A direct experimental test of founder-flush effects on the evolutionary potential for assortative mating

Transcription

1 A direct experimental test of founder-flush effects on the evolutionary potential for assortative mating J.L.REGAN,*L.M.MEFFERT &E.H.BRYANTà *Department of Biological Sciences, University of Southern Mississippi, Hattiesburg, MS, USA Department of Ecology and Evolutionary Biology, Rice University, Houston, TX, USA àdepartment of Biology and Biochemistry, University of Houston, Houston, TX, USA Keywords: artificial selection; assortative mating; bottlenecks; correlated responses; courtship behaviour; founder-flush speciation; reproductive isolation. Abstract Founder-flush speciation models propose that population bottlenecks can enhance evolutionary potential for reproductive isolation. To test this prediction, we subjected bottlenecked (three-pair founder-flush) and nonbottlenecked populations of the housefly to 18 generations of selection for assortative mating. After the selection regime, we analysed videotaped courtship bouts in these lines to identify correlated responses to the selection protocol. The realized heritabilities for assortative mating for both the bottlenecked and nonbottlenecked treatments were very low, but still significant. The founder-flush populations had thus responded to selection as well as the nonbottlenecked populations, although not significantly greater (i.e. total increases in assortative mating were 9.6 and 8.6%, respectively). Multivariate analyses on the courtship repertoires found that, although both bottlenecked and nonbottlenecked treatments attained similar levels of assortative mating, the treatments exhibited different evolutionary solutions in their correlated responses. Specifically, the bottlenecked lines demonstrated a significantly more diverse set of evolutionary trajectories (i.e. significant shifts along the second principal component for courtship). This suggests that the bottlenecked lines had greater potential for the evolution of novel phenotypes as predicted by founder-induced speciation models. Our results, however, cannot distinguish whether the more variable evolutionary responses resulted from increased heritabilities in courtship components, reduced potential to follow the convergent evolutionary trajectories noted for the nonbottlenecked lines, or some combination of both general processes in determining the resultant multivariate phenotype. Introduction The evolution of reproductive behaviour has the potential to lead to speciation by generating ethological isolation. Reinforcement and mate choice models have shown that selection for reproductive isolation could drive divergence in courtship repertoire (Dobzhansky, 1951; Lande, 1981; Kirkpatrick, 1982; Butlin, 1989; see Coyne & Orr, 1997). These models have been supported, Correspondence: Jennifer L. Regan, Department of Biological Sciences, University of Southern Mississippi, Hattiesburg, MS 39406, USA. Tel.: ; fax: ; jennifer.regan@usm.edu in general, by experiments that selected for assortative mating (Koopman, 1950; Kessler, 1966; Dobzhansky & Pavlovsky, 1971; Crossley, 1974; Hostert, 1997; but see Rice & Hostert, 1993). However, a number of these experiments focused on enhancing the isolation already present among different species or subspecies (Koopman, 1950; Kessler, 1966; Dobzhansky & Pavlovsky, 1971), as opposed to generating incipient isolation. In a more conservative approach, Crossley (1974) selected for assortative mating between populations of the same species (i.e. Drosophila melanogaster) and reported correlated responses in male courtship repertoire. The results of this experiment, however, could have been affected 302

2 Bottleneck effects on evolutionary potential 303 by still having initial differences among the populations by starting with morphological mutant stocks. Thus, there is still a lack of studies that have focused on generating isolation among wild-type subpopulations de novo. Founder-flush speciation models have proposed that the genetic changes following a drastic reduction in population size have the potential to increase evolutionary potential and open novel evolutionary trajectories for reproductive isolation (Mayr, 1963; Carson, 1975; Templeton, 1981; see Meffert, 1999 for a review). The common ground in these models is that the genetic reorganization during a founder event can promote the evolution of new coadapted gene complexes. Mayr (1963) suggested that reproductive isolation could evolve as a by-product of such genetic changes. Similarly, Carson (1975) emphasized the evolution of novel characteristics, such as those involved in mating repertoire, during the period of relaxed selection (i.e. the population flush ). Templeton s genetic transilience model (1981) described how new epistatic complexes can respond to selection and drift in driving a rapid shift to a new adaptive peak. Quantitative genetic models have detailed how bottlenecks can increase the evolutionary potential of a population (Goodnight, 1987, 1988; Cheverud & Routman, 1995, 1996). In a purely neutral model, the heritability of a trait (h 2 ) should decrease in proportion to the severity of the bottleneck (Nei et al., 1975; Falconer, 1989). In contrast, h 2 can increase when the additive component of genetic variance (V A ) is enhanced by the conversion of the nonadditive genetic components of dominance and epistasis (e.g. see reviews by Meffert, 1999, 2000). In the case of dominance, the increased frequencies of detrimental recessive alleles, because of inbreeding, can create dramatic increases in V A (Robertson, 1952; Willis & Orr, 1993; López-Fanjul et al., 2002). Such net increases in h 2, however, are unlikely to catalyse speciation, but, rather, threaten a population with extinction by reducing overall fitness (e.g. see Barton & Charlesworth, 1984). The conversion of epistatic variance (Goodnight, 1987, 1988; Cheverud & Routman, 1996; López-Fanjul et al., 2002) can also increase h 2 in bottlenecked populations. However, converted V A in epistatic traits can also have detrimental effects on overall fitness (Charlesworth, 1998). Increased evolutionary potential (as measured by V A or h 2 ) has been reported for bottlenecked populations of Musca domestica (in morphology and courtship; e.g. Bryant et al., 1986; Meffert, 1995; see Meffert, 2000 for a review), Tribolium (in fecundity; Wade & McCauley, 1988; Wade, 1991; Wade et al., 1996), and D. melanogaster (in survivorship; López-Fanjul & Villaverde, 1989). Founder-flush housefly populations also exhibited altered genetic integration of morphological traits (Bryant & Meffert, 1988) in such a way that could allow for evolution along trajectories not accessible to the ancestral population (e.g. Bryant & Meffert, 1996, 1998). Lynch (1988) suggested that apparent increases in evolutionary potential could be confounded by variation in experimental lines because of sampling and inbreeding effects. Still, Bryant & Meffert et al. (1993) noted levels of morphological V A that were higher than neutral expectation for all experimental lines. Other experiments showed no overall trend for higher h 2 in bottlenecked lines, although individual populations had profound increases (e.g. Meffert, 2000). Nevertheless, Meffert (1995) found that even statistically nonsignificant increases in V A for courtship repertoire still translated to increased evolutionary potential in terms of the realized h 2. Experimental tests of founder-flush theory have found significant levels of premating isolation in bottlenecked populations of D. pseudoobscura (Powell, 1978; Galiana et al., 1993), D. mercatorum (Templeton, 1979), D. simulans (Ringo et al., 1985, 1986), and M. domestica (Meffert & Bryant, 1991; Meffert et al., 1999), although there have been debates about the number of significant effects (Rice & Hostert, 1993; but see Slatkin, 1996; Templeton, 1996). Ethological analyses have related the patterns of assortative mating to divergence in courtship repertoire (Ringo et al., 1986; Meffert & Bryant, 1991; Meffert et al., 1999). Nevertheless, Rundle et al. (1998) found no significant cases of reproductive isolation in a separate bottleneck experiment on D. melanogaster. Arguments about the likelihood of reproductive isolation arising as a by-product of a founder event may remain moot, as all versions of the general model hold that speciation via such processes should be rare (Slatkin, 1996). Perhaps more important, several studies have shown how premating isolation among founder-flush populations can decay through time. Dodd & Powell (1985) found that only one of 11 cases of prezygotic isolation was still significant after 8 years. Similarly, Ehrman (1969), Moya et al. (1995), and Meffert et al. (1999) reported the loss of assortative mating in bottlenecked populations through time. Nevertheless, the continued behavioural evolution of such populations implicates ample or even increased genetic variation after a founder event, as hypothesized in the speciation models. In general, experiments and quantitative genetic models have supported founder-flush and reinforcement theory. Moreover, studies have identified appreciable h 2 for assortative mating and courtship repertoire (Crossley, 1974; Meffert, 1995) that could be enhanced by bottlenecks. However, there is still a lack of experimental tests on how a population could exploit an increase in evolutionary potential (but see Bryant & Meffert, 1995), particularly for traits affecting reproductive isolation. Herein, we describe an experiment to subject bottlenecked and nonbottlenecked populations to selection for assortative mating, per se, as a direct test of the effect of founder events on the evolutionary potential for

3 304 J. L. REGAN ET AL. assortative mating (as assayed by the realized h 2 ). We also evaluated shifts in courtship behaviour due to correlated responses to the selection pressure. We found that the three-pair founder-flush lines had levels of evolutionary potential for assortative mating that were comparable with those for the nonbottlenecked lines, but not greater. The same bottlenecked populations, however, exhibited a significantly more variable set of evolutionary trajectories in the correlated responses in courtship repertoire, suggesting some possible stimulus for the evolution of reproductive isolation. These results, however, cannot discern whether the bottlenecked lines had increased h 2 for courtship components, reduced potential to follow convergent evolutionary trajectories, or some combination thereof. Materials and methods The experiment involved three phases: (1) establishment of the experimental populations, (2) artificial selection Fig. 1 Schematic diagram of the experimental protocol. CTL and BTN denote control (nonbottlenecked) and bottlenecked, treatments, respectively. A and B are the replicate lines that are tested against each other for levels of reproductive isolation over the course of 18 generations. for increased premating isolation (i.e. increased assortative mating) and (3) assays of the correlated responses in courtship repertoire (see Fig. 1 for a diagram of the experimental protocol). In the first phase, we established four replicate pairs of nonbottlenecked lines (eight lines in total) and four replicate pairs of bottlenecked lines (eight lines in total) from a laboratory population of houseflies. Each nonbottlenecked line was derived from approximately 200 flies, and each bottlenecked line was derived from the progeny of exactly six flies (three males and three females). In the second phase, the replicate pairs of nonbottlenecked and bottlenecked lines then underwent 18 generations of artificial selection for increased assortative mating between the two populations of each pair (see below). In the third phase, we (J.L.R) analysed videotaped courtships of all of the lines to identify ethological divergence as a correlated consequence of the selection regime. Phase I: establishment of populations We established an outbred laboratory population of M. domestica (L.) from a single sample of approximately 100 females collected at a landfill in Pasadena, Texas. Over the course of five generations (approximately 3 weeks per generation), this population was allowed to increase in size, reaching the standard husbandry size of roughly 2000 individuals. At this time, we collected approximately 2500 eggs from this stock population, randomized the clutches, and divided the eggs among eight different cultures to establish eight control (nonbottlenecked) lines (Fig. 1). During the same generation, we established eight bottlenecked (founder-flush) lines (Fig. 1) from the progeny of three randomly selected male female pairs from the same base population. To establish the bottlenecked lines, pairs of virgin males and females were isolated in individual cages. After these isolated male female pairs mated, we collected the eggs from each pair and cultured each clutch of eggs separately in a glass vial. We then randomly drew three vials to establish each of the bottlenecked lines. The variance in clutch size among families was low as a result of our procedure of limiting 80 eggs to each vial. The bottlenecked lines were then allowed to undergo five generations of flush (i.e. rapid population growth) enabling them to reach the population sizes of the nonbottlenecked lines (approximately 2000 flies). We assigned the experimental lines within each of the bottlenecked and nonbottlenecked treatments to groups of two (i.e. four pairs of populations within each treatment). For example, bottlenecked lines 1 and 2 were paired together, conceptually, to form one replicate for examining the selection responses for increased assortative mating. Thus, these two populations were tested against each other for levels of assortative mating throughout the course of the experiment. Similarly, nonbottlenecked lines 1 and 2 were tested against each

4 Bottleneck effects on evolutionary potential 305 other throughout the selection protocol. We reasoned that this design maximized our ability to find significant evolutionary responses as compared with alternative designs to either test each line against one single central population or to test bottlenecked lines against nonbottlenecked lines. In particular, high phenotypic variation in a central population or in nonbottlenecked lines could result in weak selective forces and thus compromise the experiment. A total of four replicate pairs for each of the bottlenecked and nonbottlenecked treatments were thus established. After the flush phase for the bottlenecked lines (Generation 6), we assayed the base level of premating isolation between each of the eight pairs of lines (see Fig. 1), using mate choice tests. For these and all subsequent mate choice tests, we released 50 virgin females from each of the two lines (marked with two different colours of fluorescent dust) into a cylindrical cage (diameter 23 cm, height 26 cm) containing 100 males from each of the same two lines (also marked with different colours of fluorescent dust). Note that these tests are simultaneously testing the responses of both types of females to both types of males. After 30 min from the beginning of the assay, we anaesthetized the flies in the mating chamber with CO 2 to identify the assortatively mating pairs (mating pairs remain in copulae under anaesthesia). Housefly copulation lasts approximately 1 h, so the 30-min period was adequate to capture all the pairs that had initiated copulation (i.e. there were no break-ups or re-matings). Our mate choice protocol avoided some of the potential confounds that have been noted in related systems (e.g. see Arnold et al., 1996; Casares et al., 1998). First, we avoided the potential depletion of males from the mating pool (see Casares et al., 1998). In particular, there were twice as many males as females, and approximately 50% of the females mated in the 30-min period. Thus, all of the females could ostensibly choose the same type of male. Moreover, the potential influences of other kinds of social encounters (see Arnold et al., 1996) were greatly reduced because houseflies have no specific male male interactions other than when they court each other. We also employed several other safeguards to reduce possible environmental or experimental effects in the mate choice tests (and courtship assays, see below). First, we reared the larvae at optimal density to reduce variation in body size among individuals (80 eggs per 18 g CSMA larval medium; Bryant, 1969). We also sexed the adults within 24 h of eclosion under CO 2 anaesthesia and housed them separately by sex to ensure virginity. Moreover, we conducted the mating tests at the age of peak sexual activity (5 7 days post-eclosion). Finally, we switched the dust colour of each line each generation to control for possible differences between the two colours of fluorescent dust. To identify the correlated responses to the selection protocol, we assayed the courtship repertoires of each line before and after selection. In order to get baseline measures of courtship behaviour, we videotaped successful courtships (i.e. those ending in copulation) of an average of 25 (SE ¼ 1.22) isolated male female pairs per line (a total of 380 courtships). We employed the same controls for potential confounds of body size, virginity and age as described above. As with the mate choice assays, only courtships that resulted in successful copulation within 30 min were assayed. Using the Observer event recorder software (Noldus, 1990), the intensities of the performance of six post-mounting courtship elements (male performance of BUZZ, LUNGE, LIFT, and HOLD and female performance of WING OUT and KICK; see Table 1 for descriptions of the behaviours) were measured as the time spent in the execution of the behaviour, standardized by the observation period. Phase II: selection for assortative mating Selection for assortative mating was carried out for 18 generations (Fig. 1). Mate choice tests (described above) were used to determine which individuals would be selected as parents for the next generation. In particular, the assortatively mating male female pairs for each line were transferred to a new cage for en masse egg collection over the course of the next 3 days. In the rare event when we did not obtain enough adults for the mate choice assay (given our constraints on the age post-eclosion), we did not perform selection on the replicate pair of lines for that generation. Rather, we culled a random sample of 20 male pairs for each line to be parents for the next generation. We thus assumed Table 1 Descriptions of the courtship traits. Male courtship elements BUZZ The male beats his wings, as in the manner for flight, but without achieving lift-off LUNGE While on top of the female, the male lunges forward towards her head LIFT While in LUNGE, the male uses his forelegs to lift the female s forelegs HOLD While in LUNGE, the male can stop BUZZ, to hold his wings approximately 180 from their resting state and over the female s head Female courtship elements WING OUT The female holds her wings at right angles to her body, with the planes of the wings positioned perpendicular to the substrate KICK The female kicks her hindlegs upwards towards the male

5 306 J. L. REGAN ET AL. that no selection occurred for these events in our calculations of the heritabilities (see below). This effect was balanced across treatments (six cases of 72 tests for each treatment) and is thus unlikely to influence the conclusions. We set a lower bound for the breeding population at 40 individuals to control for the levels of inbreeding within the lines. In the case that fewer than 20 pairs assortatively mated for any line, we set up random male female pairs from the surplus virgins from that line. These random male female pairs were mated in a no choice environment (i.e. one male and one female in each cage) before being introduced to the main population cage to bring the breeding population to the minimum of 40 individuals (i.e. 20 male female pairs). Some of these random male female pairs might have been expected to assortatively mate in a multiple choice situation, such as in our assays. However, other experiments in this system have shown that the no choice environment effectively randomizes assortatively mating tendencies (Meffert & Regan, 2002). We thus assume that these random pairs are not assortatively mating for a conservative estimate of the amount of selection pressure (see below). Phase III: assays for correlated selection responses in courtship After the end of the 18 generations of selection, we assayed the courtship repertoire of each line to identify correlated responses because of the selection regime to increase assortative mating (Fig. 1). We videotaped an average of 19 (SE ¼ 1.19) isolated male female pairs per line (a total of 282 courtships). Again, we used a 30-min time limit for the initiation of copulation and the standard controls for body size, age and virginity. As a result of technical error, one of the nonbottlenecked lines was lost; therefore, only seven of the eight nonbottlenecked lines were analysed for correlated responses in courtship performance. We will refer to the potential for courtship repertoire to shift between the initial and final assays as the selection period effect. Analytical considerations In each generation, the proportion of assortatively mating flies (A) for each set of experimental lines (i.e. each mate choice assay) was calculated as the ratio of females that mated with males from their own populations relative to the total number of females that mated. Thus A ¼ number of assortatively mated females =total number of females mated. We calculated realized heritabilities (h 2 ) for A for each replicate of the bottlenecked and nonbottlenecked treatments as h 2 ¼ RR=RS; where R is the cumulative response to selection, and S is the cumulative selection differential (Falconer, 1989). The response to selection (R) for each generation is the change in the population mean for A: R t ¼ A ðt þ 1Þ A t ; where t is the generation of selection (Falconer, 1989). Assortative mating is a binary measure, so the assortative mating phenotypic value of a selected male female pair is set at 1 (i.e. they did assortatively mate). Thus, the phenotypic mean of the assortatively mating pairs that contributed to the next generation would equal 1 (with no variance). As noted above, we set the minimum population size at 20 male female pairs, so we sometimes had to add random male female pairs to the pool of selected (assortatively mated) individuals. As noted, we will assume that these random male female pairs were nonassortatively mated (i.e. the behavioural phenotype ¼ 0; see Meffert & Regan, 2002), for a conservative estimate of the degree of selection pressure. Thus, the selection differential in each generation is calculated as S ¼ð1:0 NÞ A; where N is the proportion of random (nonassortatively mating) pairs in the pool of individuals contributing to the next generation. In this way, N diluted the average degree of assortative mating away from the maximum value of 1. Consequently, S is the difference between the level of assortative mating in the pool of founders for the next generation and the level of assortative mating overall. For the analyses of the videotaped courtships, we transformed four of the six courtship elements (all but HOLD and KICK; Table 1) to meet the assumptions of normality (i.e. inverse transformation for LUNGE, log-transformation for BUZZ and LIFT, and arcsine transformation for WING OUT; Winer, 1962). For the multivariate analyses, we pooled the data from the nonbottlenecked lines before the selection protocol for condensing these variables via principal component analysis (IML in SAS, 1985). The first two principal components explained 58% of the total variance (see Appendix 1 for the loadings of the traits onto these axes). The first principal component described high positive correlations among LIFT, HOLD, WING OUT and KICK, and their negative relationships with LUNGE (see Appendix 1). The second principal component had high positive correlations of both the female courtship elements (WING OUT and KICK), which were negatively correlated with BUZZ (see Appendix 1). We thus depicted the evolutionary trajectories by graphing the first and second principal component loadings for each of the bottlenecked and nonbottlenecked treatments (averaged across lines), before and after the selection regime

6 Bottleneck effects on evolutionary potential 307 (see below), based on the eigenstructure of the nonbottlenecked preselection populations (pooled). Bootstrap methods (see Meffert, 1995) set 95% upper and lower confidence bounds for significance testing. The bottleneck and selection pressures of this experiment could have altered phenotypic variance covariance matrices through shifts in genetic variance covariance matrices (e.g. see Bryant & Meffert, 1988). Our data, however, are not amenable to evaluating the evolution of genetic variance covariance matrices (i.e. requiring parent offspring assays; see Bryant & Meffert, 1988). Moreover, a related bottleneck study on housefly courtship showed no significant differences among lines at the level of the phenotypic intercorrelation structure (i.e. see Meffert & Bryant, 1991). We thus hold that evaluating the potential for the evolution of genetic intercorrelation is beyond the scope of this study. Furthermore, our method to map the resultant courtship phenotypes onto the original nonbottlenecked, nonselected phenotypic intercorrelation structure should be adequately conservative (Meffert & Bryant, 1991), especially as intercorrelation structures should be generally more robust than variance covariance solutions (e.g. Pimentel, 1979). Fig. 2 Average levels of assortative mating over the course of 18 generations of selection. Solid lines and dotted lines denote bottlenecked and nonbottlenecked treatments, respectively. Linear regressions are depicted by the straight lines. Both regressions are highly significant (P < 0.01 and P ¼ 0.03 for the bottlenecked and nonbottlenecked treatments, respectively), but the regression slopes are not significantly different from each other. Results Within the bottlenecked treatment, analysis of variance yielded significant differences among replicates in the levels of assortative mating (F 3, 62 ¼ 4.24, P ¼ 0.01). Within the nonbottlenecked treatment, however, no significant differences were detected (F 3, 62 ¼ 0.45, P ¼ ns). Despite the significant differences among the bottlenecked replicates, we pooled the data for a conservative comparison of the changes in assortative mating between the two treatments (see Appendix 2 for the analyses on the selection responses for each pair of lines within treatments). Linear regressions were used to identify changes in premating isolation over the 18 generations of selection (Fig. 2). Both the bottlenecked and the nonbottlenecked treatments exhibited significant increases in premating isolation (bottlenecked lines: r 2 ¼ 0.13, P < 0.01; nonbottlenecked lines: r 2 ¼ 0.07, P ¼ 0.03; Fig. 2). Using the regression equations to map the total shifts in assortative mating, the bottlenecked and nonbottlenecked lines increased by a mean of 9.6 and 8.6%, respectively (SE ¼ 3.1 and 2.9, respectively) over the course of 18 generations. Moreover, the regression coefficients were not significantly different from each other (F 1, 127 ¼ 0.17, P ¼ ns). The realized heritabilities for both the bottlenecked and nonbottlenecked treatments were very low, but highly significant (bottlenecked: h 2 ¼ 0.04, r 2 ¼ 0.62, P < 0.001; nonbottlenecked: h 2 ¼ 0.05, r 2 ¼ 0.58, P < 0.001; Fig. 3). As with the proportion of assortative mating (A), there was no significant difference between the two treatments for the heritabilities (F 1, 139 ¼ 0.07, P ¼ ns). Fig. 3 Realized heritabilities for assortative mating. Solid lines (closed circles) and dotted lines (open circles) denote bottlenecked and nonbottlenecked treatments, respectively. Linear regressions are depicted (see Falconer, 1989). Both regressions are highly significant (P < 0.001), but the regression slopes are not significantly different from each other. In the univariate analyses of courtship repertoire, five of the six measures showed highly significant correlated responses to the selection protocol (male performance of BUZZ, LUNGE, LIFT, and HOLD and female performance of KICK; Table 2). Only male performance for LIFT, however, had a significant difference between bottlenecked and nonbottlenecked treatments (Table 2). Line-within-treatment effects were found for LIFT and

7 308 J. L. REGAN ET AL. Table 2 Univariate analyses of courtship repertoire treatment indicates a test between bottlenecked and control treatments. Line refers to the among line variation within treatments. Selection period identifies tests for correlated responses to selection (i.e. differences before and after the selection regime). See Table 1 for descriptions of the behaviours. (a) d.f. MS F P-value BUZZ Treatment Line (treatment) Period Treatment * period Period * line (treatment) LUNGE Treatment Line (treatment) Period Treatment * period Period * line (treatment) LIFT Treatment Line (treatment) Period Treatment * period Period * line (treatment) (b) HOLD Treatment Line (treatment) Period Treatment * period Period * line (treatment) WING OUT Treatment Line (treatment) Period Treatment * period Period * line (treatment) KICK Treatment Line (treatment) Period Treatment * period Period * line (treatment) for the female movements, WING OUT and KICK (Table 2), indicating that individual lines differed in the magnitude of response. Interactions between lines and selection period (within treatments) were significant for three of four male traits (all but LUNGE; Table 2) and female WING OUT (Table 2). Only LIFT had a significant interaction between treatment and period. Figure 4 depicts the multivariate correlated responses to selection. Seven of the eight bottlenecked lines exhibited significant shifts in courtship behaviour from their ancestral states (Fig. 4a). Of these seven lines, three Fig. 4 Multivariate depiction of the correlated responses to selection in courtship repertoire for the (a) bottlenecked and (b) nonbottlenecked treatments (note change in scale). Evolutionary trajectories along the first and second principal axes (see Appendix 1) are shown, with each arrow pointing to the resultant courtship repertoire after 18 generations of selection for assortative mating. The letters (a d) identify the pairs of populations that were tested against each other for reproductive isolation. The dotted boxes show the 95% confidence bounds for a significant deviation from the pooled data on the nonbottlenecked lines before the selection regime (determined from bootstrap simulation). had significant changes along both principal axes, whereas four changed primarily along the first principal axis (Fig. 4a). In the nonbottlenecked treatment, all seven lines had significant shifts in courtship phenotype (Fig. 4b), yet the general trend was to shift only along the first principal component (Fig. 4b). Specifically, the

8 Bottleneck effects on evolutionary potential 309 deviation from the first principal component, as assayed by the cosine of the angle with the evolutionary trajectory, yielded a difference between the bottlenecked and nonbottlenecked treatments. The average angles to the first component were (SE ¼ 12.93) and (SE ¼ 5.34) for bottlenecked and nonbottlenecked treatments, respectively (significant at P ¼ 0.03 in ANOVA, P < 0.05 in a posteriori Duncan s multiple-range tests). The lines within paired populations (i.e. those tested against each other each generation for assortative mating; see Fig. 1) generally followed different evolutionary trajectories, with several of the cases demonstrating significant shifts in different directions (e.g. population pairs a and b in Fig. 4a and pairs a and b in Fig. 4b). Discussion We found very low, but highly significant, increases in assortative mating over the 18 generations of the selection protocol (i.e. the heritabilities were 0.04 and 0.05 for the bottlenecked and nonbottlenecked treatments, respectively; Figs 2 and 3). The regressions for the levels of assortative mating and realized heritabilities had greater statistical significance for the bottlenecked treatment (Figs 2 and 3), but the overall heritability was slightly higher in the nonbottlenecked treatment (Fig. 3). Although prior housefly bottleneck experiments have found dramatically increased evolutionary potential for overall mating propensity and courtship repertoire, the average effect among bottlenecked lines was often not significantly different from the controls (Meffert, 1995, 1999, 2000; Meffert et al., 1999). Thus, our study on the evolutionary potential for assortative mating concurs with the other house fly experiments. We expected that assortative mating would have very low heritability, because of the importance that mate choice has on overall fitness (Andersson, 1994). Moreover, behaviour traits are notorious for having high environmental variance components (e.g. Mousseau & Roff, 1987; Boake, 1989; Price & Schluter, 1991). Boake & Konigsberg (1998), for example, showed how the heritabilities for morphological traits in D. silvestris can be appreciable, but be essentially undetectable for courtship behaviours. Courtship, in particular, can even show negative heritabilities, because of the genotype-byenvironment effects of one individual influencing the behaviour of its mating partner (Meffert, 1995; Meffert et al., 2002). These interactions have been modelled as a special form of epistasis (i.e. genotype-by-genotype interactions; Meffert, 1995) or indirect genetic effects (e.g. Wolf, 2000). Thus, mating behaviour can have the gamut of nonadditive genetic effects (see Meffert et al., 2002). In bottleneck models, low heritability traits should have more dramatic responses because of the conversion of V A from the nonadditive genetic components (see Meffert, 1999; Meffert et al., 2002). In other words, a low heritability trait has more room for improvement. Wade et al. (1996), for example, assayed heritabilities for viability and morphology in bottlenecked populations of Tribolium and found more significant increases in the fitness traits. Similarly, Meffert (2000) and Meffert et al. (2002) reported greater increases in bottlenecked population heritabilities for courtship behaviours (up to 330%) than for morphological traits (only up to 180%). Note that such increases in genetic variance may not necessarily represent adaptive variation. Negative fitness consequences can arise because of the inflated frequencies of deleterious recessive alleles or epistatic constructs (Meffert, 1999, 2000). The prevalence of reinforcement processes in nature is an ongoing debate (e.g. Hostert, 1997; Noor, 1997). Nevertheless, prior research has shown that reproductive isolation among populations can be modified through artificial selection (Koopman, 1950; Kessler, 1966; Dobzhansky & Pavlovsky, 1971; Crossley, 1974; Hostert, 1997), as found in this study. Most of the experiments focused on enhancing the isolation among species or subspecies (Koopman, 1950; Kessler, 1966; Dobzhansky & Pavlovsky, 1971). More germane to this study, Crossley (1974) selected for increased reproductive isolation between mutant strains of D. melanogaster and found significant correlated responses in male mating behaviour. Hostert (1997) was also able to select for assortative mating among D. melanogaster populations derived from an admixture of wild-type stock populations with introgressed eye-colour mutations. Starting with a single wild population, we produced not only significant increases in assortative mating, but also significant correlated responses in both male and female courtship elements (Table 2 and Fig. 4). Both the bottlenecked and nonbottlenecked treatments exhibited similarities in the degree of change in several of the courtship elements (Table 2). Of the six courtship elements evaluated (four male elements and two female elements) only one demonstrated any general effect because of the bottleneck: LIFT. This could be explained by the low level of incorporation of LIFT in the ancestral population courtship that ultimately converged to the levels in the bottlenecked lines. Although similar levels of assortative mating were attained by both bottlenecked and nonbottlenecked treatments, there were differences in the correlated responses. In particular, the bottlenecked lines that had significant shifts in courtship repertoire moved along both principal components (Fig. 4a), but the nonbottlenecked lines exhibited significant shifts primarily along the first axis (Fig. 4b). Importantly, the average deviation from the first principal component was significantly greater for the bottlenecked treatment than that for the nonbottlenecked treatment (a difference of 280%). The univariate analyses on courtship behaviour also revealed significant among line-variance within the bottlenecked treatment, and significant interactions of treatment with correlated selection response (Table 2). Thus, the bottlenecked

9 310 J. L. REGAN ET AL. treatment resulted in a much more diverse set of evolutionary trajectories than that seen in the nonbottlenecked lines. This greater diversity in the founder-flush lines could be interpreted as the evolution of novel phenotypes, as suggested by the founder-induced speciation models. Despite our efforts to maximize the effective population sizes of the experimental lines, inbreeding was inevitable. Given the logistical constraints, we managed 16 populations at a population size of 20 male female pairs for 18 generations. Initially, the bottlenecked treatment had a higher level of inbreeding because of the founder event (F ¼ 0.08 in bottlenecked lines; Falconer, 1989). By the end of the selection protocol, however, both treatments had experienced similar levels of inbreeding. In particular, the inbreeding coefficients for the bottlenecked and nonbottlenecked treatments were 0.23 and 0.28, respectively, based upon census size. These estimates are minimum values, however, because selection will drive additional inbreeding by favouring related individuals on the basis of shared phenotypes (e.g. Crow & Morton, 1955; Bulmer, 1971). Consequently, the similarities between the treatments in the direct selection responses (Figs 2 and 3) could be explained by a relatively small influence of the initial founder event on the bottlenecked treatment. Nevertheless, it is unclear if the inherent inbreeding of the selection protocol reduced genetic variance in accordance with neutral expectation (Falconer, 1989) or increased evolutionary potential because of the conversion of nonadditive genetic variance (e.g. Goodnight, 1988; Cheverud & Routman, 1996). The very low realized heritabilities (Fig. 3) suggest that conversion did not occur for assortative mating. However, the differences between the treatments in the correlated responses (Fig. 4) suggest that more complex processes were operating on mating behaviour. In particular, some initial effect of the founder event seems responsible for the more variable set of evolutionary trajectories in the bottlenecked treatment (in assortative mating and courtship repertoire). Conversion processes could have opened up evolutionary avenues in the bottlenecked lines. Alternatively, the bottlenecked treatment could have limited evolutionary potential to follow the convergent evolutionary trajectories of the nonbottlenecked lines. A great number of loci are likely to affect this multivariate phenotype, so these two processes need not be mutually exclusive. Some loci (or suites of loci) may have undergone conversion (e.g. López-Fanjul et al., 2002) whereas others followed the additive expectations for bottleneck effects (e.g. Falconer, 1989). In any case, founder-flush events increased the number of evolutionary solutions for the correlated responses in courtship repertoire. Our experiment has shown that selection for assortative mating can drive the evolution of courtship repertoire as a correlated response. Bottlenecked populations responded as well to selection as did nonbottlenecked populations, although our results do not suggest overall increases in the evolutionary potential for assortative mating. Nevertheless, we found evidence that bottlenecks could still promote the evolution of novel courtship repertoire, either because of increased heritabilities, the restriction of particular evolutionary corridors, or some combination of the two in determining the multivariate phenotype. Acknowledgments This research was funded by grants from the National Science Foundation (DEB and DEB to LMM). Many thanks to S. Day, B. Geary and L. Vo for technical help and to M. Noor, G. Pessoney, D. Queller, J. Strassmann and two anonymous reviewers for their comments on earlier drafts of the manuscript. References Andersson, M Sexual Selection. Princeton University Press, Princeton, NJ, USA. Arnold, S.J., Verrell, P.A. & Tilley, S.G The evolution of asymmetry in sexual isolation: a model and a test case. Evolution 50: Barton, N.H. & Charlesworth, B Genetic revolutions, founder events, and speciation. Ann. Rev. Ecol. Syst. 15: Boake, C.R.B Repeatability: its role in evolutionary studies of mating behavior. Evol. Ecol. 3: Boake, C.R.B. & Konigsberg, L Inheritance of male courtship behavior, aggressive success and body size in D. silvestris. Evolution 52: Bryant, E.H The fates of immatures in mixtures of two housefly strains. Ecology 50: Bryant, E.H., McCommas, S.A. & Combs, L.M The effect of an experimental bottleneck upon quantitative genetic variation in the housefly. Genetics 114: Bryant, E.H. & Meffert, L.M Effect of an experimental bottleneck on morphological integration in the housefly. Evolution 42: Bryant, E.H. & Meffert, L.M The effects of serial bottlenecks on quantitative genetic variation in the housefly. Heredity 70: Bryant, E.H. & Meffert, L.M An analysis of selectional response in relation to a population bottleneck. Evolution 49: Bryant, E.H. & Meffert, L.M Morphometric differentiation in serially bottlenecked lines of the housefly. Evolution 50: Bryant, E.H. & Meffert, L.M Quantitative genetic estimates of morphometric variation in wild-caught and laboratory-reared houseflies. Evolution 52: Bulmer, M.G The effect of selection on genetic variability. Am. Nat. 105: Butlin, R Reinforcement: an idea evolving. Tr. Ecol. Evol. 10: Carson, H.L The genetics of speciation at the diploid level. Am. Nat. 109:

10 Bottleneck effects on evolutionary potential 311 Casares, P., Carracedo, M.C., del Rio, B., Pineiro, R., Garcia- Florez, L. & Barros, A.R Disentangling the effects of mating propensity and mating choice in Drosophila. Evolution 52: Charlesworth, B The effect of synergistic epistasis on the inbreeding load. Genetic. Res. 71: Cheverud, J.E. & Routman, E.J Epistasis and its contribution to genetic variance components. Genetics 139: Cheverud, J.E. & Routman, E.J Epistasis as a source of increased additive genetic variance at population bottlenecks. Evolution 50: Coyne, J. & Orr, H Patterns of speciation in Drosophila revisited. Evolution 51: Crossley, S.A Changes in mating behavior produced by selection for ethological isolation between ebony and vestigial mutants of Drosophila melanogaster. Evolution 28: Crow, J.F. & Morton, N.E Measurement of gene frequency drift in small populations. Evolution 9: Dobzhansky, Th Genetics and the Origin of Species, 3rd edn. Columbia University Press, New York. Dobzhansky, Th. & Pavlovsky, O Experimentally created incipient species of the Drosophila paulistorum complex. Nature 230: Dodd, D.M.B. & Powell, J.R Founder-flush speciation: an update on experimental results with Drosophila. Evolution 39: Ehrman, L Genetic divergence in M. Vetukhiv s experimental populations of Drosophila pseudoobscura. 5. A further study of rudiments of sexual isolation. Am. Midl. Natur. 82: Falconer, D.S Introduction to Quantitative Genetics, 2nd edn. Longman, New York. Galiana, A., Moya, A. & Ayala, F.J Founder-flush speciation in Drosophila pseudoobscura: a large-scale experiment. Evolution 47: Goodnight, C.J The effect of founder events on epistatic genetic variance. Evolution 41: Goodnight, C.J Epistasis and the effect of founder events on the additive genetic variance. Evolution 42: Hostert, E.E Reinforcement: a new perspective on an old controversy. Evolution 51: Kessler, S Selection for and against ethological isolation between Drosophila pseudoobscura and Drosophila persimilis. Evolution 20: Kirkpatrick, M Sexual selection and the evolution of female choice. Evolution 36: Koopman, K.F Natural selection for reproductive isolation between Drosophila psedoobscura and Drosophila persimilis. Evolution 4: Lande, R Models of speciation by sexual selection on polygenic traits. Proc. Natl. Acad. Sci. USA 78: López-Fanjul, C. & Villaverde, A Inbreeding increases genetic variance for viability in Drosophila melanogaster. Evolution 43: López-Fanjul. C., Fernández, A. & Toro, M The effect of epistasis on the excess of the additive and nonadditive variances after population bottlenecks. Evolution 56: Lynch, M Design and analysis of experiments on random drift and inbreeding depression. Genetics 120: Mayr, E Animal Species and Evolution. Belknap Press, Cambridge, MA, USA. Meffert, L.M Bottleneck effects on genetic variance for courtship repertoire. Genetics 129: Meffert, L.M How speciation experiments relate to conservation biology. Bioscience 49: Meffert, L.M The evolutionary potential of morphology and mating behavior: the role of epistasis in bottlenecked populations. In: Epistasis and the Evolutionary Process (J. B. Wolf, E. D. Brodie & M. J. Wade, eds), pp Oxford University Press, Oxford. Meffert, L.M. & Bryant, E.H Mating propensity and courtship behavior in serially bottlenecked lines of the housefly. Evolution 45: Meffert, L.M., Hicks, S.K. & Regan, J.L Nonadditive genetic effects in animal behavior. Am. Nat. 160: S198 S213. Meffert, L.M. & Regan, J.L. A test of speciation via sexual selection on female preferences. Anim. Behav. in press. Meffert, L.M., Regan, J.L. & Brown, B.W Convergent evolution of the mating behaviour of founder-flush populations of the housefly. J. Evol. Biol. 12: Mousseau, T.A. & Roff, D.A Natural selection and the heritability of fitness components. Heredity 58: Moya, A., Galiana, A. & Ayala, F Founder-effect speciation theory: failure of experimental corroboration. Proc. Natl. Acad. Sci. USA 92: Nei, M., Maruyama, T. & Chakroborty, R The bottleneck effect and genetic variability in populations. Evolution 29: Noldus, L.P.J.J The Observer Base Package, Version 2.0. Noldus Information Technology, Wagenengen, The Netherlands. Noor, M.A How often does sympatry affect sexual isolation in Drosophila? Am. Nat. 149: Pimentel, R.A Morphometrics: The Multivariate Analysis of Biological Data. Kendall Hunt Publishing Co, Dubuque, IA, USA. Powell, J.R The founder-flush speciation theory: an experimental approach. Evolution 32: Price, T. & Schluter, D On the low heritability of lifehistory traits. Evolution 45: Rice, W.R. & Hostert, E.E Laboratory experiments on speciation: what have we learned in 40 years? Evolution 44: Ringo, J., Barton, K. & Dowse, H The effect of genetic drift on mating propensity, courtship behaviour, and postmating fitness in Drosophila simulans. Behaviour 97: Ringo, J., Wood, D., Rockwell, R. & Dowse, H An experiment testing two hypotheses of speciation. Am. Nat. 126: Robertson, A The effect of inbreeding on the variation due to recessive genes. Genetics 37: Rundle, H.D., Mooers, A.Ø. & Whitlock, M.C Single founder-flush events and the evolution of reproductive isolation. Evolution 52: SAS Institute, Inc SAS IML, Version 5. Cary, NC. Slatkin, M In defense of founder-flush theories of speciation. Am. Nat. 147: Templeton, A.R The unit of selection in Drosophila mercatorum II. Genetic revolution and the origin of coadapted genomes in parthenogenetic strains. Genetics 92: Templeton, A.R The theory of speciation via the founder principle. Genetics 94: Templeton, A.R Experimental evidence for the genetictransilience model of speciation. Evolution 50: