experimental tests of the adaptive significance of sexual recombination

EXPERIMENTAL TESTS OF THEADAPTIVE SIGNIFICANCE OFSEXUAL RECOMBINATION

William R.Rice

Numerous theories have been proposed to explain the advantages of sexual recombination — theexchange of hereditary material between different genomes or homologous chromosomes. Manyof these candidate benefits have been evaluated in controlled laboratory experiments, which,collectively, strongly indicate that sexual recombination provides important long-term advantages.

EVOLUTION OF SEXBDELLOID ROTIFERSMicroscopic organisms thatseem to have experienced aperiod ofevolution without sex,and probably without otherforms ofrecombination,formore than 80 million years.Department ofEcology,Evolution and Marine Biology,

University ofCalifornia,Santa Barbara,

California 93106,USA.e-mail: rice@lifesci.ucsb.edu

DOI:10.1038/nrg760

The wide phylogenetic and geographical patterns ofsexual recombination indicate that it has intrinsicadvantages and,indeed,identifying these benefits hasbeen the subject oflong-standing theoretical and exper-imental studies.Any satisfactory explanation for whysex evolved and is maintained,however,must accountfor the intrinsic and substantial disadvantages that areassociated with sexual recombination.These detrimentsinclude the twofold ‘cost ofproducing males’,whichrefers to the reduction in the intrinsic growth rate ofasexual population when males do not provide resourcesthat increase the fecundity oftheir mates1;the twofold‘cost ofmeiosis’,which reduces parent–offspring relat-edness from 1,in a female that reproduces parthogenet-ically,to 0.5 in a sexually reproducing female2;and thebreak-up ofco-adapted gene combinations3.Although recombination has countervailingadvantages,recombining species have not totally out-competed asexual,clonally reproducing lineages.Infact,asexual lineages are found among most ofthemain plant and animal groups4–6.Their success isshown by their persistence for thousands ofgenera-tions and geographical distributions that frequentlyfar exceed those oftheir sexual progenitors4–6.Despitethis,most asexual lineages ofplants and animals arederived only recently from sexual ancestors,and aretherefore regarded as evolutionary dead-ends that donot persist over geological time — that is,for millionsofyears1,5(FIG.1).TheBDELLOID ROTIFERS7,and possibly a few other small invertebrates8,9,seem to be rareexceptions to this pattern,having persisted as asexuallineages for millions ofyears (see the accompanyingarticle by Roger Butlin on page 311 ofthis issue).At the other extreme,some groups,such as birds andmammals,completely lack asexual lineages5.Becauserecombining lineages have adapted to aeons ofsexualreproduction,the transition to proficient,asexualreproduction might be difficult to evolve1.Nonetheless,factors such as hybridization have apparently producedinstantaneous asexual species with high competitiveability4,5.

Like recombining species,genes that are located onrecombining chromosomes persist over geological time,whereas most ofthose located on non-recombining Y chromosomes or organelle genomes do not10–12.So,just as there are rogue,ancient asexual species,there arealso some non-recombining genes that have persistedovergeological time13.

Collectively,these patterns indicate that recombi-nation is advantageous,but not universally essential.The observation that asexual species frequently out-number their sexual progenitors and persist for thou-sands ofgenerations1,4–6indicates that recombinationfrequently provides a long-term,rather than animmediate,advantage.Nonetheless,theory indicatesthat recombination can provide both short-term andlong-term advantages (see the accompanying reviewby Otto and Lenormand on page 252 ofthis issue).Below,I first identify which ofthe theoretical advan-tages to recombination are directly relevant to the

VOLUME 3 |APRIL 2002 | 241

NATURE REVIEWS |GENETICS

REVIEWSdirectly tracking the joint evolution ofspecific genes thatinfluence fitness and modifier loci that mediate recombi-nation between them.The relevant studies have beenbriefly reviewed elsewhere19.The second,‘competition-among-lineages’approach,contrasts the competitive fit-ness ofclosely related sexual and asexual lineages.Virtually all extant species recombine during at least partoftheir life cycle,and many ofthese have produced asex-ual lineages that could potentially displace their sexualprogenitors4–6.Because most ofthe experiments thathave addressed the genetic advantages ofrecombinationconcern competition between asexual lineages and theirsexual progenitors,this will be my focus here.Beforereviewing these experiments,I outline the theoreticaladvantages to recombination that they test.Mutational load.One ofthe maintheoretical advan-tages to recombination concerns its ability to reduce themutational load15–18,which is defined as the reduction inthe fitness ofa population due to the accumulation ofdeleterious mutations.Sexual lineages would beexpected to out-compete asexual lineages,all else beingequal,iftheir load ofharmful mutations was found tobe smaller.However,all else is not equal.As describedbelow,the presence ofmales in sexual lineages can leadto ANTAGONISTIC COEVOLUTIONbetween the sexes,SEXUALLYANTAGONISTIC FITNESS VARIATIONand an elevated mutationrate — all ofwhich reduce female productivity andhence the competitive ability ofsexual lineages.Sexualpopulations also accrue a higher burden oftransposableelements20–21.These factors make it difficult to translatestandard indices ofmutational load (based on themutation rate alone) into competitive exclusionbetween sexual and asexual lineages.Nonetheless,whenthe load ofmutations is too high to be sustained by anasexual lineage,whereas it can be tolerated by a sexuallineage,then this can be attributed to an unequivocaladvantage to recombination.A second problem with traditional measures ofmuta-tional load occurs because they are quantified in the currency ofmean fitness relative to a hypothetical,mutation-free genotype that is unlikely to occur in nat-ural populations — that is,relative to an undefined andunmeasurable standard.The mutational load for an asexual population can be determined,however,by mea-suring fitness relative to the most fit genotype that is actu-ally present in a finite population(BOX 1). This loadbecomes intolerable when the net reproductive rate ofthefittest class in the population cannot compensate for it,leading to a progressive loss offitness.The REQUISITE MUTA-TIONAL LOADdefines the maximum mutation rate that pre-vents the open-ended accumulation ofmutations (BOX 2).The requisite mutational load is decreased inrecombining populations when there is positive assortative mating for fitness and/or an increase in theharmful effect ofa mutation when other deleteriousmutations are simultaneously present in a genome(negative epistasis,also referred to as synergistic epistasis,BOX 2).The efficiency ofselection in eliminat-ing deleterious mutations is increased by negative epis-tasis because it increases the number ofdeleteriouswww.nature.com/reviews/geneticsSpeciesGeneraFamiliesOrdersClassFigure 1 |Typical phylogenetic distribution of asexual species.The figure represents aschematic of a typical animal phylogeny.Asexual species (green) are rare (<0.1% of all animalspecies) and their lineages are short lived on a geological timescale. With a single exception (thebdelloid rotifers; see main text), no genus of substantial size, or any higher taxonomic group, iscomposed entirely of asexual lineages5.experiments that have been carried out so far.I thendescribe how this theory has been experimentally evalu-ated with controlled laboratory experiments.Theoretical advantages of recombinationMUTATIONAL LOADThe fitness reduction ofapopulation owing toaccumulated deleteriousmutations in the gene pool.ANTAGONISTIC COEVOLUTIONA cycle ofadaptation andcounter-adaptation betweenmales and females ofthe samespecies or between a species andits enemies.SEXUALLY ANTAGONISTICFITNESS VARIATIONVariation in polymorphic genesthat increase the fitness ofonesex but decrease the fitness oftheother sex.REQUISITE MUTATIONAL LOADThe excess in the netreproductive rate ofthe fittestclass,above exact replacement,that is required to prevent open-ended mutation accumulation.Evolutionary theories for the adaptive significance ofrecombination can be classified in many ways,but here Ifocus on two main types.Ecological theories are basedon extrinsic factors that incorporate specific environ-mental or demographic contexts.For example,thepathogen ratchet theory14predicts an advantage to sexwhen recombination reduces the similarity in geneti-cally encoded resistance factors between parents andoffspring that are spatially clustered,and therebyreduces pathogen transmission between parent and off-spring.By contrast,genetic theories,such as MUTATIONALLOAD,derive from intrinsic hereditary factors,such as themutation rate15–18.Because the same phylogenetic pat-terns that pertain to recombination are seen at the levelofgenes in genomes and ofspecies in communities,asdescribed above,and because the ecological theoriesapply only to the latter,the genetic theories describe theuniversal benefits to sexual recombination and I focusexclusively on them here.There are two main experimental approaches toinvestigating the genetic advantage ofrecombination.One ofthese is to understand the causes ofvariation inrecombination rate among genomes and genomicregions.This is an important aspect ofthe adaptive sig-nificance ofsex,and involves modelling the evolution ofgenes that increase or decrease the amount ofrecombi-nation within a genome (for more on this ‘recombina-tion-modifiers’approach,see the review by Otto andLenormand on page 252 in this issue).However,the rele-vant experiments are limited to broad-scale measures ofthe rate ofrecombination observed in populations thathave been subject to intense selection,rather than242| APRIL 2002 |VOLUME 3REVIEWSmutations.Recombination re-assorts mutations thatoriginate in different lineages (by reducing linkagedisequilibrium),which allows selection to operatemore independently on individual mutations and,inturn,causes each mutation to accumulate faster.Background selection: general.A third theoreticaladvantage to recombination concerns the interactionbetween DIRECT SELECTIONon a mutation and collateralselection on its genetic background(s).Deleteriousmutations ofsmall effect occur at a high rate in meta-zoans,and these frequently persist for several genera-tions before being eliminated by natural selection28–30;recurrent mutation therefore causes gene pools to accu-mulate a burden ofmany mildly deleterious mutations.Variation in the mutational load among genomes gen-erates a diverse spectrum ofgenetic backgrounds withinwhich a mutation can arise.Variation in the number ofbeneficial mutations per genome also contributes tobackground selection,but deleterious mutations seemto be far more common and therefore are the predomi-nant factor that causes background selection.The fixa-tion or loss ofnew beneficial and deleterious mutationsis strongly influenced by background selection,asdescribed in the two following sections.Background selection and beneficial mutations.Because a population must continually adapt to achanging environment,especially to coevolving com-petitors,pathogens,parasites and predators,there isan advantage to being able to incorporate efficiently asteady stream ofnew,favourable mutations.The fateofa new beneficial mutation depends on direct selec-tion on the mutation itself,collateral selection on itsgenetic background and genetic drift (samplingerror).Temporarily ignoring the effects ofback-ground selection,the probability offixation ofa bene-ficial mutation is approximately equal to 2s,where s(the selection coefficient) is the increment by whichthe mutation increases fitness in the heterozygousstate3,31,32.This approximation assumes that the selec-tion coefficient is small and it ignores complicatingfactors such as epistatic interactions between genes,but it is a useful benchmark that is commonly used inevolutionary genetics.The probability offixation ofabeneficial mutation is less than onebecause mutationsoriginate as single copies and are therefore susceptibleto random loss by genetic drift until they accumulateto a substantial number:the larger the selection coef-ficient,the faster a mutation increases in number,andthe smaller the cumulative probability ofits loss bydrift while it is rare.The fate ofnew beneficial mutations also dependson their original genetic background and the presenceor absence ofrecombination.Mutations that originatein high/low fitness genetic backgrounds have anincreased/decreased probability ofeventually fixing,owing to collateral selection on their background.Whenrecombination is present,a mutation has only a tran-sient association with its original genetic background,and this influence rapidly diminishes as a mutationVOLUME 3 |APRIL 2002 | 243Box 1 |Calculating the mutational load for an asexual population The mutational load for an asexual population is determined by setting the relativefitness ofthe fittest extant genotype to 1.0 and solving for the equilibrium mean fitness(Wmean) that produces a stable frequency distribution offitness classes.This can be doneby focusing on the frequency ofthe fittest class (Freqbest)15.The relative fitness ofthefittest class is defined to be Wbest= 1.0 and the proportion ofoffspring that are not newlymutated is e–U,because mutations are assumed to follow a Poisson distribution and e–Uisthe proportional size ofthe zero class ofa Poisson variate with mean = Umutations pergenome per generation.The frequency ofthe best class increases each generation byreproduction and selection (that is,by a factor ofWbest/Wmean) and decreases by newmutations that occur in some ofits offspring (by a factor e–U).The frequency ofthefittest class across generations is defined byFreqbest* = Freqbest(reproduction and selection) (proportion ofunmutated offspring)= Freqbest(Wbest/Wmean) (e–U),where (*) denotes the value in the next generation.Setting Wbest= 1.0 and solving forequilibrium conditions (that is,setting Freqbest* = Freqbest),gives Wmean= e–U.When is this load intolerable? We can answer this question by reformulating themeasure ofmutational load in the currency ofabsolute fitness — that is,in terms ofthenet reproductive rates ofthe various mutational classes.In this case,Wmean= 1 because,at equilibrium,each individual leaves exactly one offspring,on average.Solving for thefitness ofthe fittest mutational class extant in a population,Wbest= 1/e–U= eU.Ifthefittest fitness class cannot achieve this net reproductive rate,then it will be lost from afinite population,as will be each successively next best class,and open-ended fitnessdecay will ensue.The maximum mutation rate that does not lead to deterministic,open-ended fitness decay is defined by the requisite mutational load18.mutations that are purged from the gene pool perselective death,and by positive assortative matingbecause it increases the variance in fitness among indi-viduals.Recent theoretical work indicates that SEXUALSELECTION,which is absent in asexual populations,might also reduce the requisite load ofa sexual popu-lation22–23.This is because sexual selection amongmales will reduce the equilibrium frequency ofharm-ful mutations in both sexes,but the cost ofsexualselection is only experienced by males.Because malesrarely contribute to the productivity ofa population,and because sexual selection can be strong,the mutational load on sexual females could be substan-tially reduced.SEXUAL SELECTIONCompetition among membersofone sex (generally males) forfertilization opportunities withthe other sex.FIXATIONThe accumulation ofa mutationto a frequency of100% in a genepool.CLONAL INTERFERENCEThe reduced competitiveadvantage ofa clone that carriesa beneficial mutation owing tothe simultaneous presence ofone or more other clones thatcarry different beneficialmutations.DIRECT SELECTIONDarwinian selection on a specificmutation.Combining beneficial mutations.The second theoreti-cal advantage to recombination is that it allowsfavourable mutations that arise in different lineages tobe united in the same genome3,24,25.By contrast,inclonally reproducing species,different beneficialmutations must occur tandemly within the same lin-eage to come together in the same genome,and thisslows the rate ofaccumulation ofbeneficial mutations(progressiveevolution).A related theoretical advan-tage to sex also occurs when different beneficial muta-tions are present simultaneously in a population26.The rate ofFIXATIONofbeneficial mutations that occurin the same asexual population is reduced throughCLONAL INTERFERENCE,which puts a ‘speed limit’on therate ofprogressive evolution27.Clonal interferenceoccurs because different beneficial mutations competeagainst each other,thereby diluting their advantagerelative to the genomes that carry no beneficial NATURE REVIEWS |GENETICSREVIEWSrecombines into new genetic backgrounds.Theoreticalanalysis33–38shows that,when recombination is present,the decline in frequency that a mutation would receivewhen it originates in an inferior genetic background ismostly compensated,on average,by the boost that itwould receive when it originates in a superior geneticbackground (BOX 3,panel a).Therefore,for a beneficialmutation,recombination causes the average effect ofbackground selection to be negative,but relatively small.When recombination is absent,background selectioncan strongly decrease the average probability offixationofa beneficial mutation33–38(BOX 3,panel b).In a non-recombining population,high-fitness genotypes gradu-ally displace low-fitness genotypes,leading to recurrentselective sweeps bylineages from the highest end ofthedistribution ofgenetic backgrounds (the progenitortail38,BOX 3,panel b).As a selective sweep proceeds,dele-terious mutations accumulate in the sweeping lineage(s)so that,at equilibrium,the distribution offitness valuesdoes not change over time.These recurrent selectivesweeps lead to the gradual extinction ofall lineages thatdo not reside in the progenitor tail;hence,lineages out-side the progenitor tail are collectively called the ‘livingdead’38(BOX 3,panel b).When recombination is absent,mutations that origi-nate in the living dead are trapped in their originalgenetic background and are doomed to eventual extinc-tion unless their selection coefficient elevates the recipi-ent genome into the progenitor tail.Beneficial muta-tions that reside in the progenitor tail,ifnot lost early onby sampling error or by competition among the fittestgenotypes,will eventually fix in the population (BOX 3,panel b).Most new beneficial mutations,however,willbe trapped in the living dead,and their loss due to back-ground trapping causes them to accumulate far moreslowly in a non-recombining population (FIG.2).Background trapping will be an important cost when-ever the heritable variance in fitness among geneticbackgrounds is substantial relative to the selection coefficient (s) ofa mutation (BOX 3,panels a,b).So,in a non-recombining population ofgenomes or chromosomes,beneficial mutations are commonlytrapped in the living dead (background trapping) and cannot become established in a population.Recombination frees beneficial mutations from theiroriginal genetic background,and thereby increases their probability offixation.Background selection and harmful mutations.Mostmutations are harmful and one ofthe main functionsofnatural selection is to continuously purge thesemutations from the gene pool.Deleterious mutationscan accumulate (retrogressive evolution) by geneticdrift when the strength ofselection is small relative torandom fluctuations in gene frequency due to sam-pling error.In the absence ofcomplicating factors,such as tight linkage to other selected genes,geneticdrift will overpower selection whenever |s| < 1/N,where Nis the census size and sis the selection coeffi-cient32,39.The probability offixation ofa neutralmutation due to genetic drift is 1/(2N),and this valuewww.nature.com/reviews/geneticsBox 2 |Requisite mutational loadConsider a finite population that is large enough to ignore sampling error.At equilibrium,the average fitness (as measured by the net reproductive rate,R0= average lifetime per-capita number ofoffspring) is equal to one.To prevent the open-ended erosion in meanfitness (defined here to be an intolerable mutational load),the distribution offitness classesmust be anchored such that,at equilibrium,the fittest class does not deterministicallydecline in frequency each generation.To be stable,the net reproductive rate ofthe fittestclass must be one after discounting for offspring that are newly mutated and therefore lostfrom the fittest class (panel a).Ifthe productivity ofthe fittest class is insufficient to offsetrecurrent mutation,it will be lost recurrently,causing an intolerable mutational load toaccrue.Unlike MULLER’S RATCHET— in which mutations accumulate due to the stochastic lossofthe fittest class — mutations,in this case,accumulate deterministically due to insufficientproductivity ofthe fittest class relative to the loss by mutation.To determine the productivity ofthe fittest class that is required to prevent an intolerablemutational load (R0(req)),it is assumed that harmful mutations occur at rate Uper genomeper generation.Because there are a large number ofmutable loci and a small independentprobability ofmutation at each locus,Uis expected to follow an approximate Poissondistribution,and a fractione–Uofoffspring will be unmutated (e–Uis the fractional size ofthe zero class ofa Poisson distribution).To compensate for the fraction ofoffspring that arenewly mutated,R0(req)must be 1/e–U= eU,so the requisite mutational load is eU – 1 (panel a).R0(req) is not the maximum reproductive rate ofthe fittest genotype;it is the average per-capita number ofoffspring produced by the fittest type under competitive equilibriumconditions when an average individual produces only a single surviving offspring.R0(req) leads to extinction in an asexual population when Uis sufficiently large relative to therealized growth rate ofthe fittest class (panel b).Panel bshows that the requisite load (R0(req) – 1) increases exponentially with the genome-wide mutation rate.With recombination,the fittest class is not produced by its own clonal reproduction,butby recombination among the population at large (panel c).When there is negative epistasis(that is,a mutation is more harmful when other harmful mutations are also present14) orpositive assortative mating for fitness (that is,when genotypes ofsimilar fitness matepredominantly among themselves18),then recombination reproduces the fittest class fasterthan it would have done by means ofits own clonal reproduction.In this case,arecombining population can persist in which an asexual population would be destroyed bythe open-ended accumulation ofdeleterious mutations — that is,by an intolerablemutational load.aNumber of individualsRequisite mutational load = eU – 1> 1 new proportion = 1 – e–UmutationR0(req) = 1R0(req) = eUNo new –Umutations proportion = eWith no mutationFittest class (initial cohort size)OffspringcWith mutationbRecombining40Requisite R0 of fittest genotypes20Probability densityNon- recombining0012Mutation rate (U)30Relative fitness1Fittest genotypes244| APRIL 2002 |VOLUME 3REVIEWSMULLER’S RATCHETRecurrent stochastic loss ofthefittest genomes in an asexualpopulation.CENSUS SIZEBox 3 |Genetic backgrounds and fixation of mutationsThe genetic background ofa mutation (shaded fitness distribution) influences its probability offixation (Prob(Fix);logscale) in a manner that depends on the presence or absence ofrecombination.As a benchmark for comparison,theprobability offixation in the absence ofbackground selection is 2sfor beneficial mutations (panels a,b) and maximally1/(2N) for harmful mutations (panels c,d),where sis the selection coefficient and Nis the CENSUS SIZE.Arrows depict theinfluence ofthe initial genetic background ofa mutation on its probability offixation:blue arrows reinforce natural selectionand red arrows oppose it.A beneficial mutation has a higher chance ofbeing fixed in a recombining population (panel a) than in a non-recombiningone (panel b).Recombination causes the original genetic background ofa mutation to be transient and,because the influ-ence ofdifferent starting backgrounds nearly cancel (∑arrows ≈0),there is only a small influence ofbackground selection,on average.By contrast,in a non-recombining population,beneficial mutations are trapped in their genetic background oforigin.A non-recombining population is functionally divided into a small ‘progenitor tail’(Prog.tail),which is composed ofgenotypes (genetic backgrounds) that have the highest or nearly highest Darwinian fitness,and the ‘living dead’,which iscomposed ofless fit genotypes — these are called the living dead because lineages ofthese genomes are destined to eventualextinction owing to selective sweeps ofthe progenitor tail.The probability offixation ofa beneficial mutation is lower in anon-recombining population because only a minority ofthese mutations that originate by chance in the progenitor tail canpotentially accumulate to fixation;all others are trapped in the living dead and are destined to eventual loss.Natural selection is sufficiently strong to prevent large-effect deleterious mutations (|s| >> 1/N) from accumulating tofixation in both recombining and non-recombining populations,but too weak to prevent the accumulation ofsmall-effectdeleterious mutations (|s| < 1/N).Deleterious mutations with intermediate effects (1/Np> |s|> 1/N,where Npis the size ofthe progenitor tail),however,accumulate only in non-recombining populations (shown here in panels cand d).Theirselection coefficient is too large to allow their accumulation in a recombining population because selection is too strongrelative to drift in the population at large (that is,|s|> 1/N).They can,however,accumulate to fixation by drift in theprogenitor tail ofa non-recombining population because drift is stronger in this smaller subpopulation (that is,|s|< 1/Np) —in which case,they will eventually spread to the entire population through a SELECTIVE SWEEP.The patterns shown are general,but the calculations to produce the figures assume a coefficient ofvariation of17% background fitness,and that the size ofthe progenitor class is 4% ofthe census size.Adapted from REFS 31–38.aBeneficial mutation (s = 0.01; N = 500)Prob(Fix) = 0.0140The total number ofindividualsin a population.SELECTIVE SWEEPThe gradual accumulation tofixation ofa genome orchromosomal region that has anet selective advantage.bBeneficial mutation (s = 0.01; N = 500)Prob(Fix) = 0.00162sRecombining Prob(Fix)2sNon-recombining Prob(Fix)Living deadProg. tail≈000.20.40.60.81≈000.20.40.60.8Genome-wide background fitness1Genome-wide background fitnesscHarmful mutation (s = 0.005; N = 500)Prob(Fix) = 0.0000dHarmful mutation (s = 0.005; N = 500)Prob(Fix) = 0.0008Non-recombining Prob(Fix)Recombining Prob(Fix)Living dead12N≈012N≈0Prog. tail00.20.40.60.8Genome-wide background fitness100.20.40.60.8Genome-wide background fitness1NATURE REVIEWS |GENETICSVOLUME 3 |APRIL 2002 | 245REVIEWSProb(Fix)0.0200.0150.0100.005–0.025–0.02–0.015–0.01–0.0050.0050.010.015recombination to prevent fixation ofsmall-effect harm-ful mutations (1/N< |s| < 1/Np) is substantial.A final point concerning harmful mutations is thecommon misconception relating to retrogressive evolu-tion in asexual populations,known as Muller’s ratchet.Itis frequently stated that Muller’s ratchet occurs only insmall non-recombining populations.This spurious con-clusion is an artefact ofthe simplifying assumption thatall harmful mutations have the same effects on fitness,set equal to the average effect ofa deleterious mutation.When the effects ofmutations are more appropriatelyassumed to be variable,with a large class ofmutationsthat have a very small effect on fitness29,30(|s|<< 0.01),then deleterious mutations will accumulate in popula-tions ofany finite size and the domain ofMuller’s ratchetis not restricted to small populations40–43.Experimental tests: mutational loadSelection coefficient (s)Infinite population/no background selectionFinite population/no background selectionFinite population/background selection and recombinationFinite population/background selection but no recombinationAdvantage of recombinationFigure 2 |Fate of a mutation depends on direct selection (s), background selection andrecombination. The probability of fixation (Prob(Fix)) of a beneficial mutation is reduced bybackground selection, but more so when recombination is absent. The probability of fixation of aharmful mutation is increased by background selection, but to a greater degree whenrecombination is absent. The patterns shown are general, and the specific values shown on thegraph are calculated as in BOX 3. Adapted from REFS 31–36. EFFECTIVE POPULATION SIZEThe equivalent number ofbreeding adults in a populationafter adjusting for complicatingfactors such as nonrandomvariation in family size orstochastic fluctuation inpopulation size.is the upper bound for the fixation probability ofdeleterious mutations31,32.Factors such as fluctuationsin population size and an unbalanced sex ratio gener-ally make the EFFECTIVE POPULATION SIZEsmaller than thecensus size.To avoid unnecessary detail here,I set thecensus size equal to the effective size and then evaluatethe influence ofrecombination on the efficacy ofnatural selection.In a recombining population,mutations freely movebetween genetic backgrounds,and deleterious mutationscan accumulate only when |s| < 1/N(BOX 3,panel c;FIG.2).In a non-recombining population,most deleteri-ous mutations originate in the living dead and they willbe deterministically eliminated owing to their inferiorgenetic background (BOX 3,panel d).In this case,back-ground selection reinforces direct selection on the muta-tion.Harmful mutations,however,also originate in theprogenitor tail and these can accumulate by drift when-ever |s| < 1/Np,where Npis the size ofthe progenitor tail(which is << N,BOX 3,panel d;FIG.2).So,harmful muta-tions with a very small effect (|s| < 1/N) accumulate inboth recombining and non-recombining populations,those with small but intermediate effects can accumulateonly in non-recombining populations (1/N < |s| < 1/Np),and mutations with large effects (|s| > 1/Np) will notaccumulate irrespective ofthe presence or absence ofrecombination.FIGURE 2summarizes the influence ofbackground selection on beneficial and deleteriousmutations.Although the advantage to recombination islarger for beneficial compared with harmful mutations(FIG.2),most mutations are deleterious with smalleffects28–30.Accordingly,the potential advantage toTheory predicts that asexual reproduction can persistonly when the mutational load is tolerable.This will betrue when the net reproductive rate ofthe fittest geno-types (R0(best)) equals or exceeds the requisite net repro-ductive rate (R0(req)) that is needed to offset recurrentdeleterious mutations — that is,when R0(best) ≥R0(req) =eU(BOX 2).Experiments to test this prediction mustevaluate both the genome-wide deleterious mutationrate (U) and the net reproductive rate ofthe fittestgenotypes (R0(best)).Ifrecombination is to rescue aspecies from an intolerable mutational load,then theremust be evidence for negative epistasis,positive assorta-tive mating for fitness and/or compensating sexualselection,which give recombining species an advantageby increasing the power ofnatural selection to removeharmful mutations.Requisite mutational load.Direct empirical estimates forR0(best)from natural populations are difficult to obtainbecause they require reliable measurements ofthe heri-table lifetime fitness ofthe fittest genotypes under equi-librium conditions.This measure,however,was recentlyestimated from a high fecundity laboratory populationofDrosophila melanogaster44.The population hadadapted to a competitive laboratory environment formore than 200 generations.Despite the fact that femalescan lay more than100 eggs per day,the empirical esti-mate ofR0(best)was only 1.82 — that is,the fittest femalegenotypes (n = 40) had a heritable net fitness that wasonly about twice as large as an average female.Althougha survey ofa larger number ofgenomes would beexpected to find a larger estimate ofR0(best),this value isstill a useful first approximation.In these experiments,the fitness ofgenomic haplotypes was measured ratherthan the fitness ofcomplete diplotypes.Extrapolating todiploid fitness,R0(best)is estimated to be ~3.3.Therefore,the maximum deleterious mutation rate that could betolerated by this population,ifasexual,would be Umax= ln(R0(best) = 3.3) = 1.2 mutations per genome per gener-ation.These data indicate that the maximum tolerablemutation rate for a high fecundity species,such asDrosophila,would be Umax≈1.Because many specieshave mutation rates far in excess ofthis value,asexualitywww.nature.com/reviews/genetics246| APRIL 2002 |VOLUME 3REVIEWSwould be predicted to lead to eventual extinction.However,comparable measures from different species innatural environments are needed to estimate,more gen-erally,the maximal tolerable mutation rate.Deleterious mutation rate.Anevaluation ofthe requi-site mutational load requires that we estimate thegenome-wide deleterious mutation rate.This rate hasbeen the subject ofmany recent reviews45–49,so here Ihighlight only a few key studies.Most data to estimatethe deleterious mutation rate come from mutationaccumulation experiments.In these experiments,a pop-ulation is recurrently bottlenecked to one or few indi-viduals so that mutations ofminor effect can freelyaccumulate by drift.The mathematical technique that isused to calculate the deleterious mutation rate fromthese experiments produces an estimate that is biaseddownwards in proportion to the coefficient ofvariationin fitness ofmutations,which will be substantial when mutations vary in their impact on fitness andwhen there are many mutations ofsmall effect.Recently,an ingenious experiment was used to esti-mate the degree ofthis bias23.The mutation rate ofCaenorhabditis eleganswas manipulated by using amutagen (ethyl methane sulphate,EMS) to produce acontrolled genome-wide mutation rate ofknown mini-mal value.Populations ofEMS-treated worms weresubject to a mutation accumulation protocol and themutation rate was estimated.Remarkably,almost allmutations (96%) were undetected.Statistical analysisindicated that most ofthe new deleterious mutationshad a very small selection coefficient (s << 1%).Ifmostspecies have a large class ofvery small effect mutations,which seems likely23,24,then the mutation accumulationprocedure will grossly underestimate the genome-widedeleterious mutation rate.An alternative to the mutation accumulation proto-col is to estimate the deleterious mutation rate inspecies for which divergence time can be determinedfrom the fossil record50.The sequence divergence ofexons,after discounting by the projected number ofneutral substitutions,is used to estimate the deleteriousmutation rate.An extension ofthis technique wasrecently applied to a wide spectrum ofspecies22.Deleterious mutation rates (adjusted for mutations innon-coding regions,transposon transpositions,andsmall insertions and deletions) increase linearly withgeneration time (0.34 for Drosophila,1.1 for the labora-tory mice and rats,3.2 for the domestic dog and cat,and 6.6 for humans and chimps).At the lower end,these estimates would produce requisite mutationalloads that — making feasible extrapolations from theavailable data — seem compatible with asexual repro-duction,but not at the higher end.Kondrashov48hascriticized these estimates (but see the rebuttle outlinedin REF.46),arguing that they are biased downwards but,nonetheless,they represent our best estimate ofthiscontentious parameter.Negative epistasis.As described above,for recomb-ination to rescue a population from an intolerable mutational load there must be negative epistasis,posi-tive assortative mating for fitness or compensating sex-ual selection.All ofthese processes increase the capacityofnatural selection to purge harmful mutations fromthe genome,and thereby increase the productivity ofarecombining species.This and the following two sec-tions discuss the tests that have been done to detect theoccurrence ofany ofthem in experimental populations.Many experiments have tested for negative epistasis(BOX 4).On balance,experimental support for the wide-spread occurrence ofepistasis is inconsistent at best,andwhen detected,it was equally likely to promote asdetract from an advantage to recombination.One criti-cism ofpast tests for negative epistasis is that they evalu-ate arbitrary spontaneous mutations41,51.Intrinsicallyinteracting mutations,such as those that code for com-ponents ofthe same enzymatic or developmental path-ways,will more plausibly produce negative epistasis52.Epistasis that is restricted to closely interacting genes isnearly as powerful in reducing mutational load as thatbetween arbitrary mutations,but experimental tests forthis kind ofepistasis are lacking18.Positive assortative mating.Positive assortative matingfor fitness is another powerful way to reduce the muta-tional load ofa recombining population18.This matingsystem increases the variance in fitness among geno-types and thereby increases the HERITABILITYoffitness andthe efficiency ofselection.There are many natural his-tory contexts that lead to positive assortative mating forfitness53,54.For example,in many species there is positiveassortative mating for body size.To the extent that largerindividuals achieved higher mass because their geno-type made them better adapted to their ecological niche,then this mating pattern will produce positive assorta-tive mating for fitness.Unfortunately,I found no experi-mental tests for the prevalence ofthis process in naturalor laboratory populations.Sexual selection.Lastly,sexual selection among malesmight reduce the mutational load ofrecombiningspecies42,43.The critical assumption that remains to betested experimentally is that sexual selection amongmales reduces the mutational load in females.The onlydirect experimental study that compared the load (mea-sured by female productivity) ofpopulations with andwithout the operation ofsexual selection is an experi-ment with D.melanogaster55.This study found that theremoval ofsexual selection increased,rather thandecreased,the productivity offemales.The reduced loadwas due to reduced male-induced harm to their mates,which is expected to occur when males and females coevolve antagonistically56–60.Another potential countervailing cost ofsexualselection is increased mutation rate.In species inwhich females have fewer mitoses per generation thanmales in their germ line (as in humans),empiricaldata indicates that males have a substantially elevatedmutation rate (for example,fourfold higher inhumans61),and this can substantially increase themutational load ofa sexual species62.VOLUME 3 |APRIL 2002 | 247HERITABILITYThe fraction ofthe phenotypicvariance that is attributable toadditive genetic variance NATURE REVIEWS |GENETICSREVIEWSBox 4 |Experimental tests of negative epistasisIn one ofthe most elegant experiments so far,Elena andLenski80inserted variable numbers oftransposableelements into the genome ofEscherichia coliand then10.0measured net fitness.They found that fitness effects ofthe inserts typically combined independently (indicatinglittle or no epistasis) and in the rarer cases when they didnot,mutations were just as likely to combine in a way5.0that favoured recombination (negative epistasis) as didnot (positive epistasis) (see figure).The figure shows thedistribution ofdeviations (observed–expected) in thefitness ofdoubly mutated E.coli .Expected values werecalculated assuming no epistasis among mutations.A0.0value ofzero indicates no epistasis,positive values indicate–0.45–0.3–0.1500.150.30.45positive epistasis and negative values indicate negativeRecombination Recombination favoureddisfavouredepistasis.The symmetry ofthe graph about zero indicatesthat there is no net advantage to recombination due toDeviation from independent fitness effectsnegative epistasis.(Data taken from REF.80.)A second experimental approach was to plot fitness against time for data collected from mutation accumulation lines.Although early studies with Drosophilathat measured a fitness component (viability) found evidence for negativeepistasis81,82,this result was not confirmed by more recent experiments,with viruses,that measured total fitness83,84.A third approach,which had particularly high experimental power,measured fitness ofnew deleterious mutations ingenetic backgrounds that had normal versus elevated numbers ofdeleterious mutations85.Data from this experiment donot support the common occurrence ofnegative epistasis.A fourth approach was to cross lines with different numbers ofmutations and compare the fitness ofparents and offspring.Data from this procedure seemed to support negativeepistasis in Chlamydomonas86,but this interpretation has been criticized87.In addition,a negative result was reported in asimilar study with yeast88.Last,a study using Drosophilatested for negative epistasis among combinations ofchromosomal regions that were markedwith homozygous recessive,visible markers51.After adjusting for the fact that one ofthe marked regions increased fitness(rather than decreasing it as was originally expected because ofthe expression ofthe visible marker),this study found thatmost (36 out of52) interactions between chromosomal regions were consistent with non-epistatic fitness interactions,and ofthose that were not,eight supported negative epistasis and eight supported positive epistasis.Number of double mutantsFor sexual selection to reduce the mutational load,there must be a positive genetic correlation for fitnessbetween the sexes:genomes that produce high/low fit-ness males must also produce high/low fitnessfemales.Some studies ofsexual selection have found acorrelation between the mating success ofa male andthe viability ofhis offspring63,.However,the crucialparameter is the genetic correlation between sexualselection in males and productivity in females.Recently,the fitness ofthe same 40 cloned genomeswas measured in both male and female Drosophila44.The intersexual genetic correlation for juvenile fitness(egg-to-adult viability) was positive.But in adults,inwhich gender roles diverge,the genetic correlationbetween male mating success and female fecunditywas negative.These data indicated that adult malesand females are selected towards different phenotypicoptima and,because most genes are expressed in bothsexes65,selection in males leads to reduced female pro-ductivity (a phenomenon known as intersexual onto-genetic conflict)44,66.Additional data are needed to resolve the degree towhich sexual selection influences the burden ofmutations in females.The available experimental dataindicates that elevated mutation rates in males61,62,antagonistic co-evolution between the sexes56–60and sex-ually antagonistic fitness variation44,66will cause the| APRIL 2002 |VOLUME 3operation ofsexual selection to increase rather thandecrease the mutational load ofa population.Experiments: combining beneficial mutations There has been limited experimental evaluation ofthehypothesis that recombination is favoured because itreduces interference between beneficial mutations thatare simultaneously accumulating in a population.Ifound no experiments that introduced two or morebeneficial mutations into populations with and withoutrecombination,and then tested for faster production ofgenomes that carry compound mutations when recom-bination was present.However,the general concept thatrecombination speeds the rate ofprogressive evolutionhas been tested in other contexts.Experiments withEscherichia coli67and RNA viruses68,69have measuredthe rate ofadaptation when a population is exposed to anew environment and compared this with a theoreticalbenchmark that assumes no clonal interference.Therewere two basic experimental designs.In the first,popu-lations ofdifferent size were exposed to the same newenvironment and the rate offitness increase was trackedover time.In the absence ofclonal interference,the rateofadaptation should increase linearly with populationsize because the rate ofproduction ofnew beneficialmutations is proportional to population size.In bothexperiments67,68,the rate ofadaptation levelled offaswww.nature.com/reviews/genetics248REVIEWSpopulation size increased,supporting the operation ofclonal interference.In the second experiment,the mag-nitude ofthe selection coefficient ofthe first mutationthat was fixed in populations ofvarying size was com-pared.Clonal interference predicts that,on average,theselection coefficient offixed mutations should increasewith increasing population size,and this prediction wasconfirmed69.The rationale for the prediction is that,aspopulation size increases,there will be more simultane-ously competing beneficial mutations,so those withsmaller selection coefficients will be competitively dis-placed.Although these experiments provide support forthe operation ofclonal interference,a more convincingcase will be made when parallel experiments show thatthe harmful effects attributed to clonal interference areameliorated or retained when recombination is presentor absent.Experimental tests: background selectionofrecombination.As predicted,the non-recombiningneo-Y chromosomes degenerated rapidly,but recombi-nation rescued the neo-X chromosomes from most ofthis fitness decay.Beneficial mutations.A large number ofexperimentshave tested the hypothesis that recombination speedsthe rate ofaccumulation ofbeneficial mutations.Mostofthese experiments do not trace the fate ofindividualbeneficial mutations,but instead measure the rate ofprogressive evolution with and without recombina-tion.In one ofthe first,Carson selected motilitybehaviour in Drosophila robustapopulations thateither had no chromosomal inversions (which sup-press recombination in heterozygotes) or had manyinversions74.The populations with lower recombina-tion had a trend towards a slower response to selec-tion,but this difference was confounded by differencesin the starting genetic variation among experimentaltreatments.McPhee and Robertson75extended thisline ofresearch by selecting for bristle number in pop-ulations ofD.melanogasterwith and without crossoveramong the autosomes (which constitute 80% ofthegenome).When crossover was present,the response toselection was 22–28% faster (18 out of20 linesresponded faster than the mean response whencrossover was absent;p< 0.01,binomial test),support-ing the hypothesis that recombination speeded the rateofaccumulation ofbeneficial alleles that influencebristle number in the selected direction.Markow76expanded this design by controllingcrossover on both the X and the autosomes,so that 20,40,60,80 or 100% ofthe genome was able to recombine.When she applied selection to phototaxis instead ofbris-tle number,she reported that recombination signifi-cantly speeded the response to selection only in certaincases.To pool all her data,I regressed the response toselection on the percentage ofthe genome that recom-bined and found a highly significant positive correlation(p = 0.0017,R2= 0.63),indicating that as more ofthegenome was allowed to recombine,the response to selec-tion was faster (W.R.R.,unpublished observations).Bycontrast,Thompson77carried out a similar experiment(but controlled crossover only on the two main auto-somes) and concluded that there was no significant effectofthe presence or absence ofrecombination.My analysisofThompson’s data indicated a significant increase inthe response to selection when recombination was pre-sent (Student’s t-test,p= 0.0163;W.R.R.unpublishedobservations),but Thompson concluded that this difference was an artefact ofthe influence ofthe geneticconstructs (balancer chromosomes) that were used tosuppress recombination.The next generation ofexperiments on the adaptivesignificance ofrecombination used bacteria,yeast,bacte-riophage and viruses as model systems.These systemshave the advantage offast generation time,but the disad-vantage ofsmall genome size (that is,low backgroundselection) and,hence,the advantage ofrecombination isexpected to be smaller than it would be in metazoanswith larger genomes (BOX 5).VOLUME 3 |APRIL 2002 | 249Recombination is predicted by theory to both slow the rate ofaccumulation ofsmall-effect deleteriousmutations and speed the accumulation ofbeneficialmutations.Both ofthese hypotheses have been experi-mentally evaluated,but most work has focused on bene-ficial mutations.Harmful mutations.Few experiments have testedwhether recombination slows the accumulation ofminor-effect harmful mutations.Numerous experi-ments with bacteria and viruses showed that recurrentlybottlenecking a population to a size ofone haploid individual leads to fitness decline (many experimentsare reviewed in REF.70).These experiments,however,aremore akin to traditional mutation accumulation studies(for example,REF.81) rather than tests ofthe Muller’sratchet process.Bottlenecks to N = 1 haploid individualscause mutations to be fixed and,therefore,mutationswill accumulate irrespective ofthe presence or absenceofrecombination.One line ofexperiments with an RNA bacterio-phage 71,72has taken the mutation accumulation proto-col one step further by first bottlenecking populations40 times and then measuring their fitness recovery withand without recombination.All populations recoveredfitness rapidly (32% offitness was recovered,on aver-age,after eight growth cycles without bottlenecks),showing the substantial potential for beneficial and/orcompensatory mutations to mitigate the accumulationofharmful mutations in a non-recombining lineage.Inaddition,when lines that had accumulated independentmutations were recombined,they recovered morerapidly (an additional 15% recovery offitness).A different approach was taken in experiments usinga D.melanogastermodel system73.Here,80% ofthegenome was made to co-segregate like a giant non-recombining neo-Y sex chromosome or a recombiningneo-X chromosome.The populations were maintainedat an effective population size of48 chromosomes eachfor 35 generations.At this small size,mildly deleteriousmutations were expected to accumulate on both chro-mosome types,but at an accelerated rate in the absenceNATURE REVIEWS |GENETICSREVIEWSSynthesisBox 5 |Recombination and adaptive evolution in microorganisms? In 1977,Malmbergadapted the bacteriophage T4 to a novel environment (proflavinein the medium) and experimentally controlled the level ofrecombination (low versusmoderate).His data indicated that the phage adapted more rapidly when recombinationwas higher.The only complication with these paradigm-setting experiments was thatthe results are expressed as deviations from control populations,and in some cases thebenefit attributed to recombination was associated with changes in the controls.The next approach with microorganisms used Escherichia colias a model system.Souza et al.90compared the rate ofadaptive evolution (to glucose-limited media) inpopulations that were,or were not,recurrently recombined with migrant,novelgenomes from an unrelated population that was not subject to the same selectionregime.The experiment tested for an advantage ofreceiving (by migration andrecombination) novel,unselected variation that might fortuitously be favoured in thenew environment.No net improvement in adaptation to limiting glucose was detectedin the recombining populations.However,there was evidence that migrant genesaccumulated in the gene pool ofthe recombining lines due to selection,but not thosethat contributed to adaptation to the low glucose environment — that is,these genesinfluenced other fitness components.Yeast have been used frequently to test the advantage ofrecombination inmicroorganisms.Birdsell and Wills91and Greig et al.92 showed that when pairs ofSaccharomyces cerevisiaestrains were recombined,at least one ofthe many recombinantprogeny lineages had a competitive advantage over the parental clones that producedthem.These experiments show that recombination can produce a competitivelysuperior genotype by recombining the parental genomes,but not that recombination,on balance,has a net advantage.Zeyl and Bell93carried out a more crucial experiment by testing whether S.cerevisiaepopulations that experience recurrent recombination (sporulation) had an adaptiveadvantage compared with strictly clonal populations.Replicate populations with andwithout periodic sporulation were exposed to a new environment (galactose as a carbonsource instead ofglucose) or allowed to evolve for the same period oftime in theancestral environment (glucose carbon source).They found that recombination had noeffect on the rate ofadaptation to the new environment,but that populations withrecombination evolved higher fitness than clonal populations when kept on theancestral environment.The authors concluded that recombination speeded theelimination ofharmful mutations but not the accumulation ofbeneficial mutations.However,ifrecombination speeded the elimination ofharmful mutations,thenrecombining populations should have had elevated fitness in both the novel and originalenvironments,and this was not observed.This inconsistency might be an artefact due tothe extra selection on the recombining populations that were recurrently selected forgrowth on a novel (pre-sporulation) medium.GENE CONVERSIONThe non-reciprocal transfer ofgenetic information betweenhomologous genes (as aconsequence ofmismatch repairafter heteroduplex formation).The most recent experiments that tested for acceler-ated progressive evolution with recombination used aD.melanogastermodel system78.In these experiments,genome-wide synthetic chromosomes were con-structed that were either recombining (neo-X) or non-recombining (neo-Y).New beneficial mutations wereintroduced into each of34 replicated experiments andthe fate ofthe beneficial mutations was traced withrecombination present (17 neo-X treatments) andabsent (17 neo-Y treatments).As predicted by theory,recombination sometimes helped and sometimes hurtthe accumulation ofthe favoured mutation in individ-ual experiments,but on average a strong advantage ofrecombination was observed.These experiments alsoshowed that the variation in fitness among geneticbackgrounds was substantial.This high-standinggenetic variance in fitness would be expected to causeboth faster progressive and slower retrogressive evolu-tion in recombining populations (BOX 3).We have made considerable experimental progress inshowing the adaptive advantages ofrecombination.Studies ofmutation rate show that it is large enough,at least among metazoans with long generation times,to create a debilitating mutational load in the absenceofrecombination.Experimental elucidation ofthespecific mechanisms that produce a load-reducingadvantage ofrecombination (synergism among muta-tions,positive assortative mating for fitness or sexualselection) is still incomplete.Nonetheless,thehypotheses based on negative epistasis among func-tionally unrelated mutations and sexual selection arenot supported by the available data.A significant challenge for future experiments will be to determinethe mechanism(s) that reduce the mutational load ofrecombining species with high genome-wide mutationrates — for example,humans.Experiments that measure the recovery offitness(in asexual populations that had been repeatedly bot-tlenecked) show a large capacity for new beneficialand/or compensatory mutations to ameliorate theharm that is produced by the accumulation ofharmfulmutations.This finding indicates that beneficial,compensatory and/or reverse mutations might substantially reduce the mutational load,due to fixeddeleterious mutations,ofasexual species.This highpotential for compensatory adaptation also might slowthe rate at which the stochastic accumulation ofdele-terious mutations (retrogressive evolution) erodes thefitness ofasexual lineages,especially when one large-effect advantageous mutation can mitigate the effectsofmultiple small-effect mutations.Experiments with model systems that range fromviruses to flies,on balance,confirm the theoretical prediction that recombination reduces backgroundtrapping and,thereby,both decreases the rate ofaccu-mulation ofharmful mutations and increases the rate ofaccumulation ofbeneficial mutations.The magnitudeofthese advantages continually accrues over time.Thereis also limited support for the hypothesis that recombi-nation reduces interference between beneficial mutations that are simultaneously segregating in a population.Future experiments need to address the rel-ative importance ofrecombination in speeding progres-sive evolution versus retarding retrogressive evolution.Last,recent experiments concerning directed evolu-tion through exon shuffling indicate a potential advan-tage to recombination at the level ofindividual genes79.Most eukaryotic genes are segmented into coding exonsthat are separated by non-coding introns.Becauseintrons tend to be much larger than exons,most intra-genic crossovers occur within introns.This pattern ofintragenic recombination shuffles intact exons fromdifferent homologous genes,or among members ofthesame gene family through GENE CONVERSION,creating newexon combinations that would require many muta-tional steps in a non-recombining gene.The relativeimportance ofthis process remains to be shown experi-mentally,but it might substantially extend the adaptivepotential ofrecombining genes.www.nature.com/reviews/genetics250| APRIL 2002 |VOLUME 3REVIEWSMaynard Smith, J. The Evolution of Sex(Cambridge Univ.Press, Cambridge, UK, 1978).2. Williams, G. C. Sex and Evolution(Princeton Univ. Press, NewJersey, 1975).3. Fisher, R. A. The Genetical Theory of Natural Selection(OxfordUniv. Press, Oxford, UK, 1930).4. Stebbins, G. L. Variation and Evolution in Plants(ColumbiaUniv. Press, New York, 1950).5. White, M. J. D. Modes of Speciation(W. H. Freeman, SanFrancisco, 1978).6. Bell, G. The Masterpiece of Nature: The Evolution andGenetics of Sexuality(Croom Helm and Univ. California Press,London and Berkeley, 1982).References 4–6 provide exhaustive reviews of theprevalence of clonal reproduction among differenttaxonomic groups.7. Welsh, D. M. & Meselson, M. Evidence for the evolution ofbdelloid rotifers without sexual reproduction or geneticexchange. Science288, 1211–1215 (2000).8. Judson, O. P. & Nomark, B. B. Ancient asexual scandals.Trends Ecol. Evol. 11, A41–A46 (1996).9. Butlin, R., Schon, I. & Martens, K. Asexual reproduction innonmarine ostracods. Heredity81, 473–480 (1998).10. Steinemann, M. & Steinemann, S. Enigma of Y chromosomedegeneration: neo-Y and neo-X chromosomes of Drosophilamiranda, a model for sex chromosome evolution. Genetica103, 409–420 (1998).11. Lahn, B. T. & Page, D. C. Four evolutionary strata on thehuman X chromosome. Science 286, 9–967 (1999).12. Lynch, M. & Blanchard, J. L. Deleterious mutationaccumulation in organelle genomes. Genetica103, 29–39(1998).13. Tilford, C. A. et al. A physical map of the human Ychromosome. Nature409, 943–945 (2001).14. Rice, W. R. Parent–offspring pathogen transmission: aselective agent promoting sexual recombination. Am. Nat.121, 187–203 (1983).15. Kimura, M. & Maruyama, T. Mutational load with epistaticgene interactions in fitness. Genetics54, 1303–1312 (1966).A classic reference to the theoretical concept ofmutational load and how it is influenced byrecombination.16. Kondroshov, A. S. Deleterious mutations as an evolutionaryfactor. 1. The advantage of recombination. Genet. Res. 44,199–217 (1984).17. Crow, J. F. Mathematical Topics in Population Genetics (ed.Kojima, K. I.) (Springer, Heidelberg, 1970).Reviews the experimental work concerning changes inrecombination rate that is associated with strongdirectional selection in small populations.18. Rice, W. R. Requisite mutational load, pathway epistasis, anddeterministic mutation accumulation in sexual versus asexualpopulations. Genetica103, 71–81 (1998).19. Otto, S. P. & Barton, N. H. Selection for recombination in smallpopulations. Evolution55, 1921–1931 (2001).20. Zeyl, C., Bell. G. & Green, D. M. Sex and the spread ofretroposon Ty3 in experimental populations ofSaccharomyces cerevisiae. Genetics143, 1567–1577(1996).21. Wright, S. & Finnegan, D. Genomic evolution: sex and thetransposable element. Curr. Biol.11, R296–R299 (2001).22. Siller, S. Sexual selection and the maintenance of sex. Nature411, 6–692 (2001).23. Agrawal, A. F. Sexual selection and the maintenance of sexualreproduction. Nature411, 692–695 (2001).24. Muller, H. J. Some genetic aspects of sex. Am. Nat. 66,118–138 (1932).25. Crow, J. F. & Kimura, M. Evolution in sexual and asexualpopulations. Am. Nat. 99, 439–450 (1965).26. Hill, W. H. & Robertson, A. The effect of linkage on limits toartificial selection. Genet. Res. 8, 269–294 (1966).27. de Visser, J. A. G. M., Zeyl, C. W., Gerrish, P. J., Blanchard, J.L. & Lenski. R. E. Diminishing returns from mutation supplyrate in asexual populations. Science283, 404–406 (1999).28. Keightley, P. D. & Eyre-Walker, A. Deleterious mutations andthe evolution of sex. Science290, 331–333 (2000).29. Davies, E. K. & Keightley, P. D. High frequency of crypticdeleterious mutations in Caenorhabditis elegans. Science285, 1748–1751 (1999).Provides an empirical appraisal of the effectiveness ofusing the mutation accumulation technique to estimatethe genome-wide mutation rate.30. Thatcher, J. W., Shaw, J. M. & Dickinson, W. J. S. Marginalfitness contributions of nonessential genes in yeast. Proc. NatlAcad. Sci. USA95, 253–257 (1998).31. Wright, S. Evolution in Mendelian populations. Genetics16,97–159 (1931).32. Kimura, M. On the probability of fixation of mutant genes in apopulation. Genetics47, 713–719 (1962).33. Peck, J. R. A ruby in the rubbish: beneficial mutations,deleterious mutations and the evolution of sex. Genetics137,597–606 (1994).34. Charlesworth, D., Charlesworth, B. & Morgan, M. T. Thepattern of neutral molecular variation under the backgroundselection model. Genetics141, 1619–1632 (1995).1. 35. Charlesworth, B. The effect of background selection againstdeleterious mutations on weakly selected, linked variants.Genet. Res. 63, 213–227 (1994).36. Barton, N. H. Linkage and the limits to natural selection.Genetics140, 821–841 (1995).References 33–36 are key theoretical referencesconcerning the adaptive significance of sexualrecombination.37. Manning, J. T. & Thompson, D. J. Muller’s ratchet and theaccumulation of favorable mutations. Acta Biotheor. 33,219–225 (1984).38. Rice, W. R. Evolution of the Y sex chromosome in animals.Bioscience46, 331–343 (1996).39. Li, W.-H. Maintenance of genetic variability under joint effect ofmutation, selection and random drift. Genetics90, 349–382(1978).40. Butcher, D. Mullers ratchet, epistasis and mutation effects.Genetics141, 431–437 (1995).41. Gessler, D. D. G. The constraints of finite size in asexualpopulations and the rate of the ratchet. Genet. Res. 66,241–253 (1995).42. Rice. W. R. Genetic polarization: unifying theories for theadaptive significance of recombination. J. Evol. Biol. 12,1047–1049 (1999).43. Gordo, I. & Charlesworth, B. On the speed of Muller’s ratchet.Genetics156, 2137–2140 (2000).44. Chippindale, A. K., Gibson, J. R. & Rice, W. R. Negativegenetic correlation for adult fitness between sexes revealsontogenetic conflict in Drosophila. Proc. Natl Acad. Sci. USA98, 1671–1675 (2001).45. Drake, J. W., Charlesworth, B., Charlesworth, D. & Crow, J. F.Rates of spontaneous mutation. Genetics148, 1667–1686(1998).46. Lynch, M. et al. Perspective: spontaneous deleteriousmutation. Evolution53, 5–663 (1999).References 45 and 46 are extensive reviews ofempirical work on the genome-wide mutation rate andthe fitness effects of spontaneous mutations.47. Bataillon, T. Estimation of spontaneous genome-widemutation rate parameters: whither beneficial mutations?Heredity84, 497–501 (2000).48. Kondrashov, A. S. Sex and U. Trends Genet. 17, 75–77(2001).49. Keightley, P. D. & Eyre-Walker, A. Sex and U: response.Trends Genet. 17, 77–78 (2001).50. Kondrashov, A. S. & Crow, J. F. A molecular approach toestimating the human deleterious mutation rate. Hum. Mutat.2, 229–234 (1993).51. Whitlock, M. C. & Bourguet, D. Factors affecting the geneticload in Drosophila: synergistic epistasis and correlationsamong fitness components. Evolution54, 1654–1660 (2000).52. Barkai, N. & Leibler, S. Robustness in simple biochemicalnetworks. Nature387, 913–917 (1997).53. Wiley, R. H. & Poston, J. Perspective: indirect mate choice,competition for mates, and coevolution of the sexes. Evolution50, 1371–1381 (1996).54. Rowe, L. & Arnqvist, G. Analysis of the causal components ofassortative mating in water striders. Behav. Ecol. Sociobiol.38, 279–286 (1996).55. Holland, B. & Rice, W. R. Experimental removal of sexualselection reverses intersexual antagonistic coevolution andremoves a reproductive load. Proc. Natl Acad. Sci. USA96,5083–5088 (1999).56. Parker, G. A. in Sexual Selection and ReproductiveCompetition in Insects(eds Blum, M. S. & Blum, N. A.)123–166(Academic, New York, 1979).57. Chapman, T. et al. Cost of mating in Drosophila melanogasterfemales is mediated by male accessory gland products.Nature373, 241–244 (1995).58. Rice, W. R. Sexually antagonistic male adaptation triggered byexperimental arrest of female evolution. Nature361, 232–234(1996).59. Rice, W. R. & Holland, B. The enemies within: intergenomicconflict, interlocus contest evolution (ICE), and the intraspecificRed Queen. Behav. Ecol. Sociobiol. 41, 1–10 (1997).60. Arnqvist, G. & Rowe, L. Sexual conflict and arms racesbetween the sexes: a morphological adaptation for control ofmating in a female insect. Proc. R. Soc. Lond. B261,123–127 (1995).61. Nachman, M. W. & Crowell, S. L. Estimate of the mutationrate per nucleotide in humans. Genetics156, 297–304(2000).62. Redfield, R. J. Male mutation rates and the cost of sex forfemales. Nature69, 145–147 (1994).63. Petrie, M. Improved growth and survival of offspring ofpeacocks with more elaborate trains. Nature371, 598–599(1994).. Drickamer, L. C., Gowaty, P. A. & Holmes, C. M. Free femalemate choice in house mice affects reproductive success andoffspring viability and performance. Anim. Behav. 59,371–378 (2000).65. Reinke, V. et al. A global profile of germline gene expression inC. elegans. Mol. Cell6, 605–616 (2000).66. Rice, W. R. & Chippindale, A. K. Intersexual ontogeneticconflict. J. Evol. Biol. 14, 685–693 (2001).67. de Visser, J. A. G. M. et al. Diminishing returns from mutationsupply rate in asexual populations. Science283, 404–406(1999).68. Miralles, R., Gerrish, P. J., Moya, A. & Elena, S. F. Clonalinterference and the evolution of RNA viruses. Science285,1745–1747 (1999).69. Miralles, R., Moya, A. & Elena, S. F. Diminishing returns ofpopulation size in the rate of RNA virus adaptation. J. Virol. 74,3566–3571 (2000).70. de la Pena, M., Elena. S. F. & Moya, A. Effect of deleteriousmutation accumulation on the fitness of RNA bacteriophageMS2. Evolution54, 686–691 (2000).71. Chao, L., Tran, T. R. & Matthews, C. Muller ratchet and theadvantage of sex in the RNA virus-phi-6. Evolution46,2–299 (1992).72. Chao, L., Tran, T. T. & Tran, T. T. The advantage of sex in theRNA virus phi 6. Genetics147, 953–959 (1997).Illustrates the capacity of non-recombining populationsto rapidly recover from fitness loss, attributed to theaccumulation of deleterious mutations, through theaccumulation of compensatory mutations.73. Rice, W. R. Degeneration of a nonrecombining chromosome.Science263, 230–232 (1994).74. Carson, H. L. Response to selection under differentconditions of recombination in Drosophila. Cold Spring Harb.Symp.Quant. Biol.23, 291–306 (1958).75. McPhee, C. P. & Robertson, A. Effect of suppressingcrossing-over on response to selection in Drosophilamelanogaster. Genet. Res. 16, 1–16 (1970).76. Markow, T. A. Genetic analysis of phototactic behavior inDrosophila melanogaster. 1. Selection in presence ofinversions. Genetics79, 527–534 (1975).77. Thompson, V. Recombination and response to selection inDrosophila melanogaster. Genetics85, 125–140 (1977).78. Rice, W. R. & Chippindale, A. K. Sexual recombination andthe power of natural selection. Science294, 555–559 (2001).79. Kolkman, J. A. & Stemmer, W. P. C. Directed evolution ofproteins by exon shuffling. Nature Biotechnol. 19, 423–428(2001).80. Elena, S. F. & Lenski, R. E. Test of synergistic interactionsamong deleterious mutations in bacteria. Nature390,395–398 (1997).81. Mukai, T. Genetic structure of natural populations ofDrosophila melanogaster. 1. Spontaneous mutation rate ofpolygenes controlling viability. Genetics50, 1–19 (19).82. Kitagawa, O. Interactions in fitness between lethal genes inheterozygous condition in Drosophila melanogaster. Genetics57, 809–820 (1967).83. Elena, S. F. Little evidence for synergism among deleteriousmutations in a nonsegmented RNA virus. J. Mol. Evol. 49,703–707 (1999).84. de la Pena, M., Elena, S. F. & Moya, A. Effect of deleteriousmutation accumulation on the fitness of RNA bacteriophageMS2. Evolution54, 686–691 (2000).85. Peters, A. D. & Keightley, P. D. A test for epistasis amonginduced mutations in Caenorhabditis elegans. Genetics 156,1635–17 (2000).This and reference 27 describe two of the bestexperiments, so far, to test for epistasis among newlyarisen mutations.86. deVisser, J. A. G. M., Hoekstra, R. F. & VandenEnde, H. Theeffect of sex and deleterious mutations on fitness inChlamydomonas. Proc. R. Soc. Lond. B263, 193–200(1996).87. West, S. A., Peters, A. D. & Barton, N. H. Testing for epistasisbetween deleterious mutations. Genetics149, 435–444(1998).88. Wloch, D. M., Borts, R. H. & Korona, R. Epistatic interactionsof spontaneous mutations in haploid strains of the yeastSaccharomyces cerevisiae. J. Evol. Biol. 14, 310–316 (2001).. Malmberg, R. L. Evolution of epistasis and advantage ofrecombination in populations of bacteriophage-T4. Genetics86, 607–621 (1977).90. Souza, V., Turner, P. E. & Lenski, R. E. Long-term experimentalevolution in Escherichia coli. 5. Effects of recombination withimmigrant genotypes on the rate of bacterial evolution. J. Evol. Biol.10, 743–769 (1997).91. Birdsell, J. & Wills, C. Significant competitive advantageconferred by meiosis and syngamy in the yeastSaccharomyces cerevisiae. Proc. Natl Acad. Sci. USA93,908–912 (1996).92. Greig, D., Borts. R. H. & Louis, E. J. The effect of sex onadaptation to high temperature in heterozygous andhomozygous yeast. Proc. R. Soc. Lond. B265, 1017–1023(1998).93. Zeyl, C. & Bell, G. The advantage of sex in evolving yeastpopulations. Nature388, 465–468 (1997).Online linksFURTHER INFORMATIONWilliam Rice’s lab:http://lifesci.ucsb.edu/EEMB/faculty/rice/index.htmlAccess to this interactive links box is free online.NATURE REVIEWS |GENETICSVOLUME 3 |APRIL 2002 | 251