Recombination Rate Evolution and the Origin of Species

1 Recombination rate evolution and the origin of species

2 Daniel Ortiz-Barrientos1,¶,*, Jan Engelstädter1,¶, and Loren H. Rieseberg2,3

4 1The University of Queensland, School of Biological Sciences, St. Lucia,

5 Queensland, Australia. 2 University of British Columbia, Department of Botany,

6 Vancouver, British Columbia, Canada, 3 Indiana University, Biology

7 Department, Bloomington, Indiana, USA

9 ¶ These authors contributed equally.

11 Corresponding author: * Ortiz-Barrientos, D. ([email protected])

14 Abstract

15 A recipe for dissolving species into a continuum of phenotypes is to recombine

16 their genetic material. Similarly, speciation is impeded if diverging populations

17 exchange their genetic material. Students of speciation have become

18 increasingly interested in the mechanisms by which recombination between

19 locally adapted lineages is reduced. Evidence abounds that chromosomal

20 rearrangements, via their suppression of recombination during meiosis in

21 hybrids, play a major role in adaptation and speciation. In contrast, genic

22 modifiers of recombination rates have been largely ignored in studies of

23 speciation. Here, we show how both types of reduction in recombination rates

24 facilitate divergence in the face of gene flow, including the early stages of

25 adaptive divergence, the persistence of species after secondary contact, and

26 reinforcement.

28 Introduction

29 In the absence of geographic barriers between populations, sexual

30 reproduction is considered the greatest obstacle to the origin of new species [1].

31 While divergent natural selection creates distinct populations, sexual

32 reproduction, through the homogenising effects of genetic recombination,

33 dissolves them. Cognizance of this antagonism has prompted many

34 evolutionary biologists to deem speciation within freely (sympatric) or between

35 partially (parapatric) interbreeding populations, improbable [2]. However, recent

36 theoretical and empirical results suggest that speciation in the face of gene flow

37 should be feasible, and perhaps common in nature [3, 4, 5]. Unfortunately, due

38 to our incomplete understanding of the genetics of speciation, the plausibility of

39 key assumptions underpinning theoretical models of the process remain

40 controversial. In the few instances where we do have some limited

41 understanding (e.g., Drosophila and Rhagoletis fruit flies [6, 7], Ficedula

42 flycatchers [8, 9], Heliconius butterflies [10], Mimulus monkeyflowers [11], and

43 Helianthus sunflowers [12]), the genetics of speciation with gene flow appears

44 to revolve around a key process: reduction of recombination between the genes

45 responsible for reproductive isolation and those contributing to local adaptation

46 [13, 14].

47 Recombination generates new genetic combinations every generation,

48 making it a rapid source of genetic variability upon which natural selection can

49 operate [15]. However, recombination also breaks apart favourable

50 combinations of alleles, potentially reducing the average fitness of a population

51 [16, 17]. This two-sided evolutionary effect of recombination is at the heart of

52 both the evolution of sex [18] and the formation of new species when there is

53 gene flow [19]. More specifically, the conditions that impede the evolution of sex

54 and recombination are largely the same as those that facilitate speciation with

55 gene flow. Alas, this connection is infrequently made and therefore biological

56 connections between the two processes have been often overlooked [but see

57 13, 20].

58 To visualize this connection, consider a situation where chromosomes carrying

59 alleles ABCD or abcd confer high fitness, but chromosomes carrying any other

60 combination of alleles produce low fitness. Now assume a population with only

61 ABCD and abcd chromosomes, i.e., strong linkage disequilibrium (LD, see

62 Glossary). Recombination would then reduce mean population fitness, and

63 mutations that reduce recombination can be favoured if they are linked to the

64 ABCD loci [17]. During speciation with gene flow, a similar situation arises: if

65 alternative combinations of alleles are favoured in two populations (abcd and

66 ABCD, respectively), then gene flow and recombination between them will

67 generate new genotypic combinations of low fitness, reducing levels of LD in

68 each population. Therefore, we expect that modifiers that prevent or reduce

69 recombination would also be favoured during speciation. Such modifiers would

70 preserve the original genotypes and maintain high fitness in the population via

71 the increase and maintenance of high levels of LD within populations [18]. On

72 the other hand, speciation with gene flow is dependent on divergent natural

73 selection, and rates of adaptation can be greater in sexual than asexual

74 populations [18]. Therefore, it is important to keep in mind that during speciation

75 driven by natural selection, sex and recombination can be favoured within

76 populations, but not between diverging lineages.

77 Here we review important aspects of the fundamental link between the

78 origin of sex and the origin of new species. We start by describing the various

79 forms in which LD arises within a population and produces selection on

80 recombination rates. Next, we review the various mechanisms that could evolve

81 to change recombination rates, including structural (e.g., chromosomal

82 rearrangements) and allelic modifiers of recombination. We then show how

83 selection on recombination rates is expected to play out in various stages of

84 speciation. We finish by outlining some of the consequences of recombination

85 rate evolution for our understanding of divergence and adaptation.

86 The why and how of recombination rate evolution

87 Natural selection on recombination rates

88 Recombination reduces LD within a population but normally does not

89 alter allele frequencies. Therefore, for recombination rates to be under natural

90 selection LD must be present in a population and altering levels of LD must

91 have fitness consequences. LD can be generated by a number of evolutionary

92 forces, three of which are illustrated in Figure 1. Perhaps most importantly in the

93 context of speciation, LD within a population will be induced by migration

94 between two populations with different allele frequencies at two or more loci

95 [21]. For instance, positive LD will build up when there is divergent selection in

96 the two populations but polymorphism is maintained through migration (Figure

97 1a). This is most clearly seen in the extreme case where alleles a and b are

98 fixed in one population and A and B in the other so that one round of migration

99 creates populations with both the AB and ab genotypes but no Ab or aB

100 genotypes.

101 Second, LD is generated by natural selection in single populations when

102 there is epistasis [22]: negative epistasis produces negative LD and positive

103 epistasis produces positive LD (Figure 1b). This is rather intuitive as natural

104 selection will produce an overabundance of intermediate genotypes (Ab and

105 aB) when they are fitter than expected based on the average of the extreme

106 genotypes ab and AB (negative LD), and vice versa (positive LD). Third, LD can

107 arise due to random genetic drift or mutation. Over time and across many loci,

108 drift and mutation per se will generate both positive and negative LD in equal

109 amounts. However, when selection also acts on these loci, this distribution

110 becomes biased towards negative LD through a mechanism known as the Hill-

111 Robertson effect. According to this, genetic variation characterized by positive

112 LD (e.g., only ab and AB genotypes present) will be eliminated more rapidly

113 than genetic variation characterized by negative LD (e.g., only Ab and aB). This

114 is because the former implies greater variance in fitness and hence more

115 efficient natural selection. As a consequence, negative LD will predominate in

116 the population (Figure 1c) [23]. Finally, LD can also be created by a number of

117 other factors, including assortative mating and sexually antagonistic selection

118 [19, 24].

119 Provided that one or several evolutionary forces create LD in a

120 population, will recombination be favoured or disfavoured by natural selection?

121 Positive LD implies an overabundance of individuals with either very high or

122 very low fitness compared to individuals with intermediate fitness values. This

123 high variance in fitness means that natural selection will be very efficient, so

124 that there will be indirect selection against recombination to maintain the

125 positive LD. More precisely, a modifier allele reducing recombination (see

126 Glossary) can spread by preserving and hitchhiking along with genotypes

127 carrying multiple beneficial mutations, provided linkage between the modifier

128 and the selected loci is sufficiently strong. Conversely, negative LD implies a

129 low variance in fitness and therefore impedes natural selection, so that

130 recombination modifiers increasing recombination rates can be favoured.

131 Based on this effect alone, divergent selection in two populations

132 connected by migration as well as positive epistasis are predicted to produce

133 selection against recombination, whereas negative epistasis and the Hill-

134 Robertson effect would create selection for recombination. However,

135 recombination also has the effect of breaking up co-adapted gene

136 combinations, with immediate and potentially strong fitness effects. As a

137 consequence, when there is strong negative epistasis, strong selection against

138 recombination can outweigh selection for recombination resulting from indirect

139 effects on facilitating adaptation. For in-depth treatments of the principles of

140 selection on recombination rates, see [15, 18, 25, 26].

141 Mechanisms to change recombination rates

142 So far, we have discussed selection on recombination without specifying

143 how genetic variation in recombination rates can arise. We will now focus on the

144 two main mechanistic ways in which recombination rates between a pair of loci

145 can change: through changing the rate at which crossovers occur between

146 linked loci, or though genomic re-arrangements that alter their physical location.

147 An overview of different mechanisms affecting recombination that involves one

148 or both of these principles is given in Table 1.

149 Most mathematical models for the evolution of recombination employ the

150 modifier approach introduced by Nei [27]. Here, alleles at a modifier locus

151 produce different recombination rates between other loci that might be under

152 direct natural selection (see Glossary). Assuming such recombination modifiers

153 is justified in that there is ample evidence for heritable variation in crossover

154 rates within populations [e.g., 28]. One of the most striking examples are

155 humans, where crossover hotspot positions differ considerably between

156 individuals of African vs. European ancestry [29]. Similarly, recombination rates

157 and patterns often vary considerably between closely related species, indicating

158 fast evolution [reviewed in 30]. However, it is less clear how common locally

159 acting recombination modifiers of the type assumed in the population genetics

160 models are relative to other ‘modifiers’ that are more firmly established

161 empirically but exhibit more complex dynamics (Box 1).

162 In addition to changes in crossover frequencies between different loci,

163 recombination rates will also change as a result of chromosomal

164 rearrangements that produce new linkage relationships. The most important

165 mechanism in the context of speciation is chromosomal inversions, which can

166 affect recombination rates in a number of ways [13, 14, 41, 42]. At the most

167 basic level, by changing gene order an inversion increases recombination rates

168 between some loci that were previously tightly linked and reduces

169 recombination rates between other loci that were less tightly linked before the

170 inversion occurred (Figure 2a). Moreover, crossover events in individuals that

171 are heterozygous for the inversion will be suppressed in the region flanking the

172 breakpoints of the inversion (sometimes extending several megabases away

173 from the breakpoint), further reducing recombination rates between genes in the

174 vicinity of the breakpoints [43, 44, 45, 46]. Finally, in heterozygous individuals,

175 crossover events can lead to unbalanced meiotic products (i.e., deletions and

176 duplications). As a consequence, no recombinant offspring might be recovered

177 because they are inviable or because recombinant meiotic products fail to

178 develop into gametes. These effects are the basis for underdominant selection

179 against inversions occurring in the heterozygous state.

180 Translocations and chromosome fusions are similar to inversions in that

181 they also produce large-scale changes in linkage relationships (Figure 2b) but,

182 in the heterozygous state, often result in unbalanced meiotic products and are

183 therefore selected against when rare [e.g., 47]. Finally, recombination rates can

184 change on a more limited scale through transpositions of short DNA segments

185 (e.g., containing only individual genes). For example, such transpositions can

186 arise as a result of active transposable elements. Unless transpositions disrupt

187 functional genes or strongly affect gene expression, they will generally not to be

188 expected to be under strong negative selection. Transpositions will generally

189 result in strongly increased recombination rates between genes that were

190 formally tightly linked.

191 Table 1: Mechanisms to alter recombination

192 Figure 2: Effects of chromosomal rearrangement on linkage

193 The evolution of recombination rates during speciation

194 Adaptive divergence and ecological speciation

195 How are recombination rates expected to evolve during speciation when

196 there is on-going gene flow between populations? Although this is a complex

197 and underexplored question, some light can be shed through previous

198 theoretical results. Consider a scenario of ecological speciation where two

199 parapatric populations adapt to different environmental conditions. Assuming

200 fitness trade-offs between alleles contributing to adaptation, selection against

201 recombination is expected. This is because selection for locally adapted alleles

202 at different loci will be more efficient when these alleles remain together rather

203 than recombine with non-adaptive migrant alleles. Selection against

204 recombination in this scenario has been explored quantitatively in a number of

205 recombination modifier models [26, 48, 49]. The same principle can also been

206 applied to selection for inversions [20], chromosome fusions [50], and

207 transpositions [51]. All else being equal, such chromosomal rearrangements

208 should spread more easily than modifier alleles because rearrangements are

209 completely linked to the loci under selection and recombination is suppressed

210 only in heterozygotes (see Box 2).

211 It is important to note that selection against recombination is not

212 inevitable. For example, negative epistasis between loci can produce selection

213 for modifier alleles that increase recombination rates even in the face of gene

214 flow of locally maladapted alleles [26]. Similarly and perhaps of more general

215 importance, random genetic drift can also produce selection for recombination

216 that alleviates Hill-Robertson interference among selected mutations at different

217 loci. The Hill-Robertson effect has previously been shown to produce strong

218 selection for recombination in both unstructured and subdivided populations

219 [54, 55]. It is conceivable that selection against recombination caused by

220 differential adaptation with gene flow could be overcome by selection for

221 recombination due to the Hill-Robertson effect, but this possibility remains to be

222 investigated.

223 Secondary contact and intrinsic reproductive isolation

224 These considerations also apply to scenarios where previously allopatric

225 populations come into secondary contact. As populations diverge in allopatry,

226 they might accumulate genetic differences that would fail to work in a hybrid.

227 One example comes from the evolution of interacting mutations that function in

228 each population but fail to properly interact in the alternative genetic

229 background of the diverging population (see Dobzhansky-Muller Model in

230 Glossary, and [1]). For instance, the ancestral interaction aabb might evolve into

231 AAbb in one descendant population and into aaBB in a second one. Although

232 each allelic interaction is functional in each population, the hybrid combination

233 AaBb might be defective since the functionality of A and B together has never

234 been tested. Provided the hybrids are not completely inviable or sterile, this

235 genetic disassembly in hybrids creates strong negative epistasis for fitness and

236 thereby strong selection for the evolution of modifiers that reduce

237 recombination, or for the ancestral genotypes and thus the breakdown of

238 reproductive barriers. These modifiers of recombination could then spread to

239 the two populations and partially favour the completion of speciation [56].

240 One potential case where reductions in recombination might have

241 evolved in response to negative epistasis occurs in Drosophila pseudoobscura

242 and D. persimilis. These fruit flies co-occur and hybridize [57] in the

243 northwestern region of North America, while D. pseudoobscura expands its

244 range into southern forests of the United States and Mexico. Genetic

245 experiments have found that genes contributing to postzygotic isolation localise

246 exclusively to chromosomal inversions that separate the two hybridising species

247 [58]. In their hybrids, these inversions lock into place the untested A to B

248 interaction we mentioned above. Therefore, these inversions preserve the

249 evolved configurations, AAbb and aaBB, that function properly within each

250 species and prevent the dissolution of the Dobzhansky-Muller incompatibilities

251 that would occur in the presence of recombination [6]. This might provide

252 evidence for selection against recombination as outlined above. Unfortunately,

253 the alternative possibility that the inversions had been present already before

254 secondary contact cannot be currently rejected. Moreover, in their sympatric

255 range, D. pseudoobscura females strongly discriminate against heterospecific

256 males (prezygotic isolation) compared to those females found in the southern

257 allopatric ranges [59]. Thus, mating discrimination and reductions in

258 recombination might evolve concurrently, a matter that we turn our attention to

259 below.

260 Interaction with other modes of reinforcement

261 Overall, reduction in recombination rates after secondary contact helps

262 maintain the integrity of co-adapted gene clusters. Adopting a broad definition

263 of reinforcement as an increase in reproductive isolation by natural selection in

264 response to maladaptive gene flow, this reduction in recombination rates can be

265 viewed as a mechanism of reinforcement (see also Table 1 in [56]). This is

266 because even though reduced recombination does not prevent the formation of

267 hybrids, it does reduce the production of offspring with different parental gene

268 combinations and thus enhances population differentiation. Other mechanisms

269 of reinforcement include reductions in dispersal [60 (pp. 127-131), 61],

270 increased mate preference for conspecifics [62], and the evolution of alternative

271 mating systems (e.g., selfing or asexual reproduction; [63]). These mechanisms

272 reduce genetic exchange between incipient species but act at different stages

273 of the life-cycle of the organism [dispersal, syngamy and meiosis, respectively;

274 see also Ref. 56]. Some degree of competition between the different

275 mechanisms of reinforcement can be expected, with a hierarchy that reflects the

276 order of the different life stages. As an extreme example, consider the case

277 where complete suppression of dispersal between two locally adapted

278 populations has evolved. Here, selection for reduced recombination rates (and

279 for assortative mating) in response to maladapted migration will come to a full

280 stop. The reverse however is not necessarily true: even when complete

281 suppression of recombination between a set of loci has evolved, there might still

282 be on-going selection for reduced dispersal rates or assortative mating through

283 selection against migrants, heterozygotes, or recombination at other loci.

284 An interesting interaction is predicted to occur between modifiers for

285 assortative mating and recombination. In the classic model by Felsenstein [19],

286 assortative mating can evolve as a consequence of selection against

287 recombinants between two locally adapted genotypes. The higher the

288 recombination rate between the two loci under selection, the easier it is for the

289 assortative mating alleles to become associated with the locally adapted

290 genotypes. As a consequence, modifiers that reduce recombination between

291 the two loci under selection impede the evolution of assortative mating [64],

292 illustrating the competition between assortative mating and recombination

293 reinforcement. Conversely, recombination between the assortative mating locus

294 and the loci under selection constrains speciation, and modifiers suppressing

295 recombination are favoured [64].

296 In general, the different mechanisms of reproductive isolation generated

297 by reinforcement could evolve at the same time; which one is most dominant

298 would depend on available genetic variation, biological constraints and

299 pleiotropic effects of the respective modifiers. Of course, recombination

300 reinforcement alone is unable to bring about complete reproductive isolation

301 due to inherent limitations on genome-wide reductions in recombination.

302 However, species can behave as gene clusters where a subset of genes

303 controls species identity despite some levels of gene flow between groups.

304 Under this species concept (Genotypic Clusters; [65]), or in a Biological Species

305 concept where we relaxed the criterion of complete reproductive isolation to

306 describe the culmination of the speciation process [1], recombination

307 reinforcement can aid in the formation and persistence of well-defined species.

308 Some consequences of altered recombination rates

309 The spread of modifier alleles during speciation with gene flow also

310 affects the genetic architecture of speciation. A recombination modifier will

311 spread if it is linked to alleles conferring local adaptation, or by facilitating

312 reinforcement. We expect that regions of low recombination will tend to harbour

313 genes for various forms of reproductive isolation, as well as modifiers of

314 recombination, during the early stages of speciation or secondary contact. A

315 chromosomal inversion can also play this role, and therefore, inversions are

316 predicted to be disproportionally associated with these stages of speciation [6].

317 Furthermore, these regions might be partially responsible for genomic patterns

318 of species differentiation. Although other genomic regions of low recombination

319 (such as centromeres, Y-chromosomes, and mtDNA genomes) could potentially

320 be conducive to adaptive differentiation, it is not clear how important these

321 regions are for ecological speciation with gene flow given their low gene

322 density. (But note that these regions could be important with other modes of

323 speciation [66, 67, 68, 69].)

324 Natural selection reduces genetic diversity around favoured alleles within

325 populations, while increasing the divergence of these regions between

326 populations [70]. As a consequence, genomic divergence between populations

327 is expected to be heterogeneous [71, 72]. Gene flow intensifies these patterns

328 by homogenising neutrally evolving but not divergently selected loci and their

329 surrounding regions [73, 74, 75]. The extent to which gene flow can

330 homogenise regions around selected loci is a function of the strength of

331 selection, the frequency of hybridisation, and the recombination rates

332 experienced in hybrid meiosis [76]. If selection is strong, a selected locus can

333 create large swaths of loci with strong genetic differentiation between

334 populations, often known as islands of genomic divergence [71, 75].

335 Measures like Fst are often used to explore patterns of heterogeneous

336 genomic divergence during speciation with gene flow, where clusters of loci with

337 high Fst values should indicate the presence of islands of genomic divergence

338 [72]. Unfortunately, Fst is sensitive to low levels of heterozygosity, which are

339 likely to arise from selection acting on regions of reduced recombination [71,

340 76]. Therefore, variance in Fst might reflect the interplay between natural

341 selection and gene flow, but also selective sweeps, or the action of background

342 selection, in regions of reduced recombination occurring in local populations

343 that do not exchange genes. However, as we have argued here, these regions

344 of reduced recombination can in fact be a consequence of divergent selection

345 during speciation with gene flow. In general, it is challenging to argue in favour

346 of speciation with gene flow from patterns of heterogeneous genomic

347 divergence alone, at least in studies where recombination rate variation within

348 species is not considered (but see [45]).

349 Conclusions

350 We have revisited the strong and natural link between the evolution of

351 sex and the origin of new species. As populations adapt to new conditions while

352 still exchanging genes with other populations, and also during secondary

353 contact, recombination rates might evolve in response to maladaptive gene

354 exchange. Generally, recombination is expected to be selected against,

355 resulting in the evolution of coadapted gene complexes, favouring the evolution

356 of various forms of reproductive isolation and thereby helping advance the

357 speciation process. It is important to stress, however, that there can also be

358 conditions where increased rates of recombination are favoured. Moreover,

359 there are multiple scenarios where recombination rate evolution is not expected

360 to play a role during speciation. For instance, strong selection could prevent

361 gene flow almost instantaneously between parapatric populations, and

362 modifiers other than those affecting recombination (especially dispersal and

363 assortative mating) could predominate as mechanisms of reinforcement. These

364 scenarios shorten the window of time in which recombination rates can evolve

365 in response to gene flow between populations adapting to contrasting

366 environments.

367 Chromosomal inversions should play a major role during speciation with

368 gene flow. However, we have argued that our focus should also encompass

369 allelic modifiers of recombination that might create more subtle patterns of

370 recombination rate variation across the genome. Genetic mapping exercises

371 within species are likely to reveal the genetic basis of recombination rate

372 variation, and whether different mechanisms operate across the tree of life.

373 Finally, we have suggested that the genetic architectures arising from the

374 interplay between selection and gene flow are likely to produce islands of

375 genomic divergence in regions of reduced recombination, thus complicating

376 interpretations of patterns of genomic divergence between populations. Studies

377 that scrutinize both genetic differentiation across space and variation in

378 recombination rates within species are therefore needed to gain a better

379 understanding of the causes of genomic variation in nature. More generally, we

380 expect that frequent dialogue between students of recombination rate evolution

381 and those of speciation will lead to better-integrated theories for the origin and

382 maintenance of genetic and phenotypic variability in nature.

383 Acknowledgements

384 We thank Mohamed Noor, Maddie E. James and three anonymous reviewers

385 for constructive comments on previous versions of this manuscript. Daniel Ortiz-

386 Barrientos and Jan Engelstädter are funded by the Australian Research Council

387 (Discovery Project Grant, DP140103774). Loren Rieseberg’s work on speciation

388 is supported by a Discovery Grant from the Natural Sciences and Engineering

389 Research Council of Canada.

390

391 References

392 1. Coyne, J.A. and Orr, H.A. (2004) Speciation. Sinauer Associates, Inc.

393 2. Futuyma, D.J. and Mayer, G.C. (1980) Non-Allopatric Speciation in 394 Animals. Syst Zool 29, 254-271.

395 3. Gavrilets, S. (2004) Fitness landscapes and the origin of species. 396 Princeton University Press.

397 4. Nosil, P. (2012) Ecological speciation. Oxford University Press.

398 5. Bolnick, D.I. and Fitzpatrick, B.M. (2007) Sympatric speciation: Models 399 and empirical evidence. Annual Review of Ecology Evolution and 400 Systematics 38, 459-487.

401 6. Noor, M.A.F., et al. (2001) Chromosomal inversions and the reproductive 402 isolation of species. Proc. Natl. Acad. Sci. USA 98, 12084-12088.

403 7. Feder, J.L., et al. (2003) Allopatric genetic origins for sympatric host- 404 plant shifts and race formation in Rhagoletis. Proc Natl Acad Sci U S A 405 100, 10314-10319.

406 8. Saetre, G.P. and Saether, S.A. (2010) Ecology and genetics of 407 speciation in Ficedula flycatchers. Mol. Ecol. 19, 1091-1106.

408 9. Burri, R., et al. (2015) Linked selection and recombination rate variation 409 drive the evolution of the genomic landscape of differentiation across the 410 speciation continuum of Ficedula flycatchers. Genome Res in press.

411 10. Joron, M., et al. (2011) Chromosomal rearrangements maintain a 412 polymorphic supergene controlling butterfly mimicry. Nature 477, 203- 413 206.

414 11. Lowry, D.B. and Willis, J.H. (2010) A Widespread Chromosomal 415 Inversion Polymorphism Contributes to a Major Life-History Transition, 416 Local Adaptation, and Reproductive Isolation. PLoS Biol. 8, pii: 417 e1000500.

418 12. Strasburg, J.L., et al. (2009) Genomic Patterns of Adaptive Divergence 419 between Chromosomally Differentiated Sunflower Species. Mol. Biol. 420 Evol. 26, 1341-1355.

421 13. Butlin, R.K. (2005) Recombination and speciation. Mol. Ecol. 14, 2621- 422 2635.

423 14. Ortiz-Barrientos, D., et al. (2002) Recombination and the divergence of 424 hybridizing species. Genetica 116, 167-178.

425 15. Otto, S.P. (2009) The Evolutionary Enigma of Sex. Am. Nat. 174, S1- 426 S14.

427 16. Maynard Smith, J. (1978) The Evolution of Sex. Cambridge University 428 Press.

429 17. Altenberg, L. and Feldman, M.W. (1987) Selection, generalized 430 transmission and the evolution of modifier genes. 1. The Reduction 431 Principle. Genetics 117, 559-572.

432 18. Otto, S.P. and Lenormand, T. (2002) Resolving the paradox of sex and 433 recombination. Nat Rev Genet 3, 252-261.

434 19. Felsenstein, J. (1981) Skepticism towards Santa Rosalia, or Why are 435 there so few kinds of animals? Evolution 35 124-138.

436 20. Kirkpatrick, M. and Barton, N. (2006) Chromosome inversions, local 437 adaptation and speciation. Genetics 173, 419-434.

438 21. Li, W.H. and Nei, M. (1974) Stable linkage disequilibrium without 439 epistasis in subdivided populations. Theor Popul Biol 6, 173-183.

440 22. Kimura, M. (1956) A Model of a Genetic System Which Leads to Closer 441 Linkage by Natural-Selection. Evolution 10, 278-287.

442 23. Hill, W.G. and Robertson, A. (1966) Effect of Linkage on Limits to 443 Artificial Selection. Genet. Res. 8, 269-294.

444 24. Ubeda, F., et al. (2011) Stable linkage disequilibrium owing to sexual 445 antagonism. Proc. R. Soc. Lond. B 278, 855-862.

446 25. Barton, N.H. (1995) A general model for the evolution of recombination. 447 Genet. Res. 65, 123-144.

448 26. Lenormand, T. and Otto, S.P. (2000) The evolution of recombination in a 449 heterogeneous environment. Genetics 156, 423-438.

450 27. Nei, M. (1967) Modification of linkage intensity by natural selection. 451 Genetics 57, 625-641.

452 28. Coop, G., et al. (2008) High-resolution mapping of crossovers reveals 453 extensive variation in fine-scale recombination patterns among humans. 454 Science 319, 1395-1398.

455 29. Hinch, A.G., et al. (2011) The landscape of recombination in African 456 Americans. Nature 476, 170-U167.

457 30. Smukowski, C.S. and Noor, M.A.F. (2011) Recombination rate variation 458 in closely related species. Heredity 107, 496-508.

459 31. Myers, S., et al. (2008) A common sequence motif associated with 460 recombination hot spots and genome instability in humans. Nat. Genet. 461 40, 1124-1129.

462 32. Baudat, F., et al. (2010) PRDM9 is a major determinant of meiotic 463 recombination hotspots in humans and mice. Science 327, 836-840.

464 33. Myers, S., et al. (2010) Drive against hotspot motifs in primates 465 implicates the PRDM9 gene in meiotic recombination. Science 327, 876- 466 879.

467 34. Parvanov, E.D., et al. (2010) Prdm9 controls activation of mammalian 468 recombination hotspots. Science 327, 835.

469 35. Oliver, P.L., et al. (2009) Accelerated evolution of the Prdm9 speciation 470 gene across diverse metazoan taxa. PLoS Genet 5, e1000753.

471 36. Schwartz, J.J., et al. (2014) Primate evolution of the recombination 472 regulator PRDM9. Nat Commun. 5, 4370.

473 37. Berg, I.L., et al. (2010) PRDM9 variation strongly influences 474 recombination hot-spot activity and meiotic instability in humans. Nat 475 Genet 42, 859-863.

476 38. Mihola, O., et al. (2009) A mouse speciation gene encodes a meiotic 477 histone H3 methyltransferase. Science 323, 373-375.

478 39. Nowick, K., et al. (2013) A prominent role of KRAB-ZNF transcription 479 factors in mammalian speciation? Trends in genetics : TIG 29, 130-139.

480 40. Boulton, A., et al. (1997) The hotspot conversion paradox and the 481 evolution of meiotic recombination. Proc. Natl. Acad. Sci. USA 94, 8058- 482 8063.

483 41. Faria, R., et al. (2011) Role of Natural Selection in Chromosomal 484 Speciation. eLS.

485 42. Hoffmann, A.A. and Rieseberg, L.H. (2008) Revisiting the Impact of 486 Inversions in Evolution: From Population Genetic Markers to Drivers of 487 Adaptive Shifts and Speciation? Annual review of ecology, evolution, and 488 systematics 39, 21-42.

489 43. Barb, J.G., et al. (2014) Chromosomal evolution and patterns of 490 introgression in Helianthus. Genetics 197, 969-979.

491 44. Stevison, L.S., et al. (2011) Effects of Inversions on Within- and 492 Between-Species Recombination and Divergence. Genome biology and 493 evolution 3, 830-841.

494 45. McGaugh, S.E. and Noor, M.A.F. (2012) Genomic impacts of 495 chromosomal inversions in parapatric Drosophila species. Phil. Trans. R. 496 Soc. B 367, 422-429.

497 46. Machado, C.A., et al. (2007) Evaluation of the genomic extent of effects 498 of fixed inversion differences on intraspecific variation and interspecific 499 gene flow in Drosophila pseudoobscura and D. persimilis. Genetics 175, 500 1289-1306.

501 47. Stathos, A. and Fishman, L. (2014) Chromosomal rearrangements 502 directly cause underdominant F1 pollen sterility in Mimulus lewisii- 503 Mimulus cardinalis hybrids. Evolution 68, 3109-3119.

504 48. Charlesworth, D. and Charlesworth, B. (1979) Selection on 505 Recombination in Clines. Genetics 91, 581-589.

506 49. Pylkov, K.V., et al. (1998) Migration versus mutation in the evolution of 507 recombination under multilocus selection. Genet. Res. 71, 247-256.

508 50. Guerrero, R.F. and Kirkpatrick, M. (2014) Local Adaptation and the 509 Evolution of Chromosome Fusions. Evolution 68, 2747-2756.

510 51. Yeaman, S. (2013) Genomic rearrangements and the evolution of 511 clusters of locally adaptive loci. Proc. Natl. Acad. Sci. USA 110, E1743- 512 E1751.

513 52. Schultz, J. and Redfield, H. (1951) Interchromosomal effects on crossing 514 over in Drosophila. Cold Spring Harb Symp Quant Biol 16, 175-197.

515 53. Farré, M., et al. (2013) Recombination Rates and Genomic Shuffling in 516 Human and Chimpanzee-A New Twist in the Chromosomal Speciation 517 Theory. Mol. Biol. Evol. 30, 853-864.

518 54. Keightley, P.D. and Otto, S.P. (2006) Interference among deleterious 519 mutations favours sex and recombination in finite populations. Nature 520 443, 89-87.

521 55. Martin, G., et al. (2006) Selection for recombination in structured 522 populations. Genetics 172, 593-609.

523 56. Lenormand, T. (2012) From Local Adaptation to Speciation: 524 Specialization and Reinforcement. International Journal of Ecology 2012, 525 508458.

526 57. Powell, J.R. (1983) Interspecific cytoplasmic gene flow in the absence of 527 nuclear gene flow: evidence from Drosophila. Proc Natl Acad Sci U S A 528 80, 492-495.

529 58. Noor, M.A.F., et al. (2001) The genetics of reproductive isolation and the 530 potential for gene exchange between Drosophila pseudoobscura and D- 531 persimilis via backcross hybrid males. Evolution 55, 512-521.

532 59. Noor, M.A. (1995) Speciation driven by natural selection in Drosophila. 533 Nature 375, 674-675.

534 60. Fisher, R.A. (1930) The Genetical Theory of Natural Selection. Oxford 535 University Press.

536 61. Yukilevich, R. and True, J.R. (2006) Divergent outcomes of 537 reinforcement speciation: The relative importance of assortative mating 538 and migration modification. Am. Nat. 167, 638-654.

539 62. Servedio, M.R. and Noor, M.A.F. (2003) The role of reinforcement in 540 speciation: Theory and data. Annual Review of Ecology Evolution and 541 Systematics 34, 339-364.

542 63. Hopkins, R. (2013) Reinforcement in plants. New Phytol. 197, 1095- 543 1103.

544 64. Trickett, A.J. and Butlin, R.K. (1994) Recombination Suppressors and the 545 Evolution of New Species. Heredity 73, 339-345.

546 65. Mallet, J. (1995) A species definition for the modern synthesis. Trends 547 Ecol Evol 10, 294-299.

548 66. Johnson, N.A. (2010) Hybrid incompatibility genes: remnants of a 549 genomic battlefield? Trends in genetics : TIG 26, 317-325.

550 67. Henikoff, S., et al. (2001) The centromere paradox: stable inheritance 551 with rapidly evolving DNA. Science 293, 1098-1102.

552 68. Burton, R.S., et al. (2013) Cytonuclear Genomic Interactions and Hybrid 553 Breakdown. Annu Rev Ecol Evol S 44, 281-302.

554 69. Sweigart, A.L. (2010) Simple Y-Autosomal Incompatibilities Cause 555 Hybrid Male Sterility in Reciprocal Crosses Between Drosophila virilis 556 and D. americana. Genetics 184, 779-U225.

557 70. Begun, D.J. and Aquadro, C.F. (1992) Levels of naturally occurring DNA 558 polymorphism correlate with recombination rates in D. melanogaster. 559 Nature 356, 519-520.

560 71. Cruickshank, T.E. and Hahn, M.W. (2014) Reanalysis suggests that 561 genomic islands of speciation are due to reduced diversity, not reduced 562 gene flow. Mol Ecol 23, 3133-3157.

563 72. Nosil, P., et al. (2009) Divergent selection and heterogeneous genomic 564 divergence. Mol Ecol 18, 375-402.

565 73. Turner, T.L., et al. (2005) Genomic islands of speciation in Anopheles 566 gambiae. PLoS Biol 3, e285.

567 74. Harr, B. (2006) Genomic islands of differentiation between house mouse 568 subspecies. Genome Res 16, 730-737.

569 75. Nosil, P. and Feder, J.L. (2012) Genomic divergence during speciation: 570 causes and consequences. Philosophical transactions of the Royal 571 Society of London. Series B, Biological sciences 367, 332-342.

572 76. Nachman, M.W. and Payseur, B.A. (2012) Recombination rate variation 573 and speciation: theoretical predictions and empirical results from rabbits 574 and mice. Philosophical transactions of the Royal Society of London. 575 Series B, Biological sciences 367, 409-421.

576 77. Avelar, A.T., et al. (2013) Genome architecture is a selectable trait that 577 can be maintained by antagonistic pleiotropy. Nat Commun 4, 2235.

578

579 Figure legends

580 Figure 1. The evolution of linkage disequilibrium in a population.

581 Illustration of three different ways in which LD can be created in a population: a)

582 migration, b) epistasis and c) the Hill-Robertson effect. In all plots, two biallelic

583 loci A and B are considered, and the two axes give the frequencies of alleles a /

584 A and b / B at these loci, respectively. The four areas in the plots then give the

585 proportion of the four possible haploid genotypes (ab, Ab, aB and AB), with

586 fitness indicated by increasing intensity of green.

587 Figure 2. The effect of chromosomal rearrangement on linkage.

588 (a) A chromosomal inversion (inverted purple box g-c) increases recombination

589 rates between some loci that were previously tightly linked (b,c and g,h) and

590 reduces recombination rates between other loci that were less tightly linked

591 before the inversion occurred (b,g and c,h). (b) Translocations (mixture of

592 chromosome blocks of different colors) also produce large-scale changes in

593 linkage relationships, as denoted by arrows. Note that these changes in

594 recombination rate arise in homokaryotypes; in contrast, in heterokaryotypes,

595 additional factors influence patterns of recombination (see Table 1).

596

597 Glossary box

598  Chromosomal rearrangement: A structural modification within or between

599 chromosomes that affects the spatial location of genetic material.

600  Ecological speciation: The development of reproductive isolation

601 between populations as a result of adaptation to different environments.

602 It can take place in the face of gene flow.

603  Epistasis: Non-independent effects of alleles at different loci. In

604 particular, epistasis in fitness between two loci with alleles A/a and B/b

605 can be defined in haploid organisms as E = wabwAB - wAbwaB, where w is

606 the fitness of the respective genotype. Negative epistasis implies that the

607 double mutant AB has a lower fitness than expected from the fitness

608 effects of the two mutations A and B on the wild type backgrounds b and

609 a, respectively. Conversely, positive epistasis implies that AB has a

610 higher fitness than expected from the individual effects of A and B.

611  Fitness trade-offs: Alleles improving fitness in one environment reduce

612 fitness in the other environment.

613  Hill-Robertson effect: Negative linkage disequilibrium arising from the

614 interplay between random genetic drift and selection acting on several

615 linked loci. Can produce selection for increased recombination rates.

616  Linkage disequilibrium (LD): Non-random associations of alleles at

617 different loci in a population. With two loci with alleles A/a and B/b, LD

618 can be expressed as D = pabpAB - pAbpaB, where p is the frequency of the

619 respective genotype in the haploid phase. Negative LD thus indicates an

620 overabundance of the Ab and aB genotypes, whereas positive LD

621 indicates an overabundance of the ab and AB genotypes. Note that

622 labelling of genotypes and thus the sign of D is often arbitrary, but in the

623 case of natural selection acting on both loci the convention is to assign

624 genotype labels AB and ab to the genotypes with the highest and lowest

625 fitness, respectively, whereas the Ab and aB genotypes have

626 intermediate fitness.

627  Recombination hotspots: Regions of the genome where crossovers

628 happen with high frequency.

629  Recombination modifier: A gene that affects the rate of crossovers

630 between other genes. A modifier allele can increase or decrease in

631 frequency by hitchhiking with the gene combinations it creates, thereby

632 changing recombination rates within the population.

633  Reinforcement: The strengthening of reproductive isolation by natural

634 selection in response to maladaptive hybridization. Reinforcement aids

635 the completion of speciation.

636  Underdominance: Individuals heterozygous at a given locus are at a

637 selective disadvantage compared to homozygous individuals.

638 Table 1. Mechanisms affecting recombination and their consequences on fitness in a single individual

MECHANISM EFFECTS

Change in crossover Change in linkage Pleiotropic fitness frequency relationships effects

Modifier alleles Yes No Possible

Acting locally (on linked or Possibly underdominant unlinked sites), or globally effects [38, 39]

Gene transpositions No Yes Possible

Leads, by definition, to new Might insert into another gene order along a gene and disrupt it; changes chromosome, or movement in expression level also to new chromosome possible

Inversions Yes Yes Possible

Particularly near breakpoints Between a gene within and a Underdominant effects due gene flanking the inversion to aneuploidy in gametes; fitness effects due to altered

gene expression [77]

Translocations Yes Yes Yes

(including chromosome Particularly near breakpoints, Free recombination between Underdominant effects due fusions) and apparent linkage translocated region and to aneuploidy in gametes; between different donor chromosome; linkage changes in expression level chromosomes between translocated region also possible [77] and recipient chromosome

639

640

40 Trends Box

Trends box

 An often-overlooked connection between the evolution of sexual

recombination and the origin of new species with gene flow suggests

that the conditions for speciation with gene flow may be less restrictive

than previously anticipated.

 Evolutionary scenarios for which mathematical models predict selection

for reduced recombination can provide insights into how ecological

speciation and reinforcement can proceed.

 Recent advances in our understanding of the molecular mechanisms of

recombination in eukaryotes have provided first insights into how

recombination rates can be modified within and between species.

 In addition to the well-established role of chromosomal inversions

during speciation, more subtle changes in recombination rates through

modifier genes influencing the frequencies and distributions of

crossover across the genome might also be important.  A better understanding of genomic patterns of differentiation during

speciation could be gained by taking into account that recombination

rates can themselves evolve. Outstanding Questions

Outstanding questions

 Migration-selection balance favours the suppression of recombination.

However, in finite populations, there may also be selection for increased

recombination as this alleviates the Hill-Roberson effect. Can the Hill-

Robertson effect cancel out selection for suppressed recombination, thereby

hampering divergence? This may have relevance for the question of how

population size impacts the likelihood of speciation.

 Migration between adapted populations may favour reductions in

recombination between locally adaptive loci. Will these regions become

islands of speciation?

 Are chromosomal inversions more effective than genic modifiers of

recombination rates in favouring the speciation process?

 How much genetic variation for recombination rates at different genomic

scales is there within and between species, and what is the genetic basis for

this variation?  There are multiple forms of reinforcing selection. How do these interact, and

what are the conditions under which the evolution of suppressed

recombination fails to facilitate speciation with gene flow?

 What are the consequences of the evolution of suppressed recombination for

our understanding of genomic patterns of differentiation between populations? (a) Population 1 Population 2 (c) a A a A a A b b

B B

Migration Drift

a A a A a A a A b b b

B B B

B ! Positive LD ! Positive LD Selection

a A a A b (b) Negative epistasis Positive epistasis b a A a A

b b B B

B B ! Negative LD ! Positive LD persists disappears

Selection

a A a A

b b

B B

! Negative LD ! Positive LD

Figure 1 (a) = =

a b g f e d c h i

- +

(b) = =

a b c d e q r m n o p f g

- + Figure 2 (a) m a b M A B

Meiosis

m a B

M A b

Figure B1a (b) Individual 1: m m Individual 2: M M

Figure B1b (c) a b A B

Meiosis

a B

A b Figure B1c