1. Introduction
Since the seminal work of Maynard Smith & Haigh (Reference Maynard Smith and Haigh1974), the ‘hitch-hiking’ effect has been recognized as an important cause of genome-wide variation in neutral molecular diversity (Fay & Wu, Reference Fay and Wu2000). For instance, it has been put forward as a major explanation for the observed correlation between genomic diversity and local recombination rates in Drosophila and humans (Begun & Aquadro, Reference Begun and Aquadro1992; Payseur & Nachman, Reference Payseur and Nachman2002). In the presence of linkage disequilibrium between loci, hitch-hiking can in fact change the frequency of an allele depending on selection acting at a distinct but closely linked locus. If positive selection is acting, the favoured allele increases its frequency in the population, reducing allelic diversity at the target locus. When such a selective sweep occurs in regions of low recombination, neutral variability at flanking linked genes follows the evolutionary trajectory of the selected locus. A genomic region in which neutral variation is lower than expected under the mutation–drift balance could thus represent the signature of a recent selective sweep. Most of the studies so far that suggested evidence for an hitch-hiking effect were considering a decrease in neutral molecular diversity due to purifying or positive selection at a linked locus (reviewed in Fay & Wu, Reference Fay and Wu2000).
However, hitch-hiking to a locus subject to balancing selection would have the opposite effect. Since balancing selection maintains alternative alleles at the selected locus within populations for much longer evolutionary times than neutral alleles (Gillespie, Reference Gillespie1991; Takahata, Reference Takahata1990), closely linked neutral variants would also be actively maintained, thereby substantially increasing local levels of neutral molecular diversity (Charlesworth, Reference Charlesworth2006; Hudson & Kaplan, Reference Hudson and Kaplan1988). The predicted increase in nucleotide diversity near the selected locus is due to the evolutionary divergence among allelic subpopulations linked to distinct alleles at that locus, because of their long coalescence times (Kamau et al., Reference Kamau, Charlesworth and Charlesworth2007; Takahata & Satta, Reference Takahata and Satta1998). By analogy to the island model of migration, recombination allows ‘migration’ between allelic subpopulations, disrupting the coupling between the evolutionary fate of the target locus and its nearby genomic region. As a consequence, the diversity-enhancing effect of balancing selection is predicted to decline with chromosomal distance, at a rate dependent on local recombination rates.
One of the strongest forms of balancing selection described in natural populations is negative frequency-dependent (NFD) selection with a rare-allele advantage. NFD selection is predicted to maintain many alleles in populations for long evolutionary times, generating high allelic diversity and consequently high levels of sequence polymorphism (Takahata, Reference Takahata1990). A less recognized feature of systems under NFD selection is a noticeable reduction in among-population genetic differentiation at the selected locus, compared with neutral references, due to the high invasion success of immigrating rare alleles in the recipient population, causing a rapid increase in their frequency (Muirhead, Reference Muirhead2001; Schierup et al., Reference Schierup, Vekemans and Charlesworth2000b). By studying hitch-hiking to a locus subject to balancing selection in a subdivided population, Schierup et al. (Reference Schierup, Charlesworth and Vekemans2000a) also showed that a ‘valley’ in F ST values for neutral loci was expected in the genomic region surrounding the selected locus, extent of which was dependent on the local recombination rate and selection strength.
A well-known example of NFD selection is that acting on homomorphic self-incompatibility (SI) systems in hermaphroditic plants (Wright, Reference Wright1939), where a mating occurs only if the allelic specificity expressed at the stigma surface is different from the one borne by the pollen grain (de Nettancourt, Reference de Nettancourt2001). In Brassicaceae, pollen and stigma allelic specificities are encoded by two genes (respectively SCR and SRK) located in a region commonly referred to as the S-locus (Kusaba et al., Reference Kusaba, Dwyer, Hendershot, Vrebalov, Nasrallah and Nasrallah2001). The two genes are thought to be under very strong linkage, as recombination would decouple pollen and pistil specificities, thereby leading to a breakdown of SI (Casselman et al., Reference Casselman, Vrebalov, Conner, Singhal, Giovannoni, Nasrallah and Nasrallah2000; Awadalla & Charlesworth, Reference Awadalla and Charlesworth1999). This genomic region is thus experiencing very low rates of recombination. Consistent with strong NFD selection, an extremely high allelic diversity, with elevated levels of divergence between allelic classes, has been observed at SRK in several Brassicaceae species (reviewed in Castric & Vekemans, Reference Castric and Vekemans2004; Charlesworth et al., Reference Charlesworth, Vekemans, Castric and Glémin2005; Takebayashi et al., Reference Takebayashi, Brewer, Newbigin and Uyenoyama2003). Kamau & Charlesworth (Reference Charlesworth and Vekemans2005) showed that a peak of synonymous nucleotide diversity was present in the genomic region surrounding the S-locus in a single population of Arabidopsis lyrata, consistent with a hitch-hiking scenario and confirming that recombination could be suppressed in the region. This result was confirmed by a more detailed analysis using a larger set of flanking genes and characterizing the extent of linkage disequilibrium between those and the S-locus across several populations (Kamau et al., Reference Kamau, Charlesworth and Charlesworth2007). Further support for strong long-term hitch-hiking effects was also provided by the observation of high numbers of trans-specific polymorphisms at two S-locus flanking genes between the species A. lyrata and A. thaliana (Charlesworth et al., Reference Charlesworth, Kamau, Hagenblad and Tang2006). In contrast, in Nicotiana alata, low diversity was found in a gene located just a few kilobases from the S-locus (Takebayashi et al., Reference Takebayashi, Brewer, Newbigin and Uyenoyama2003).
Based on a species-wide sampling approach, we examined sequence diversity and population genetic structure in the genomic region surrounding the S-locus in Arabidopsis halleri, a sister species to A. lyrata with which it shares the same SI system (Bechsgaard et al., Reference Bechsgaard, Castric, Charlesworth, Vekemans and Schierup2006; Castric & Vekemans, Reference Castric and Vekemans2007). We obtained nucleotide sequences for four genes in the S-locus genomic region as well as five unlinked control genes, in a sample composed of individuals from six geographically distant populations. For S-linked genes, we checked for co-segregation with the SRK locus by analysing a large backcross family. For each gene we estimated synonymous nucleotide diversity as well as among-population differentiation, checking for differences between the sets of S-linked and control genes. Our results are consistent with the hitch-hiking model affecting both patterns of molecular diversity and population structure.
2. Materials and methods
(i) Sampling
Individual DNAs were chosen from a collection of European accessions of Arabidopsis halleri (Pierre Saumitou-Laprade, University of Lille 1). Based on a previous study that used this collection (Pauwels et al., Reference Pauwels, Saumitou-Laprade, Holl, Petit and Bonnin2005) we have chosen six populations representative of the species-wide diversity. Populations were from Germany (Harz, abbreviated as AL13, located at 51°55′ N, 10°19′ E), France (Auby, AU: 50°25′ N, 3°03′ E), Italy (St Leonhard in Passeier, I5: 46°49′ N, 11°15′ E), Poland (Katowice, PL1: 50°15′ N, 18°57′ E), Slovenia (Stojnci, SLO5: 46°22′ N, 16°00′ E) and the Czech Republic (Zaton, TC8: 48°57′ N, 13°48′ E). Five individuals from each population were randomly chosen, except for the population of Auby from which 6 individuals were taken, for a total of 31 samples.
Attempts to genotype individuals at the SRK gene were performed a posteriori using PCR primer pairs designed to amplify specifically each of the 25 S-haplotypes currently identified in A. halleri (Castric & Vekemans, Reference Castric and Vekemans2007; primer sequences given in Supplementary Table S1). We identified two S-haplotypes at SRK in 11 individuals, a single S-haplotype in 14 individuals, and none in six individuals (Supplementary Table S2). Missing data is probably due to the occurrence of S-haplotypes that have not been characterized yet. Genotypes with a single S-haplotype could also be homozygotes for recessive alleles. Because of these uncertainties, and because we had only access to DNA samples, it was not possible to investigate the association between particular S-haplotypes and observed haplotypes at the S-locus flanking genes.
Table S1. Sequences of the primers used to genotype individuals at SRK
Table S2. Genotypes at SRK for the 31 individuals sampled. For 11 individuals the complete genotype was determined (heterozygous individuals); for 14 individuals only one allele was identified, these genotypes could either be homozygotes or heterozygotes with an unknown allele; and for six individuals no allele could be identified.
Table S3. Number of sequences obtained from each sampled individual. For control genes (HAT4, CAUL, scADH, Aly9 and CHS), only three of the five sampled individuals per population were used. nd: not determined. f: failed, no sequence obtained after two consecutive cloning-sequencing operations. 1: heterozygous individual but only one complete sequence could be obtained. 1ho: presumably homozygous individual. 2: heterozygous individual with two complete sequences obtained
(ii) Loci surveyed
We obtained sequences for nine genes. Four of them are located in the genomic region flanking the S-locus at various distances from each side of SRK, according to the physical maps available for A. thaliana (www.arabidopsis.org) and A. lyrata (Kusaba et al., Reference Kusaba, Dwyer, Hendershot, Vrebalov, Nasrallah and Nasrallah2001). The genes flanking the S-locus are: ARK3 (At4g21380 according to the A. thaliana genome annotation, 1·7 kb from SRK), B120 (At4g21390, 7·2 kb) and B160 At4g21430, 20·7 kb) on one side of SRK and B80 (At4g21350, 26·6 kb) on the other side.
To test for the occurrence of hitch-hiking, these flanking genes were compared with five presumably unlinked loci, namely: HAT4 (At4g16780), CAULIFLOWER (CAUL) (At1g26310), CHS (At5g13930), scADH (At4g05530) and Aly9 (called SLR1 in A. thaliana, At3g12000, which belongs to the same gene family as SRK; Charlesworth et al., Reference Charlesworth, Mable, Schierup, Bartolomé and Awadalla2003). Although HAT4 and scADH are located on the same chromosome as the S-locus, they are sufficiently distant from SRK to be considered as genetically unlinked. For genes flanking SRK, all individuals sampled were sequenced, while for control genes a subset of individuals (3 per population) was used. Sequences for B80 and Aly9 consisted of a single exon.
In order to get a control for genome-wide population differentiation, seven microsatellite loci (AthZFPG, GC22, H117, Ice 13, MDC16, nga 112 and nga 361) were also genotyped on the whole sample, using a multiplex PCR procedure according to Llaurens et al. (submitted).
(iii) PCR amplification and sequencing
Primer sequences for amplification of B120 and B160 were taken from Kamau & Charlesworth (Reference Charlesworth and Vekemans2005). For B80, the reverse primer was as in Kamau & Charlesworth (Reference Charlesworth and Vekemans2005), and the forward primer was newly designed based on the A. thaliana sequence: B80-Ah-F 5′CGATCGGGTCTCTATCCAAC3′. Primers for ARK3 were taken from Hagenblad et al. (Reference Hagenblad, Bechsgaard and Charlesworth2006). For HAT4, primer sequences were taken from Wright et al. (Reference Wright, Lauga and Charlesworth2003), for CAUL from Purugganan & Suddith (Reference Purugganan and Suddith1998) and for CHS from Ramos-Onsins et al. (Reference Ramos-Onsins, Stranger, Mitchell-Olds and Aguadé2004). For scADH, primers were designed from A. thaliana and were as follows: scADH_Ah_F2 5′ACATCGCCGCAATCTTGT3′ and scADH_Ah_R2 5′CAGAAGAACCCTTCTCTAGGTGA3′. Annealing temperatures were as follows: 54°C for B120 and B160, 57°C for CAUL and CHS, 58°C for B80 and for ARK3, 65°C for scADH, HAT4 and Aly9. PCR amplification consisted of 1 min at 94°C, 40 s at the annealing temperature and 40 s at 72°C, for 35 cycles.
PCR fragments were cloned into the PCR 2.1 vector using the TA cloning kit (Invitrogen Life Technologies), and at least eight clones were sequenced individually using the BigDye Terminator Kit 3.1 (Applied Biosystems) and run on an ABI-3100 capillary sequencer (Applied Biosystems). The universal M13 primers were used to sequence cloned PCR products. All sequences have been deposited in GenBank, with accession numbers EU273946–EU274288.
(iv) Co-segregation analysis
Although a physical map of the S-locus in A. halleri is not available, it is available for two haplotypes in its sister species A. lyrata. Kusaba et al. (Reference Kusaba, Dwyer, Hendershot, Vrebalov, Nasrallah and Nasrallah2001) showed high levels of macrosynteny in the S-locus region between A. lyrata and A. thaliana, suggesting that the relative position of the genes is well conserved in the genus, although intergenic distances are variable. In order to verify co-segregation of flanking genes with SRK in A. halleri, we genotyped individuals from a large first-generation backcross family available in the laboratory (Willems et al., Reference Willems, Dräger, Courbot, Gode, Verbruggen and Saumitou-Laprade2007). In detail, the F1 of an interspecific cross A. halleri×A. lyrata was backcrossed against a different A. lyrata individual and 331 progeny were genotyped for B80, B120, B160, ARK3 and HAT4. This backcross family had been used previously to obtain a saturated genetic map in which three markers were located at less than 10 cM from SRK (Willems et al., Reference Willems, Dräger, Courbot, Gode, Verbruggen and Saumitou-Laprade2007), allowing us to estimate recombination rates in the S-locus genomic region.
Nucleotide sequences for each flanking gene and for HAT4 were obtained for each parent, and the web-based software WebCutter 2.0 (http://rna.lundberg.gu.se/cutter2/) was then used to generate a restriction map for each allele. Restriction enzymes were chosen so as to distinguish parental alleles in the progeny based on expected banding patterns. All restriction reactions were carried out for at least 4 h at 37° C, using 1 u enzyme/100 ng DNA. Fragments were separated through electrophoresis on 2% agarose gels and visualized using ethidium bromide under ultraviolet light. A single restriction enzyme per gene was sufficient to distinguish the A. halleri allele from that of A. lyrata.
(v) Data analyses
Sequences were aligned using ClustalW software (Thompson et al., Reference Thompson, Higgins and Gibson1994), implemented in BioEdit v.7.0.4.1 (Hall, Reference Hall1999). BioEdit was also used to manually adjust alignments when necessary. Coding regions were identified according to the A. thaliana genome annotation provided by the TAIR database (www.arabidopsis.org). In the case of homozygous individuals, two copies of the allele were included in the alignment. For some heterozygous individuals it was not possible to obtain the second allele, and only one was included in the data set.
Total, synonymous, non-synonymous and silent nucleotide diversity values (π; Nei & Gojobori, Reference Nei and Gojobori1986; Nei, Reference Nei1987) were calculated for each gene using DNAsp v.4.10 (Rozas et al., Reference Rozas, Sánchez-Delbarrio, Messeguer and Rozas2003). The average within-population diversity was also calculated, weighted by the sample sizes (Wright et al., Reference Wright, Lauga and Charlesworth2003).
Under the neutral model, within-species polymorphism should be proportional to between-species divergence (Hudson et al., Reference Hudson, Kreitman and Aguadé1987). In order to test for deviation from neutral expectations in flanking genes, a multilocus HKA test (Hudson et al., Reference Hudson, Kreitman and Aguadé1987) was used, testing the significance of the ratio of polymorphism within A. halleri to divergence between A. halleri and A. thaliana on all loci. Sequences for A. thaliana for each gene were retrieved from the TAIR database and used to estimate average nucleotide divergence between A. halleri and A. thaliana. To test for significance, 5000 coalescent simulations were run using the HKA program (http://lifesci.rutgers.edu/wheylab/HeylabSoftware.htm). We also used the maximum-likelihood multilocus HKA framework developed by Wright & Charlesworth (Reference Wright and Charlesworth2004) to test for an overall difference in polymorphism between the set of four S-locus flanking genes against the control genes. Specifically, we used the MLHKA program distributed by Stephen Wright to compare a model with free mutation at each locus and no selection against a model with free mutation and selection on the four S-locus flanking genes. Since these two models are nested, we used a log-likelihood ratio test with 4 degrees of freedom to compare their likelihood. Chain length was set to 100 000. Deviation from neutrality was also tested using Tajima's D for each gene (Tajima, Reference Tajima1989), for which an excess or a lack of intermediate frequency polymorphisms suggests the presence, respectively, of balancing selection (positive values of D) or purifying selection (negative values of D). All these analyses were conducted with DNAsp, unless specified otherwise.
The genetic differentiation between the six sampled populations was assessed with the statistic F ST computed according to Hudson et al. (Reference Hudson, Slatkin and Maddison1992a), as implemented in DNAsp. Significance of population differentiation was assessed through the Kst* test (Hudson et al., Reference Hudson, Boos and Kaplan1992b) using 1000 permutations. We tested for differences between mean values of F ST for S-linked versus control genes using the Mann–Whitney non-parametric test.
3. Results
Due to differences in sample sizes among genes and populations, and to technical problems with the cloning procedure, we obtained two to 12 sequences for each gene in each population, for a total ranging from 28 to 58 (Table 1 and Supplementary Table S3). Few homozygous individuals were found (8 in Aly9, 6 in CAUL, 5 in CHS, 2 in HAT4, 2 in scADH and 9 in B160). The length of the aligned sequences varied from 497 bp for Aly9 to 1477 bp for scADH. For ARK3, two different groups of sequences were found, differing by a long indel of 346 bp in the intron as was previously reported in A. lyrata (Hagenblad et al., Reference Hagenblad, Bechsgaard and Charlesworth2006). Two smaller indels were also found within the coding sequence: an in-frame 3 bp indel, and an 8 bp-long indel introducing a premature stop codon. For the latter, the associated sequence was thus considered as a pseudogene (it occurred in a single individual for which three distinct ARK3 sequences were obtained) and discarded from further analyses. In a few other cases, three haplotypes were found per individual. The presence of pseudogenes, characterized by indels in the coding parts of the gene and the existence of duplicated copies in ARK3 have also been reported in A. lyrata (Hagenblad et al., Reference Hagenblad, Bechsgaard and Charlesworth2006), although it is not clear whether the duplicated copies do co-segregate. For the purpose of conservativeness, we eliminated from our data set all putative pseudogenes and individuals with duplicated copies of ARK3. Numerous indels of up to 91 bp were also observed in the introns of scADH.
Table 1. Length of the alignments and number of sequences obtained from each population for each gene
Sample size per population is 5 individuals for S-locus flanking genes (6 in the population from Auby, AU) and 3 individuals for control genes. For detailed information on the number of sequences obtained for each individual, see Supplementary Table S3.
(i) Co-segregation analysis
Table 2 shows the results of the co-segregation analysis in a first-generation backcross progeny of A. halleri and A. lyrata between SRK and each of the B80, ARK3, B120, B160 and HAT4 genes together with three previously studied marker loci of the genomic region surrounding the S-locus (TSB2, At4-TC1 and FCA; Willems et al., Reference Willems, Dräger, Courbot, Gode, Verbruggen and Saumitou-Laprade2007). This region spans about 2·2 Mb on each side of SRK, as inferred from A. thaliana's Col-0 full genomic sequence. In a total progeny size of 331 individuals, 55 individuals (16%) showed a recombination event within the 4·4 Mb region surveyed. Among the flanking genes, no recombinants were detected between SRK and each of the ARK3, B120 and B160 genes, whereas one recombination event was observed with B80. For HAT4, about 2 Mb distant from SRK (according to the A. thaliana genome), 25 recombinants were detected. Our results confirm thus that the four putative S-locus flanking genes studied here (B80, ARK3, B120 and B160) are closely linked to SRK, while HAT4 cannot be considered as S-linked. The region between markers TSB2 and FCA spanned over 7·85 cM on one side of SRK and 8·76 cM on the other side, which amounts to an overall recombination rate of 3·8 cM/Mb, a value very close to the average genomic rate in non-centromeric regions reported by Kawabe et al. (Reference Kawabe, Hansson, Forrest, Hagenblad and Charlesworth2006) for A. lyrata (4 cM/Mb). Additionally, within the 2 Mb region lying between SRK and marker TSB2, we found very similar estimates of recombination rate per physical distance in the 1 Mb segment adjacent to SRK (3·52 cM/Mb in the region between the S-locus and the marker At4-TC1) and the next 1 Mb segment (3·63 cM/Mb in the region between At4-TC1 and TSB2). These results are only compatible with a narrow region of suppressed recombination around the S-locus.
Table 2. Results of the co-segregation analysis: multilocus haplotypes observed in the backcross progeny are listed, together with the number of individuals sharing them
For each locus, the distance in kilobases from SRK (according to the A. thaliana physical map), the number of recombinant individuals and the genetic distances are shown. A total of 331 individuals have been genotyped.
l, allele from A. lyrata; h, allele from A. halleri.
a Expected number of recombination events with SRK occurring under the hypothesis of a 4 cM/Mb recombination rate in the region.
(ii) Patterns of polymorphism in S-linked and control genes
Patterns of nucleotide polymorphism in S-linked and control genes are shown in Table 3. S-locus flanking genes showed on average nucleotide diversities about twice as high as those in control genes, both at the overall and at the within-population levels, and for total nucleotide diversity as well as for synonymous or silent diversity. The two genes directly flanking the S-locus on each side, B80 and ARK3, are those that show the highest level of polymorphism (about 4 times higher than the average of control genes). It should be noted, however, that the polymorphism previously reported at SRK in A. halleri is still an order of magnitude higher than at these two polymorphic genes (πtot=0·388 and πs=0·819; Castric & Vekemans, Reference Castric and Vekemans2007). The average ratio of πn (non-synonymous sites only) over πs (synonymous sites) was about 0·1 in both types of genes, suggesting no apparent overall difference in the level of adaptive constraints between them.
Table 3. Analysis of nucleotide polymorphism, population differentiation in A. halleri and divergence versus A. thaliana at four S-locus flanking genes and at five control genes. Values of Tajima's D statistic are also shown (all are non-significant)
a Sequence length excluding all gaps from the alignment.
b Overall Watterson's θ statistic computed over all sites.
c Overall nucleotide diversity computed over all sites.
d Average within-population nucleotide diversity computed over all sites.
e Overall synonymous nucleotide diversity.
f Overall silent diversity (synonymous sites+introns).
g Ratio of overall non-synonymous over synonymous nucleotide diversity.
h Estimates of FST and results from the Kst* test: ns, not significant; *P<0·5; **P<0·1; ***P<0·001.
i Number of substitutions per total site versus A. thaliana.
In the HKA test we compared nucleotide polymorphism within A. halleri with divergence from A. thaliana for the four S-locus flanking genes, and for all control genes. The multilocus HKA test showed a significant departure from neutral expectation (χ2=22·57, P=0·004), indicating that some loci differed in their relative patterns of polymorphism and divergence. In Fig. 1, the relative contribution of polymorphism (filled diamonds) and divergence (open squares) to the overall χ2 test statistics at each locus is shown. S-linked genes showed a clear excess of intraspecific polymorphism with respect to neutral expectations and a deficit in interspecific divergence, while the opposite situation was found in control genes. One control gene, Aly9, showed a substantial excess of divergence relative to polymorphism, suggesting a signature of positive directional selection that could bias our results. We performed the multilocus HKA test by excluding this gene and still found significant departure from neutral expectation (χ2=18·50, P=0·010). Comparison of a free-mutation model allowing for selection at the four S-locus flanking genes, with a strictly neutral model, under a maximum likelihood framework, gives log likelihood-ratio statistics of 27·0 and 20·8, for runs with or without Aly9, respectively, which are highly significant (P<0·001) against the χ2 distribution with 4 degrees of freedom. The estimates of the selection parameter for B160, B120, ARK3 and B80 are, respectively, 2·48, 3·20, 4·50 and 5·48. This parameter corresponds here to the relative increase in polymorphism due to genetic hitch-hiking, taking into account differences across loci in patterns of divergence between A. halleri and A. thaliana. Hence, these results provide evidence for a higher overall level of polymorphism at the four S-locus flanking genes compared with control loci. Tajima's D values, however, did not detect any deviations from neutrality for any of the genes, either S-linked or unlinked, and showed no differences between mean values for S-linked and control genes (Mann–Whitney test, P=0·46).
Fig. 1. Results of the multilocus HKA test. Filled diamonds represent deviation from the neutral expectation for polymorphism and open squares represent divergence. Points above the line indicate deviation toward an excess, points below the line indicate deviation towards a deficit. The test rejects neutrality (P=0·004).
(iii) Population genetic differentiation
Results for population subdivision show a clear difference between S-linked and control genes, with the expected valley in F ST values around the S-locus (Table 3). We obtained consistent estimates of F ST for the seven unlinked microsatellite loci (F ST=0·204) and for the control genes (F ST=0·206; range 0·09–0·38). Such close agreement between these two types of loci is typically not found because of large differences in mutation rates (Charlesworth, Reference Charlesworth1998), but is consistent with the observation of moderate polymorphism at the studied microsatellite loci (heterozygosity values ranging from 0·28 to 0·64, data not shown). This also suggests that the small sample size used for control genes has not produced a biased estimate of F ST. In contrast, the average F ST for S-locus flanking genes was about 4 times lower (F ST=0·054; range −0·02 to 0·10). Still, most of the loci, except B160, showed significant population differentiation with the Kst* test (Table 3). Overall, control loci showed significantly higher levels of population differentiation than S-locus flanking genes (Mann–Whitney test, P=0·027). Hence, these results thus clearly demonstrate the expected valley in F ST values around the S-locus.
4. Discussion
In the presence of balancing selection, genetic hitch-hiking is expected to cause a higher polymorphism in the flanking region compared with unlinked regions, but the size of this affected region can be small, depending on the local recombination rate (Wiuf et al., Reference Wiuf, Zhao Innan and Nordborg2004, Takahata & Satta, Reference Takahata and Satta1998). In this paper, we confirmed in A. halleri the observation of Kamau & Charlesworth (Reference Charlesworth and Vekemans2005) that hitch-hiking in the region of the S-locus in A. lyrata produced a local increase in polymorphism, and showed that it is also causing a local decrease in population genetic structure, compared with control loci unlinked to the S-locus. Such a localized effect on population structure was predicted by theoretical models (Schierup et al., Reference Schierup, Charlesworth and Vekemans2000a) and is due to the fact that strong NFD selection acting on the S-locus within subpopulations counteracts allele frequency divergence due to random genetic drift and causes higher effective migration rate for S-alleles than for neutral alleles (Schierup et al., Reference Schierup, Vekemans and Charlesworth2000b, Muirhead, Reference Muirhead2001).
(i) How large is the genomic region linked to the S-locus?
Co-segregation analyses based on 331 offspring from a first-generation backcross between A. halleri and A. lyrata showed that the four putative S-locus flanking genes investigated here are indeed closely linked to SRK. Moreover our data provided some estimates of recombination rate in the S-locus region. Kawabe et al. (Reference Kawabe, Hansson, Forrest, Hagenblad and Charlesworth2006) analysed recombination data in a mapping population of 99 individuals from a F2 intraspecific cross in A. lyrata subsp. petraea. They found evidence for a significantly lower recombination rate in a region spanning over about 600 kb from each side of the S-locus, compared with more distant flanking regions. In contrast, our results do not show indication of a reduced recombination rate in a 4 Mb region centred on the S-locus compared with the average genomic rate in non-centromeric regions reported by Kawabe et al. (Reference Kawabe, Hansson, Forrest, Hagenblad and Charlesworth2006) for A. lyrata (4 cM/Mb). We have no definite explanation for the discrepancy between our results and those of Kawabe et al. (Reference Kawabe, Hansson, Forrest, Hagenblad and Charlesworth2006), but note several differences between the two studies. First, our mapping population consisted of an interspecific first-generation backcross, which could generate some bias. However, any bias expected would be of the opposite direction, i.e. generating lower rather than higher recombination rates, compared with intraspecific crosses (Williams et al., Reference Williams, Goodman and Stuber1995). Moreover, Willems et al. (Reference Willems, Dräger, Courbot, Gode, Verbruggen and Saumitou-Laprade2007) showed for the same mapping family that recombination between A. halleri and A. lyrata genomes was as efficient in the interspecific hybrid as in the A. lyrata intraspecific crosses, based on the studies by Kuittinen et al. (Reference Kuittinen, de Haan, Vogl, Oikarinen, Leppala, Koch, Mitchell-Olds, Langley and Savolainen2004) and Yogeeswaran et al. (Reference Yogeeswaran, Frary, York, Amenta, Lesser, Nasrallah, Tanksley and Nasrallah2005). Secondly, the number of markers in the study of Kawabe et al. (Reference Kawabe, Hansson, Forrest, Hagenblad and Charlesworth2006) was higher than in ours, with notably many markers in the 600 kb region on both sides of the S-locus, whereas we had no marker in the interval between 25 kb and 1 Mb. Thus, due to the paucity of markers, we could have missed the region of reduced recombination. However, the high overall rate of recombination that we observed would still not be consistent with an extended region of suppressed recombination around the S-locus. Hence, a more detailed study with large mapping populations and large numbers of markers would be necessary to better characterize the size of the genomic region in close linkage to the S-locus. Uncertainty in the estimates of recombination rate also arises because the physical sizes of the S-locus regions in A. halleri and A. lyrata are not known precisely. The physical map of A. thaliana was used for calculations in this study and that of Kawabe et al. (Reference Kawabe, Hansson, Forrest, Hagenblad and Charlesworth2006), which could lead to overestimates of recombination rates because the genome of A. thaliana is about 40% smaller than that of its sister species (Johnstone et al., Reference Johnstone, Pepper, Hall, Chen, Hodnett, Drabeck, Lopez and Price2005). Notably the size of the S-locus itself has been found to be larger in two S-haplotypes of A. lyrata compared with that of the A. thaliana Col-0 accession (Kusaba et al., Reference Kusaba, Dwyer, Hendershot, Vrebalov, Nasrallah and Nasrallah2001), and to vary extensively between the two S-haplotypes. This could lead to differences among S-haplotypes in the size of the region with reduced recombination, as suggested by observed variation in the strength of linkage between ARK3 and SRK (Hagenblad et al., Reference Hagenblad, Bechsgaard and Charlesworth2006) or between B80 and SRK (Kamau et al., Reference Kamau, Charlesworth and Charlesworth2007) across S-haplotypes. As discussed in Kawabe et al. (Reference Kawabe, Hansson, Forrest, Hagenblad and Charlesworth2006), better knowledge of the size of the region in linkage disequilibrium with the S-locus is of critical importance to evaluate the evolutionary significance of S-haplotype-specific genetic load, a feature that would be generated in non-recombining regions surrounding the S-locus because of high heterozygosity and long divergence times of S-haplotypes (Uyenoyama, Reference Uyenoyama1997). Experimental observations consistent with the existence of a linked genetic load in the S-locus region have been reported by Stone (Reference Stone2004) in Solanum carolinense, and by Bechsgaard et al. (Reference Bechsgaard, Bataillon and Schierup2004) in A. lyrata. By fitting data on diversity at flanking neutral sites within and between different functional S-allele classes to a model of hitch-hiking under balancing selection, Kamau et al. (Reference Kamau, Charlesworth and Charlesworth2007) obtained indirect estimates of the recombination rate in the S-locus region. Their results pointed out that recombination is highly reduced in only a very narrow region flanking the S-locus, so that the number of highly linked genes is expected to be low. We suggest that this conclusion could still be compatible with the empirical observations of linked genetic load if this were to be caused by few deleterious mutations with large effect. The approach of Kamau et al. (Reference Kamau, Charlesworth and Charlesworth2007) could not be used in this study because we could not resolve the phases between SRK and the flanking genes.
(ii) Patterns of polymorphism and deviations from neutrality
Patterns of total, synonymous and silent nucleotide diversity observed in A. halleri in the present study were all consistent with the predicted hitch-hiking effect. Results from HKA tests indicated significant heterogeneity among genes in relative patterns of polymorphism and divergence, and more specifically showed that average polymorphism at the four S-locus flanking genes was higher than at the control genes. Moreover, the two genes closely flanking SRK on each side, i.e. ARK3 and B80, showed the highest levels of polymorphism, suggesting that the hitch-hiking effect is highly localized. These results are in close agreement with those reported in A. lyrata, which also showed a strong but localized signature of the hitch-hiking effect in the S-locus region (Kamau & Charlesworth, Reference Kamau and Charlesworth2005; Kamau et al., Reference Kamau, Charlesworth and Charlesworth2007). It is remarkable that a localized peak in nucleotide polymorphism has also been found in the S-locus region in A. thaliana (Shimizu et al., Reference Shimizu, Cork, Caicedo, Mays, Moore, Olsen, Ruzsa, Coop, Bustamante, Awadalla and Purugganan2004), and was interpreted as a transient signature of past hitch-hiking effects that occurred before the breakdown of self-incompatibility along the A. thaliana lineage (Charlesworth & Vekemans, Reference Charlesworth and Vekemans2005). Our results are also consistent with observations by Castric & Vekemans (Reference Castric and Vekemans2007) of a strong hitch-hiking effect on neutral polymorphism within the SRK gene itself, with exons 2 to 6 of this gene exhibiting extremely high polymorphism (πS=0·3–0·7) although they are not involved in the allelic-specificity determination. In contrast to the nucleotide polymorphism/divergence data, the Tajima's D-tests did not suggest any deviation from neutrality. Non-selective effects such as demographic events or population structure, however, have been demonstrated to affect this estimator (e.g. Schierup et al., Reference Schierup, Charlesworth and Vekemans2000a), reducing its reliability in detecting selection in flanking regions. An alternative explanation for the increased diversity at the S-locus flanking genes would be a direct effect of balancing selection on those genes. Surprisingly, it has been shown recently that one of the studied genes, B80, is a modifier of the expression of self-incompatibility through its role in regulation of endogenous SRK transcript levels in the stigmas (Liu et al., Reference Liu, Sherman-Broyles, Nasrallah and Nasrallah2007). However, phenotypic effects were shown to depend on variation occurring in the 5′ promoter region of the gene and not within the coding region surveyed in the present study.
(iii) Patterns of population structure
Schierup et al. (Reference Schierup, Vekemans and Charlesworth2000b) have shown theoretically that the effective migration rates are higher at loci under strong balancing selection than at neutral loci. This is intuitively sound, due to the underlying NFD selection, which prevents allele frequency divergence due to drift and gives an advantage to the migrant allele not previously present in the recipient population. Few empirical studies have compared population genetic structure between the S-locus and unlinked control genes or markers, but the results seem to confirm theoretical expectations. In Brassica insularis, Glémin et al. (Reference Glémin, Gaude, Guillemin, Lourmas, Olivieri and Mignot2005) found a significantly lower population structure at the S-locus compared with microsatellite markers. In A. lyrata, Charlesworth et al. (Reference Charlesworth, Mable, Schierup, Bartolomé and Awadalla2003) did not detect significant genetic differentiation among five populations at the SRK gene, whereas significant genetic structure was found at four of six control genes (see also Wright et al., Reference Wright, Lauga and Charlesworth2003 for F ST at control loci). In A. halleri, F ST at SRK has been found to be threefold lower than at unlinked microsatellite markers (Castric et al., in prep.). In a separate theoretical paper, Schierup et al. (Reference Schierup, Charlesworth and Vekemans2000a) predicted that F ST values for neutral sites flanking a locus subject to balancing selection should increase with increasing recombination rate from the selected locus. We found significantly lower F ST values for the S-locus flanking genes compared with control loci, confirming the theoretical expectations of a reduced population subdivision in genomic regions subject to NFD selection. A similar trend has recently been reported in A. lyrata, where high diversity at two S-locus flanking genes was found to be caused by sequence differences among allelic classes at the S locus, rather than among populations (Kamau et al., Reference Kamau, Charlesworth and Charlesworth2007).
Altogether, our results suggest that strong balancing selection can be considered as a candidate process causing heterogeneity in polymorphism and population genetic structure across genomes, but its effects would only be detected in high-resolution genomic scans because of its rather local influence.
We thank A.-C. Holl and A. Courseaux for technical assistance, P. Saumitou-Laprade for DNA samples from A. halleri, G. Willems for kindly providing the genotypic data for markers TSB2, At4-TC1 and FCA, and E. Kamau and D. Charlesworth for useful discussions. We also thank the two anonymous reviewers and the associate editors for numerous comments on the manuscript. M.V.R. was supported by a postdoctoral grant from CNRS-Environmental Sciences and Sustainable Development department. This study was funded by an ATIP grant from CNRS-Life Science Department, a FEDER fund from the EU, an ARCIR grant from the Région Nord-Pas de Calais, and grant ANR-06-BLAN-0128-01 from the French National Research Agency.
