sign in sign up
<<
Home , Allopatric speciation, Cline (biology), Ecological selection, Parapatric speciation, Speciation, Sympatry

330 Review TRENDS in & Vol.16 No.7 July 2001 Theory and

Michael Turelli, Nicholas H. Barton and Jerry A. Coyne

The study of speciation has become one of the most active areas of selection, and the relative rates of evolution of evolutionary , and substantial progress has been made in documenting premating and postmating isolation, is more likely to and understanding phenomena ranging from and emerge from empirical than from mathematical reinforcement to the evolutionary of postzygotic isolation. This analyses4–6. However, some important questions are progress has been driven largely by empirical results, and most useful mathematically tractable, such as the plausibility of theoretical work has concentrated on making sense of empirical patterns. reinforcement7,8 and the evolutionary consequences Given the complexity of speciation, mathematical theory is subordinate to of genetic incompatibilities that cause postzygotic verbal theory and generalizations about data. Nevertheless, mathematical isolation9,10. theory can provide a useful classification of verbal theories; can help determine We first discuss , and then the biological plausibility of verbal theories; can determine whether alternative proceed to PARAPATRIC and sympatric speciation. Given mechanisms of speciation are consistent with empirical patterns; and can the recent flurry of empirical and theoretical studies occasionally provide predictions that go beyond empirical generalizations. We of sympatric speciation11–13 (Via14, this issue) and of discuss recent examples of progress in each of these areas. rapid PHENOTYPIC DIVERGENCE driven by (Schluter15, this issue), our traditional This review covers both verbal and mathematical biogeographical approach requires some defense. We theories of SPECIATION (see Glossary) in sexually offer two rationales, one conceptual, the other reproducing diploid . Adopting the empirical. BIOLOGICAL CONCEPT, we equate speciation to the The conceptual rationale is simply that, given evolution of reproductive isolating mechanisms that enough time, speciation is an inevitable consequence essentially prevent exchange between newly of evolving in allopatry. Because there arising taxa. However, our use of this definition does are no forces acting to enforce reproductive not imply that we believe to be compatibility between geographically isolated essential for morphological, ecological or genetic populations, they will eventually become divergence, or that we regard other aspects of reproductively incompatible. Thus, in contrast to the divergence between sympatric or allopatric groups as balance of forces required to produce sympatric or being less interesting than the evolution of , allopatric speciation requires reproductive isolation. only geographical isolation and time. As with most areas of biology, theories In addition, several lines of empirical evidence of speciation are generally verbal, describing support the view that allopatric speciation is conditions or mechanisms that are thought to cause pervasive; whereas current data suggest that the reproductive isolation. Classic examples include opposite extreme, sympatric speciation, is far less Dobzhansky’s suggestion1 that evolution in ALLOPATRY common. First, we have many empirical examples of leads to POSTZYGOTIC ISOLATION through the allopatric speciation (e.g. geminate sister pairs and accumulation of incompatibilities between in island endemics). Second, despite recent data and different LINEAGES, and Mayr’s idea2 that population theory indicating that sympatric speciation almost BOTTLENECKS can produce rapid speciation through certainly occurs, few examples are unambiguous or ‘genetic revolutions’. Mathematical analyses serve widely accepted. Because it is harder to demonstrate mainly to test the plausibility of such conjectures. sympatric than allopatric speciation, these first two Recent treatments of SYMPATRIC SPECIATION and lines of evidence are compromised by an Michael Turelli* Section of Evolution and REINFORCEMENT are obvious examples. ascertainment bias. However, recent comparative Ecology, University Discussions of have been analyses5 (Barraclough and Nee16, this issue) show of California, Davis, dominated by mathematically based theory. By that, in several taxa, the most recently evolved CA 95616, USA. contrast, speciation is dominated by verbal theories species generally have allopatric ranges, supporting *e-mail: [email protected] because the process involves so many complex Mayr’s view that allopatric speciation might be most mechanisms, including ecology, behavior and common. These tests are conservative, because Nicholas H. Barton ICAPB, Division of interactions between multilocus . related sympatric species might falsely appear to be Biological Sciences, Moreover, speciation focuses on a composite trait – sister species because of genetic homogenization University of Edinburgh, reproductive isolation – that is a property of pairs of resulting from hybridization after divergence in Edinburgh, UK EH9 3JT. taxa and hence is inherently more complex than are allopatry. Third, if sympatric speciation were Jerry A. Coyne traits evolving within a single lineage3. Because of extremely common, we would expect to often see Dept of Ecology and this complexity, progress on major issues, including sister taxa of highly mobile species on islands, but the Evolution, University of 6,17 Chicago, 1101 E. 57th St, the of speciation, the frequency of few existing studies do not show such a pattern. An Chicago, IL 60637, USA. reinforcement, the roles of sexual versus natural alternative review of speciation could be based on a

http://tree.trends.com 0169–5347/01/$ – see front matter © 2001 Ltd. All rights reserved. PII: S0169-5347(01)02177-2 Review TRENDS in Ecology & Evolution Vol.16 No.7 July 2001 331

Box 1. Evolution of prezygotic isolation via

Sexual selection can produce sexual females who mate with males possessing prezygotic isolation. As in other ‘quasi- isolation between allopatric populations ‘good ’. There can often be multiple neutral’ models, many combinations of because female preferences can evolve for stable states that produce sexual isolation. trait and preference are compatible, and arbitrary male characteristics. In Lande’s Slight perturbations (e.g. the effect of transitions between them can be produced seminal modela, both trait and preference environmental differences between by natural and/or sexual selection. Paths of are additive polygenic traits. The female isolated populations) might then lead to high that connect incompatible preference is assumed to have no direct rapid evolutionary changes in trait and genotypes also occur in the effect on fitness, evolving solely through preferencec,f. However, Dobzhansky–Muller model of epistasisn its association with the male trait. When need not be involved in speciation via and in Gavrilets’ ‘holey landscapes’o. the mean trait value is plotted against the sexual selection: examples are changes in References mean strength of female preference, this male genitalia or post-mating, prezygotic a Lande, R. (1981) Models of speciation by sexual model produces a line of neutral reproductive isolation (e.g. sperm–egg selection on polygenic traits. Proc. Natl. Acad. equilibria, corresponding to the infinite incompatibility), both of which can be Sci. U. S. A. 78, 3721–3725 number of ways that natural and sexual driven by male–male or by b Lande, R. (1982) Rapid origin of sexual isolation and character divergence in a . Evolution selection on the male trait can female behavioral or biochemical 36, 213–223 counterbalance. Isolated populations can ‘preference’. c Lande, R. and Kirkpatrick, M. (1988) Ecological move freely along the line, leading to ‘Arms races’ set off by a divergence speciation by sexual selection. J. Theor. Biol. sexual isolation. between the reproductive interests of 133, 85–98 d Payne, R.J.H. and Krakauer, D.C. (1997) Sexual Many models have built upon Lande’sa, males and females can also cause non- selection, space, and speciation. Evolution 51, invoking various mechanisms to drive ecological sexual selection that yields 1–9 populations to different points on the speciationk,l. However, other types of arms e Pomiankowski, A. and Iwasa, Y. (1998) equilibrium line; these mechanisms (e.g. –parasite ) could Runaway ornament diversity caused by include , selection on the male also drive divergence, and if sexual Fisherian sexual selection. Proc. Natl. Acad. Sci. U. S. A. 96, 5106–5111 trait, and changes in the intensity of female selection is important in speciation, one f Price, T. (1998) Sexual selection and natural a–g preferences . In many of these models, must explain why it is especially likely to selection in speciation. Philos. Trans. R. however, the evolution of reproductive be involved in such coevolutionary arms Soc. London Ser. B 353, 251–260 isolation is facilitated by the unrealistic races. One might argue that the arbitrary g Schluter, D. and Price, T. (1993) Honesty, assumption that female preferences are of sexual signals facilitates perception and population divergence in sexually selected traits. Proc. R. Soc. London B a–d,h–j not subject to direct selection . differences that lead to complete sexual Biol. Sci. 253, 117–122 a Lande’s invocation of drift as a isolation when allopatric populations h Turner, G.F. and Burrows, M.T. (1995) A model possible mechanism for population meetf,l; however, it is not obvious that of sympatric speciation by sexual selection. divergence and sexual isolation seems sexual signals are any less likely to be Proc. R. Soc. London B Biol. Sci. 260, 287–292 implausible for two additional reasons. subject to direct selection than, say, the i Van Doorn, G.S. et al. (1998) Sympatric speciation and driven by First, when divergent populations meet, antigenic signals that trigger host immune environment dependent sexual selection. Proc. incomplete sexual isolation will response. R. Soc. London B Biol. Sci. 265, 1915–1919 collapseb,d. This is a common problem for The most straightforward explanation j Wu, C.I. (1985) A stochastic simulation study on schemes in which divergence is not of female preferences for extreme and speciation by sexual selection. Evolution 39, 66–82 opposed by selection. However, arbitrary male traits is that these k Gavrilets, S. (2000) Rapid evolution of reproductive isolation can persist if sexual preferences are pleiotropic side effects of reproductive barriers driven by sexual conflict. isolation is complete. A more serious alleles selected for other reasonsf,g,m, for Nature 403, 886–889 problem is that slight changes in the instance, selection for finding prey or l Parker, G.A. and Partridge, L. (1998) Sexual assumptions (e.g. a slight cost to female mates, or for mating with conspecifics. In conflict and speciation. Philos. Trans. R. Soc. London Ser. B 353, 261–274 choice) reduce the neutral line to a single this scenario, females can evolve strong m Kirkpatrick, M. and Ryan, M.J. (1991) The point, preventing divergence by drift preferences even for traits that do not exist evolution of mating preferences and the paradox alone. One must then invoke other forces in the population. This ‘sensory drive’ of the lek. Nature 350, 33–38 to produce sexual isolation. mechanism can lead naturally to n Coyne, J.A. et al. (1997) A critique of Sewall These other forces can include a cost of prezygotic isolation between isolated Wright’s shifting balance theory of evolution. Evolution 51, 643–671 g exercising preference , direct selection on populations. o Gavrilets, S. and Gravner, J. (1997) Percolation female preferences through paternal care, In all these models, it is easy to evolve on the fitness hypercube and the evolution of and indirect genetic benefits accruing to population differences that produce reproductive isolation. J. Theor. Biol. 184, 51–64

classification of evolutionary forces; however, even Allopatric speciation then the conditions required for speciation would Both pre- and postzygotic isolating mechanisms arise involve considering biogeography (e.g. SEXUAL as inevitable by-products of in SELECTION by itself is unlikely to lead to speciation in allopatry, and their evolution can be accelerated by but can easily produce species in allopatry; DIVERGENT SELECTION18 (Schluter15, this volume). The Box 1). main problem with understanding the origin of

http://tree.trends.com 332 Review TRENDS in Ecology & Evolution Vol.16 No.7 July 2001

isolating mechanisms during allopatric speciation is incipient reproductive isolation between allopatric not theoretical but empirical: in most cases, we do not populations adapting to different physical and biotic know which forms of reproductive isolation evolved habitats21,22, such evidence is especially compelling first, and which forms evolved only after other forms when parallel divergence is seen in naturally had already prevented . occurring ‘replicates’ (Schluter15, this issue). When examining a pair of species isolated by both Mathematical theory is hardly needed to understand pre- and postzygotic isolation, one often finds that such cases. Similarly, mathematical demonstrations factors acting earliest in the cycle of the that selectively driven substitutions can accelerate (e.g. sexual or isolation) restrict gene flow reproductive isolation23 simply restate the obvious more than those acting after hybridization unless they yield novel and testable predictions. (e.g. sterility and inviability). But the current There is substantial evidence that sexual selection importance of isolating mechanisms need not reflect has caused speciation. It is often noted that ADAPTIVE their importance during speciation: as taxa continue RADIATIONS, such as Hawaiian or the New to diverge over time, PREZYGOTIC ISOLATION, which acts Guinea of paradise, are accompanied by first, will always restrict current gene flow more than spectacular divergence of male SECONDARY SEXUAL does postzygotic isolation. New species of polyploid TRAITS. This does not prove that sexual selection plants, for example, originate entirely through caused speciation, for we do not know whether those postzygotic isolation (chromosomally based hybrid traits are actually involved in reproductive isolation, sterility), but can later develop prezygotic or whether they evolved before or after other isolating mechanisms that prevent the formation of hybrids. mechanisms. Better evidence comes from Moreover, reinforcement operates when pre-existing comparative studies showing a correlation between postzygotic isolation accelerates the later evolution of the species richness of taxa and various proxies for prezygotic isolation. We are almost completely the intensity of sexual selection24,25, and from species ignorant of which isolating mechanisms are involved in which traits demonstrably subject to intraspecific in the origin of species. Comparative studies of sexual selection also produce interspecific sexual allopatric taxa in different stages of evolutionary isolation26. divergence can address this problem19. Although there are many verbal and Verbal theory usually suffices to explain or mathematical theories of sexual selection27, describe the origin of reproductive isolation in understanding how this process causes behavioral allopatry. Unlike sympatric or parapatric speciation, isolation, mechanical isolation based on male in which the probability of divergence depends on the genitalia, and postmating, prezygotic isolation (such precise nature and strength of selection, the lack of as sperm–egg incompatibility) is, in principle, gene flow between allopatric populations allows them straightforward. If two geographically isolated to diverge by any evolutionary force, regardless of its lineages diverge in male traits and female nature or strength. Reproductive isolation might be preferences, they are likely to be sexually isolated functionally related to adaptive divergence (e.g. the when their ranges subsequently overlap. With sexual presence of different pollinators can lead to floral selection, as with some forms of natural selection isolation, or to different prey can cause (e.g. adaptation to a particular habitat or hybrids to be less efficient predators). Alternatively, also used as a mating site), there can be a direct and isolation might result not from the adaptive traits intuitive connection between changes within a themselves, but from the pleiotropic byproducts of lineage and reproductive isolation between lineages. those traits that can cause incompatibilities in Allopatric populations can diverge via sexual hybrids (e.g. the developmental breakdown of hybrids isolation for many reasons: for example, ecological discussed in Box 2). differences between can affect either male traits (e.g. through differential selection by predators) Prezygotic isolation and sexual selection or female preferences (e.g. different habitats that Under natural selection, prezygotic isolation can affect the ability to perceive sights or sounds), evolve directly via ecological forces acting differently different occurring in different places, and in different places (e.g. isolation via habitat or host- random GENETIC DRIFT in small populations. Some plant use). Natural selection can also cause sexual mathematical studies of allopatric SPECIATION BY isolation if selection-driven changes in are SEXUAL SELECTION28,29 are simply analyses of pure accompanied by selection to prefer the new high- anagenic change, followed by the verbal assertion fitness as potential mates. Evidence that if different lineages evolve in different directions, connecting prezygotic isolation to selection response sexual isolation will occur (Box 1). [However, comes from several sources: (1) laboratory divergent sexual selection need not cause substantial experiments showing that such isolation can be a by- reproductive isolation. For example, divergence of product of artificial selection18; (2) observations that male traits with open-ended (i.e. psychophysical30) closely related, sympatric species can be isolated by female preferences for extreme phenotypes will lead characters that clearly evolved in response to only to asymmetric isolation.] Mathematical models different local environments20; and (3) observations of of sexual selection become more important when

http://tree.trends.com Review TRENDS in Ecology & Evolution Vol.16 No.7 July 2001 333

Box 2. Evolution of intrinsic postzygotic isolation

Incompatibilities of the type postulated by simple predictions about the relative only on the number of heterozygous loci. Dobzhansky and Muller, especially those fitness of all possible hybrid genotypes by This assumption seems implausible for involving only two loci, provide a simple considering the relative effects of different postzygotic isolation, which is expected to framework for modeling both the decline genetic classes of D–M incompatibilitiesd. be mediated by specific deleterious of hybrid fitness through timea,b and the When two loci are involved, there are epistatic interactions rather than selection patterns of postzygotic isolation seen three classes of incompatibilities to against heterozygosity per se. Gavrilets among hybridsc,d. In this model, allopatric consider: between heterozygous loci, treats both the causes and consequences lineages diverge without opposition from between a heterozygous and a of sister-group divergence in a single selection, and a small fraction of hybrid homozygous (or hemizygous) , and mechanistic framework. However, there is gene combinations reduce fitness. This between homozygous loci. Both Haldane’s no empirical basis for assuming a contrasts with models, such as those rule and the large-X effect will result if connection between the intraspecific and involving ‘peak shifts’, in which the making an incompatible interspecific effects of alleles that cause divergence is directly opposed by homozygous (or hemizygous) reduces intrinsic reductions in hybrid viability and selection (for instance, under a fixed fitness significantly more than does fertilityl. regime of stabilizing selection) and a large making such an allele heterozygous at one References proportion of intraspecific substitutions of the interacting loci (Muller’s ‘dominance a Orr, H.A. (1995) The of lead to selection against hybridse,f. theory’)c,d. However, to explain Haldane’s speciation: the evolution of hybrid Orra modeled the accumulation of rule for sterility, one must also take into incompatibilities. Genetics 139, 1805–1813 Dobzhansky–Muller (D–M) account the fact that different sets of genes b Orr, H.A. and Turelli, M. The evolution of postzygotic isolation: accumulating incompatibilities by assuming that each affect fecundity in males and females, so Dobzhansky–Muller incompatibilities. substitution differentiating two taxa has a that incompatibilities causing sterility in Evolution (in press) small chance of causing fitness problems hybrids of each sex can accumulate at c Turelli, M. and Orr, H.A. (1995) The dominance in their hybrids. A fundamental prediction different ratesg. In particular, if male- theory of Haldane’s rule. Genetics 140, 389–402 d Turelli, M. and Orr, H.A. (2000) Dominance, emerges: if genetic divergence between sterilizing incompatibilities accumulate epistasis and the genetics of postzygotic taxa increases linearly with time, the faster (possibly because they are driven by isolation. Genetics 154, 1663–1679 number of incompatibilities contributing sexual selection via male–male e Barton, N.H. The role of hybridisation in to postzygotic isolation should increase at competitionh), this will tend to produce evolution. Mol. Ecol. (in press) least as fast as the square of the Haldane’s rule for sterility in male- f Taylor, C.F. and Higgs, P.G. (2000) A population genetics model for multiple quantitative traits divergence time between the two heterogametic species and, at the same exhibiting and epistasis. J. Theor. a,b groups – the ‘snowball effect’. This effect time, reduce the frequency of cases of Biol. 203, 419–438 implies that analyses of long-separated Haldane’s rule for sterility versus g Orr, H.A. (1989) Localization of genes causing species pairs will seriously overestimate inviability in female-heterogametic postzygotic isolation in two hybridizations the number of incompatibilities species. Data from male-heterogametic involving . 63, 231–237 contributing to the initial postzygotic and female-heterogametic species h Wu, C-I. et al. (1996) Haldane’s rule and its a i,j isolation . Testing this prediction will support both predictions . legacy: Why are there so many sterile males? require analyzing the genetics of Gavriletsk proposed an alternative Trends Ecol. Evol. 11, 281–284 postzygotic isolation between recently class of models that makes explicit i Presgraves, D.C. and Orr, H.A. (1998) Haldane’s diverged sister taxa, as well as between assumptions about the forces driving the rule in taxa lacking hemizygous X. Science 282, 952–954 more distantly related pairs. Such data are substitutions that cause reproductive j Turelli, M. (1998) The causes of Haldane’s rule. only beginning to emerge, and are isolation. These models explore the Science 282, 889–891 confined to the Drosophila. development of pre- and postzygotic k Gavrilets, S. (1999) A dynamical theory of By assuming that postzygotic isolation isolation within a single theoretical speciation on holey adaptive landscapes. Am. Nat. 154, 1–22 is attributable to numerous D–M framework. His model involves sexually l Coyne, J.A. et al. (2000) Is Wright’s shifting incompatibilities between loci scattered reproducing haploids, and assigns balance process important in evolution? throughout the , one can make fitnesses to (diploid) mating pairs based Evolution 53, 306–317

dealing with reinforcement or sympatric speciation. measure, such as the intensity of natural selection In such cases, the exact form of the model and its acting on female preferences. The unfortunate result biological assumptions become crucial in assessing is that well-known cases of interpopulation the likelihood of reproductive isolation7,8. divergence in male traits and female preferences are One potential role of theory, then, is to determine often compatible with virtually every theory of sexual which of the various forms of sexual selection can lead selection31. to patterns observed in nature. This task is difficult. There are many competing theories of sexual Postzygotic isolation selection that can produce reproductive isolation27, Postzygotic isolation can be characterized as either and the behavior of these theories depends crucially EXTRINSIC or INTRINSIC. In extrinsic isolation, the on many parameters whose values are difficult to relative viability and fertility of hybrids varies with

http://tree.trends.com 334 Review TRENDS in Ecology & Evolution Vol.16 No.7 July 2001

the environment in which organisms are tested, when we appreciate how many substitutions separate whereas intrinsic isolation depends on developmental lineages that remain isolated for, say, half a million problems that are relatively independent of the years – a reasonable timescale for the development of environment. These categories are not sharply appreciable postzygotic isolation4,35. With non- demarcated. For instance, even in the laboratory, synonymous substitution rates of about 10–9 bp−1 reduced viability of young hybrids might reflect year−1, after one million years, each non-synonymous higher frequency of developmental abnormalities, site will experience a substitution with probability reduced feeding efficiency of hybrid phenotypes, or about 10−3. For having roughly 104 loci, with both. Conversely, behavioral dysfunction of hybrids each locus containing approximately 103 non- can stem from either intermediate phenotypes being synonymous sites, the number of non-synonymous unable to find mates, or from reduced vigor caused by sites differing between allopatric populations after intrinsic incompatibilities19. half a million years of separation can easily be around Laboratory studies of the genetics of postzygotic 104. Thus, on average, every locus could have isolation have concentrated on intrinsic undergone about one non-synonymous substitution36. incompatibilities expressed under even ‘optimal’ The many potentially novel interactions among conditions, simply because these incompatibilities are loci that differ between species supports the view that genetically most tractable. Recently, however, field substantial postzygotic isolation can result from only studies have concentrated on how niche differences a tiny fraction of these interactions – which might still produce extrinsic ‘ecological’selection against hybrids20 number in the tens or hundreds. We know that some (Schluter15, this issue). These two broad classes of hybrid loci maintain the same basic over hundreds dysfunction differ both in their genetic basis and in the of millions of years, even after having evolved way that selection producing evolutionary divergence is different regulatory and amino acid sequences37,38. related to the selection against hybrids [selection during Thus, the breakdown of viability and fertility seen in divergence and selection against hybrids are often hybrids between lineages separated for hundreds of expected to be directly (i.e. functionally) connected in thousands rather than hundreds of millions of years extrinsic isolation but indirectly (i.e. pleiotropically) must arise either because some parts of the genome connected in intrinsic isolation]. evolve functionally much faster than do conserved Specifying the fitness of hybrid genotypes is aspects of development and , or because central to models of both intrinsic and extrinsic many slight incompatibilities (possibly including selection. Models of intrinsic selection are constrained those that result from pleiotropic effects of genes – and hence simplified – by two widespread genetic whose ‘basic’ function is conserved) have a large patterns (HALDANE’S RULE and the LARGE X EFFECT, cumulative effect. Box 2) and by empirical evidence for epistatic Making predictions about the consequences of incompatibilities that accumulate between lineages. D–M incompatibilities that arise in allopatry is By contrast, the genetics of phenotypic differences relatively simple because, to a reasonable first between taxa that contribute to extrinsic postzygotic approximation, one can ignore the evolutionary forces isolation do not generally display either driving the underlying substitutions. The relative

disproportionate X- effects or sex-specific fitness of different hybrid genotypes (F1s, effects analogous to Haldane’s rule (Orr32, this issue). backcrosses, and introgressions of particular Intrinsic postzygotic reproductive isolation arises segments) can be predicted by assuming that the because alleles that increase fitness (or are neutral) in fitness loss caused by individual D–M one genetic background can decrease fitness in hybrid incompatibilities in hybrids is far greater than are the backgrounds. The potential origin of such epistatic fitness gains produced by the individual substitutions interactions, which are expected to accumulate as a within the evolving lineages10. Similarly, the result of favorable (or, less likely, neutral) accumulation of incompatibilities can be studied with substitutions within lineages, was first described by a simple stochastic model of the substitution process Bateson, but is usually associated with Dobzhansky that assumes that each genetic difference between and Muller33. The key point is that deleterious lineages has a small chance of producing a D–M epistatic interactions [including DOBZHANSKY–MULLER incompatibility. Such models show that the number of (D–M) INCOMPATIBILITIES] in hybrids might occur D–M incompatibilities must rapidly ‘snowball’ between alleles that never appeared together in a through time, increasing at least as fast as the square single lineage. Thus, the fitness loss experienced by of divergence time (Box 2). No comparable hybrids does not imply that any ‘unfit’ alleles were simplification is possible when studying extrinsic fixed within a lineage, or that lineages passed hybrid fitness loss or many forms of prezygotic through adaptive valleys34. isolation in which the fitness of both parental species This view does not deny the existence of epistasis and their hybrids are affected by identical factors within populations, but simply postulates that (e.g. resource use or ). Similarly, the complex genomes provide many potential routes to specific evolutionary forces driving divergence adaptation that do not traverse maladaptive between lineages cannot be ignored when trying to intermediates. This idea becomes more compelling understand how gene flow inhibits the accumulation

http://tree.trends.com Review TRENDS in Ecology & Evolution Vol.16 No.7 July 2001 335 of D–M incompatibilities in parapatry or sympatry, much attention because they offer a tractable theory because the nature of selection driving substitutions with few parameters, and because some researchers determines how rapidly alleles spread within have been strong proponents of chromosomal and subpopulations and how readily they will spread past founder-effect speciation. However, the verbal and barriers to dispersal. Box 2 compares models that do mathematical theories have developed and do not assume specific mechanisms of independently. The result is that verbal theories have evolutionary divergence. invoked several factors that do not appear in the mathematical models, including general relaxation of Mechanisms driving allopatric speciation: drift versus selection during population flushes, changes of fitness selection effects of alleles in alternative genetic backgrounds, One of the first mathematical models for the evolution and modifiers of major genes. Unless verbal theories of reproductive isolation was Wright’s analysis39 of are modeled mathematically, it is hard to determine the probability of fixation of UNDERDOMINANT whether such phenomena can actually facilitate chromosome rearrangements by random drift. speciation. Although Wright claimed that his shifting He showed that the probability of such fixation is balance process was ubiquitous in adaptive ~ exp(−Ns), where s represents selection against evolution34, Templeton47,48 – the major advocate of heterozygotes, and N is the effective . drift-associated speciation – has always conceded Later analyses have shown that this is a very general that drift very rarely affects speciation. Nevertheless, result. Whenever a single population drifts through Templeton has proposed from circumstantial an adaptive valley, the rate of shifts between adaptive evidence that the process could produce a significant peaks is proportional to W2N2N, where W is the mean fraction of all morphological innovations and adaptive fitness of a population in the valley relative to the radiations in certain groups, such as the Hawaiian original peak40. Because the strength of reproductive Drosophila47. isolation depends primarily on the mean fitness of a Although the existence of adaptive radiations hybrid population, this result shows that divergence by supports the view that speciation is accelerated when drift is unlikely to yield strong reproductive isolation in a limited number of individuals colonize a new a single step41. In a sudden population bottleneck, the habitat, there is little evidence that drift is as dynamics of peak shifts are rather different: selection is important as novel selection pressures in producing negligible, and all that matters is whether the reproductive isolation. Indeed, much experimental population will drift into the domain of a new and theoretical evidence shows that isolated equilibrium. This is feasible only if there is substantial populations under intense, novel selection can segregating variation for alleles that can produce undergo rapid and substantial changes in reproductive isolation. Because it is unlikely that and mating behavior18,49–51. Conversely, populations can maintain levels of fitness variation experiments using repeated bottlenecks52 have failed associated with such epistatic variation, strong to produce substantial reproductive isolation in the reproductive isolation (especially postzygotic isolation) laboratory, and such extreme bottlenecks might not via bottlenecks seems improbable. even have occurred in some of the classic examples of These considerations suggest that if drift is to be speciation on islands40,53. The continuing responsible for reproductive isolation, there must be experimental and theoretical interest in drift-based little or no opposition from selection30,42–44. We call models of speciation is curious in view of the very such models ‘quasi-neutral’. Several such models weak empirical support44,45,52–55. (including the D–M model33) invoke a specific genetic A question separate from, but related to, the basis for reproductive isolation23,30,42–45, but all evolution of reproductive isolation is why evolution behave in a similar way. The genetic differences produces morphologically distinct species rather than causing reproductive isolation might be established a multivariate continuum of phenotypes and by random drift; because the valley is shallow or genotypes (Box 3). nonexistent, divergence can occur in large populations. Alternatively, the origin of novel Parapatric and sympatric speciation favorable mutations in specific populations is a Hybrid speciation and powerful cause of divergence that can be effective in Many botanists have argued that hybridization very large populations (N ~1/µ, where µ, the provides variation that facilitates adaptation and rate to favorable alleles, is likely to be very small – that hybrids might even evolve into new species56–59. probably ≤10−7 locus−1). However, fluctuating Such ‘hybrid speciation’ must, by necessity, begin in selection might be at least as effective as drift or sympatry. We expect recombinant hybrid genotypes mutation34,46 in producing interpopulation genetic to be, on average, less fit than individuals from the differences. The important feature of these models is parental species, simply because most have never that the absence of any substantial selective barrier been tested by natural selection. However, some allows ready divergence – and consequent hybrid genotypes might be fitter than the parentals: reproductive isolation – between populations. given some degree of gene interaction, it is unlikely Models of speciation by random drift have received that the fittest possible gene combination would occur http://tree.trends.com 336 Review TRENDS in Ecology & Evolution Vol.16 No.7 July 2001

Box 3. Why species?

As Dobzhansky noteda, ‘the living world is either sympatry or allopatry, is likely (and, reproducing organisms inevitably breaks not a single array of individuals in which given enough time, almost certain) to lead up into discrete clusters. When put into the any two variants are connected by directly or indirectly to the evolution of context of resource competition without unbroken series of intergrades, but an reproductive isolation. Such isolation can geographical isolation, this question is array of more or less distinctly separate allow both permanent coexistence closely related to understanding sympatric arrays, intermediates between which are between taxa in sympatry and future speciation. Noesth addresses this with a absent or usually rare’. The most distinct evolutionary divergence without the simple model incorporating competition arrays are species. This raises a question: impediment of gene flowf. This between phenotypes, , what properties of sexually reproducing explanation is also related to the existence outbreeding depression, genetic variability organisms and their environments of ecological niches, because divergent and an . The Allee effect, inevitably lead to the evolution of discrete adaptation to such niches could impede outbreeding depression and assortative species? In some ways, this topic is more gene flow by producing either prezygotic mating all tend to produce clusters difficult than understanding how species or extrinsic postzygotic isolation (i.e. make the sexual continuum unstable). arise because it is more abstract, (e.g. adaptation in plants to discrete soil This process is facilitated by geographical addressing not just the natural history of types or pollinators). In addition, there are isolationi,j, by the intrinsic heterogeneity of species differentiation but trying to ‘developmental niches’ that arise because the global environment and by the understand whether and why such development requires the joint action of evolutionary pressures towards differences must occur. many coadapted genes. Sufficiently specialization and range limits (which lead The clustering of organisms into groups diverged developmental systems cannot to distinct taxa in different geographic is evident not only in morphology, but also work properly within hybrids, leading to regions)k–m. in gene sequence and reproductive intrinsic hybrid sterility or inviability. References compatibility, with these factors usually Finally, sexual itself leads to a Dobzhansky, T. (1937) Genetics and the Origin of coinciding. (We do not deal with clustering the evolution of , which, in turn, Species, Columbia University Press in asexual taxa, because it is not clear creates the possibility for sexual selection. b Maynard Smith, J. and Szathmáry, E. (1995) whether they form clumps as discrete as Such selection operating in isolated The Major Transitions in Evolution, W.H. those seen in sexual speciesb–d. Various populations leads almost inevitably to Freeman & Company/Spektrum c Holman, E.W. (1987) Recognizability of sexual theoretical explanations of clumping in behavioral or gametic isolation. and asexual species of rotifers. Syst. Zool. 36, asexuals have been suggested, based on According to recent theoretical models, 381–386 distinct ecological niches, phylogenetic the ‘distinct niche’ explanation seems d Cohan, F. (2000) Genetic structure of bacterial history, and the consequences of selection necessary for the existence of species populations. In Evolutionary Genetics from Molecules to Morphology (Singh, R.S. and on organisms that rarely exchange arising sympatrically, or those arising Krimbas, C.B., eds), pp. 475–489, Cambridge e genes .) allopatrically whose later distinctness in University Press Many explanations have been offered sympatry is based on ecological or extrinsic e Majewski, J. and Cohan, F.M. (1999) Adapt for such co-incident clusters in sexually postzygotic isolation. However, although globally, act locally: The effect of selective sweeps reproducing taxa, the two most common discrete niches might be necessary to on bacterial sequence diversity. Genetics 152, 1459–1474 being the supposition that ecological explain species coexistence in sympatry, f Futuyma, D.J. (1987) On the role of species in niches are discrete, and that adaptation to they are sometimes unnecessary to explain . Am. Nat. 130, 465–473 the environment (or simple divergence by species distinctness. Full species can arise g Hopf, F.A. and Hopf, F.W. (1985) The role of the drift) inevitably produces the reproductive in allopatry via non-ecological processes Allee Effect in species packing. Theor. Popul. Biol. isolation necessary to keep species causing intrinsic postzygotic isolation 27, 27–50 h Noest, A.J. (1997) Instability of the sexual discrete in sympatryb. The ecological (e.g. polyploidy or antagonistic coevolution continuum. Proc. R. Soc. London B Biol. Sci. 264, explanation sees clusters as resulting from of males and females). Such species, 1389–1393 intrinsic discreteness in ways of exploiting although completely genetically isolated, i Higgs, P.G. and Derrida, B. (1992) Genetic physical resources (e.g. the mechanisms might not be able to coexist in sympatry. distance and species formation in evolving by which microbes use alternative carbon The few mathematical analyses of why populations. J. Mol. Evol. 35, 454–465 j Gavrilets, S. et al. (1998) Rapid parapatric sources or capture energy might impose might produce speciation on holey adaptive landscapes. Proc. R. distinct phenotypic ‘solutions’ in the same morphological clusters have dealt largely Soc. London B Biol. Sci. 265, 1483–1489 way that different jaw morphologies are with the evolution of clusters in sympatry. k Kawecki, T.J. et al. (1997) Mutational collapse of needed to efficiently handle different prey). Hopf and Hopf g showed that, given the pre- fitness in marginal habitats and the evolution of ecological specialisation. J. Evol. Biol. 10, This effect cascades as clumps at one existence of reproductive isolation 407–430 foster clumps at higher levels. between species, ‘Allee effects’ l Whitlock, M.C. (1996). The Red Queen beats the The ecological explanation also rests on (specifically, reduced fitness of rare jack-of-all-trades: the limitations on the evolution the inevitability of tradeoffs that create phenotypes caused by difficulty in finding a of and niche width. Am. disruptive selection, with hybrids who fall mate) will produce distinct clusters on a Nat. 148, S65–S77 m Case, T.J. and Taper, M.L. (2000) Interspecific ‘between niches’ being less fit. resource gradient. However, their paper competition, environmental gradients, gene flow, The sexual-isolation explanation relies does not address the question of whether and the coevolution of species borders. Am. Nat. on the fact that , in an initial continuum of sexually 155, 583–605

http://tree.trends.com Review TRENDS in Ecology & Evolution Vol.16 No.7 July 2001 337 in the lineage of either parental species. Yet, without a because rearrangements are unlikely to be specific model of selection, it is impossible to predict established if they cause enough sterility to the distribution of hybrid fitnesses or the potential significantly reduce gene flow. These models are gain in fitness that might be realized by selection on based on chromosomal rearrangements, but the same hybrids. A simple but highly idealized possibility is considerations apply to genic incompatibilities. Orr’s elaboration of Fisher’s model of The time taken to fix a hybrid can be multidimensional selection60. Hybridization between estimated from the extent of recombination; in populations that accumulate different favorable sunflowers, it is roughly 20–60 generations, assuming alleles can yield small fitness gains as a result of both a single isolated population69. This scenario can be introgression of single genes and fixation of seen as a special kind of founder-effect speciation recombinant genotypes61. However, given the model- triggered when a novel combination of alleles, dependence of predictions about this process, the produced by hybridization, allows a new adaptive potential contribution of hybridization to adaptation peak to be reached. Similarly, a novel highly fit and speciation is best treated as an empirical recombinant genotype can also be established at high question. frequency in a narrow , when clines at The key difficulties for hybrid speciation are different loci are out of phase, producing different singling out a particularly fit recombinant genotype combinations of alleles at different points in the after the parental species hybridize, and then keeping hybrid zone70,71. However, further divergence is this genotype intact. The simplest solution is required if a new species is to form: for example, immediate reproductive isolation of the genotype, reinforcement of prezygotic isolation between the which can be attained through , recombinant and parents, or ecological divergence selfing, or allopolyploidy. In flowering plants, that allows the recombinant to spread into a new, polyploidy (usually, allopolyploidy) is a frequent mode widely distributed niche. The immediate effect of an of speciation (2–4% of speciation events62); similarly, increased fitness of the hybrid population is to many parthenogenetic animal species are derived weaken reproductive isolation between the parental from interspecific hybrids62. The novel phenotypes populations70. attendant on combining two disparate genomes might also provide sufficient ecological divergence to allow Parapatric speciation polyploids to coexist with their ancestors63. There has been much argument about whether In contrast to polyploidy, diploid hybrid speciation speciation can occur between adjacent populations (the origination of a new species through selection occupying a broadly continuous habitat – that is, acting on a diploid interspecific hybrid) is rare64. populations in parapatry. Darwin’s observations in Because most putative diploid hybrid species , together with his gradualist view of outcross65, the question arises of how such species are evolution, convinced him that new species could established despite the possibility of backcrossing to evolve in this way, an assertion that later led to much the parental species. One explanation is that mixed dispute72. Population genetics, however, supports pollen stimulates selfing of the new recombinant, but Darwin’s view. Given a sufficiently broad the single existing study did not show this effect65. geographical range, any mechanism that can produce Alternatively, divergence in a new , or divergence among allopatric populations can also colonization of a novel habitat, might lead to cause divergence in parapatry. Indeed, even if most parapatric or allopatric isolation after initial genetic divergence occurs in allopatry, the diverging hybridization. Indeed, diploid hybrid species are populations are likely to be spatially extended, and usually associated with novel habitats66. each genetic difference that contributes to isolation is Even with random mating, fit recombinants can be likely to originate in a local region within a broader established by selection, and this might lead to some range, either as a single mutation or as a localized reproductive isolation. The simplest case is a single adaptation. If different alleles arise in different population segregating for several underdominant places, and if these prove incompatible with each chromosome rearrangements58,67,68. Under weak other when they meet, then they will contribute to the selection, any homozygous combination of build up of reproductive isolation in parapatry – just chromosome arrangements is equally likely to be as they would in allopatry. fixed and will show a degree of postzygotic isolation The ‘isolation by distance’ necessary for parapatric from the polymorphic parental species. With stronger speciation depends on the strength of selection acting selection, LINKAGE DISEQUILIBRIA tend to favor recovery during population divergence. If strong selection is of parental combinations. In their analysis of this involved (either causing adaptation to local model, Buerkle et al.68 show that, even with multiple conditions, or maintaining alternative adaptive rearrangements, gene flow between populations peaks), then divergence of traits leading to remains high at NEUTRAL LOCI. Although new adaptive reproductive isolation can occur over small spatial gene combinations can be established, especially if scales ( σ / 2 s , where σ measures average favored in a novel environment, there is little per-generation dispersal distance and s measures immediate reproductive isolation. This is simply selection); this is true whether divergence is driven by http://tree.trends.com 338 Review TRENDS in Ecology & Evolution Vol.16 No.7 July 2001

drift or by selection73. If divergence is ‘quasi-neutral’, probable if substantial premating and/or postmating then larger spatial scales might be involved, although divergence have previously evolved7. However, the they might still be small compared with the range of effects on reinforcement of alternative population the species. Parapatric divergence of reproductive structures, forms of selection against hybrids and compatibility seems most difficult when it is based on mechanisms of nonrandom mating are difficult to alleles that are favorable everywhere against the disentangle. This is partly because many theoretical ancestral genetic background but incompatible with studies explore unique combinations of assumptions, each other. Such divergence requires that two or more making it difficult to sort out which factors are alleles be established in different parts of the range important. For example, Cain et al.81 claim to have before any one allele has spread over the whole demonstrated that reinforcement occurs more readily range74. in a MOSAIC HYBRID ZONE, in which fitnesses vary The existence of narrow clines and hybrid zones spatially across a patchy environment, than in a demonstrates that selection can dominate gene flow ‘TENSION ZONE’, in which two equally fit genotypes meet over quite small scales, allowing parapatric and produce less-fit hybrids. This claim is based on a divergence. Reproductive isolation might arise as a comparison of their numerical results with those of a pleiotropic by-product of locally selected alleles, just model by Sanderson82. However, Cain et al.’s model as in allopatric speciation. However, gene flow can confounds the role of population structure (single alter the course of parapatric divergence in two ways. zone of contact versus many zones) with that of the First, alleles that are favorable everywhere can form of selection (hybrid dysfunction versus spatially readily spread across hybrid zones75, slowing varying selection on parentals). Moreover, the divergence and producing confusing genealogies. interaction between and Such single-gene introgression might be much more ASSORTATIVE MATING in the model of Cain et al. provides widespread than is usually appreciated. Examples an indirect force favoring the evolution of assortment, are known in sunflowers76 and Drosophila77, and even if there is no hybrid dysfunction8 (Box 4). seem especially common for cytoplasmically inherited Theory has helped us to better understand the genomes, which frequently show genealogies that factors that can inhibit or prevent reinforcement, differ from those of nuclear loci66,78. Nevertheless, including: (1) lack of phenotypic differentiation on such introgression might not prevent continuing which mating discrimination can act83; (2) direct divergence between populations. Second, gene flow forces opposing the spread of ‘reinforcement’ alleles, might lead to increased prezygotic isolation through such as migration from populations lacking the reinforcement. alleles84 or natural selection acting directly to oppose their spread85; (3) restriction of hybrids to a narrow Reinforcement portion of the species range7,82; and (4) lack of Noor79 provides an excellent review of the empirical sufficient ecological and/or mating differentiation to literature on reinforcement. As he emphasizes, there allow the taxa to coexist long enough for is no longer any question that the pattern of reinforcement to occur7. Factors (3) and (4) illustrate ‘REPRODUCTIVE ’ – greater the difficulty of reaching unambiguous conclusions interspecific mate discrimination between sympatric about which factors promote or inhibit reinforcement. than between allopatric populations – is common. The If divergent populations compete and mate across a question is whether this pattern is driven, as wide region, the rarer might go extinct (despite Dobzhansky claimed, by reinforcement, or by other pervasive selection favoring reinforcement); yet, if mechanisms that do not involve direct selection they meet in a narrow hybrid zone, gene flow and lack against hybridization80. Following Coyne and Orr’s4 of can hinder reinforcement, even compilation of Drosophila data, many theoretical though each has a ‘refuge’ that prevents its analyses, both numerical and analytical, have extinction in the region of sympatry7. Theoretical demonstrated that reinforcement can occur under conclusions about such complex interactions are biologically plausible conditions (Box 4). Although likely to be model dependent. limited biogeographical data5 suggest that Although the pattern of reproductive character reinforcement is not as common a route to speciation displacement has generally been interpreted as an as is strictly allopatric divergence, reinforcement has evolutionary response to spawned a much richer theoretical literature because (reinforcement), the same pattern can emerge even if its plausibility is less obvious, the process can be there is no initial stage of allopatry, but there is affected by many factors and it is appealing to believe continual gene flow across an environmental that natural selection can play a direct role in discontinuity80,86. In general, future theoretical creating new species. Sorting out the relative analyses should aim at producing empirical importance of different biological factors remains a predictions that can distinguish alternative major theoretical challenge. evolutionary forces producing reproductive character From the diverse studies described in Box 4, we displacement83,85. For instance, Kelly and Noor85 conclude that the evolution of reinforcement is show that reinforcement occurring with different theoretically plausible and becomes even more types of change in female mate-choice criteria can

http://tree.trends.com Review TRENDS in Ecology & Evolution Vol.16 No.7 July 2001 339

Box 4. Models of reinforcement

Over the past decade, at least seven assumptions of previous analyses, making incipient speciesc. This eliminates the theoretical studiesa–g have supported the it difficult to know which individual problem of selection opposing the spread plausibility of reinforcement. These differ assumptions are crucial in determining the of ‘reinforcement’ alleles outside areas appreciably in assumptions about: outcomes. Furthermore, the effect of each where hybridization occursh. However, (1) genetics (one locus per characterc,e,f assumption might vary depending on this model seems biologically versus polygenic charactersa,d,g); other assumptionsb,f, making a succinct unreasonable unless ecological (2) population structures (a single mixed summary of the results of alternative divergence has already proceeded far populationa,b,e versus narrow contact studies impossible. For instance, Kelly and enough to allow wide sympatry. Finally, zonea,e,h versus two islandsc,f versus Noorb explore several alternative genetic explicit consideration of population island–continentd,g); (3) the nature of assumptions about the nature of dynamics, so that extinction is possible selection against hybrids (e.g. all hybrids postzygotic isolation, and find that a through maladaptive hybridization, seems equally unfita versus spatially varying combination of factors can promote to restrict the conditions for selection favoring specific phenotypesc,e; reinforcement even though numerical reinforcementa,c. Population structure can versus sterilitya); (4) the explorations suggest that each individual alter this conclusiona,h. If hybridization can potential basis for reinforcement factor inhibits reinforcement. produce , partial allopatry (assortative matinge,f,g versus preference One factor that affects the likelihood of can aida (rather than inhibith) and trait coevolutiona,c,d versus increasing reinforcement is the genetic basis of reinforcement by providing a ‘refuge’ from discriminationb); (5) whether the prezygotic isolation. Felsensteini argued global extinction. ‘reinforcement’ alleles are under direct that speciation is generally facilitated if a In summary, theoretical studies show natural selection (yesb,e versus noa,c,d,f,g); single allele (or, more generally, a specific that reinforcement is clearly plausible, and (6) density regulation of populationsa trait value) causes assortative mating in which is reassuring given the growing (specifically, whether lineages can go each INCIPIENT SPECIES (see Glossary; evidence for its occurrence. However, the extinct). i.e. allele A causes its carrier to mate with available studies provide few unambiguous Because of the analytical complexities phenotypically similar individuals). In predictions about when it is most probable. associated with strong selection acting on contrast, speciation is opposed by References multiple loci in spatially distributed recombination if hybridization is reduced a Liou, L.W. and Price, T.D. (1994) Speciation by populations, most analyses are based on only when sets of co-adapted genes reinforcement of premating isolation. Evolution numerical calculations. By contrast, become associated with different alleles 48, 1451–1459 Kirkpatrick and Servediod,g produced (or traits). However, even this conclusion b Kelly, J.K. and Noor, M.A.F. (1996) Speciation by analytical, weak-selection results for might depend on population structuref. In reinforcement: a model derived from studies of Drosophila. Genetics 143, 1485–1497 polygenic models by allowing only small general, non-random mating can produce c Servedio, M.R. and Kirkpatrick, M. (1997) The changes in mating patterns, assuming that associations between mating-system effects of gene flow on reinforcement. Evolution selection against hybrids is weak, and alleles and alleles favored by local 51, 1764–1772 positing limited one-way migration from a selection that either promote or inhibit d Kirkpatrick, M. and Servedio, M.R. (1999) The ‘continent’ to an ‘island’ population. They reinforcementf. These indirect effects reinforcement of mating preferences on an island. Genetics 151, 865–884 note, however, that their weak-selection produced by are e Cain, M.L. et al. (1999) Reinforcing selection is predictions might not apply to more analogous to Fisher’s runaway effect for effective under a relatively broad set of conditions realistic levels of migration or selection or sexual selection, as both are driven by in a mosaic hybrid zone. Evolution 53, 1343–1353 to the spread of alleles that induce large associations between male traits and f Servedio, M.R. (2000) Reinforcement and the genetics of nonrandom mating. Evolution 54, changes in mating discriminationf,g. Most female preferencesa,f. 21–29 other investigations have the typical Population structure also affects the g Kirkpatrick, M. (2000) Reinforcement and weaknesses of numerical studies: only a likelihood of reinforcement. divergence under assortative mating. Proc. R. limited range of parameters can be Reinforcement seems to occur more Soc. London B Biol. Sci. 267, 1649–1655 explored and the results can be difficult to readily when there is migration between h Sanderson, N. (1989) Can gene flow prevent reinforcement? Evolution 43, 1123–1235 interpret. These difficulties are two island populations, corresponding i Felsenstein, J. (1981) Skepticism towards Santa compounded because alternative studies effectively to secondary contact occurring Rosalia, or why are there so few kinds of often modify several biological throughout the entire range of each animals. Evolution 35, 124–138

lead to qualitatively different outcomes of models (Via14, this issue), motivated by the interpopulation mate-choice experiments. observation of host RACES in the tephritid fly Rhagoletis pomonella89, involved ecologically driven Sympatric speciation reproductive isolation associated with adaptation to The emergence of strong empirical support for alternative discrete resources and/or habitats. A sympatric speciation in a study of Cameroon crater- novel feature of more recent analyses is their lake cichlids11 has generated a spate of models emphasis on speciation driven by competition for supporting the theoretical plausibility of sympatric continuously distributed resources (e.g. a continuum speciation12,13,87,88. Many earlier sympatric speciation of prey-size categories versus alternative discrete

http://tree.trends.com 340 Review TRENDS in Ecology & Evolution Vol.16 No.7 July 2001

Box 5. Models of sympatric speciation

Sympatric speciation is driven by habitat and alleles favoring behavioral In contrast to this model, which allows disruptive selection. Recent theoretical selection of that habitat. This provides the the intensity of assortative mating to research has concentrated on three indirect force that drives the evolution of evolve, Kondrashov and Kondrashova classes of models: ecological models habitat isolation and hence speciation. assume a fixed assortment rule, based on based on competition for resourcesa–c; These models, which produce habitat either the ecological character or a neutral sexual-selection models based on races more easily than do models marker. Their analysis incorporates a highly competition for matesd,e; and models of requiring balanced polymorphisms, can in symmetric polygenic model (loci have habitat-race formation based on habitat- principle be tested by determining the equal effects and all alleles that increase a specific deleterious or beneficial allelesf,g. prevalence of alleles whose effects are trait value have equal frequencies across Of these, the sexual selection models are limited to specific habitats. these loci) and artificially imposes on the least convincing. Instead of invoking Given the current interest in the ecological trait a fixed level of disruptive disruptive natural selection, they invoke ecology of speciation, most attention has selection that does not weaken as the disruptive sexual selection with different been given to models in which incipient species diverge. (When disruptive females preferring different extreme male competition for resources causes selection is based on niche overlap, trait phenotypes. This assumption might be disruptive natural selection. The important divergence should often lead to reduced plausible on its own, but the published lesson from these models is that competition and weaker disruptive analyses suggest that additional (and sympatric speciation appears plausible selection.) Drossel and McKanec treat a implausible) assumptions are required for even when it requires the evolution of simplified version of the Dieckmann and such selection to sunder a population into genetic associations between ecologically Doebeli model analytically, but they use reproductively isolated groups rather than important traits and ‘neutral markers’ that several approximations whose validity to simply move the selected trait to one organisms use to discriminate among cannot be assessed without explicit genetic extreme phenotype throughout the entire potential mates. The most convincing simulations. species. For instance, the model of Higashi study is that of Dieckmann and Doebelib, in et al.d yields sympatric speciation using which disruptive selection emerges from References a Kondrashov, A.S. and Kondrashov, F.A. (1999) artificially symmetrical initial conditions an explicit model of competition for a Interactions among quantitative traits in the (corresponding to females showing no continuum of resources. With a unimodal course of sympatric speciation. Nature 400, preference on average for alternative distribution of resources, disruptive 351–354 values of the male trait) and invoking selection results when the width of the b Dieckmann, U. and Doebeli, M. (1999) by sympatric speciation. Nature abrupt changes in parameter values resource distribution exceeds the average 400, 354–357 governing the intensity of stabilizing range of resources used by individuals. c Drossel, B. and McKane, A. (2000) Competitive selection on the male trait or the female For asexual haploids, this condition speciation in quantitative genetic models. preference. Their simulations suggest that suffices to produce a bimodal distribution J. Theor. Biol. 204, 467–478 relaxing this initial symmetry prevents of phenotypes. For sexuals, roughly the d Higashi, M. et al. (1999) Sympatric speciation by sexual selection. Nature 402, 523–526 speciation. same condition suffices to drive the e Turner, G.F. and Burrows, M.T. (1995) A model f,g Kawecki presents models of evolution of assortative mating based of sympatric speciation by sexual selection. sympatric habitat- (or host-) race directly on the ecologically important Proc. R. Soc. London B Biol. Sci. 260, 287–292 formation driven by the accumulation of character, again leading to speciation. f Kawecki, T.J. (1997) Sympatric speciation by mutations with habitat-specific beneficial When mating is based instead on a neutral habitat specialization driven by deleterious mutations. Evolution 51, 1751–1763 or deleterious effects. By assuming mating marker, the resource distribution must be g Kawecki, T.J. (1996) Sympatric speciation within habitats, linkage disequilibrium significantly wider (for a fixed level of driven by beneficial mutations. Proc. R. Soc. develops between alleles beneficial in a competition) for speciation to occur. London B Biol. Sci. 263, 1515–1520

prey types) and/or by sexual selection acting on a earlier models that focused on assortative mating as a continuum of phenotypes. The central driving force in by-product of niche differences (Via14, this issue). these newer models is selection against intermediate The idea that disruptive selection can drive phenotypes. Such intermediates are deleterious sympatric speciation was championed by Darwin in because they accrue fewer resources as a result of Chapter 4 of the Origin of Species. However, he used density- and frequency-dependent selection or an explicitly phenotypic definition of species and because they procure fewer mates as a result of focused only on the origin of distinguishable varieties preferences for extreme phenotypes90. The resulting under disruptive selection. Unlike recent models of disruptive selection can produce bimodal sympatric speciation, Darwin’s model failed to distributions of phenotypes for asexual taxa13; it can recognize the key role of reproductive isolation also drive the evolution of reproductive isolation for (usually via assortative mating) in maintaining sexual taxa in sympatry. Box 5 discusses some recent sympatric species; alone does models of sympatric speciation. The available data not necessarily impede gene flow between taxa. More provide no basis for determining whether these new convincing arguments for speciation in the face of models have wider biological applicability than do (parapatric) gene flow were provided by Fisher91

http://tree.trends.com Review TRENDS in Ecology & Evolution Vol.16 No.7 July 2001 341

(pp. 125–129), who did recognize the importance of must be paid to models in which reproductive assortative mating, and proposed that selection could divergence occurs despite continual but limited gene plausibly achieve it through habitat selection or mate flow. Existing models of sympatric speciation (Box 5) discrimination. and the evolution of range limits95 can be easily The quantitative predictions of the recent adapted to study parapatric speciation. We would like mathematical treatments are certainly sensitive to to understand how much gene flow is needed to their genetic and ecological assumptions, but the inhibit the accumulation of D–M incompatibilities (in qualitative conclusion that competition-driven particular, under what conditions will sympatric speciation can occur seems secure. interpopulation hybrids have lower fitness in a widely Superimposing disruptive sexual selection on distributed organism with limited dispersal?). Most selection mediated by competition for resources is recent theoretical work on speciation consists of likely to facilitate the process92, but it is not clear numerical analyses demonstrating the feasibility of whether disruptive sexual selection alone will some narrowly defined speciation scenario, often produce new species in sympatry (Box 5). Models of without sufficient exploration of the robustness of the reinforcement tell us that conditions for sympatric conclusions or their relation to alternative published speciation are likely to depend on the criteria for non- analyses (Boxes 4,5). To gain general insights, we random mating (e.g. assortative mating between need to explore the biologically plausible regions of males and females of similar phenotype versus parameter space more carefully. This demands more matching male traits to female preferences83). analytical treatments and fewer simulations. tells us that the details of When simple models seem reasonable, such as the competition models can significantly alter their evolution of hybrid inviability or sterility via D–M quantitative predictions93. Hence, although incompatibilities, one can derive testable quantitative analytical treatments would be preferable to the predictions9,10. However, it might be unrealistic to numerical simulations that have dominated studies expect quantitative predictions to emerge from of sympatric speciation, predictions must surely complex models of sympatric speciation and depend on empirically elusive genetic, behavioral reinforcement that involve ecology, behavior and and ecological details. The challenge will be to multilocus genetic interactions. Nevertheless, we can uncover robust predictions that go beyond intuitive expect testable qualitative predictions. For instance, conjectures, such as the frequent claim that we expect that reinforcement and sympatric ecological sympatric speciation is most likely when speciation would be facilitated more by assortative competitor-free, resource-rich habitats are invaded. mating based on characters subject to divergent Until then, theory will take a back seat to empirical than by assortative mating based studies in revealing and understanding sympatric on secondary sexual characters13,18. Similarly, models speciation. of sympatric speciation mediated by disruptive selection lead to the prediction that speciation driven Conclusions and prospects by ‘extrinsic’ selection against hybrids is likely to occur Because it often includes sophisticated mathematics, faster than does largely allopatric speciation non-verbal theory has a special air of authority associated with the gradual accumulation of D–M Acknowledgements among . Many experimentalists are incompatibilities. In general, we expect that rapid We thank D. Bolnick, unequipped to judge the limitations or weakness of speciation, e.g. 104 years or faster, is more likely to be B. Fitzpatrick, S. Gavrilets, R. Haygood, C.D. Jones, such theory, which has occasionally been misleading. associated with sexual selection and extrinsic M. Kirkpatrick, For example, the results of Spencer et al.94 convinced selection against hybrids than with intrinsic selection A. Kondrashov, many that reinforcement was nearly impossible to against hybrids. J.B. Mullet, S.V. Nuzhdin, H.A. Orr, T.D. Price, obtain, even under optimal conditions. Renewed As well as providing testable predictions, T. Prout, D.W. Schemske, empirical interest emerged only when new data mathematical theory can play at least two other roles D. Schluter, M.R. Servedio indicated that reinforcement was plausible4, and in understanding speciation. First, it can help and P.S. Ward for when subsequent theoretical treatment – using organize verbal theories, showing how they are discussion and comments. Some of these assumptions different from those of Spencer et al. – interrelated and focusing research on the relative reviewers disagree with supported this plausibility7. With theory, as with importance of different mechanisms (e.g. natural our conclusions. This architecture, God is in the details, and the conclusions versus sexual selection, extrinsic versus intrinsic work was supported by of theories about speciation might be highly sensitive selection against hybrids, divergence in allopatry US National Science Foundation grants DEB to only slight changes in their assumptions. versus parapatry). Second, they can suggest new 9527808 and DEB 0089716 Because of this, we believe that significant ways to think about evolution in the high- to MT, grants from the understanding of speciation is more likely to be dimensional spaces defined by multilocus genotypes Darwin Trust of Edinburgh and the produced by new data than by new theory. and multivariate phenotypes (Box 3). Such theory and Biological Sciences Nevertheless, our review suggests several lines of need not produce empirical predictions, nor lead Research Council useful theoretical research. We need convincing directly to new observations or experiments. Two such (GRJ/76057, GR/H/09928) general models of sympatric speciation that delimit cases are the influential papers on speciation of to NHB, and National 96 97 Institutes of Health grant testable or observable conditions under which it can Maynard Smith and Felsenstein , which clarified R01 GM58260 to JAC. and cannot occur. Similarly, additional attention central biological issues about the origin of species.

http://tree.trends.com 342 Review TRENDS in Ecology & Evolution Vol.16 No.7 July 2001

References 26 Wiernasz, D.C. and Kingsolver, J.G. (1992) Wing 51 Schluter, D. (2000) The Ecology of Adaptive 1 Dobzhansky, T. (1936) Studies on hybrid sterility. melanin pattern mediates species recognition in Radiation, Oxford University Press II. Localization of sterility factors in Drosophila Pieris occidentalis. Anim. Behav. 43, 89–94 52 Charlesworth, B. (1995) Down the bottleneck? Curr. pseudoobscura hybrids. Genetics 21, 113–135 27 Andersson, M. (1994) Sexual Selection, Princeton Biol. 5, 995–996 2 Mayr, E. (1954) Change of genetic environment and University Press 53 Vincek, V. et al. (1997) How large was the founding evolution. In Evolution as a Process (Huxley, J. et al., 28 Pomiankowski, A. and Iwasa, Y. (1998) Runaway population of Darwin’s ? Proc. R. Soc. London eds), pp. 157–180, Allen and Unwin Press ornament diversity caused by Fisherian sexual B Biol. Sci. 264, 111–118 3 Coyne, J.A. (1994) and the origin of selection. Proc. Natl. Acad. Sci. U. S. A. 96, 54 Rundle, H.D. et al. (1999) Experimental tests of species. Evolution 48, 19–30 5106–5111 founder-flush: a reply to Templeton. Evolution 53, 4 Coyne, J.A. and Orr, H.A. (1989) Patterns of 29 Gavrilets, S. (2000) Rapid evolution of reproductive 1632–1633 speciation in Drosophila. Evolution 43, 362–381 barriers driven by sexual conflict. Nature 403, 55 Nagl, S. et al. (1998) Persistence of neutral 5 Barraclough, T.G. and Vogler, A.P. (2000) Detecting 886–889 polymorphisms in fish. Proc. the geographical pattern of speciation from species- 30 Lande, R. (1981) Models of speciation by sexual Natl. Acad. Sci. U. S. A. 95, 14238–14243 level phylogenies. Am. Nat. 155, 419–434 selection on polygenic traits. Proc. Natl. Acad. Sci. 56 Anderson, E. (1949) Introgressive Hybridization, 6 Coyne, J.A. and Price, T.D. (2000) Little evidence for U. S. A. 78, 3721–3725 Chapman & Hall sympatric speciation in island birds. Evolution 54, 31 Endler, J.A. and Houde, A.E. (1995) Geographic 57 Stebbins, G.L. (1950) Variation and Evolution in 2166–2171 variation in female preferences for male traits in Plants, Columbia University Press 7 Liou, L.W. and Price, T.D. (1994) Speciation by Poecilia reticulata. Evolution 49, 456–468 58 Grant, V. (1958) The regulation of recombination in reinforcement of premating isolation. Evolution 48, 32 Orr, H.A. (2001) The genetics of species differences. plants. Cold Spring Harbor Symp. Quant. Biol. 23, 1451–1459 Trends Ecol. Evol. 16, 343–350 337–363 8 Servedio, M.R. (2000) Reinforcement and the 33 Orr, H.A. (1996) Dobzhansky, Bateson, and the 59 Arnold, M. (1996) Natural Hybridization and genetics of nonrandom mating. Evolution 54, 21–29 genetics of speciation. Genetics 144, 1331–1335 Introgression, Princeton University Press 9 Orr, H.A. (1995) The population genetics of 34 Coyne, J.A. et al. (2000) Is Wright’s shifting balance 60 Orr, H.A. (1998) The population genetics of speciation: the evolution of hybrid incompatibilities. process important in evolution? Evolution 53, adaptation: The distribution of factors fixed during Genetics 139, 1805–1813 306–317 adaptive evolution. Evolution 52, 935–949 10 Turelli, M. and Orr, H.A. (2000) Dominance, 35 Sasa, M.M. et al. (1998) Patterns of postzygotic 61 Barton, N.H. The role of hybridisation in evolution. epistasis and the genetics of postzygotic isolation. isolation in . Evolution 52, 1811–1820 Mol. Ecol. (in press) Genetics 154, 1663–1679 36 Orr, H.A. and Turelli, M. The evolution of postzygotic 62 Otto, S.P. and Whitton, J. (2000) Polyploid incidence 11 Schliewen, U.K. et al. (1994) Sympatric speciation isolation: accumulating Dobzhansky–Muller and evolution. Annu. Rev. Genet. 34, 401–437 suggested by of crater lake . incompatibilities. Evolution (in press) 63 Bretagnolle, F. and Lumaret, R. (1995) Bilateral Nature 368, 629–632 37 Chervitz, S.A. et al. (1998) Comparison of the polyploidization in Dactylis glomerata L. subsp. 12 Kondrashov, A.S. and Kondrashov, F.A. (1999) complete protein sets of worm and yeast: Orthology lusitanica: occurrence, morphological and genetic Interactions among quantitative traits in the course and divergence. Science 282, 2022–2028 characteristics of first polyploids. Euphytica 84, of sympatric speciation. Nature 400, 351–354 38 Neumann, C.J. and Nuesslein-Volhard, C. (2000) 197–207 13 Dieckmann, U. and Doebeli, M. (1999) On the origin Patterning of the zebrafish by Sonic Hedgehog 64 Rieseberg, L.H. (1997) Hybrid origins of plant of species by sympatric speciation. Nature 400, activity. Science 289, 2137–2139 species. Annu. Rev. Ecol. Syst. 28, 359–389 354–357 39 Wright, S. (1941) On the probability of fixation of 65 Rieseberg, L.H. et al. (1998) Patterns of mating in 14 Via, S. (2001) Sympatric speciation in animals: the reciprocal translocations. Am. Nat. 75, 513–522 wild sunflower hybrid zones. Evolution 52, 713–726 ugly duckling grows up Trends Ecol. Evol. 16, 40 Barton, N.H. and Charlesworth, B. (1984) Genetic 66 Rieseberg, L.H. (1995) The role of hybridization in 381–390 revolutions, founder effects, and speciation. Annu. evolution: old wine in new skins. Am. J. Bot. 82, 15 Schluter, D. (2001) Ecology and the origin of species. Rev. Ecol. Syst. 15, 133–164 944–953 Trends Ecol. Evol. 16, 372–380 41 Walsh, J.B. (1982) Rate of accumulation of 67 McCarthy, E.M. et al. (1995) A theoretical 16 Barraclough, T. and Nee, S. (2001) reproductive isolation by chromosome assessment of recombinational speciation. Heredity and speciation. Trends Ecol. Evol. 16, 391–399 rearrangements. Am. Nat. 120, 510–532 74, 502–509 17 Diamond, J.M. (1977) Continental and insular 42 Bickham, R.J. and Baker, J.W. (1986) Speciation by 68 Buerkle, C.A. et al. (2000) The likelihood of speciation in Pacific land birds. Syst. Zool. 26, monobrachial centric fusions. Proc. Natl. Acad. Sci. homoploid hybrid speciation. Heredity 84, 441–451 263–268 U. S. A. 83, 8245–8248 69 Ungerer, M.C. et al. (1998) Rapid hybrid speciation 18 Rice, W.R. and Hostert, E.E. (1993) Laboratory 43 Barton, N.H. (1989) The divergence of a polygenic in wild sunflowers. Proc. Natl. Acad. Sci. U. S. A. 95, experiments on speciation: what have we learned in system under stabilising selection, mutation and 11757–11762 40 years? Evolution 47, 1637–1653 drift. Genet. Res. 54, 59–77 70 Virdee, S.R. and Hewitt, G.M. (1994) Clines for 19 Coyne, J.A. and Orr, H.A. (1998) The evolutionary 44 Barton, N.H. (1996) Natural selection and random hybrid dysfunction in a grasshopper hybrid zone. genetics of speciation. Philos. Trans. R. Soc. London genetic drift as causes of evolution on islands. Philos. Evolution 48, 392–407 Ser. B 353, 287–305 Trans. R. Soc. London Ser. B 351, 785–795 71 Hatfield, T. et al. (1992) A model of a hybrid zone 20 Schemske, D.W. and Bradshaw, H.D. (1999) 45 Gavrilets, S. and Boake, C.R.B. (1998) On the between two chromosomal races of the common Pollinator preference and the evolution of floral evolution of premating isolation after a founder shrew (Sorex araneus). Evolution 46, 1129–1145 traits in monkeyflowers (Mimulus). Proc. Natl. event. Am. Nat. 152, 706–716 72 Mayr, E. (1982) The Growth of Biological Thought: Acad. Sci. U. S. A. 96, 11910–11915 46 Whitlock, M. (1997) Founder effects and peak shifts Diversity, Evolution and Inheritance, Belknap Press 21 Clausen, J. et al. (1940) Experimental Studies on the without genetic drift: Adaptive peak shifts occur 73 Slatkin, M. (1973) Gene flow and selection in a cline. Nature of Species. I. Effect of Varied Environments easily when environments fluctuate slightly. Genetics 75, 733–756 on Western North America Plants, Carnegie Evolution 51, 1044–1048 74 Barton, N.H. (2000) Genetic hitch-hiking. Philos. Institute of Washington Publ. No. 520 47 Templeton, A. (1989) The relation between Trans. R. Soc. London Ser. B 355, 1553–1562 22 Johnson, S.D. (1997) Pollination of speciation mechanisms and macroevolutionary 75 Barton, N.H. and Bengtsson, B.O. (1986) The Satyrium hallackii (Orchidaceae) in South Africa. patterns. In Evolutionary Processes and Theory barrier to genetic exchange between hybridising Bot. J. Linn. Soc. 123, 225–235 (Karlin, S. and Nevo, E., eds), pp. 497–512, populations. Heredity 57, 357–376 23 Gavrilets, S. (1999) A dynamical theory of speciation Academic Press 76 Kim, S.C. and Rieseberg, L.H. (1999) Genetic on holey adaptive landscapes. Am. Nat. 154, 1–22 48 Templeton, A.R. (1999) Experimental tests of genetic architecture of species differences in annual 24 Barraclough, T.G. et al. (1995) Sexual selection and transilience. Evolution 53, 1628–1632 sunflowers: Implications for adaptive trait taxonomic diversity in birds. Proc. R. Soc. 49 Hill, W.G. and Caballero, A. (1992) Artificial selection introgression. Genetics 153, 965–977 London B Biol. Sci. 259, 211–215 experiments. Annu. Rev. Ecol. Syst. 23, 287–310 77 Wang, R.L. et al. (1997) Gene flow and natural 25 Arnquist, G. et al. (2000) Sexual conflict promotes 50 Reznick, D.N. et al. (1997) Evaluation of the rate of selection in the origin of Drosophila speciation in insects. Proc. Natl. Acad. Sci. U. S. A. evolution in natural populations of guppies (Poecilia pseudoobscura and close relatives. Genetics 147, 97, 10460–10464 reticulata). Science 275, 1934–1937 1091–1106

http://tree.trends.com Review TRENDS in Ecology & Evolution Vol.16 No.7 July 2001 343

78 Harrison, R.G. (1989) Animal mitochondrial DNA as 85 Kelly, J.K. and Noor, M.A.F. (1996) Speciation by 92 van Doorn, G.S. (1998) Sympatric speciation and a genetic marker in population and evolutionary reinforcement: a model derived from studies of extinction driven by environment dependent sexual biology. Trends Ecol. Evol. 4, 6–12 Drosophila. Genetics 143, 1485–1497 selection. Proc. R. Soc. London B Biol. Sci. 265, 79 Noor, M.A.F. (1999) Reinforcement and other 86 Stam, P. (1982) The evolution of reproductive 1915–1919 consequences of sympatry. Heredity 83, 503–508 isolation in closely adjacent plant populations 93 Abrams, P.A. (1980) Consumer functional response 80 Day, T. (2000) Sexual selection and the evolution of through differential flowering time. Heredity 50, and competition in consumer-resource systems. costly female preferences: spatial effects. Evolution 105–118 Theor. Popul. Biol. 17, 80–102 54, 715–730 87 Drossel, B. and McKane, A. (2000) Competitive 94 Spencer, H.G. et al. (1986) A theoretical investigation 81 Cain, M.L. et al. (1999) Reinforcing selection is speciation in quantitative genetic models. J. Theor. of speciation by reinforcement. Am. Nat. 128, effective under a relatively broad set of conditions in Biol. 204, 467–478 241–262 a mosaic hybrid zone. Evolution 53, 1343–1353 88 Higashi, M. et al. (1999) Sympatric speciation by 95 Case, T.J. and Taper, M.L. (2000) Interspecific 82 Sanderson, N. (1989) Can gene flow prevent sexual selection. Nature 402, 523–526 competition, environmental gradients, gene flow, reinforcement? Evolution 43, 1123–1235 89 Bush, G.L. (1994) Sympatric speciation in animals: and the coevolution of species borders. Am. Nat. 155, 83 Kirkpatrick, M. and Servedio, M.R. (1999) The new wine in old bottles. Trends Ecol. Evol. 9, 285–288 583–605 reinforcement of mating preferences on an island. 90 Turner, G.F. and Burrows, M.T. (1995) A model of 96 Maynard Smith, J. (1966) Sympatric speciation. Genetics 151, 865–884 sympatric speciation by sexual selection. Proc. R. Am. Nat. 100, 637–650 84 Servedio, M.R. and Kirkpatrick, M. (1997) The Soc. London B Biol. Sci. 260, 287–292 97 Felsenstein, J. (1981) Skepticism towards Santa effects of gene flow on reinforcement. Evolution 51, 91 Fisher, R.A. (1930) The Genetical Theory of Natural Rosalia, or why are there so few kinds of animals. 1764–1772 Selection, Clarendon Press Evolution 35, 124–138

The genetics of species differences

H. Allen Orr

Species are separated by reproductive isolation as well as by more ‘ordinary’ logical status: if species are things that are differences in morphology and behavior that play no necessary role in blocking reproductively isolated, the genetic study of gene flow. Although a great deal is now known about the genetics of SPECIATION will sensibly start with such isolation. reproductive isolation, we are only beginning to understand the genetic basis However, genetic studies of ordinary species of ordinary phenotypic differences between species. I review what is known differences have now progressed far enough to allow about the number of genes involved in such differences, as well as about the at least some preliminary conclusions. role of major genes and epistasis in the evolution of these differences. I also The first survey of the genetics of species compare and contrast these findings with those on the genetics of differences appeared in 1938, with J.B.S. Haldane’s reproductive isolation. ‘The nature of interspecific differences’3. Although the problem he discussed was largely dropped (little of If we compare individuals from the same species with much relevance appeared over the next 45 years; but those from different species, we see two kinds of see Ref. 4) two important developments occurred over differences. First, members of separate species are the past 15 years. The first was the rise of reproductively isolated and, second, they often differ QUANTITATIVE TRAIT LOCUS (QTL) and related dramatically in morphology and behavior. According association-mapping strategies. We now possess the to the BIOLOGICAL (see Glossary), these polymorphic neutral molecular markers and the body differences are themselves different. The first is a of statistical theory required to map the genes necessary part of what it means to be species whereas underlying arbitrary trait differences in arbitrary the second is not. Despite this, it is clear that species, at least roughly5. The second was the REPRODUCTIVE ISOLATION and ‘ordinary’ phenotypic appearance of molecular tools (e.g. germline differences tend to go hand-in-hand, and there is good transformation) that allow us, in favorable cases, to reason to think that this association is partly causal. confirm the identity of a particular gene underlying a In fact, there is good reason to think the causes run in phenotypic difference and, in especially favorable both directions: phenotypic evolution might often give cases, to identify the actual nucleotide changes rise to reproductive isolation1, whereas reproductive (‘Quantitative Trait Nucleotides’ or QTNs) involved. isolation might often preserve phenotypic differences Despite these technical advances, current when taxa come into contact with each other2. questions about species differences largely remain Although the genetics of reproductive isolation has the same as in Haldane’s day. They are questions been studied intensively, that of ordinary species about genetic architecture: How many genes are differences (i.e. differences in traits expressed within involved? How large are their phenotypic effects? H. Allen Orr Dept of Biology, pure species and that play no necessary role in Where are these genes and what are their functions University of Rochester, blocking gene flow) has not. The reason does not seem in development? And what are the roles, if any, of Rochester, NY 14627, primarily technical as the two kinds of study largely dominance and epistasis? This emphasis on numbers USA. e-mail: involve the same experimental approaches. Instead, of genes and sizes of effects reflects one of the oldest [email protected] the reason probably reflects the above difference in problems in – the complexity of

http://tree.trends.com 0169–5347/01/$ – see front matter © 2001 Elsevier Science Ltd. All rights reserved. PII: S0169-5347(01)02167-X