Hybrid breakdown in fitness components as a correlated response to selection for dessication resistance.
Evidence for coadaptation of gene pools has come mainly from interpopulational crosses between natural populations (reviewed in Endler 1977). A decrease in the magnitude of the measured character in the |F.sub.2~ and subsequent generations from these crosses suggests that genes that work well in one genetic background do not function as well in another. Each gene pool is said to be coadapted because genes seem to be selected not only on the basis of their individual properties but also on how they interact with other genes segregating in the population (Wright 1956; Wallace 1959).
Little direct experimental evidence exists for multiple-peak epistasis constituting a general mode of gene action (Enfield and Anklesaria 1986). Cohan et al. (1989) describe two classes of artificial selection experiments that have provided evidence for multiple-peak epistasis. The first class requires that the response to selection is greater when a population is divided into smaller lines with some migration and interdemic selection between them, compared with when the same population is mass selected as a single entity (reviewed in Barker 1989). The most recent and comprehensive test using this design was made by Wade and Goodnight (1991), who imposed interdemic selection via migration (as opposed to extinction) between replicate lines of a laboratory stock of Tribolium castaneum. Interdemic selection was implemented by lines (initial N = 20) contributing to a migrant pool in direct proportion to their productivity (census size after 60 d). Three treatments of interdemic selection were used; interdemic selection every generation, every second generation, and every third generation. Significant realized interdemic heritabilities were found in each of the treatments (0.207, 0.383, and 0.205, respectively). The significant heterogeneity found between these heritabilities indicated that the response was not proportional to the selection differential, suggesting indirectly that nonadditive genetic variation in fitness was involved in the larger response when interdemic selection was applied every second generation.
The second class of experiments requires selection to push replicate lines from a single population to different adaptive peaks as measured by an epistatic loss of the selected character in the |F.sub.2~ generation after these lines are crossed (e.g., King 1955; Enfield 1977; Cohan 1984). As Cohan et al. (1989) emphasize, these two classes of experiments require drift to take replicate lines of a single population into the domains of different adaptive peaks before selection can produce multiple peak epistasis. The authors then described an experiment that overcame this limitation by using five natural populations of Drosophila melanogaster that may have already been in different domains of attraction before imposing selection for knockdown resistance to ethanol vapors. However, even though correlated responses to this stress were different from the five populations, suggesting that different genes were involved in the responses of the different populations (Cohan and Hoffmann 1986), no evidence was found for epistasis contributing to the selection response on the basis of between-population crosses (Cohan et al. 1989).
The experiment described here extends the design of Cohan et al. (1989), as suggested by the authors, in one important respect. Although epistasis did not contribute to the response to selection for ethanol knockdown in their study, the authors suggested that the genes that had been selected may have contributed epistatically to fitness, because gene action can be completely additive for a character under selection and still produce multiple fitness peaks (Wright 1968, p. 425; Barker 1979). That is to say, multiple-peak epistasis for fitness may have been a correlated response of selection for ethanol knockdown. The present experiment tests whether selection for an abiotic stress with additive gene action (desiccation resistance) can increase nonadditive genetic variation between populations in fitness and therefore push different geographical populations to different fitness peaks as a correlated response. Price et al. (1993), have recently modeled a similar situation between two characters (as distinct from a character and fitness in the present context). They investigated whether a peak shift may occur in one character as a correlated response to selection on another genetically correlated character. The success of the peak shift depended on the genetic correlation between the two characters, the magnitude of the fitness deficit between the two peaks, and the strength of selection. Furthermore, when a peak shift did occur, the transition was rapid.
In the present experiment, selection was carried out on four populations of Drosophila serrata from central and marginal areas of its distribution. Central and marginal populations were chosen in an attempt to maximize the genetic variation between populations, which may result in these populations residing in the domains of attraction of different fitness peaks. These four populations have previously been shown to differ in their ability to respond to selection for desiccation resistance and in their correlated responses to selection for this stress (Blows and Hoffmann 1993).
MATERIALS AND METHODS
A detailed description in the Drosophila serrata stocks and the selection procedure are given in Blows and Hoffmann (1993), thus only a brief outline is given here.
Stocks. -- D. serrata Malloch is an Australian endemic species in the melanogaster species group and is found along the east coast of Australia. Its range extends from 200 km north of Sydney to New Guinea and the surrounding islands (Dobzhansky and Mather 1961). Four populations of D. serrata were sampled from the center of its distribution and the southern margin between August 1989 and January 1990. These populations were (from the most southerly to the most northerly) Forster, Coffs Harbour (300 km north of Forster), Brunswick Heads (250 km north of Coffs Harbour), and Townsville (1550 km north of Brunswick Heads). Each population was rounded by 16 inseminated females from each locality and were maintained at a census size greater than 300 in six culture bottles for six mo at 25 |degrees~ C prior to selection. Six lines from each population were created for the selection experiment, three control lines and three selected lines. Each line was maintained throughout the experiment in three culture bottles with a total population size of approximately 100 individuals. A fifth laboratory population was founded from a single inseminated female collected at Cairns (700 km north of Townsville) in November 1990 and was maintained for use in the fitness experiments described below.
Selection. -- Resistance to a desiccation stress of 5%-10% humidity at 25 |degrees~ C was the selected trait. One population was stressed on any particular day. From each selected line, 100 virgin females were placed singly in test tubes in one desiccator, and 100 virgin males from each selected line were placed in another. Tubes were held in the desiccators until about 50% of the flies had died. Survivors founded the next generation. All lines were selected for a total of 14 generations.
Genetics of Desiccation Resistance. -- The genes contributing to extreme environmental stresses are usually additive in gene action (Parsons 1983). Desiccation resistance was the trait that underwent selection, thus it was important to establish if gene action for this trait in D. serrata was additive when the four populations were crossed. A control line from each of the populations was crossed to control lines from the other three populations (i.e., six population crosses). This was done for the three control lines of each population, giving three replicates of each of the six possible population combinations (i.e., 18 population crosses in total). The six population crosses are denoted throughout this paper as FC (Forster x Coffs Harbour), FB (Forster x Brunswick Heads), FT (Forster x Townsville). CB (Coffs Harbour x Brunswick Heads), CT (Coffs Harbour x Townsville), and BT (Brunswick Heads x Townsville).
Because it was desirable to test seven types of families (i.e., parental crosses, reciprocal |F.sub.1~, |F.sub.2~ and both backcrosses) for all 18 population crosses at the same time, an |F.sub.1~ for each cross was generated first by crossing 20 virgin males and 20 virgin females. These |F.sub.1~ flies and the parental lines were then used to generate all seven types of families for each of the 18 population crosses simultaneously by crossing 20 virgin males with the appropriate 20 virgin females. From the 102 crosses generated (this is fewer than the 18 x 7 = 126 crosses one might expect, because parental crosses within one replicate could be used in 3 of the 6 population crosses in that replicate), nine virgin males were sexed under ether and left to age for seven days. Flies were then scored for desiccation resistance in three desiccators (each desiccator representing a complete randomized block) until all flies were dead.
Fitness Experiments. -- Three fitness traits, egg to pupa developmental time, egg to pupa viability, and fecundity were scored to determine the affect of selection on hybrid breakdown. It was not possible to measure these traits simultaneously on all the crosses resulting from the 24 experimental lines (a total of 204 crosses was required), thus three separate experiments were run sequentially. Each experiment contained one selected line and one control line from each of the four populations. This design allowed a direct comparison of the results of crosses from selected and control lines within each experiment but did not allow a test of whether selection had affected fitness interactions between replicate lines of a single population because a day effect was compounded with the replicate line term.
Within each of the three experiments, each selected line was crossed to the three other selected lines from the other populations, and each control line was crossed to the three other control lines. Thus, for experiment 1, for example, the six population crosses of FC, FB, FT, CB, CT, and BT were created by crossing the selected line 1 of each population to its selected line 1 counterpart of the other populations. The same was done for the control line 1 of the four populations. The |P.sub.1~, |P.sub.2~, reciprocal |F.sub.1~, |F.sub.2~, and both backcrosses for each population cross were generated as described for the desiccation-resistance experiment above. This gave a total of 68 crosses for each of the three experiments.
The following procedure measured the three fitness trails in the three experiments: once the seven types of family for each population cross were set up, they were left for 7 d to allow the 20 females in each cross to reach peak egg production. Flies were then tipped into an empty culture bottle that was closed with a watch glass with a little medium on the undersurface. The females were allowed to lay for 6 h. For each cross, five eggs were placed in each of nine vials. One vial from each cross was placed in a tray giving nine complete randomized blocks for each of the three experiments. Vials were placed at 20 |degrees~ C and after nine d pupation began. Time to pupation was scored at 12-h intervals to measure developmental time. The number of eggs reaching pupation was scored, just before first emergence, as the measure of viability.
TABLE 1. Means and their standard errors for desiccation resistance in control crosses. Desiccation resistance is measured as the time in hours for flies to die. Means are based on three replicate lines. The first |F.sub.1~ value is for the cross with the female parent from the most southerly population. Population combination Cross FC FB FT CB CT BT |P.sub.1~ 7.3 7.3 7.3 8.3 8.3 8.8 1.4 1.4 1.4 1.1 1.1 1.9 |BC.sub.1~ 7.4 8.0 6.7 8.1 7.1 7.5 1.1 1.3 1.2 1.4 1.3 1.9 |F.sub.1~ 7.4 8.3 7.1 7.9 8.1 8.9 1.7 1.6 1.6 1.2 1.7 1.9 |F.sub.1~ 7.2 7.9 8.7 8.2 8.2 8.2 1.6 1.3 1.9 1.4 1.4 1.6 |F.sub.2~ 8.8 8.3 8.0 8.5 8.2 7.9 1.4 1.5 1.5 1.4 1.4 1.9 |BC.sub.2~ 7.3 8.3 8.0 7.5 8.1 7.8 1.3 1.5 1.3 1.4 2.3 2.0 |P.sub.2~ 8.3 8.8 8.4 8.8 8.4 8.4 1.1 1.9 1.5 1.9 1.5 1.5
To measure fecundity, one virgin female was taken from each vial and placed in a fresh vial with a virgin Cairns male from the isofemale line described above. The isofemale Cairns line was used for two reasons. (1) The variation in fecundity between females as a consequence of their mate should be reduced by using an inbred line. (2) Because Cairns is more than 700 km from the nearest of the four populations used in the experiment, a potential bias toward any of the four populations was minimized. These vials were placed at 25 |degrees~ C for 4 d. On day five, the male and female pairs were placed in vials that contained a spoon with medium dyed with pink food dye and arranged once again in nine complete randomized blocks. The spoons were changed every 24 h for another 3 d and scored for eggs. This measured period of laying extends from early fecundity up to the beginning of peak fecundity for D. serrata at 25 |degrees~ C (Birch et al. 1963).
The experimental design for tests conducted on the developmental time and fecundity data for each experiment is given by
|Y.sub.ijkl~ = a + |b.sub.i~ + |c.sub.j~ + |bc.sub.ij~ + |d.sub.k~ + |e.sub.l(ijk)~,
where a is the grand mean, |b.sub.i~ the population term |c.sub.j~ the selection term, |bc.sub.ij~ the population by selection interaction term, |d.sub.k~ the block term, and |e.sub.l(ijk)~ the error. In the case of viability, the data from the nine blocks within each experiment had to be combined because a highly skewed distribution was present when values from individual replicates were examined. Therefore, the block term was used as the error and the interaction term could not be tested.
Hybrid Breakdown in Desiccation Resistance. -- The means and their standard errors for each type of family for the six population crosses, calculated across the three replicate lines, are presented in table 1. The method used to determine if hybrid breakdown had occurred in desiccation resistance and in the fitness traits below is that described by Mather and Jinks (1982) and involves the calculation of three individual scaling tests, which are then combined in Cavalli's (1952) joint scaling test. These test for a significant deviation from the expectations of the additive-dominance model. The three individual scaling tests are
A = 2|B.sub.1~ - |P.sub.1~ - |F.sub.1~
B = 2|B.sub.2~ - |P.sub.2~ - |F.sub.1~
C = 4|F.sub.2~ - 2|F.sub.1~ - |P.sub.1~ - |P.sub.2~.
Orthogonal comparisons were used to estimate each of the individual scaling tests for each population cross and were tested for significance from zero with a two-tailed t-test (not presented). Because several tests were conducted on these data, probabilities were recalculated using the sequential Bonferroni test (Rice 1989) to adjust for multiple comparisons.
Significant values were obtained for scaling tests from the population crosses FC (test C, P |is less than~ 0.001), FT (test A, P = 0.014), CB (test B, P = 0.006), CT (test A, P = 0.002), and BT (test A, P = 0.004), although only the FC and CT values remain significant after applying the sequential Bonferroni technique over these 18 tests. These results suggest that the additive model may not explain the gene action observed in two cases.
However, the joint scaling test employs a least-squares technique to estimate the expected family means (i.e., |P.sub.1~, |P.sub.2~, reciprocal |F.sub.1~, |F.sub.2~, and both backcrosses) under an additive-dominance model and is considered more informative than the individual scaling tests (Mather and Jinks 1982). |X.sup.2~s with 3 df, testing for a difference between observed and expected family means for each population cross, were calculated, and none were significant (not presented). Therefore, little evidence exists to suggest that the additive-dominance model is not sufficient to explain gene action in desiccation resistance.
TABLE 2. Developmental time for selected and control lines of the four parental populations. Development time is measured in 12-h intervals from the beginning of pupation. A selected and control line from each population was tested in each of the three experiments. Means and their standard deviataions are based on nine replicates. Mean squares are presented from ANOVAs for each experiment. Developmental time Experiment 1 Experiment 2 Experiment 3 Mean (SD) Forster Selected 2.99 (0.57) 3.61 (0.30) 3.18 (0.63) Control 3.30 (0.63) 2.96 (0.47) 2.95 (0.45) Coffs Harbour Selected 3.07 (0.33) 4.62 (0.81) 3.19 (0.51) Control 2.72 (0.52) 2.97 (0.41) 3.58 (0.60) Brunswick Heads Selected 3.11 (0.39) 4.13 (0.53) 3.39 (0.38) Control 3.37 (0.54) 3.90 (0.69) 3.01 (0.35) Townsville Selected 2.84 (0.77) 3.88 (0.63) 2.89 (0.29) Control 4.17 (0.49) 3.57 (0.56) 3.56 (0.76) Mean Squares (df) Population (3) 1.13 1.57(*) 0.31 Selection/drift (1) 2.82(*) 8.36(***) 0.23 Population x selection/drift (3) 2.17(*) 1.92(*) 1.15(**) Block (8) 0.53 0.36 0.15 Error (55) 0.26 0.32 0.28 * P |is less than~ 0.05; ** P |is less than~ 0.01; *** P |is less than~ 0.001.
Population Differences and the Correlated Response to Selection in Fitness. -- Population differences and the correlated response to selection of development time, viability, and fecundity were tested using ANOVA. The sequential Bonferroni technique was applied separately to each term in each of the three ANOVAs. A correlated response to selection was suggested by the significant selection term in experiment two for developmental time. However, the interpretation of this selection term must also take into account the possibility that drift may be responsible for any difference between the selected and control lines. Significant population by selection interactions were found in experiments two and three for developmental time, but there is no consistent pattern between experiments. A significant difference between populations was found in experiment three for viability.
Hybrid Breakdown in Fitness. -- Two types of statistical approaches were used to determine if hybrid breakdown in the between-population crosses in the control and selected lines had occurred for developmental time, viability, and fecundity. The first follows the procedure of Mather and Jinks (1982) described above. Results of the individual scaling tests on each population cross in the three experiments are not presented because they involve a many t-tests and are limited in power when compared with the joint scaling test (although the mean values of each test calculated across the three replicate lines for each population cross were of considerable use in the analysis with distance presented below). An overall pattern in hybrid breakdown in each of the three fitness traits was tested using ANOVA on the values of the three scaling tests. The significant population cross by selection interaction for development time suggests that selection has changed the level of hybrid breakdown expressed by the population crosses to different extents.
Joint scaling tests were calculated for each population cross in each of the three experiments, and the sequential Bonferroni technique was applied to the 36 tests produced for each of the three traits. The uncorrected results indicated a significant departure from the additive expectations in the selected lines for development time in the following two cases (|X.sup.2~ tests have 3 df): FT, experiment one (|X.sup.2~ = 15.101, P |is less than~ 0.005); and CT, experiment 2 (|X.sup.2~ = 73.47, P |is less than~ 0.001), although only the second test remains significant after correction. The control lines for developmental time displayed no significant hybrid breakdown. The joint scaling tests for viability and fecundity indicated no significant departure from the additive-dominance expectations in either the control or selected lines. Therefore, there is virtually no evidence for hybrid breakdown from the joint scaling tests for the three fitness traits.
TABLE 3. Viability for the parental populations measured as the number of eggs out of five surviving to pupation. Nine blocks within each experiment were pooled to give the means displayed, and therefore no standard deviations were available. A two-way ANOVA without replication was carried out for each experiment. Viability Experiment Experiment Experiment 1 2 3 Means (SD) Forster Selected 4.56 4.67 Control 4.78 4.38 3.89 Coffs Harbour 3.89 Selected 4.67 4.56 4.44 Control 5.00 4.11 4.44 Brunswick Heads Selected 4.67 4.67 4.78 Control 5.00 4.67 4.56 Townsville Selected 4.44 4.00 4.78 Control 4.33 4.67 4.67 Mean Squares (df) Population (3) 0.09 0.05 0.29(**) Selection (1) 0.08 0.00 0.01 Error (3) 0.02 0.12 0.01 ** P |is less than~ 0.01.
Although the joint scaling tests indicate little evidence for hybrid breakdown between these populations, it was important to directly consider whether selection had any affect on the expression of nonadditive genetic variation. To test whether selection had changed the amount of hybrid breakdown exhibited by a particular population cross, a paired t-test (pairing was between the control and selected line form each population within each of the three experiments) with 2 df was performed between the three control values and three selected values from each of the three individual scaling tests for development time and fecundity (a total of 18 tests for each trait). A two-tailed t-test between two means with 2 df was performed between the mean selected and control values from each of the three scaling tests for viability. A significant increase in hybrid breakdown in the selected lines was found for development time in the population crosses FT (test C; P |is less than~ 0.05) and BT (test B; P |is less than~ 0.02), but after applying the sequential Bonferroni technique across these 18 tests, no significant values remained. No significant differences between selected and control scaling test values were found for viability or fecundity.
Relationship between Hybrid Breakdown and Distance. -- A second approach to determine if selection had changed the amount of hybrid breakdown between selected and control lines was to plot each individual scaling test against the distance between the parent populations in the field (following Lynch 1991). As Lynch (1991) suggested, this analysis enables changes in the way genes interact as they diverge over time to be investigated. The three scaling tests are plotted against distance in figures 1 through 3 for developmental time, viability, and fecundity, respectively. The sequential Bonferroni technique was applied over the six regressions conducted for each trait. A significant regression slope was found in the control lines for developmental time (test B; b = -0.00059, P = 0.008) and a marginally nonsignificant slope was also present for test C (b = -0.00092, P = 0.060), although only the slope for test B was significant after correction, suggesting an increase in hybrid breakdown as distance between the populations increased. It should be noted that slow development, rather than fast development, seems to be associated with fitness, because hybrid breakdown occurred in the direction of faster development.
Similar patterns are found in figure 4, which plots a measure of |F.sub.2~ performance relative to the midparent value against distance. Lynch (1991) showed that this difference is equal to
|F.sub.2~ - midparent (MP) = (scaling test C) - (|F.sub.1~ - |F.sub.2~),
where scaling test C is the measure of hybrid breakdown described above and the difference |F.sub.1~ - |F.sub.2~ is the mean dominance crossbreeding effect or measure of heterosis. A significant increase in |F.sub.2~ performance with distance is present for developmental time in the control lines (b = 0.0027, P = 0.020) and a nonsignificant decrease in performance is present for the selected lines (b = -0.00019, P = 0.165). The |F.sub.2~, performance in viability also follows this pattern, with the relationship for the selected lines being only marginally nonsignificant (b = -0.00021, P = 0.068). After applying the sequential Bonferroni technique to each pair of regressions for the three traits, the regression for development time in the control lines remains significant.
TABLE 4. Number of eggs laid over a four-day peeriod for the four parental populations. A selected and control line from each population was tested in each of the three experiments. Means and their standard deviations are based on nine replicates. Mean squares are present from ANOVAs for each experiment. Fecundity Experiment 1 Experiment 2 Experiment 3 Mean (SD) Forster Selected 78.8 (26.2) 55.8 (13.6) 53.3 (17.6) Control 72.1 (26.3) 62.1 (29.6) 59.3 (6.8) Coffs Harbour Selected 58.6 (15.3) 39.3 (26.1) 55.5 (14.5) Control 55.4 (31.7) 54.0 (21.3) 40.6 (20.7) Brunswick Heads Selected 44.6 (34.4) 57.1 (11.8) 52.0 (11.3) Control 62.9 (30.0) 54.2 (25.5) 49.3 (16.3) Townsville Selected 59.7 (41.3) 46.7 (31.1) 42.0 (24.6) Control 54.0 (15.2) 58.7 (20.3) 61.5 (24.5) Mean Squares (df) Population (3) 1401 433 366 Selection/drift (1) 92 981 62 Population x selection/drift (3) 438 176 785 Block (8) 1622(*) 727 671(*) Error (df) 695 (41) 494 (45) 278 (47) * P |is less than~ 0.05.
ANCOVA was used to determine whether selection had changed the way the relative distances between the population crosses covaried with hybrid breakdown and |F.sub.2~ performance (i.e., distance was used as the covariate). ANCOVAs were carried out on the mean values displayed in figures 1 through 4 and the sequential Bonferroni technique was applied across the four ANCOVAs conducted for each trait. For developmental time, significant selection x distance interactions were found for scaling test B (fig. 1, P = 0.003) and |F.sub.2~ performance and a marginally nonsignificant interaction was found for test C, with the first two interactions remaining significant after correction. These results indicate that selection has changed the way distance influences the magnitude of hybrid breakdown. No significant interaction terms were found for the viability or fecundity data. Even though |F.sub.2~ performance in viability displays a trend similar to that in developmental time, a negative marginally nonsignificant slope for selected lines, ANCOVA indicated no significant change between control and selected lines (selection x distance interaction, P = 0.180).
TABLE 5. ANOVAs for the three fitness traits run over the three scaling tests. Mean squares (df) Developmental time Viability Fecundity Scaling test (2) 2.00 6.90(***) 853 Population cross (5) 1.34 0.34 849 Selection (1) 2.30 0.23 333 Population cross x selection (5) 2.76(**) 0.77 352 Error (72) 0.76 0.90 783 ** P |is less than~ 0.01; *** P |is less than~ 0.001.
Relationship between |F.sub.2~ Variance and Distance. -- As populations diverge, the |F.sub.2~ variance generated when the populations are crossed would be expected to increase (see Hedrick et al. 1978). The |F.sub.2~ variance is plotted against distance for the three fitness traits. No significant relationship between |F.sub.2~ variance and distance is apparent in either the selected or control lines for these three traits, nor does selection appear to have influenced the magnitude of these variances.
Relationship between Heterosis and Distance. -- The measure of the mean dominance crossbreeding effect or heterosis (Lynch 1991) is plotted against distance in figure 6. If it is assumed that heterosis is caused by dominance, as allele frequencies diverge, heterosis would be expected to increase (Falconer 1981). No significant relationship was found between heterosis and distance for these traits.
The experiments presented here were designed to test whether selection for an additively controlled character would increase the amount of epistatic genetic variance in fitness found between different geographic populations. Hybrid breakdown was measured in three fitness traits (developmental time, viability, and fecundity) in crosses between four populations of Drosophila serrata, which had undergone 14 generations of selection for desiccation resistance. Virtually no evidence for the existence of hybrid breakdown, before or after selection, was found when the individual crosses were examined.
However, when distance between populations in the field is considered, some patterns begin to emerge. When developmental time is measured in the control lines, scaling test B and |F.sub.2~ performance display a linear relationship with distance and a similar but nonsignificant relationship is present for scaling test C. Thus, there appears to be some evidence in the control lines for hybrid breakdown increasing as distance between populations increase.
It should be emphasized, that the four populations have probably also undergone selection for adaptation to the laboratory environment in the 18 mo before the measurement of hybrid breakdown. Crosses between the Brunswick Heads laboratory population and two populations sampled from the same site in August 1992 and October 1992, which had been in the laboratory environment for only two generations, displayed large levels of hybrid breakdown in developmental time (Blows, unpubl. data). It therefore seems prudent to interpret the results from the control lines as reflecting the result of selection for the laboratory environment as well as differences in the environment in the field.
Nevertheless, after selection for desiccation resistance, any relationship between hybrid breakdown and distance disappears. This pattern is reflected in the analysis of covariance, which suggests that selection has changed the way distance influences the level of hybrid breakdown in developmental time. It is also consistent with the ANOVA suggesting that selection for desiccation resistance seems to have differentially changed the level of hybrid breakdown in developmental time expressed by each population cross.
Thus, genes originating closer together acted more beneficially in concert in the control lines for developmental time than genes originating farther away. This may be a consequence of differences between environments in the field, or alternatively, the responses of the populations to 18 mo in the laboratory environment. However, the level of hybrid breakdown, when measured in the individual crosses, seems too small to be consistently statistically significant. After only 14 generations of selection for desiccation resistance, this relationship has been lost. A major limitation of this experimental design was that the fitness measurements were made only 4-6 generations after selection had ceased. Therefore, fitness had only the 14 generations during selection and a maximum of 6 generations after selection to evolve in response to an increase in alleles conferring desiccation resistance. Furthermore, a relatively moderate level of selection was used (50% mortality). It may be, that if selection for desiccation resistance had been maintained for a longer period, a new relationship with distance in developmental time may have been formed.
The relationship between genetic divergence and epistatic loss of fitness has been addressed theoretically by Hedrick et al. (1978), Hedrick (1985), and Lynch (1991). These models also predicted an associated change in the variance of the |F.sub.2~ (an increase as distance increases) and the heterotic effect observed in the |F.sub.1~ (an increase as distance increases up to an unidentified distance). No evidence was found for these relationships in either the selected or control lines.
Unfortunately, the interpretation of the evolution of developmental time in these populations is complicated by an inconsistency among the three replicate experiments. The population x selection interaction terms in the ANOVAs for developmental time from two experiments are significant (this term was not significant in experiment 1 after the sequential Bonferroni correction), indicating that a combination of selection and drift has affected the speed of development to different extents in the four populations. However, the extent to which selection has affected developmental time in the four populations is not consistent across the three experiments. This may suggest that some day-to-day fluctuations between experiments have differentially affected the four populations.
Such a day effect between experiments may also explain the opposing directions of the correlated response of developmental time to selection in experiments one and two (in experiment three, selected lines from two populations developed faster than their controls and two developed slower). Another, and perhaps more likely explanation, is that different replicates may be recovering from an initial pleiotropic cost suffered in developmental time during selection at varying speeds as a consequence of drift. These same experimental populations have previously been shown to exhibit significant replicate line effects in almost all traits measured to investigate correlated responses (Blows and Hoffmann 1993). Evidence for the evolution of maladaptive pleiotropic costs associated with selection for resistance has been found in resistance to pesticides (McKenzie et al. 1982). After 10 yr of continual exposure to diazinon, field populations of the blowfly Lucilia cuprina resistant to the pesticide had recovered an initial loss of fitness that had been associated with resistance. Lenski (1988) found that after 400 generations in the absence of T4 phage, resistant populations of Escherichia coli had maintained their resistance and also almost recovered an initial loss of fitness associated with resistance.
In summary, the shifting-balance theory of evolution requires that multiple-peak epistasis be an important mode of gene action between populations. Virtually no evidence was found to suggest that hybrid breakdown occurred among the four populations before or after selection for desiccation resistance when individual crosses were examined. However, some evidence was presented to suggest that selective differences in the field, or the responses of the populations to the laboratory environment, have created a relationship between hybrid breakdown before selection for desiccation resistance and distance: the farther away populations were in the field, the more hybrid breakdown was generated. After selection for desiccation resistance, this relationship was lost. Selection may need to be maintained for a longer period than that used in this study, and fitness may need to be monitored at regular intervals, if an association between multiple-peak epistasis and uniform selection is to be verified.
This work would not have been possible without the guidance and advice of A. Hoffmann. I would also like to thank R. Lenski, I. Marschner, M. Schwarz, M. Turelli, and S. Ward for many useful discussions on this topic. S. Ward wrote a computer program for use in the joint scaling tests. This work was supported by an Australian Postgraduate Research Scholarship.
Barker, J.S.F. 1979. Inter-locus interactions: a review of experimental evidence. Theoretical Population Biology 16:323-346.
-----. 1989. Population structure. Pp. 75-80 in W. G. Hill and T.F.C. Mackay, eds. Evolution and animal breeding. C.A.B. International. Wallingford, Oxon, U.K.
Birch. L. C., Th. Dobzhansky, P. O. Elliot, and R. C. Lewontin. 1963. Relative fitness of geographic races of Drosophila serrata. Evolution 17:72-83.
Blows, M. W., and A. A. Hoffmann. 1993. The genetics of central and marginal populations of Drosophila serrata. I. Genetic variation for stress resistance and species borders. Evolution 47:1255-1270.
Cavalli, L. L. 1952. An analysis of linkage in quantitative inheritance. Pp. 135-144 in E.C.R. Reeve and C. H. Waddington, eds. Quantitative inheritance. HMSO, London.
Cohan, F. M. 1984. Genetic divergence under uniform selection. I. Similarity among populations of Drosophila melanogaster in their responses to artifical selection for modifiers of |ci.sup.D~. Evolution 38:55-71.
Cohan, F. M., and A. A. Hoffmann. 1986. Genetic divergence under uniform selection. II. Different responses to selection for knockdown resistance to ethanol among Drosophila melanogaster populations and their replicate lines. Genetics 114:145-163.
Cohan, F. M., A. A. Hoffmann, and T. W. Gayley. 1989. A test of the role of epistasis in divergence under uniform selection. Evolution 43:766-774.
Dobzhansky, Th., and W. B. Mather. 1961. The evolutionary status of Drosophila serrata. Evolution 15:461-467.
Endler, J. A. 1977. Geographic variation, speciation, and clines. Princeton University Press, Princeton, NJ.
Enfield, F. D. 1977. Selection experiments in Tribolium designed to look at gene action issues. Pp. 177-190 in E. Pollack et al., eds. Proceedings of the International conference on quantitative genetics. Iowa State University Press, Ames.
Enfield, F. D., and F. Anklesaria. 1986. An evaluation of multiple peak epistasis and population structure in directional selection program. Pp. 283-297 in G. E. Dickerson and K. R. Johnson, eds. Third World Congress on Genetics Applied to Livestock Production. University of Nebraska Institute of Agriculture and Natural Resources, Lincoln.
Falconer, D. S. 1981. Introduction to quantitative genetics, 2d ed. Longman Scientific and Technical, Harlow, U.K.
Hedrick, P. 1985. Genetics of populations. Jones and Bartlett, Boston.
Hedrick, P., S. Jain, and L. Holden, 1978. Multilocus systems in evolution. Evolutionary Biology 11:101-184.
King, J. C. 1955. Evidence for the integration of the gene pool from studies of DDT resistance in Drosophila. Cold Spring Harbor Symposia on Quantitative Biology 20:311-317.
Lenski, R. E. 1988. Experimental studies of pleiotropy and epistasis in Escherichia coli. II. Compensation for maladaptive effects associated with resistance to virus T4. Evolution 42:433-440.
Lynch, M. 1991. The genetic interpretation of inbreeding depression and outbreeding depression. Evolution 45:622-629.
Mather, K., and J. L. Jinks. 1982. Biometrical genetics, 3d ed. Chapman and Hall, New York.
McKenzie, J. A., M. J. Whitten, and M. A. Adena. 1982. The effect of genetic background on the fitness of the diazinon resistance genotypes of the Australian sheep blowfly, Lucilia cuprina. Heredity 49:1-9.
Parsons, P. A. 1983. The evolutionary biology of colonizing species. Cambridge University Press, Cambridge.
Price, T., M. Turelli, and M. Slatkin. 1993. Peak shifts produced by correlated response to selection. Evolution 47:280-290.
Rice, W. R. 1989. Analyzing tables of statistical tests. Evolution 43:223-225.
Wade, M. J., and C. J. Goodnight. 1991. Wright's shifting balance theory: an experimental study. Science 253:1015-1018.
Wallace, B. 1959. Influence of genetic systems on geographic distributions. Cold Spring Harbor Symposia on Quantitative Biology 24:193-204.
Wright, S. 1932. The roles of mutation, inbreeding, cross-breeding, and selection in evolution. Proceedings of the Sixth International Congress of Genetics 1:356-366.
-----. 1956. Modes of selection. American Naturalist, 90:5-24.
-----. 1968. Evolution and the genetics of populations, vol. 1. Genetic and biometric foundations. University of Chicago Press, Chicago.
-----. 1970. Random drift and the shifting balance theory of evolution. Pp. 1-31 in K. Kojima, ed. Mathematical topics in population genetics. Springer, Heidelberg.
|Printer friendly Cite/link Email Feedback|
|Title Annotation:||The Genetics of Central and Marginal Populations of Drosophila serrata, part 2|
|Author:||Blows, Mark W.|
|Date:||Aug 1, 1993|
|Previous Article:||Genetic variation for stress resistance and species borders.|
|Next Article:||Disciplining evolutionary biology: Ernst Mayr and the founding of the Society for the Study of Evolution and Evolution (1939-1950).|