Molecular Ecology (2008) 17, 4177–4180

© 2008 The AuthorJournal compilation © 2008 Blackwell Publishing Ltd

Blackwell Publishing LtdNEWS AND VIEWS

PERSPECTIVE

Hitching a lift on the road to speciation

CAROLE SMADJA , JUAN GALINDO and ROGER BUTLINAnimal and Plant Sciences, The University of Sheffield, Western Bank, Sheffield S10 2TN, UK

Abstract

Understanding how speciation can take place in the presenceof homogenizing gene flow remains a major challenge in evo-lutionary biology. In the early stages of ecological speciation,reproductive isolation between populations occupying dif-ferent habitats is expected to be concentrated around genesfor local adaptation. These genomic regions will show highdivergence while gene exchange in other regions of thegenome should continue relatively unimpaired, resulting inlow levels of differentiation. The problem is to explain howspeciation progresses from this point towards completereproductive isolation, allowing genome-wide divergence.A new study by Via and West (2008) on speciation betweenhost races of the pea aphid, Acyrthosiphon pisum, introducesthe mechanism of ‘divergence hitchhiking’ which can generatelarge ‘islands of differentiation’ and facilitate the build-up oflinkage disequilibrium, favouring increased reproductiveisolation. This idea potentially removes a major stumblingblock to speciation under continuous gene flow.

Keywords: ecological speciation, gene flow, hitchhiking, in-direct selection, islands of differentiation

Received 30 May 2008; revision received 28 July 2008; accepted30 July 2008

Via & West (2008) combined an AFLP-based genome scanfor differentiation with a QTL map for key traits involved inlocal adaptation. This verified that the FST outliers detected bythe genome scan are linked to QTLs for key ecologicallyimportant traits in pea aphid host races (Hawthorne & Via2001). This association has been assumed in many previousstudies but was only demonstrated in one other system(whitefish, Coregonus; Rogers & Bernatchez 2007). However, themost exciting contribution is that Via and West are the first touse this information to estimate the size of the genomic regionsaffected by the QTLs under selection. They show that, onaverage, outliers are 10.6 centimorgans (cM) from the nearestmapped QTL and they suggest these outliers reveal surprisinglylarge genomic islands of differentiation. Divergence hitchhiking

is the mechanism they propose to explain how differentiationcan be generated and maintained over such large regions. Ithas major implications for ecological speciation.

The idea of hitchhiking is usually associated with the spreadof an advantageous allele through a population which gener-ates a transient reduction in neutral genetic diversity at sitesclose to the selected locus, a selective sweep (Fig. 1a). Theeffect is genomically localized because of recombination and istransient because of new mutations (Maynard Smith & Haigh1974). Divergence hitchhiking is different (Fig. 1c). Early in theprocess of ecological speciation, genetic differentiation betweenpopulations occupying different habitats is expected to beconcentrated around genes for local adaptation while geneexchange in other regions of the genome should continuerelatively unimpaired, creating a genetic mosaic of diver-gence. Between populations, the effective rate of recombinationbetween selected loci and neighbouring neutral loci is reducedbecause recombination is only possible when haplotypes arebrought together in the same individual. Habitat choice, assor-tative mating, immigrant inviability and selection againsthybrids all make the production of recombinant haplotypesunlikely. Within populations, normal recombination cancontinue. As a result, divergence between populations that isgenerated by the spread of locally adapted alleles is main-tained over large genomic regions and the balance betweendrift and gene flow in these regions also produces greaterdifferentiation because the effective migration rate is lower.

An important theoretical study by Charlesworth et al. (1997)had previously explored how equilibrium neutral diversitylevels are affected by selection at neighbouring sites in thepresence of population subdivision. They showed that popu-lation subdivision increases the genomic extent of the impactof local selection on neutral sites. The appropriate comparisonhere is not with a selective sweep, where the effect is transient,but with balancing selection (Fig. 1b). This generates an areaof elevated polymorphism due to hitchhiking which is equiv-alent to differentiation between two subpopulations, charac-terized by the alternative alleles (such as the F and S alleles in theclassic case of alcohol dehydrogenase in Drosophila mela-nogaster: Kreitman & Hudson 1991) but connected by maximalgene flow (m = 0.5). While divergent selection between subdi-vided populations (m = 0.001) leads to large islands (at 10 cM,FST = 0.21 compared to the neutral expectation of 0.11, basedon the Charlesworth et al. model), the same selection pressurewould create a much smaller island when selection acts withina population (at 10 cM, FST = 0.01).

Via and West show that their results are consistent withthese theoretical predictions. However, while the Charles-worth et al. model has been invoked in connection with usingoutliers to find genes for adaptation (Storz 2005), its implicationsfor speciation in the presence of gene flow have not previouslybeen recognized. One important implication of divergence

Correspondence: Roger Butlin, Fax: +44 114 2220002; E-mail: r.k.butlin@sheffield.ac.uk

4178 N E W S A N D V I E W S

© 2008 The AuthorJournal compilation © 2008 Blackwell Publishing Ltd

Fig. 1 The effect of hitchhiking under different conditions and its role in speciation. The figure shows the effect of hitchhiking (HH) on thegenetic variation at neutral loci physically linked to selected ones (a) during a selective sweep in a single population, (b) under balancingselection, and (c) when the population is subdivided (divergence hitchhiking). In all cases, the initial situation (t1) represents a sample ofhaplotypes with a certain degree of polymorphism and the presence of an advantageous allele which originates from recent mutation orstanding variation. (a) During a selective sweep, the advantageous allele increases in frequency due to selection and closely linked neutralregions hitchhike, generating a valley of diversity (t2). After a longer period of time (t3), the decay of HH at the selected and closely linkedloci due to new mutations and drift tends to rebuild the original levels of diversity within the population. (b) Under balancing selection,the new allele spreads to an equilibrium frequency, defining a fraction of the population within which closely linked neutral regionshitchhike in a similar way to the selective sweep (t2). Diversity within allelic classes (subpopulations) is reduced (although less than duringa selective sweep) while diversity increases between subpopulations. Over time, the effect of HH abates, leaving only a narrow peak ofbetween-subpopulation diversity close to the selected locus (t3). (c) In a subdivided population (subpopulations in habitats I and II), an alleleunder divergent selection increases in frequency in I and closely linked regions hitchhike (t2). The initial hitchhiking effect depends on allelefrequencies at polymorphic sites in the regions flanking the selected locus (see text). Because selection reduces gene flow (and thereforeeffective recombination) between subpopulations in this genomic region, the extent of HH also affects loosely linked regions, increasing thesize of the ‘island of differentiation’. New mutations accumulate over time (t3), and if a new mutation responsible for a further reduction ofgene flow (e.g. due to habitat choice) arises within the ‘island’, its frequency is likely to increase and additional regions will hitchhike withit, further extending the island of differentiation (t4).

N E W S A N D V I E W S 4179

© 2008 The AuthorJournal compilation © 2008 Blackwell Publishing Ltd

hitchhiking is that it may overcome a major obstacle to speci-ation in sympatry or parapatry. If divergence hitchhiking canaffect loosely linked loci, it is expected to facilitate linkage dis-equilibrium (LD) between genes underlying different traitsimportant in speciation. This LD is crucial in most scenarios ofspeciation with gene flow, for example for habitat choice orassortative mating to evolve and complete speciation that wasinitiated by local adaptation (Felsenstein 1981), and it remainsa challenge to understand how LD can evolve in the presenceof gene flow (Gavrilets 2004). For LD to be maintained undercontinuing gene flow, the destructive effect of recombinationthat breaks down the association between genes involved inlocal adaptation and genes influencing premating isolation mustbe overcome. Pleiotropy, very close physical linkage or reductionof the actual recombination rate in these regions (e.g. due toinversions) are usually considered to be prerequisites, at leastin ‘two-allele’ models (Kirkpatrick & Ravigné 2002; Bolnick &Fitzpatrick 2007). Via and West suggest that divergence hitch-hiking removes this biggest perceived difficulty with sympat-ric speciation, by offering a much more widely applicablemechanism for the build-up of LD between remote loci in theface of gene flow (Fig. 1c at time t4). In the pea aphid, hostplant preference — a key component of premating isolation —could have evolved through divergence hitchhiking. Theprocess may underlie other cases of divergence-with-gene-flow, and if confirmed in other taxa, divergence hitch-hiking could turn out to be a fundamentally important partof speciation.

Detecting divergence hitchhiking and estimating the sizesof regions of differentiation it creates during the speciationprocess is not an easy task. First, it is necessary to analyse pop-ulations at the appropriate stage of divergence. As speciationprogresses and the background level of genetic divergenceincreases, it becomes more difficult to distinguish islands ofdifferentiation. Second, if there is a period of allopatry fol-lowed by secondary contact, differentiation may be associatedwith other barriers to gene exchange, such as incompatibilitiesdue to genetic drift or divergent responses to uniform selectionpressures. In this case, islands of differentiation cannot beattributed confidently to selection for local adaptation (viadivergence hitchhiking) but they may still be important infacilitating the further evolution of reproductive isolation, forexample, via reinforcement. This means that information on thehistory of population divergence (time of divergence, geo-graphical settings, etc.) is needed for the proper interpretationof the genomic pattern of differentiation.

Description of the differentiation pattern requires anappropriate methodology. Via and West’s approach ofmeasuring the map distance between differentiated ‘outlier’loci and the nearest QTL for an adaptive trait is a powerfulone. When they re-analysed data from the only other studythat has combined QTL and outlier analysis (in whitefish,Coregonus: Rogers & Bernatchez 2007), they found a compara-ble mean distance of 16.5 cM. However, unless levels of differ-entiation are consistently high in the regions surrounding theQTLs, it might be difficult to disentangle the role of divergencehitchhiking from other factors. Via and West’s results actuallyshow wide variation among loci within the islands of differen-tiation (for a mean FST of 0.25, individual locus values range

from 0.0 to ~0.72). How can this variability be explained? Itcould result from the stochastic effects of drift in the history ofthe two populations (and, to a small extent, experimentalsampling) (as commonly simulated in ‘population genomics’studies, e.g. Nosil et al. 2008). It could also be a transient effectof the spread of new advantageous alleles within subpopula-tions on ancestral polymorphisms: if the new advantageousmutation is associated with a common allele at a neutral locus,its spread within one subpopulation creates less differentiationbetween subpopulations than it does if it is associated with arare allele (Fig. 1c at t2). The effect of ancestral polymorphismwill decay over time, due to mutation, drift and gene flow(Fig. 1c at t3), as it does after a selective sweep within onepopulation (Fig. 1a at t3) (Maynard Smith & Haigh 1974).However, the variation that Via and West observe may alsosuggest an alternative hypothesis: rather than large areas ofdifferentiation around single loci under strong selection, theislands observed may actually be composites of multiplesmaller areas of differentiation. This would imply that thestudy by Hawthorne & Via (2001) did not identify all QTLslikely to influence reproductive isolation between host-plantassociated populations. Via and West consider it unlikely thatthe nine out of 12 outlier loci observed > 5 cM from the nearestQTL could be explained in this way when the mapping studyincluded two composite traits, acceptance and performance,which, together, are expected to capture all key components oflocal adaptation.

Two recent studies, also based on genome scans, have sug-gested that islands of differentiation are much smaller than the10 cM proposed by Via and West (in Helianthus Yatabe et al.2007, and in Littorina Wood et al. 2008). In Helianthus, anextended period of allopatry means that processes other thanlocal adaptation contribute much of the observed differentiationbut in Littorina any period of allopatry must have been shortand the background level of differentiation is lower than inpea aphids. Either study might have interpreted an area of lowdifferentiation close to a putatively selected locus as the limitof the hitchhiking effect when it is actually part of the variabilityexpected within islands of differentiation (Via & West 2008).The two interpretations of large, noisy islands around singleloci vs. composites of smaller islands will be hard to distin-guish. Since a QTL should always be interpreted as a region ofchromosome rather than a single locus, and the effects ofneighbouring selected loci on gene exchange will not be inde-pendent (as in hybrid zones, e.g. Szymura & Barton 1991), thesetwo explanations are really ends of a continuum. Indeed, ifdivergence at one locus facilitates differentiation at nearbyloci, we actually expect to see clusters of selected loci. Distin-guishing the interpretations may depend on a better under-standing of population history, effective size and geneexchange since these factors determine the effects of drift andselection on differentiation. An insightful new study of Helianthussunflowers focuses attention on these parameters. Using the‘isolation-with-migration’ modelling approach, Strasburg &Rieseberg (2008) estimate very large effective population sizes(N = 1.8 and 2.4 × 106) and very low migration rates (mbetween 1.3 and 3.3 × 10–07) for H. annuus and H. petiolaris,respectively. Despite the very low m, the number of migrantsNm is enough to prevent divergence due to drift. However,

4180 N E W S A N D V I E W S

© 2008 The AuthorJournal compilation © 2008 Blackwell Publishing Ltd

weak selection can produce strong divergence because therequirement is only for s > m. This is consistent with the smallislands of differentiation suggested by Yatabe et al. (2007). Itwill be important to obtain similar parameter estimates for peaaphids, and for Littorina and Coregonus, in order to improveour understanding of the processes generating observed genomicpatterns of differentiation. Moreover, while the approachesusing AFLP-based genome scans and QTL mapping alreadyprovide exciting results on this topic, the development of newsequencing methods and the increasing availability of sequencedgenomes is likely to offer additional and complementaryoptions to detect selected loci and assess the sizes of genomicregions of differentiation that they create (e.g. microarrays,Turner et al. 2005).

In conclusion, Via and West have made a major contribution byrevealing a surprising and complex pattern of differentiationbetween pea aphid host races and by emphasizing the potentialfor divergence hitchhiking to spread the influence of individual,strongly selected loci across large genomic regions. It will beexciting to see how further studies on the pea aphid willimprove the estimation of the size of these islands of differen-tiation. Also, it will be interesting to see how the apparentdifferences between this system and others are resolved.Whatever the outcome, it is clear that recognition of themechanism of divergence hitchhiking is a highly significantstep forward in our understanding of ecological speciation.

References

Bolnick DI, Fitzpatrick BM (2007) Sympatric speciation: models andempirical evidence. Annual Review of Ecology, Evolution, and Systematics,38, 459–487.

Charlesworth B, Nordborg M, Charlesworth D (1997) The effects oflocal selection, balanced polymorphism and background selectionon equilibrium patterns of genetic diversity in subdivided populations.Genetical Research, 70, 155–174.

Felsenstein J (1981) Skepticism towards Santa Rosalia, or why are thereso few kinds of animals? Evolution, 35, 124–138.

Gavrilets S (2004) Fitness Landscapes and the Origin of Species. PrincetonUniversity Press, Princeton, New Jersey.

Hawthorne DJ, Via S (2001) Genetic linkage of ecological specializationand reproductive isolation in pea aphids. Nature, 412, 904–907.

Kirkpatrick M, Ravigné V (2002) Speciation by natural and sexual

selection: models and experiments. The American Naturalist, 159,S22–S35.

Kreitman M, Hudson RR (1991) Inferring the evolutionary histories ofthe Adh and Adh-Dup loci in Drosophila melanogaster from patterns ofpolymorphism and divergence. Genetics, 127, 565–582.

Maynard Smith J, Haigh J (1974) The hitchhiking effect of a favourablegene. Genetical Research, 23, 23–35.

Nosil P, Egan SP, Funk DJ (2008) Heterogeneous genomic differentia-tion between walking-stick ecotypes: ‘isolation by adaptation’ andmultiple roles for divergent selection. Evolution, 62, 316–336.

Rogers SM, Bernatchez L (2007) The genetic architecture of ecologicalspeciation and the association with signatures of selection in naturallake whitefish (Coregonus sp. Salmonidae) species pairs. MolecularBiology and Evolution, 24, 1423–1438.

Storz JF (2005) Using genome scans of DNA polymorphism to inferadaptive population divergence. Molecular Ecology, 14, 671–688.

Strasburg JL, Rieseberg LH (2008) Molecular demographic history ofthe annual sunflowers Helianthus annuus and H. petiolaris — largeeffective population sizes and rates of long-term gene flow. Evolution.62, 1936–1950.

Szymura JM, Barton NH (1991) The genetic structure of the hybridzone between the fire-bellied toads Bombina bombina and B. Variegata:comparisons between transects and between loci. Evolution, 45, 237–261.

Turner TL, Hahn MW, Nuzhdin SV (2005) Genomic islands of speciationin Anopheles gambiae. Public Library of Science Biology, 3, 1572–1578.

Via S, West J (2008) The genetic mosaic suggests a new role for hitchhikingin ecological speciation. Molecular Ecology, 17, 4334–4345.

Wood HM, Grahame JW, Humphray S, Rogers J, Butlin RK (2008)Sequence differentiation in regions identified by a genome scan forlocal adaptation. Molecular Ecology, 17, 3123–3135.

Yatabe Y, Kane NC, Scotti-Saintagne C, Rieseberg LH (2007) Rampantgene exchange across a strong reproductive barrier between theannual sunflowers, Helianthus annuus and H. petiolaris. Genetics,175, 1883–1893.

Carole Smadja, Juan Galindo and Roger Butlin are working at theuniversity of Sheffield. Their research focuses on speciation in variousorganisms and they are interested in using population genomicsapproaches to gain insights into the genetic basis of adaptive speciation.

doi: 10.1111/j.1365-294X.2008.03917.x