A microsatellite linkage map for Drosophila montana shows large variation in recombination rates, and a courtship song trait maps to an area of low recombination

Authors


Martin A. Schäfer, Zoologisches Museum, Universität Zürich, Winterthurerstrasse 190, CH-8057 Zürich, Switzerland. Tel.: +41 (0)44 635 4772; fax: +41 (0)44 635 4780; e-mail: m_schaefer_de@yahoo.de

Abstract

Current advances in genetic analysis are opening up our knowledge of the genetics of species differences, but challenges remain, particularly for out-bred natural populations. We constructed a microsatellite-based linkage map for two out-bred lines of Drosophila montana derived from divergent populations by taking advantage of the Drosophila virilis genome and available cytological maps of both species. Although the placement of markers was quite consistent with cytological predictions, the map indicated large heterogeneity in recombination rates along chromosomes. We also performed a quantitative trait locus (QTL) analysis on a courtship song character (carrier frequency), which differs between populations and is subject to strong sexual selection. Linkage mapping yielded two significant QTLs, which explained 3% and 14% of the variation in carrier frequency, respectively. Interestingly, as in other recent studies of traits which can influence speciation, the strongest QTL mapped to a genomic region partly covered by an inversion polymorphism.

Introduction

The genetics of differences between closely related species and subspecies has been an important focus of research since the modern synthesis. Summaries of the genetics of speciation (e.g. Coyne & Orr, 2004) emphasize the extent to which a few species of Drosophila dominate our current knowledge, but fortunately the diversity of study species both within Drosophila and in general, is now increasing in parallel with recent advances in molecular methodology (Bhutkar et al., 2008; Butlin & Ritchie, 2009). Genomic divergence between species and populations is likely to be a patchwork with islands of genetic differentiation within more freely introgressing stretches of the genome (Wu, 2001; Nosil et al., 2009). Such patches can be identified by genome scans as long as genetic markers or quantitative trait loci (QTLs) are still in disequilibrium with loci, which contribute to isolation or are under selection (Via & West, 2008; Wood et al., 2008).

Studies of Drosophila led Dobzhansky (1951) and colleagues to argue that polygenic adaptations could be protected from recombination if they were contained within inversions, effectively allowing larger ‘islands’ of genetic differentiation. Several interspecific QTL studies have recently found that genes encoding for speciation-relevant traits often co-localize to regions of suppressed recombination (e.g. Rieseberg et al., 1999; Williams et al., 2001; Feder et al., 2003), which have stimulated new models of the process (Noor et al., 2001a; Rieseberg, 2001; Kirkpatrick & Barton, 2006). For empirical QTL studies, however, an important corollary to these observations is that low recombination rates can lead to overestimates of the magnitude of QTLs if these regions contain numerous small effect loci, the combined effect of which are easier to detect than solitary QTLs (Noor et al., 2001b). Hence, QTL studies often cross homokaryotypic strains to maximize recombination (e.g. Etges et al., 2007), but which potentially bias studies away from finding differentiated genomic regions.

Drosophila montana, a species belonging to the Drosophila virilis group, is distributed across the northern hemisphere adapting to various kinds of environmental conditions. Recent phylogenetic studies have documented substantial population differentiation across the species range with populations from Asia, Europe and different parts of Northern America forming distinct genetic clades (Mirol et al., 2007). Finnish, Canadian and USA populations also show significant differences in several traits linked with behaviour and local adaptation (e.g. reproductive diapause and cold tolerance; T. Salminen and L. Vesala, personal communication), some of which are not simply correlated with patterns of genetic differentiation at ‘neutral’ markers (Routtu et al., 2007). Furthermore, D. montana has been characterized by bearing a large number of inversions, making this species interesting to study patterns of local recombination and genetic divergence. However, a detailed genetic map of the species has not been available – and the species is only distantly (about 9.0 Ma) related to the well-characterized and sequenced D. virilis (e.g. Nurminsky et al., 1996; Bhutkar et al., 2008).

One of the most interesting phenotypic traits in D. montana is the carrier frequency of male courtship song. Studies have not only found that males from Colorado (USA) produce songs with a much higher carrier frequency than the males from the Finnish and Vancouver (Canada) populations, but also that female preferences for this trait do not co-evolve in a simple straightforward manner (Klappert et al., 2007). In Finnish populations, both field studies and playback experiments implicate the carrier frequency and pulse length of the male courtship song to be the most important song characteristics subject to female choice (Aspi & Hoikkala, 1995; Ritchie et al., 1998). In this population females preferentially mate with males producing high-frequency song (Ritchie et al., 1998) perhaps because a high frequency correlates positively with male condition and offspring survival (Hoikkala et al., 1998, 2008), whereas females from Colorado show no or even negative preferences for males singing at high carrier frequency (Klappert et al., 2007). In comparison with detailed information on the effects of courtship song on female preferences, the genetic control of naturally occurring song variation in Drosophila remains poorly understood, yet critical to understanding its potential influence on species diversification.

In this study, we used 42 microsatellite markers evenly distributed over the genome to construct a linkage map for crosses between lines of D. montana from Finland and Colorado by taking advantage of the D. virilis genome and available cytological maps of both species. Both strains were collected quite recently and were checked for differences in chromosome structure during the course of the study. We also performed a linkage mapping analysis for male song carrier frequency, known to be divergent between populations and subject to strong sexual selection within populations. Our study therefore represents an example of genetic linkage mapping of an out-bred, nonhomokaryotypic cross in a nonmodel species of Drosophila, showing substantial population genetic variation in behaviour, morphology and inversions. Our data indicate large variation in recombination rates across the genome, and that a highly significant QTL for male song carrier frequency co-localizes with a genomic region of suppressed recombination partly covered by a polymorphic inversion.

Methods

Fly strains and crossing scheme

We used two recently collected D. montana strains (O66, collected 2003 from Oulanka in Finland and C13, collected in 2003 in the surroundings of Crested Butte, Colorado) for the construction of the linkage map and the QTL analysis. Both strains maintained a relatively high allelic diversity with a mean number of 2.40 ± 0.21 microsatellite alleles per locus for the Colorado strain and a mean number of 2.36 ± 0.20 alleles for the Finnish strain, respectively (range from 1 to 6 alleles per locus for both populations). Most importantly, these strains differed markedly from each other in song carrier frequency, representing the song types of their home population (Fig. 1). Reciprocal crosses between the parental lines were performed to generate the F1 generation, which in turn was intercrossed in all four combinations to obtain the recombining F2 generation. Crosses were based on multiple individuals to obtain sufficient numbers of offspring. For the genetic map and the QTL analysis we genotyped all parental flies of the F0 generation (= 148) and 422 offspring, which represented the upper and lower 30% extremes of the song carrier frequency distribution of the F2 progeny (total of 846 males).

Figure 1.

 Mean male courtship song carrier frequency (±SE) in the parental lines (F0) and the reciprocal crosses of the first (F1) and second (F2) generation. Parental females are always listed first and numbers in brackets indicate the sample size.

Courtship song analysis

The parental strains and crosses were maintained in vials containing malt medium in continuous light in a culture room (19 ± 1 °C), and the emerging progeny flies were sexed within 3 days of emergence to ensure virginity. Males and females were stored separately in groups of about 10 flies in malt vials until they were sexually mature (21 ± 3 days).

To record the songs produced by male wing vibration, we placed the males individually in a recording chamber with a virgin D. montana female of the same strain or cross. The chambers consisted of petri dishes (diameter 55 mm, height 13 mm) with a nylon net roof and a moistened piece of tissue paper covering the floor. The flies walked upside down on the roof of the chamber and the song of the courting male was recorded by holding the microphone (AKG C 1000 S) directly above the roof. The songs were recorded on a Marantz Professional (model no. 74PDM 502/02B) tape recorder at a temperature of 20 ± 1 °C.

The songs of D. montana males consist of trains of polycyclic sound pulses with a short interval between the pulses (for graphical illustration and discussion of the song structure, see Ritchie et al., 1998). The carrier frequency of the song (FRE) was measured as the highest peak in the frequency spectrum (FFT) for three randomly chosen pulse trains per male, using a Signal Sound Analysis system (Engineering Design, Belmont, MA, USA). For statistical analyses we calculated the mean carrier frequency over three pulse trains of each male.

Microsatellites

To construct a linkage map for D. montana we took advantage of the D. virilis genome sequence and published correspondences between cytological maps of both species (Vieira et al., 1997; Morales-Hojas et al., 2007). In brief, we located D. virilis scaffolds onto the cytological map of Gubenko & Evgen’ev (1984) using the alignment tool BLAT (Kent, 2002), and published markers for which the corresponding DNA sequence is available and that have been previously located on this cytological map. These anchored maps are shown in Fig. S1 of Supporting Information. These maps, together with the information given by Vieira et al. (1997) and Morales-Hojas et al. (2007) for the correspondences of the D. virilis cytological map of Gubenko & Evgen’ev (1984) and the D. montana cytological map of Moorhead (1954), were used to infer the location of D. virilis microsatellites on the D. montana polytene chromosomes, under the assumption that the DNA content per cytological subdivision is roughly equal (see Table S1 of Supporting Information and Fig. 2). Based on these inferences, microsatellite primers were designed using the region of the D. virilis scaffolds of interest, and tested whether they cross-amplify in D. montana. Only informative markers that segregated for alternative alleles in the parental lines were selected for genotyping of the offspring. Most importantly, we selected regions that were cytologically predicted to be roughly evenly spaced along the X chromosome and the four autosomes of D. montana. This procedure yielded a total of 19 informative microsatellites, which segregated among the parental lines. We also used 17 microsatellites that were originally isolated from P1 clones in D. montana (Orsini & Schlötterer, 2004) and another set of six markers that contained microsatellites that were previously described for D. virilis and also successfully cross-amplified in D. montana (Huttunen & Schlötterer, 2002). The cytological location of these markers was inferred using the same procedure as described before. The primer sequences of all microsatellites, the repeat motifs and the inferred cytological positions in D. virilis are listed in Table S1 of Supporting Information. DNA extraction and polymerase chain reaction amplification followed standard protocols given in Huttunen & Schlötterer (2002) and Orsini & Schlötterer (2004). Alleles were visualized by autoradiography and scored manually. Individuals of both parental strains were used as internal size standards on the gels to identify the origin of parental alleles in the mapping population.

Figure 2.

 Recombinational (in cM) and cytological maps (Moorhead, 1954) for the five chromosomes of Drosophila montana. Recombination distances are based on Kosambi’s mapping function. Lines indicate the predicted marker and cytological positions across maps. Letters refer to known inversions present in the Standard, Alaskan-Canadian and giant forms of D. montana (Morales-Hojas et al., 2007).

Chromosomal inversions

Differences in the chromosome structure (inversions) between the parent strains were traced in the salivary gland chromosomes of F1 hybrid larvae from reciprocal mass-mating crosses between the study lines. The vials with fresh larvae were transferred into +4° C for 15 days with extra yeast added into the vials. After the cold treatment the salivary glands of the third instar larvae were dissected in Drosophila ringer solution, transferred into a drop of 45% acetic acid and fixed for 3 min. The salivary glands were then squashed on an object slide to study the structure of polytene chromosomes. The slides were stored for further inspection by freezing them in liquid nitrogen and slipping off the cover lip. After this, the slides were put in 96% cold ethanol and stored in a freezer.

Linkage map and QTL analysis

We assigned markers onto the five chromosomes according to their predicted cytological location (Table S1 of Supporting Information). The relative order and the recombination distances between markers within each of the five chromosomes (in centiMorgan) were calculated with the computer program MapMaker, version 3.0 (Lander et al., 1987) using the Kosambi mapping function to account for interference. Because only males of the F2 generation were genotyped, which are hemizygous for the X chromosome, we used a backcross-coding scheme for the linkage analysis of X chromosomal markers. Grouping was performed with a minimum LOD score (logarithm of the odds to the base 10) of 3.0 and a distance of 50 cM as threshold values. Haldane’s mapping function yielded the same order of markers, but linkage groups were on average 16.6% longer (range from 14.2% to 18.7%). We also calculated separate linkage maps for each of the reciprocal crosses to test for possible differences. However, map orders were basically identical and recombination distances between adjacent pairs of markers were highly correlated (anova: F35,108 = 62.6, < 0.001, R2 = 0.95), so that we used a consensus map for the QTL analysis of male song carrier frequency.

To detect genomic regions associated with variation in song carrier frequency we first performed single marker regressions. To examine Y-chromosomal effects we included a hypothetical, Y-linked marker that segregated between the parental strains in our analysis. QTL localization was completed using composite interval mapping (CIM). CIM tests for QTL effects in a given interval while statistically controlling for other QTLs outside that interval (Jansen & Stam, 1994; Zeng, 1994). Forward and backward regression procedures were applied to screen for significant markers to be used as background in CIM model 6 of the Zmapqtl program as implemented in the QTL Cartographer software (Basten et al., 1997). Forward and backward regressions resulted in the same set of five informative markers, which were used as background (vir28ms, vir103ms, mon7, vir90ms located on the X, second and third chromosomes respectively, plus the hypothetical marker on the Y chromosome; see also Table 1). Likelihood ratios of autosomal QTL were calculated by testing for additive and dominance components, whereas likelihood ratios for X-linked QTL considered the additive component only, because of the hemizygosity of the X chromosome in Drosophila males. Genome-wide significance thresholds were adjusted by permuting the trait data against the marker data 1000 times (Churchill & Doerge, 1994). Redundant microsatellite markers with recombination distances smaller than 3 cM were excluded prior to the QTL analysis. We tested the robustness of CIM by using different numbers of background markers (ranging from 3 to 7) and window sizes (ranging from 5 to 20 cM). The precision of QTL positions was estimated using the 1- and 2-LOD support intervals. For strong QTLs these intervals approximate the 95% and 99% confidence limits, but support intervals tend to underestimate confidence limits for weak QTLs (e.g. Darvasi & Soller, 1997). We further examined if the detected QTL differed between the four crosses by testing for genotype by cross interactions. As none of the interaction terms was significant (all > 0.1), we present the results for the pooled data only. All QTL analyses were performed using the program QTL Cartographer, version 2.5 (Basten et al., 1997) and Map Manager QTX b20 (Manly et al., 2001).

Table 1.   Single marker regressions on carrier frequency (in Hz) of courtship song in Drosophila montana. Only significant associations with P < 0.05 are shown. Markers are ordered in their recombinational position along chromosomes.
MarkerChromosomePR2
  1. BSignificant after sequential Bonferroni correction procedure.

hypYY0.001B0.02
vir103msX0.0470.01
mon30X0.0270.01
mon26X0.0420.01
vir97msX0.0320.01
vir28msX0.0050.02
vir71msX0.0060.02
mon402<0.001B0.05
vir32ms2<0.001B0.06
vir3ms2<0.001B0.09
mon62<0.001B0.11
v11–232<0.001B0.10
vir38ms2<0.001B0.10
mon72<0.001B0.10
vir4ms2<0.001B0.11
vir90ms30.0160.02

Results

Song carrier frequency

Consistent with previous studies (e.g. Klappert et al., 2007; Routtu et al., 2007) males from the Colorado line had a much higher carrier frequency (mean = 269.81, SE ± 3.57) than males originating from the Finnish line (mean = 237.20, SE ± 3.21; Fig. 1). This difference explained 44.2% of total variation in male song carrier frequency (F1,56 = 43.6, P < 0.001), and corresponded to a 1.33-fold difference in standard deviation units between both parental strains. Reciprocal crosses indicated that female origin had a strong influence on variation in male carrier frequency (Fig. 1). F1 males from the reciprocal crosses showed a carrier frequency spectrum significantly shifted towards their own maternal strain (F1,64 = 19.79, P < 0.001, R2 = 0.24), implying X-linked inheritance or maternal effects. Song carrier frequency also differed significantly between the four reciprocal F2 crosses (F3,842 = 10.6, P < 0.001, R2 = 0.04). Pairwise Bonferroni post hoc tests revealed that the progeny of one cross (F1B × F1A; see Fig. 1) had a significantly higher carrier frequency than those from the other three crosses (all: P < 0.01), whereas all remaining comparisons were nonsignificant (all: P > 0.3). However, the difference between the reciprocal crosses was quite small (R2 = 0.04) and the frequency range of the F2 generation was largely intermediate between that of the parental strains and of the F1 progeny.

Linkage map

Recombination mapping of 42 informative microsatellites resulted in fourteen markers located on the X chromosome, eight on chromosome 2, eight on chromosome 3, six on chromosome 4 and six on chromosome 5 (Fig. 2). The X chromosome showed two linkage groups separated by recombination distances greater than 50 cM resulting in a total map length of more than 242.3 cM, whereas the markers on all autosomes clustered in a single linkage group. Within linkage groups the ordering of markers was highly significant over alternative arrangements (log-likelihood ratios from 10−3 to 10−12) when markers with distances greater than 5 cM were considered. Also, the placement of microsatellites by recombination analysis was roughly consistent with that expected from cytological prediction, except for some markers located on the X chromosome and three markers (mon 40, vir34 and mon5) on chromosomes 2, 3 and 5, respectively. The lack of collinearity between the inferred physical map and the placement of these markers on the recombination map could be the result of translocation of the microsatellites, wrong inference and/or segregating inversions which had been present in the individuals used for the crosses, but could not be detected in parent strains during our cytological investigations. The average spacing between markers in the two linkage groups of the X chromosome and the four autosomes was 12.40 cM ± 1.72 SE (range: 0–38.8; = 36).

Inversion on chromosome 2

Cytological observations on the salivary gland chromosomes of F1 hybrid larvae between the parent strains revealed a large chromosomal rearrangement, corresponding to inversion 2Y on chromosome 2 (Morales-Hojas et al., 2007; see Figs 2 and 3). In 15 slides (larvae) examined, the inversion was found to be present in 13 and absent in 2 larvae. Other chromosomes showed no structural differences leading to inversion loops between the parent lines. However, the structure of these chromosomes was clearly observable in only 8–12 slides depending on the chromosome and we cannot reject a possibility that the strains have some other inversions segregating at a low frequency.

Figure 3.

 Example of an inversion loop detected in crosses between Drosophila montana strains from Finland and Colorado. The loop maps to inversion 2Y located on chromosome 2 (see Fig. 2).

QTL mapping of carrier frequency

Single marker regression analysis showed that of the 42 microsatellites scored, 15 were significantly associated with variation in song carrier frequency (P < 0.05; Table 1). Six of the significant marker-trait regressions involved microsatellites located on the X chromosome and seven on the 2nd chromosome. Only one significant association was found on chromosome 3, whereas none of the markers located on chromosomes 4 and 5 had any detectable association with carrier frequency. In addition to the genotyped markers, the hypothetical Y-chromosomal marker significantly correlated with carrier frequency. After sequential Bonferroni correction, only the Y-linked marker and the seven associations on chromosome 2 remained statistically significant.

The CIM revealed two significant QTL on the X chromosome and the second chromosome, which explained 3% and 14% of the total variance in carrier frequency, respectively (Fig. 4 and Table 2). In addition, there were several suggestive QTL on the X chromosome and chromosome 3, all of which, however, were nonsignificant. Also, when controlling for the same genetic background in CIM the Y-chromosomal effect was far below the genome-wide significance threshold (Fig. 4). Genetic components of the QTL on the second chromosome were largely additive and positive in sign, whereas the significant QTL on the X chromosome was transgressive with an effect in the opposite direction to that predicted from the phenotypic means of the parental strains (Fig. 5 and Table 2). The significant QTL on chromosome 2 was located in a chromosomal region of very low recombination, partly covered by the 2Y inversion polymorphic within the Finnish D. montana population (Morales-Hojas et al., 2007; see Figs 2 and 3; Table 2).

Figure 4.

 Composite interval mapping of song carrier frequency against the marker positions for the two linkage groups (LG) of the X-chromosome, and the four autosomes in Drosophila montana. The log-likelihood ratio for a hypothetical Y-chromosomal marker after controlling for the same background is indicated by an asterisk. The genome-wide significance threshold corresponds to a log-likelihood ratio of 13.46.

Table 2.   Quantitative trait loci (QTL) locations (in cM) and effects (in Hz) on variation of courtship song carrier frequency in Drosophila montana.
QTLChromosomePeak position1-LOD; 2-LOD support intervalsAdditivity†Dominance†LR†R2
  1. †Given for the peak position of the QTL.

1X (LG2)182177.4–196.3; 164.1–205.4−6.3115.80.03
225547.8–60.3; 47.8–60.315.642.4362.90.14
Figure 5.

 Effects of representative microsatellite markers of two significant quantitative trait loci located on the X chromosome (cX) and the second chromosome (c2) on courtship song carrier frequency in Drosophila montana. O66 and C13 refer to the parental origin of alleles in the mapping population and the sample sizes are given in brackets.

Discussion

Our linkage mapping study of an out-bred cross between two geographically isolated populations of D. montana leads us to three conclusions. First, the linkage map is roughly consistent with cytological predictions, but indicates large heterogeneity in recombination rates. Second, differentiation in song carrier frequency involves one QTL with relatively large effect located in a genomic region of very low recombination partly covered by a polymorphic inversion, and probably a number of loci with much smaller effects. Third, although female origin had a strong influence on the song carrier frequency distribution of F1 male progeny, suggestive of X-linked inheritance, we detected only one transgressive QTL of minor effect on the X chromosome. In the following, we first discuss sources of the heterogeneity in recombination rates and then we focus on the implications of our QTL mapping results for understanding evolutionary divergence in courtship song.

Our linkage map, from allelic variation at 42 microsatellite loci, represents the first molecular marker map constructed for D. montana. The placement of markers was found to be largely consistent with cytological predictions derived from the D. virilis comparison, which highlights the strength of combining comparative chromosome maps with genomic information for using linkage-mapping approaches in species with no sequenced genomes. However, our linkage map also indicated large heterogeneity in recombination distances along chromosomes, which cannot be explained by uneven spacing of markers. The map for the third chromosome appeared to be roughly of comparable size (110.3 cM) to the D. virilis homologue (145 cM; Gubenko & Evgen’ev, 1984), whereas in chromosomes 2, 4 and 5 recombination was absent or very low over large physical distances. This resulted in much smaller genetic maps (ranging from 27.1 cM up to 62.5 cM; Fig. 2) compared with the maps derived for the homologous chromosomes in D. virilis (range from 198 cM up to 257.6 cM; Gubenko & Evgen’ev, 1984). The X chromosome represented the other extreme: parts of this chromosome showed high recombination rates, which led to two independent linkage groups and a total map length of more than about 242.3 cM (Fig. 2). In comparison, the D. virilis homologue shows a map length of only about 170 cM (Gubenko & Evgen’ev, 1984).

The lack of recombination in chromosome 2 can, to a large extent, be explained by a paracentric inversion 2Y (Figs 2 and 3). This inversion has earlier been detected in Finnish populations, but not in North American populations (Morales-Hojas et al., 2007), and so it is likely to be segregating in the Finnish QTL line. There are also several other inversions segregating within and between Finnish and/or North American populations in other chromosomes (Vieira et al., 1997; Morales-Hojas et al., 2007), but the study lines did not show differences in these inversions (the presence of these inversions at low frequency cannot, however, be excluded). Reduced recombination on large chromosomal areas in hybrid chromosomes could be a result of several factors in addition to inversions. One explanation could be genomic incompatibility between the parent lines, as the divergence time between the Scandinavian and North American clades of D. montana has been estimated to be 450 000–900 000 years (Mirol et al., 2007), and populations of this species from different continents also show some degree of post-zygotic isolation (J. Jennings, personal communication). Another possibility would be the presence of specific mutations reducing crossing over. There is, for example, an increasing amount of evidence of genetic variation in crossover frequencies in natural populations of Drosophila melanogaster (Tatsuta & Takano-Shimizu, 2009). Recombination has also been found to be reduced on chromatin dense regions near the centromeres, but this may not be an important factor in telocentric chromosomes like the ones of D. montana. Finally, lack of recombination in some genomic regions (owing to nonpairing) may lead to an increased rate of recombination in collinear (paired) regions of the genome, like on the X chromosome in the present study. For example, in the Drosophila pseudoobscuraDrosophila persimilis cross, higher recombination rate in hybrids compared with that in either of the parental species has been suggested to be owing to fixed chromosomal inversions between these two species causing interchromosomal effects (Ortiz-Barrientos et al., 2006; see also Schultz & Redfield, 1951).

The carrier frequency of male song is subject to strong sexual selection in at least the Finnish D. montana population, and this trait also varies between populations (Klappert et al., 2007; Routtu et al., 2007). This could favour rapid divergence of sex-linked genes involved in the song (Charlesworth et al., 1987). In our study, the reciprocal F1 crosses showed large differences, consistent with an X-linked effect (Fig. 1), and so we expected to find a QTL on the X chromosome. The only QTL explaining more than 10% of the phenotypic variation was, however, found on the second chromosome. This QTL could be either one relatively large effect gene, or several smaller loci with ‘pooled’ effects owing to the low recombination in this region. There was also a smaller QTL on the X chromosome, but it explained only 3% of the phenotypic variation. The additive effect of the X chromosomal QTL was in fact transgressive (Table 2 and Fig. 5), and so it is unlikely to have evolved under direct selection (Orr, 1998) and cannot explain the reciprocal difference seen in the F1 generation. The contrast between the F1 crosses and the QTL analysis is surprising and implies that either numerous QTLs of individual, small but cumulatively large effect remained undetected (i.e. owing to low statistical power associated with the high recombination rates and/or the presence of multiple alleles in the studied lines), or that maternal effects explain the pattern. Diallelic crosses between inbred D. montana strains suggested that genes influencing carrier frequency are predominantly autosomal (Aspi, 2000; Suvanto et al., 2000), although these were performed using flies of different Japanese, Finnish and Alaskan origin, involving no high-frequency singers from Colorado.

One central question concerning the genetic basis of reproductive isolation is whether species differences evolve by fixation of genetic variants segregating within species or whether different loci are involved (cf. Orr, 2001; Stern & Orgogozo, 2009). Although there are studies showing that intraspecific and interspecific QTL positions sometimes overlap (e.g. Nuzhdin & Reiwitch, 2000; Kopp et al., 2003; Tatsuta & Takano-Shimizu, 2006), there is little if any supporting evidence of this for song characteristics in Drosophila. Rather, QTLs for traits potentially influencing reproductive isolation show little overlap between intraspecific and interspecific studies. For example, Gleason et al. (2002) found none of three QTLs for interpulse interval in D. melanogaster to correspond with QTL positions identified in hybrid crosses between Drosophila simulans and Drosophila sechellia or with candidate genes predicted from mutational analysis (Gleason & Ritchie, 2004). Similar findings have been reported between D. virilis and its close relatives (Huttunen et al., 2004). Using hybrids between D. virilis and Drosophila littoralisHoikkala et al. (2000) found a highly significant QTL affecting song carrier frequency on the proximate end of the X chromosome near the centromere, whereas no QTL was detected on the second chromosome as seen in our study. However, differences between the species were more pronounced in temporal song traits, like the lengths of the sound pulses and the pauses between them, than in song frequency. A lack of correspondence between interspecific and intraspecific genetic variation might mean that different processes influence speciation and sexual selection (Arbuthnott, 2009), or perhaps more likely that the genes influencing complex polygenic behavioural traits are haphazard and show low congruence between related species.

The fact that the QTL with the strongest effect on song carrier frequency was located in a region including an inversion polymorphism may be owing to the fact that the chances of detecting QTLs is greater in regions of low recombination (Noor et al., 2001b). However, a number of recent QTL studies on a variety of traits including pre- and post-mating isolation have suggested that traits often co-localize with inversions. For example, Noor et al. (2001a) describe how loci influencing hybrid sterility and female mating preferences co-localize with fixed inversions (and flanking sequences which also show signs of reduced recombination) between Drosophila pseudoobscura and Drosophila persimilis. Also, Williams et al. (2001) found that song differences between the same species were strongly associated with two inversions, which differ between these species. There are several scenarios under which linkage of ‘speciation genes’ within inversions could occur. Dobzhansky (1951) originally proposed that reduced recombination within heterozygotes can protect co-adapted gene complexes (and earlier models proposed that underdominance of the inversions contribute to isolation). If inversions become fixed (e.g. during allopatry), then genetic differences will be maintained against gene flow in such regions (Noor et al., 2001b; Rieseberg, 2001). Alternatively, if alleles promoting local adaptation arise in regions of low recombination, they are more likely to become established (Kirkpatrick & Barton, 2006). Irrespective of the mechanism involved, the importance of genomic areas of suppressed recombination in species diversification appears to become becoming firmly established in a wide range of species.

Acknowledgments

This work was funded by the EU Research Training Network (HPRN-CT-2002–00266). The authors thank Claus Vogl for statistical advice and valuable discussions and members of the EU RTN for helpful comments on the project.

Ancillary