof varying size,forming amosaic at the Cain et al. 1999; Servedio2000). Becausewe quantified scaleof a mountainslope. However, some host patches female mating discrimination in multiple populations of

Proc.R. Soc.Lond. B (2003) Natural selection andspeciation P.Nosil andothers 1913 differing sizes,we were able totest these key predictions inhabitedby study population, the ‘rarer’ the individualsof the ofthe reinforcement hypothesis. study population becomerelative to individualsof the neigh- Collectively, weprovide anassessment of the joint bouring population. effectsof host adaptation, reinforcing selectionand gene The rarity ofeachstudy population was calculatedas (sizeof flowon the evolution of reproductive isolation, predicting neighbouring patch)/(sizeof study patch 1 sizeof neighbouring that: (i) reinforcing selectionstrengthens premating iso- patch). The areas of eachpopulation and its neighbour (inthe lation; (ii) high levels ofgene flow counteract the effects of caseof parapatric patches) werecalculated using aerialphoto- reinforcing selection;and (iii) thestrongest reproductive graphs. Patch sizehas beenshown to bestrongly and positively isolation is foundunder the combined influence of diver- correlatedwith population size( r2 = 0.63and 0.53for Ceonothus genthost-plant adaptation andreinforcement. and Adenostoma patches, respectively; n = 13patches of each host; Sandoval 1994 a).The strength of femalemating discrimi- nation against foreignmales was calculatedfor eachof the 12 2. MATERIALAND METHODS study populations as the absolute valueof the meancopulation (a) No-choice matingexperiment frequencyof femaleswith foreignmales subtracted fromthe Timemacristinae werecollected from multiple study sitesin meancopulation frequency of femaleswith malesfrom their the Santa YnezMountains, California, USA, inFebruary 2001 own population. and 2002using sweep nets.Study siteswere chosen such that awiderange of geographical arrangements of populations was (ii) DNAsequencing and estimatesof gene flow represented.Other species of Timema do not occurin sympatry Atotal of 107mtDNA (cytochromeoxidase 1) sequences, with populations fromthese sites.Animals were captured in the 467bp inlength, werecollected from the 12study populations first instar and rearedto maturity inthe laboratory onthe foliage and fromeach of the populations that neighbour the parapatric of eithertheir native or the alternativehost. study populations (meannumber of individualsper Protocolsfor the no-choicemating trialsused in this study population = 6.3,range of 3 –11;protocols and 40sequences have beenpreviously published (Nosil et al. 2002) (n = 3320 from Nosil et al. (2002)).Haplotypes fromthe 67sequences trials;1024 of these fromNosil et al. (2002);median number acquiredin this study have beendeposited in GenBank under of mating trialsper pairwise comparison = 275,range of 75 – accessionnumbers AY311411 –AY311477. 497).One male and onefemale were placed in a 10cmPetri Wethen estimatedlevels of geneflow betweenadjacent dish and at the endof 1hwe scoredwhether the maleand the patches to test whether geneflow intoa parapatric population, femalewere paired (male on female without genitalcontact) or fromits neighbouring population ofthe alternatehost, increases not, and copulating ornot. Individualswere selected randomly with increasingrelative size of its neighbouring population. Gene fromeach population, such that mating trialswere conducted flow was estimatedusing the coalescent-basedmethods of using natural morph frequencies.In mating trialsconducted for Beerli& Felsenstein(2001). First, we useddefault settings in the currentstudy, eachindividual was testedonly once. the program M igrate to obtain estimatesof the numberof

Copulationfrequencies (proportion of male –femalepairs that migrants pergeneration ( Nm)intoeach of the parapatric study werecopulating at the endof 1h)were analysed using logistic populations, fromtheir neighbouring population ofthe alternate regression,assessing significanceusing likelihoodratio tests host. Wethen estimatedmigration rates using two independent (LR).We tested for reinforcementin amodelthat examinedthe approaches. First,we estimated m (theproportion of the popu- dependenceof copulationwith malesfrom each of the 12study lationconsisting of migrants, migrationrate) from Nm by calcu- populations on malehost, femalehost, malepopulation, female lating total population size(using previously published population, allopatry (femalefrom an allopatricor parapatric regressionequations for patch sizeversus population size;San- population)and allpossible interaction terms. Host-specific sex- doval 1994a),dividingthis numberby two to obtain female ualisolation is indicated by asignificantinteraction between population size(mtDNA ismaternallyinherited) and then mul- malehost and femalehost. Reinforcementis indicatedby asig- tiplying by 0.10to obtain effectivepopulation sizes(Frankham nificantallopatry term.We donot report the significancesof the 1995)(changing this finalscaling value alters only the absolute other termsin the model,as they areperipheral to the topicof estimatesof m,whereas resultsof ouranalyses and their the study. Allresults are from a reducedregression model interpretationdepend only on variation in relative migration derivedusing backward elimination(the model initially included rates).Second, we alsoreport the migrationparameter M (M allfactors and interactionsbut then removedall terms for which = m/mutation rate),obtained fromM igrate.Wenote that our the significanceof –2log LRwas morethan 0.10).Including analyses of the relationshipbetween gene flow and neighbouring- rearingenvironment (insect raised on its nativeor the alternative population sizedepend on relativemigration rates and thus esti- host) inthe logisticregression models yielded no significant mating m from M using differentmutation rates willnot affect interactions(all main effects and interactions: p . 0.25),indi- ourresults. cating that differencesin mate preference probably have agen- eticbasis. Analogous analyses conductedusing males(male (b) Selection againstthe intermediate pattern fromparapatric orallopatric populations) didnot yieldevidence morph of reinforcement(all main effects and interactionswith allopa- The intermediatemorph israrein most populations of T. cris- try: p . 0.25). tinae (lessthan 2%).However, one site (Refugio) has relatively high frequenciesof the intermediatemorph, providingan (i) Populationrarity and theopportunity for reinforcement opportunity to test for selectionagainst the intermediatemorph In T.cristinae ,the areaof the host-plant patch of the study within anatural population. This sitecontained both host population relativeto that of its parapatric neighbour servesas plants, eachof which was sampledduring March and Aprilin ameasureof the study population ’srarity:the largerthe patch 1996and 1997.Captured individualswere scored for colour inhabitedby neighbouring population and the smallerthe patch pattern (unstriped,striped, intermediate) and age class(juvenile,

Proc.R. Soc.Lond. B (2003) 1914P. Nosil andothers Natural selection andspeciation

253 288 504

h 318 p r o y c m

0.40 n

e 0.4 92 e t 686 693 a u i q d e e r f

m r 0.2 110 104 n e o 0.30 t i t n i a 831

l

f 141 u o p

o y 0.1 c c n

e 0.20 u q

e 1 2 3 4 r f 0 juveniles adults adults hosts : same samedifferent 1996 1996 1997 populations: same different different

Figure 2. Thefrequency of theintermediate colour-pattern type of comparison morph declined in successive ageclasses within asample period and in successive samplemonths (blackbars, March; Figure 3. Copulation frequencies for male –femalepairs of grey bars, April). Numbers aboveeach bar refer to the Timema cristinae walking-stickinsects. Consistent with samplesize. reinforcement, between-population copulation frequencies were higher for mating trials involving femalesfrom allopatric populations (rectangles) thanfor trials involving adult;we found juvenilesonly during March 1996) and released femalesfrom parapatric populations (circles) ( p , 0.001). wherecaptured. All specimens were scored by oneindividual Theincreased mating discrimination of parapatric females (C.P.S.). wasindependent of host-specific sexual isolation (male host ´ femalehost, LR = 11.70, d.f. = 1, p , 0.001; all other Wefirst assessedwhether the frequencyof the intermediate interactions, p . 0.15). Thusreproductive isolation has morph was dependenton age classby testing for an association evolved both viareinforcement and asa by-product of betweenmorph (intermediateor other)and age class(using the adaptation to different habitats.Numbers of mating trials for March1996 sample) in a three-way log-linearcontingency eachcomparison are shown aboveeach 95% confidence analysis that includedmorph, age classand host as factors. interval. Thenumbers abovethe x-axisrefer to thetype of Second,we tested whether the frequencyof the intermediate between-population mating trial asoutlined in figure 1. All morph within adults was dependentupon samplemonth using combinations of mating trials were conducted, such that afour-way log-linearanalysis that alsoincluded sample year as some femalesfrom both parapatric and allopatric afactor.For both analyses,we examinedthe effectsof the inter- populations were tested withmales from their own actionsin questionindependent from the effectsof other factors population, some withmales from different populations using thesame host and some withmales from different by using partial x2-valuesand by assessing the significanceof populations using thealternate host. higher-orderinteractions (Norusis 1993).

p = 0.08; all higher-order interactions: p . 0.10). These 3. RESULTS resultsdemonstrate selection against theintermediate (a) Gene flowand hybridization in thewild morph during thejuvenile stages, possibly continuinginto Adjacentpairs ofpopulations usingdifferent host plants adult life. Ongoing interbreeding betweenthe parental are weakly or notdifferentiated at mtDNA (mean forms(i.e. striped and unstriped morphs) couldaccount

Fst = 0.07, range of0.00 –0.25, n = 7pairs) while geo- for thepersistence of theintermediate morph at thestudy graphically separatedpopulations are strongly differen- site,despite selection against it. tiated (mean Fst = 0.31, range of0.00 –0.79, n = 129 pairs; Mantel’s t = 2.33, p , 0.01). Thesedata suggestthat sub- (c) Reinforcement ofpremating isolation stantial geneflow between neighbouring populations Female mating discrimination against foreign males is occursin thewild (seealso coalescent-basedestimates of significantly stronger whenthe females are from popu- gene flow § 3d).Thus, although incompletelineage sort- lations wherethe two host-adapted forms hybridize than ing (betweenneighbouring populations only) couldpro- whenthe females are from geographically isolated popu- ducesimilar patterns,the requirement for reinforcement lations (figure 3; meanbetween-population copulation fre- ofrecent hybridization appears tobe fulfilled. quencyin mating trials involving femalesfrom parapatric populations = 28%, s.d. = 0.45; in trials usingfemales (b) Selection againstintermediate colour-pattern from allopatric populations = 35%, s.d. = 0.47; allopatry morphs term: LR = 11.93, d.f. = 1, p , 0.001). Notably, thecopu- The frequencyof the intermediate morph washigher in lation frequenciesof females with males from their own juvenilesthan in adultssampled during thesame time per- population weresimilar for femalesfrom parapatric and iod(partial x = 53.40, d.f. = 1, p , 0.001; higher-order allopatric populations (allopatry term in amodel including interaction: p . 0.10; figure 2). In addition,the frequency within-population mating trials only:LR = 1.40, d.f. = 1, ofthe intermediate morph within theadult age class p = 0.24), indicating that reinforcementhas strengthened decreasedbetween March andApril. This reduction female mating discrimination against foreign males with- occurredin both sample years,but the overall resultswere outreducing mating frequencieswith local males.More- marginally non-significant(partial x = 2.96, d.f. = 1, over,between-population copulation frequencieswere

Proc.R. Soc.Lond. B (2003) Natural selection andspeciation P.Nosil andothers 1915

geographical scenario (e) Combinedeffects ofhost-adaptation andreinforcement Host-plantadaptation andreinforcement contributed independentlyand additively tothe evolution of repro- 0.40 ductiveisolation (figure 3). Thus,copulation frequency n i

wasreduced when the sexes were from populations using n o

i 0.30 alternate hosts(for both allopatric andparapatric females) t a

n andmating discrimination ishighest whenboth adap- i m

i tation andreinforcement occur. r c

s 0.20 i d (f ) Alternative hypotheses e l a Numerousprocesses have beenpresented that could m

e 0.10 f accountfor increasedmating discrimination in parapatric 0 0.35 0.650.85 0.95 orsympatric populations relative toallopatric populations rarity of relative to (reviewedin Howard1993; Butlin 1995; Noor 1999). Eachof these hypotheses can be viewed as an alternative Figure 4. Female mating discrimination againstmales from toreinforcement. As describedbelow, each of these alter- other populations is strongest when therarity of thestudy native hypotheseswas unsupported. population isintermediate (malesfrom thealternative and First, whensimilar phenotypesfrom differentpopu- samehost, respectively; t = 22.37, 22.25, both p , 0.05, lations competemost strongly for resources,frequency- quadratic term in regression model including both linear and dependentdisruptive selectiondrives population diver- quadratic terms; r 2 change between alinear and quadratic gence(Slatkin 1980). This phenomenon,called ecological model = 0.37, 0.33, both p , 0.05, partial F-test). Shown here is therelationship between therarity of thestudy character displacement,results in populations that are in population (relative to its neighbouring population on the geographical contactwith oneanother exhibiting greater alternative host; values for eachstudy population denoted by adaptive divergencethan allopatric pairs ofpopulations. blackcircles on the x-axis)and femalemating discrimination Asa consequenceof this greater trait divergence,mating againstforeign malesthat use thealternative host (absolute discrimination canbe stronger betweenparapatric than value of mean copulation frequency withforeign males betweenallopatric pairs ofpopulations.Previous workhas subtracted from mean copulation frequency withresident shownthat ecological character displacementin mor- males, for eachof the12 study populations). Boxes illustrate phology doesnot occur in T.cristinae :divergencesin thedifferent geographical scenarios, where thegrey box morph frequency,body sizeand body shapeare greater denotes thestudy population and theblack box denotes the betweenallopatric pairs ofpopulations usingdifferent neighbouring population. Thecurve wasestimated using the hoststhan betweenparapatric pairs (Sandoval 1994 a; non-parametric cubic spline (dashed lines showstandard errors from 1000 bootstrap replicates) (Schluter 1988). Nosil &Crespi2003). In this study,we testedfor charac- ter displacementin resting behaviour andin acorrelate ofphysiology (survival tomaturity in thelaboratory) by assessingthe effects of host ( Ceonothus versus Adenostoma),allopatry (allopatry versusparapatry) andan lowerfor femalesfrom theparapatric population than the interaction term onvariability among the12 studypopu- average from thefour allopatric populations in sevenout lations.Resting behaviour refersto whether the insects ofthe eight population comparisons(Wilcoxon ’s signed werefound resting wherevisible from theside (versus ranks test(WSR): Z = 2.24, p , 0.05). Thesedata provide above or below)in 1073 host-preferencetrials (seeNosil strong evidencefor reinforcementof premating isolation. et al. (2002) for details).For thesebinary variables, we assessedthe effects of host, allopatry andthe interaction (d) Populationsize, gene flowand reinforcement term usinglogistic regression. Coalescent-basedanalyses indicate that theopportunity The character-displacement hypothesisis untenable for geneflow to erode the effects of reinforcing selection becauseindividuals from parapatric populations usingdif- increaseswith increasing population rarity: levels ofgene ferenthosts are lessdivergent from oneanother than are flowinto the parapatric studypopulations, from their individuals from allopatric populations usingdifferent neighbouring populations,were highly variable (range of hosts(mean trait valuesfor individuals from parapatric numberof migrants ( Nm ) = 0.68–14.7, mean = 3.13; range versusallopatric populations;resting behaviour, 8% and ofmigration rates estimatedusing population sizes 36%, respectively; LRfrom interactionsterm = 28.03, (m) = 0.001–0.232, mean = 0.043; range ofmigration p , 0.001; survival in laboratory, 4% and5%, respect- parameter (M) = 22.70–300.43, mean = 177.83) and ively; LR = 0.25, p . 0.10). increasedwith increasing population rarity ( r = 0.86, 0.62, Second,we evaluated whetherpopulation ancestry(i.e. 0.92, p , 0.01, 0.05, 0.01, respectively; Spearman rank time sincedivergence) might contributeto levels ofrepro- correlations).Consistent with thebalancing effectsof fre- ductiveisolation by testingwhether values ofan index of quency-dependentreinforcing selectionand gene flow, reproductiveisolation ( IP S I)(Rolan-Alvarez &Caballero mating discrimination is lowwhen the study population is 2000) werecorrelated with neutral mtDNA differentiation allopatric or large relative toits neighbour, it increases (range ofnucleotide divergence of 0 –5%) in the66 pairs rapidly until thesizes of the study and neighbouring popu- ofstudy populations, or with differentiation at anuclear lations are similar, andit thendecreases when the study locusin eight outof the12 populations usedin this study population is relatively rare (figure 4). (n = 28 pairs ofpopulations;data from Nosil et al. (2002)).

Proc.R. Soc.Lond. B (2003) 1916P. Nosil andothers Natural selection andspeciation

The population-ancestryhypothesis is unsupported weremet, and as predicted females from populations becauselevels ofreproductive isolation are uncorrelated wherethe two host-adapted forms interbreed exhibit with neutral differentiation (mtDNA, r = 20.11, greater mating discrimination against foreign males than p . 0.50, Manteltest; nuclear DNA, r = 0.29, p = 0.10, dofemales from geographically isolated populations. data from Nosil et al. (2002)). Moreover,levels ofgene Moreover,each of the alternative explanations for the diversity, definedas the probability that tworandomly increasedmating discrimination ofparapatric femaleswas chosenhaplotypes from thesample are different,tend to unsupported(Noor 1999). By contrast,two additional key becorrelated with population age (Nei 1987) butdo not predictionsof the reinforcement hypothesis were sup- differbetween parapatric andallopatric populations ported(Sanderson 1989; Servedio& Kirkpatrick 1997; (meanof 0.75, s.d.0.14 versusmean of 0.82, s.d.0.05, Cain et al. 1999; Servedio2000). Specifically, migration respectively; t-test: p . 0.25). betweendivergent populations actedas both ahomogeniz- Third, thebiased-extinction hypothesis predicts that ing anda diversifying force,such that reinforcementwas non-allopatric pairs ofpopulations tendto exhibit greater mostlikely whenmigration washigh enoughto facilitate mating discrimination becausenon-allopatric populations reinforcementbut low enough to prevent gene flow from withoutstrong reproductive isolation fuseupon secondary eroding adaptive divergencein mate choice.These are, to contact,or onepopulation goesextinct (seeNoor (1999) ourknowledge, the first empirical data todemonstrate the for areview).We tested the key prediction ofthis hypoth- rolesof relative population sizesand levels ofgene flow esis,that someallopatric populations will exhibit levels of in theevolution of reinforcement.We note that, although mating discrimination similar tothose observed in para- maternal effectson mating tendencieshave notbeen patric populations. explicitly ruledout, such maternal effectsare unlikely to The biased-extinctionhypothesis is unsupported producepatterns of mating discrimination that are con- becauseall fourallopatric populations usedin this study sistentwith abalance betweenselection and gene flow. exhibited lower discriminations than theaverage para- Reinforcementrequires a costto hybridization. In T. patric population (one-tailedWSR: Z = 1.83, p , 0.05). cristinae,thecosts of between-host mating are twofold. However,this hypothesiscannot be unequivocally rejected First, colourpattern in T.cristinae is genetically determ- usingthis approach becauseonly fourallopatric popu- ined(Sandoval 1993) andimmigrant males from thealter- lations weresampled (seeNoor (1999) for adiscussion). natehost plant tendto exhibit thelocally less-cryptic Wedo notethat biasedextinction doesnot apply asread- colourpattern (Sandoval 1994 a,b).Thusfemales that ily tocases with geneflow among conspecificpopulations mate with males from thealternate hostplant tendto pro- (versuscases with little or nogene flow between distinct ducea higher frequencyof less-cryptic offspring (i.e.the species),where populations are definedby thelocal locally non-matching morph oramorph that is intermedi- geography oftheir hostplant, andwhere populations are ate in colourpattern) than dofemales that mate with resi- parapatric rather than sympatric (i.e. T.cristinae ) (Noor dentmales. Second, females pairing with males from the 1995, 1999). alternative hostmight themselvessuffer increased pre- Fourth, wetested for male preferencefor allopatric dationrates, favouring increasedfemale mating discrimi- femalesbecause such male preferencescould result in nation.Under both thesescenarios, natural selection higher copulation frequencieswhen males are paired with favours mating discrimination against foreign males (see allopatric versusparapatric females.For asubsetof the Servedio(2001) for adiscussion).Although weare not mating trials ( n = 160), werecorded the position of the able todisentangle these two costs of hybrid mating com- male every 10 min,over a4hinterval. For trials where pletely, wenote that directbenefits would probably be pairing occurred,we assessed whether males paired more obtainedat thelevel ofmale –female pairing, rather than rapidly with femalesfrom threeallopatric populations willingness tocopulate in aconfinedarea. Mating trials (n = 74) than with femalesfrom aparapatric population wereconducted in small Petri dishes,allowing usto assess (n = 22). Male post-copulatory guarding behaviour whetherfemales discriminate against males oncethe towardsallopatric versusparapatric femaleswas examined opportunity for directbenefits is reduced (i.e. males can by observing single male –female pairs for twoweeks after easily pair with thefemale butcannot force copulations; thefirst copulation eventand noting whenthe male first thefrequency of pairing is65%), rather than testing stoppedguarding thefemale (pairs observedtwice a day, whetherfemales discriminate against males prior topair- n = 30, half ofthetrials conductedwith femalesfrom allo- ing (i.e.by actively fleeing). patric populations,all pairs werefrom populations using Previous workhas demonstratedthat indirect effectsof differenthosts). adaptation toalternative hostplants also increaserepro- The male-preferencehypothesis is unsupportedbecause ductiveisolation between T.cristinae populations (Nosil et males donot pair more rapidly with allopatric than with al. 2002). Our augmenteddataset affirms thesefindings, parapatric females(mean time until first pairing of55 and anddemonstrates that host-plantadaptation and 35 min respectively; Mann –Whitney U-test: Z = 0.67, reinforcementcontribute independently and additively to p . 0.50) andmales donot preferentially guard allopatric theevolution of reproductive isolation (i.e.mating dis- versusparapatric females(mean time until males crimination is highest whenboth processesoccur). The dismountof 3.38 and3.88 days,respectively; Mann – indirect effectsof suchhost-plant adaptation may provide Whitney U-test: Z = 0.35, p . 0.50). theinitial degreeof divergence in mate preferencethat has beenpredicted to make reinforcementevolve more readily 4. DISCUSSION (Liou& Price 1994; Kelly &Noor 1996). Wedetectedstrong evidencefor reinforcementin T. cri- Consistentwith theindependence of the effects of stinae:theassumptions of the reinforcement hypothesis reinforcementand host adaptation, reinforcing selection

Proc.R. Soc.Lond. B (2003) Natural selection andspeciation P.Nosil andothers 1917 exerted a ‘universal’ effecton mating preferences.Thus, Dobzhansky,T. 1951 Genetics and the originof species . New although femalesare selectedto be more discriminating York: Columbia University Press. against males from anadjacent population, this selection Doebeli, M.&Dieckmann, U.2003 Speciation along environ- has indirectly resultedin increasedmating discrimination mental gradients. Nature 451, 259–264. against foreign males from multiple other populations that Endler, J.A.1986 Natural selection in the wild .Princeton Uni- versity Press. useeither host(trials involving femalesfrom parapatric Frankham, R.1995 Effectivepopulation size/adult population populations exhibited lowercopulation frequenciesthan size ratios in wildlife: areview. Genet. Res. 66, 253–280. thoseinvolving allopatric femalesfor males from 10 out Funk, D.J.1998 Isolating arole for natural selection in speci- ofthe 12 populations,one-tailed WSR: Z = 1.83, p , ation: host adaptation and sexual isolation in Neochlamisus 0.05; significant differencesdetected only in this direction bebbianae leafbeetles. Evolution 52, 1744–1759. andin six comparisons,LR: all p , 0.05). Such ‘univer- Funk, D.J., Filchak, K.E.&Feder, J.L.2002 Herbivorous sal’ effectsof reinforcement may becaused by females insects: model systemsfor thecomparative study of speci- recognizing andpreferring males from their ownpopu- ation ecology. Genetica 116, 251–267. lation basedon a ‘population-specific ’ trait (rather than a Higgie, M., Chenoweth, S.&Blows, M.W.2000 Natural ‘host-specific ’ trait) (Kelly &Noor 1996; Higgie et al. selection and thereinforcement of materecognition. Science 290, 519–521. 2000). Similar findingshave beenreported in Drosophila Howard, D.J.1993 Reinforcement: origins, dynamics and fate ,wherefemales from mainland populations dis- mojavensis of an evolutionary hypothesis. In Hybrid zones and the evol- criminate against conspecificmales from other regions, utionary process (ed. R.G.Harrison), pp. 46 –69. NewYork: andevidence is presentedthat this discrimination is aby- Oxford University Press. productof selectionfor sexual isolation betweenmainland Jiggins, C.D., Naisbit,R. E.,Coe, R.L.&Mallet, J.M.2001 D.mojavensis andits sympatric sibling species D. ari- Reproductive isolation caused bycolour pattern mimicry. zonensis (Zouros & D’Entremont1980). If such ‘universal’ Nature 411, 302–305. effectsof reinforcement are common,then reinforcement Kelly, J.K.&Noor, M.A.F.1996 Speciation byreinforce- will contributeto speciation between both ecologically ment: amodel derived from studies of Drosophila. Genetics similar andecologically divergent pairs ofspecies and 143, 1485–1497. betweenconspecific populations. Liou, L.W.&Price, T.D.1994 Speciation byreinforcement of premating isolation. Evolution 48, 1451–1459. Our findingshave broad implications for thestudy of Mayr, E.1963 Animal species and evolution .Cambridge, MA: speciation.First, ourresults indicate that theoutcome of Harvard University Press. reinforcementdepends on the spatial distribution ofpopu- Nei, M. 1987 Molecular evolutionary genetics . New York: lations andon relative population sizes.These results con- Columbia University Press. cordwith recenttheoretical developmentsindicating that Noor, M.A.1995 Speciation driven bynatural selection in ecological interactionsbetween populations are likely to Drosophila. Nature 375, 674–675. drive speciation(Doebeli &Dieckmann 2003). Second, Noor, M.A.F.1999 Reinforcement and other consequences ourresults indicate that, evenwithin asingle species,natu- of sympatry. Heredity 83, 503–508. ral selectioncan favour theevolution of reproductive iso- Norusis, M.N.1993 SPSS for windows.Advanced statistics . lation in twodistinct ways, via indirect pleiotropic effects Chicago, IL: SPSSInc. Nosil, P.&Crespi, B.J.2003 Doesgene flow constrain trait ofecological adaptation andvia directselection for divergence or vice-versa? Atestusing ecomorphology and increasedpremating isolation. The greatest progress sexual isolation in Timema cristinae walking-sticks. Evolution towardsspeciation occurs when both processesoperate. (In thepress.) Nosil, P., Crespi, B.J.&Sandoval, C.P.2002 Host-plant We thankT. Reimchen, H.Rundle, A.Mooers, F.Breden, J. adaptation drives theparallel evolution of reproductive iso- Endler, R.Vos, C.Parent, J.Joy, S.Springer, D.Bolnick, M. lation. Nature 417, 441–443. Turelli, B.Fitzpatrick, D.Ortiz-Barrientos, J.Endler and Rolan-Alvarez, E.&Caballero, A.2000 Estimatingsexual especially D.Schluter for discussion and comments on the selection and sexual isolation effectsfrom mating fre- manuscript. D.McLaren, B.Mickelson, M.Vankoeveringe, T. quencies. Evolution 54, 30–36. Luchin and T.Leversageprovided field and technical assist- Rundle, H.D.&Schluter, D.1998 Reinforcement of stickle- ance. J.Endler provided laboratory spacefor all themating backmating preferences: sympatrybreeds contempt. Evol- experiments. Financial support wasprovided bythe Natural ution 52, 200–208. Sciences and Engineering Research Council of Canadaand the Entomological Society of Canada. Rundle, H.D., Nagel, L., Boughman, J.W.&Schluter, D. 2000 Natural selection and parallel speciation in sympatric sticklebacks. Science 287, 306–308. REFERENCES Saetre, G.P., Moum, T., Bures, S., Kral, M., Adamjan, M.& Moreno, J.1997 Asexually selected character displacement Beerli, P.&Felsenstein, J.2001 Maximum-likelihood esti- in flycatchers reinforces premating isolation. Nature 387, mation of amigration matrix and effective population sizes 589–592. in n subpopulations byusing acoalescent approach. Proc. Sanderson, N.1989 Can gene flow prevent reinforcement? Natl Acad. Sci. USA 98, 4563–4568. Evolution 43, 1223–1235. Butlin, R.K.1995 Reinforcement: an idea evolving. Trends Sandoval, C.P.1993 Geographic, ecological and behavioral Ecol. Evol. 10, 432–433. factors affecting spatialvariation in colour or morph fre- Cain, M.L., Andreasen, V.&Howard, D.J.1999 Reinforcing quency in thewalking-stick Timema cristinae .PhDthesis, selection iseffective under arelatively broad setof conditions University of California, SantaBarbara. in amosaic hybrid zone. Evolution 53, 1343–1353. Sandoval, C.P.1994 a Theeffects of relative geographic scales Crespi, B.J.&Sandoval, C.P.2000 Phylogenetic evidence for of gene flow and selection on morph frequencies in thewalk- theevolution of ecological specialization in Timema walking- ing stick Timema cristinae . Evolution 48, 1866–1879. sticks. J.Evol. Biol. 13, 249–262. Sandoval, C.P.1994 b Differential visual predation on morphs

Proc.R. Soc.Lond. B (2003) 1918P. Nosil andothers Natural selection andspeciation

of Timema cristinae (Phasmatodeae: Timemidae) and its Servedio, M.R.&Kirkpatrick, M.1997 Theeffects of gene consequences for host range. Biol. J.Linn.Soc. 52, 341–356. flow on reinforcement. Evolution 51, 1764–1772. Schluter, D.1988 Estimatingthe form of natural selection on Slatkin, M.1980 Ecological character displacement. Ecology aquantitative trait. Evolution 42, 849–861. 61, 163–177. Schluter, D.2000 The ecology ofadaptive radiation .Oxford Uni- Zouros, E.&D ’Entremont, C.J.1980 Sexual isolation among versity Press. populations of Drosophila mojavensis :response to pressure Servedio, M.R.2000 Reinforcement and thegenetics of non- from arelated species. Evolution 34, 421–430. random mating. Evolution 54, 21–29. Servedio, M.R.2001 Beyond reinforcement: theevolution of premating isolation bydirect selection on preferences and postmating, prezygotic incompatibilities. Evolution 55, As this paper exceedsthe maximum lengthnormally permitted, the 1909–1920. authors have agreedto contributeto production costs.

Proc.R. Soc.Lond. B (2003)