Mike Ritchie Department of Enviromental and Evolutionary Biology, University of St Andrews, Bute Medical Building, St Andrews, Fife, KY16 9TS, UK. Tel.: +44 133 4463495; fax: +44 133 4463600; e-mail: email@example.com
For many years it was thought that Drosophila melanogaster was relatively panmictic, without differentiation in the Mate Recognition System. Recent studies have demonstrated that flies from Africa vary in pheromones and assortative mating. Strains from Zimbabwe show strong sexual isolation from others. We show that the interpulse interval (IPI) of courtship song, an important mating signal, is unusually short among African flies. Zimbabwean flies have the shortest IPI, but there is no correlation with assortative mating, suggesting little direct role in sexual isolation. Chromosome replacements show that the IPI difference is largely due to genes on chromosome III, with significant interactions involving other chromosomes. Several traits potentially influencing sexual isolation among the melanogaster group of Drosophila seem to be localized to this chromosome. A concentration of important genetic differences might mean that the interaction effects reflect secondary coadaptation of the genetic background to changes associated with chromosome III.
Differences in courtship behaviour can play an important role in preventing gene flow between species ( Butlin & Ritchie, 1994). However, the way in which courtship differences arise and their actual role in the process of speciation is much debated ( Dobzhansky, 1951; Coyne & Orr, 1989; Paterson, 1993; Price, 1998). Analysis of intraspecific geographical variation in courtship signals is of great interest because such variation provides a means of investigating the initial stages of divergence and often allows study of the genetic changes involved.
Initial studies of variation in the mating behaviour of Drosophila melanogaster suggested that the Mate Recognition System (MRS) of this species was relatively invariant on a global scale, with little evidence of divergence between populations ( Henderson & Lambert, 1982). More recently this view has been questioned ( Marín, 1994) and there is a growing body of evidence suggesting that strains from Africa show variation in the MRS ( Welbergen et al., 1992 ; Paillette et al., 1993 ). There are also unexpectedly high levels of molecular variation and population structuring among populations within Africa ( Begun & Aquadro, 1993). Such studies have culminated in the discovery of lines of D. melanogaster from two populations in Zimbabwe (the Z strains) which show extreme premating isolation from flies from Europe, North America and North Africa (Cosmopolitan or M strains) and also a lesser degree of premating isolation from strains from the rest of Southern Africa including flies from Southern Zimbabwe ( Wu et al., 1995 ; Hollocher et al., 1997a ). Despite the high level of premating isolation, there is no apparent postmating isolation, with Z strain flies producing perfectly viable offspring in a variety of crosses with Cosmopolitan flies ( Lawhorne, 1998). This has led to Wu & Hollocher (1998) describing the extreme Z strains as a potential example of the process of speciation being caught in flagrante delicto.
The existence of strains of D. melanogaster that show extreme premating isolation, but produce viable offspring, opens up exciting possibilities for investigation of the genetic basis of assortative mating, and such studies are underway ( Wu et al., 1995 ; Hollocher et al., 1997b ). Genes carried on chromosome III play the major role in causing the assortative mating. However, the traits involved in the premating isolation have not yet been identified. Determination of the traits responsible and their genetics would be invaluable in understanding the processes that have driven this divergence.
Assortative mating in Drosophila can be caused by a number of factors including divergence associated with habitat choice ( Etges, 1998), male ( Scott, 1994) or female ( Cobb & Jallon, 1990; Coyne et al., 1994 ) cuticular hydrocarbons, vigour and mating speed interactions ( Welbergen et al., 1992 ) and courtship songs ( Kyriacou & Hall, 1986; Tomaru et al., 1995 ). There is known to be intraspecific variation in the cuticular hydrocarbon profiles of D. melanogaster, with flies from sub-Saharan Africa and the Caribbean having different pheromone blends from Cosmopolitan strains ( Jallon, 1984; Jallon & David, 1987; Cobb & Jallon, 1990; Ferveur et al., 1996 ). In sub-Saharan males, the 7-tricosene component is largely replaced by the 7-pentacosene component. This pheromone difference has been shown to potentially influence assortative mating as Canton-S females mate more slowly with males with 7-pentacosene (the effect is eliminated by adding synthetic 7-tricosene to the mating chamber, Scott, 1994). In Afro-Caribbean females, 5,9-heptacosa- diene largely replaces the 7,11-heptacosadiene compo- nent, and again there is some evidence that this correlates with mating speed ( Ferveur et al., 1996 ; but see Coyne et al., 1999 ). However, these pheromone blend differences seem unlikely to explain the assortative mating exhibited by the Z and M strains because Z strains show at least some assortative mating with other African flies which are likely to possess the Afro-Caribbean pheromone blends ( Hollocher et al., 1997a ).
This paper describes variation in the IPI of courtship song of D. melanogaster from Africa, including several of the lines from Zimbabwe which show assortative mating with lines from elsewhere. We also examine covariation of IPI divergence with the strength of assortative mating, allowing us to infer whether song variation contributes to assortment. In addition, the genetic basis of song divergence was examined by measuring the IPI of a series of chromosome replacement lines between one of the Z strains and a Cosmopolitan strain.
Materials and methods
The courtship songs of males from 22 isofemale lines of D. melanogaster were recorded. Ten of these lines were from the two Zimbabwe populations that show extreme sexual isolation (four from Sengwa Wildlife reserve (ZS lines), and six from near Harare (ZH lines)) from M strains (as described in Hollocher et al., 1997a ). Two lines were from Okavanga in Botswana (OK lines) ( Hollocher et al., 1997a ), two from Matobo National Park, Zimbabwe (MM lines), and one each from Chobe National Park, Botswana (CH5), Victoria Falls National Park, Zimbabwe (VF15), and Kariba, Zimbabwe (K16) (as described in Dubill, 1996). The remaining five lines consisted of one each from Ivory Coast (Tai), France (FrV-3) and California (HGCA) (the same three lines used in Hollocher et al., 1997a ), Italy (C80) and Spain (SM27). Prior to the experiment, all lines were cultured at the University of St Andrews at 25 °C, 12 h light/dark cycle, for at least three generations. Most of the African collections were made in 1995, though the two extreme Z populations were collected in 1990 (experiments were carried out in 1998).
To obtain males for recording, five females and one male were taken from stock cultures and placed in a 75 × 23-mm vial with standard cornmeal medium. The flies were left to oviposit for 48 h only, leading to low densities of larvae during development. Once adults began to emerge, males were collected under CO2 anaesthesia within 12 h of emergence and placed individually in 95 × 16.5-mm vials containing medium where they were kept until recorded. Collecting was limited to days 2–4 after emergence began.
To record song, 4- to 8-day-old males were placed with a silenced (wingless) female of arbitrary strain (female strain does not influence male IPI, Ritchie, unpublished observation). Once courtship began, 5 min of song was recorded using a customized ‘Insectavox’ microphone ( Gorczyca & Hall, 1987) and a Marantz CP430 cassette recorder. Temperature inside the insectavox was noted (±0.1 °C) at the beginning and the end of the recording period. The song was digitized using a Cambridge Electronic Design 1401 A/D converter (at 2 kHz following bandpass filtering at 350 Hz to 1 kHz) and mean IPI calculated using the ‘Spike2’ software package (Copyright C.E.D) and custom written programs. This procedure has been checked for accuracy (see Ritchie & Kyriacou, 1994).
Between five and eight males were recorded from each vial, and two replicate vials were used for each line. However, any song that did not contain at least 50 IPIs was discarded. Recordings were carried out in three separate blocks, with each line limited to a single block. It was not possible to record each line in all blocks but, in order to monitor possible systematic differences in IPI measurements between blocks, a control line, C80, was recorded in each block.
A nested ANOVA ( Sokal & Rohlf, 1995; Ch.10) was used to examine the level at which variability in the IPI data occurred. Nesting levels were geographical region (Zimbabwe, elsewhere in Africa and Cosmopolitan, this division reflecting current knowledge of variation in behavioural traits), isofemale line and vial. The significance of each level of nesting is tested using the variance among the next lower level as the denominator mean square (e.g. the variance among lines is used to test whether variation among regions is significant, not the error (within vial) mean square).
Data on the sexual isolation of the lines used were taken from Hollocher et al. (1997a ) and Dubill (1996). These data are based on a series of multiple-choice mating trials and provide two measures of premating isolation (discrimination indices) for each line. DI(U,M) measures the degree of assortative mating between the focal line (U) and a reference Cosmopolitan line (M) whilst DI(Z,U) measures the degree of assortative mating between the focal line (U) and a reference Zimbabwean line (Z). In both cases values range from about –1 to 5, with higher values indicating a higher degree of assortative mating (and hence premating isolation). Thus a typical Z line has a high DI(U,M) (shows strong assortative mating when tested with Cosmopolitan flies), but a low DI(Z,U) (shows little assortment when tested with Z flies). Further details can be found in Hollocher et al. (1997a ).
To examine the genetic basis of divergence in mean IPI, the songs of a series of chromosome replacement lines between one of the Z lines (ZS30) and an M line from the USA (HGCA) were recorded. These lines match those used in the behavioural analyses performed by Hollocher et al. (1997b ). Chromosome replacement lines are lines in which one or more pairs of chromosomes from one of the parental lines have been swapped for the equivalent chromosome pair from the other parental line (see Ashburner, 1989, pp. 543–545, for general techniques; Hollocher et al., 1997b , for details of the particular lines used). Lines were not made co-isogenic before these crosses were carried out, which will have countered inbreeding.
Retaining the classification of Hollocher et al. (1997b ), each line is designated by a three-letter code indicating the origin of its X, second (II) and third (III) chromosomes, respectively, with Z used for ZS30-derived, and M for HGCA-derived chromosomes. For example, the line ZMZ would have X and III from ZS30, and II from HGCA. The origin of the fourth or ‘Dot’ chromosome was not monitored for any of these lines, but as it contains less than 0.5% of the genome of D. melanogaster it was unlikely to have an important effect. Males from all six chromosome replacement lines (representing all possible homozygous combinations of the three sets of chromosomes) were recorded, as were males from both parental lines.
Males were obtained and recorded using the same procedures as before. Recording was carried out in three blocks. Two replicate vials of each chromosome replacement line and also the HGCA parental line were set up in each block and between five and 10 males from each vial were recorded. Unfortunately, due to culturing problems, the Z strain parental line, ZS30, was only recorded in one of the blocks.
The genetics of the IPI differences between the chromosome replacement lines was examined by fitting a mixed linear model to our data using Restricted Maximum Likelihood (REML) with Proc Mixed in SAS ( Littell et al., 1996 ). Each of the three main chromosome effects (X, II and III), along with all possible interactions between them, were fitted as fixed effects, whilst recording block and rearing vial were fitted as random effects. Significance of the fixed effects was determined using Wald’s F statistics with the between vial within line variance component as the denominator mean square. Denominator degrees of freedom were adjusted using Satterthwaite’s approximation ( Sokal & Rohlf, 1995, p. 282) due to the unbalanced nature of the design. All means and standard errors reported in this paper are calculated using the vial means as the independent data points.
Correcting IPI for temperature
IPI varies with temperature, decreasing as a fly warms up ( Shorey, 1962), so all IPI measures need to be adjusted to a standard temperature. For both experiments an analysis of covariance was used to examine the relationship between IPI and recording temperature (taken as the mean of the start and finish temperature). In each case the regression coefficients did not differ among the strains or lines used (geographical survey: ANOVA; strain by temperature interaction, F21,162=1.48, P=0.10; genetic analysis: ANOVA; line by temperature interaction, F7,215=1.60, P=0.14) and so these were used to correct all IPI values to 25 °C (geographical survey: b=–1.03, F1,183=229.7, P < 0.001; genetic analysis: b=–1.10, F1,222=523.68, P < 0.001). These corrected IPI values were used in all further analyses.
Mean IPI of the control line (C80) did not differ between recording blocks ( ANOVA; F2,25=0.33, P=0.7) suggesting that block effects are small in this experiment. Thus, the data from the three recording blocks were pooled for analysis, and C80 data from blocks two and three discarded to maintain balance.
Mean IPI did not differ between replicate vials from the same line, but differed both between lines from the same region, and also between regions (Table 1, Fig. 2; percentage variance component for each source: region 87%, line within region 11%, vial within line 0%, flies within vials 2%). In general, the lines from the two Z populations had the shortest IPI (mean ± SE=29.62 ± 0.12), and non-African lines the longest (32.86 ± 0.57) with the other African lines intermediate (31.06 ± 0.15). Thus there is very significant geographical patterning of IPI, and this broadly corresponds to the pattern of premating isolation observed by Hollocher et al. (1997a ). However, at the level of individual lines there was no relationship between either measure of sexual discrimination and mean IPI ( Fig. 3; DI(U,M): r=0.003, N=20, P=0.99; DI(Z, U): r=0.37, N=19, P=0.12). So mean IPI is not a good predictor of the degree of premating isolation. We therefore conclude that Africa flies show differences in their mean IPI from other, Cosmopolitan, strains of D. melanogaster ( Ritchie et al., 1994 ), but that variation within Africa does not support the conclusion that the sexual isolation between flies from Zimbabwe and elsewhere is due to a pronounced difference in song.
Table 1. Analysis of variance of mean IPI for the geographical survey. The significance of each term was determined using the mean square of the level below as the denominator of the F ratio (terms marked with an asterisk have been adjusted using Satterthwaite’s approximation due to the unbalanced nested nature of the design; Sokal & Rohlf, 1995, p. 282).
Table 2 shows the predicted means from the REML analysis for each chromosome separately (i.e. pooled over all possible genetic backgrounds), as well as the predicted means for each individual genotype (i.e. replacement line). Overall, chromosome III had the largest main effect on mean IPI, with HGCA-derived chromosomes lengthening mean IPI by about 2.5 ms relative to ZS30-derived chromosomes (F1,31.7=86.23, P < 0.001). The effect of chromosome II was smaller and, surprisingly, in the opposite direction with the HGCA second chromosomes shortening mean IPI by about 1 ms (F1,32.7=7.63, P=0.01). Substitution of the X chromosome had no overall effect on mean IPI (F1,30.9=0.50, P=0.48). However, the presence of large interactions between the effects of all three chromosomes (X by II by III interaction; F1,31.1=8.13, P=0.01, X by II interaction; F1,31.4=6.94, P=0.01; all other interactions nonsignificant) makes interpretation of these effects difficult, as the actual effect of substituting each chromosome depends on the genetic background in which the substitution occurs. In particular, inspection of the individual genotype means (Table 2) shows that both the magnitude and the direction of the effect of chromosome II depends on the other chromosomes present, whilst the effect of chromosome III is always in the same direction, but does vary in magnitude. Thus, the difference in IPI between ZS30 and HGCA appears to be mainly controlled by genes on III, but with large interactions involving genes on II. The X chromosome has no, or only a minor, role in the difference. Although we have only studied a single pair of lines, the fact that the genetic architecture of differences in IPI is qualitatively similar to that of assortative mating ( Hollocher et al., 1997b ) suggests this may reflect general differences between Z and M strains.
Table 2. Predicted means from REML analysis of the chromosome replacement lines: (a) predicted means of individuals with either the ZS30-derived (Z) chromosomes or HGCA-derived (M) chromosomes at each site (pooled over all possible combinations of chromosomes at other sites); (b) predicted means for each of the eight possible genotypes represented by the two parental lines and the six chromosome replacement lines.
For many years it has been believed that D. melanogaster as a species is relatively panmictic ( Powell, 1997). There is clinal variation in allozymes and morphology, but on the whole Fst values are low ( Singh & Long, 1992), and significant behavioural differentiation lacking ( Henderson & Lambert, 1982; Lambert & Henderson, 1982; van den Berg et al., 1984 ). Recent detailed study has shown much more population substructure than previously expected, especially among African strains ( Begun & Aquadro, 1993). The current study supports suggestions that the Mate Recognition System of African strains may also be more variable than previous studies have implied ( Paillette et al., 1993 ; Wu et al., 1995 ). We have found that flies from sub-Saharan Africa are characterized by unusually short mean IPI. The two populations from Zimbabwe, previously shown to have strong sexual isolation from Cosmopolitan strains, have the shortest IPI with many lines having an IPI below 30 ms. Cosmopolitan D. melanogaster typically have a mean IPI longer than 32 ms, and artificial selection experiments have not succeeded in shortening the IPI of Cosmopolitan strains to the levels observed among Z lines ( Ritchie & Kyriacou, 1996; Pugh, 1997). However, shorter IPIs are not limited to the strongly sexually isolated Z lines, with many of the lines from Southern Africa and the line from Ivory Coast having IPIs in the intermediate range 30–32 ms. Thus, courtship song in D. melanogaster is far more variable than previously thought ( Ritchie et al., 1994 ).
The overall pattern of geographical variation for IPI observed in this study is in general accordance with that of sexual isolation described by Hollocher et al. (1997a ), with flies from two Z populations having the most divergent courtship song and the most extreme assortative mating with flies from outside Africa, and flies from other parts of Southern Africa having an intermediate IPI value and an intermediate level of assortative mating. However, this pattern is not supported by the more detailed analysis which shows no relationship between the individual IPI value measured for each line, and either measure of assortative mating obtained by Hollocher et al. (1997a ). Thus, although the IPI divergence may play some role in the overall pattern of assortative mating, it is certainly not the major factor.
It is notable that the covariance between the Afro-Caribbean pheromone blends ( Cobb & Jallon, 1990; Ferveur et al., 1996 ) and low IPI is probably quite high. Ferveur et al. (1996 ) show that the predominant pheromone component of cosmopolitan female D. melanogaster, 7,11-heptacosadiene, is largely replaced by 5,9-heptacosadiene in females from sub-Saharan Africa (and the Caribbean, where it is thought to have been introduced from Africa via the slave trade). D. melanogaster originated in Africa ( Lachaise et al., 1988 ; Powell, 1997), so it is likely that the population substructure now being described in the species represents ancient polymorphisms around Zimbabwe ( Begun & Aquadro, 1993) contrasting with more genetically depauperate Cosmopolitan strains, perhaps originating via transhumance. The song differentiation is on a larger geographical scale than the behavioural differences contributing to sexual isolation between Z and M races. It seems likely that the song and pheromonal blend differences alone do not have a major effect on assortative mating, though more detailed studies of assortative mating between non-Zimbabwean African or Caribbean with Cosmopolitan flies might be rewarding ( Scott, 1994; Ferveur et al., 1996 ; Coyne et al., 1999 ). Significant sexual isolation is also found in mating tests among flies from Brazzaville ( Paillette et al., 1993 ).
The results from the chromosome replacement lines suggest a complex genetic basis for the courtship song difference between the strains, with chromosome III having the largest effect, but with significant epistatic interactions between all three chromosomes pairs. The X lacks any main effect, and chromosome II actually acts in the opposite direction to the difference between the strains. This architecture contrasts markedly with other studies of the genetic basis of courtship song, in both D. melanogaster and other species of Drosophila ( Ritchie & Phillips, 1998). Using conventional crossing schemes, Ritchie et al. (1994 ) found that difference in IPI between an Italian and Greek strain of D. melanogaster could be explained by an additive polygenic model, and a similar result was found by Cowling (1980) for a variety of laboratory strains of D. melanogaster. Pugh & Ritchie (1996) found that differences in courtship song between D. simulans and D. mauritiana were due to the additive action of genes spread throughout the genome, and variation within a population of D. melanogaster was similar ( Ritchie & Kyriacou, 1996). Chromosome replacement lines are probably a more powerful technique for disentangling main effects and interactions as the genetic background can be standardized more easily than in conventional crosses. However, there still remains the possibility that the unusual effect of chromosome II found here is a property of these two lines rather than a general difference between Z and M strains.
That a main effect of chromosome III with strong interactions was found is highly intriguing given that a number of studies in the literature also suggest that genes on this chromosome play a role in the other differences between African and Cosmopolitan flies. The 5,9-heptacosadiene/7,11-heptacosadiene pheromone blend polymorphism of females is mainly determined by chromosome III ( Ferveur et al., 1996 ), and a single candidate gene here has recently been identified ( Coyne et al., 1999 ). The major differences in blend components between female D. simulans and D. sechellia which affect sexual isolation ( Coyne et al., 1994 ) are also linked to chromosome III. Canton-S females’ preference for the 7-tricosene pheromone blend involves genes on the third chromosome ( Scott, 1994). Genes influencing the sexual isolation between Z and M strains of D. melanogaster are also predominantly found on III ( Hollocher et al., 1997b ), though our analysis of covariance among isofemale lines shows this to be linkage rather than causation. If there is a concentration of genetic differences influencing sexual signalling on one chromosome, interaction effects such as we have found might represent secondary coadaptation of the X and chromosome II to major changes associated with chromosome III.
If variation in courtship song or pheromonal blend are unlikely to be responsible for the assortative mating between Z and M strains, what is? Time did not permit detailed study of many aspects of courtship other than mean IPI in most of the strains we examined. For a limited sample, we examined sine song frequency and the period of the cycle in mean IPI ( Kyriacou & Hall, 1986; Alt et al., 1998 ), with no obvious differences being found. A preliminary observational study of the courtship repertoire of individual pairs of flies from two strains of each type revealed that Z flies spend a greater proportion of courtship during the ‘Orientation’ stage (approximately 36% of focal samples involved this behaviour, vs. 20% in M strains, there were no differences in the incidence of wing vibration, touching, licking, attempted copulation or mating speeds; behaviours defined as in Welbergen et al., 1992 ). This implies that an extra stage of signal exchange is occurring between these flies early in courtship, but what these signals are remains unsolved. Sexual isolation between Z and M flies persists under different temperature and lighting regimes ( Wu et al., 1995 ). It seems likely that detailed characterization of the genetics involved, beyond that already available ( Hollocher et al., 1997b ), will require further clarification of the signals and preferences contributing to the sexual isolation.
This study adds to the increasing body of evidence showing that components of the Mate Recognition System of D. melanogaster are far more variable than previously thought, largely due to variation present in flies from sub-Saharan Africa ( Paillette et al., 1993 ; Wu et al., 1995 ; Ferveur et al., 1996 ). This seems to be a common conclusion to studies of the variability of courtship behaviour in a range of organisms ( Butlin, 1993; Herring & Verrell, 1996), and raises severe doubt about the effectiveness of stabilizing selection to constrain geographical variation in mating signals. However, the behavioural and evolutionary significance of the variation and the forces responsible for the divergence remain to be elucidated.
This work was funded by the Natural Environment Research Council, UK (grant GR9/02910 to M.G.R.). H.H. is supported by grants from the Alfred P. Sloan and National Science Foundations. Adrian Pugh, Tanya Hamill and Jenny Gleason provided important assistance with the research. Jenny Gleason and two reviewers gave helpful comments on the manuscript. We would also like to thank C. F. Aquadro, T. Mutangadura and A. J. Dubill for their help in securing many of the lines used in this study and Margaret MacKinnon for statistical advice.