Fluctuating asymmetry, fecundity and development time in Drosophila: is there an association under optimal and stress conditions?

Authors

  • R. E. Woods,

    1. Department of Genetics, Centre for Environmental Stress and Adaptation Research (CESAR), Department of Genetics, La Trobe University, Bundoora, Victoria, Australia
    Search for more papers by this author
  • C. M. Sgrò,

    1. Department of Genetics, Centre for Environmental Stress and Adaptation Research (CESAR), Department of Genetics, La Trobe University, Bundoora, Victoria, Australia
    Search for more papers by this author
  • M. J. Hercus,

    1. Department of Genetics, Centre for Environmental Stress and Adaptation Research (CESAR), Department of Genetics, La Trobe University, Bundoora, Victoria, Australia
    Search for more papers by this author
  • A. A. Hoffmann

    1. Department of Genetics, Centre for Environmental Stress and Adaptation Research (CESAR), Department of Genetics, La Trobe University, Bundoora, Victoria, Australia
    Search for more papers by this author

Richard E. Woods, Department of Genetics, Centre for Environmental Stress and Adaptation Research (CESAR), La Trobe University, Bundoora, Victoria 3083, Australia. Tel.: 613 9479 1209; fax: 613 9479 2361; e-mail: r.woods@latrobe.edu.au

Abstract

A number of studies have reported a significant negative association between fluctuating asymmetry (FA) of bilateral morphological traits and individual fitness traits, but almost all of these are unreplicated and based on small sample sizes using single trait estimates of FA. We therefore tested if there was a relationship between the FA of five bilateral traits and fecundity and development time in Drosophila in a multiple replicated experimental design. Stressed treatments were included to increase the variability of traits and to test whether associations among traits were affected by changes in the environment. Significant positive relationships were found between the size of wing characters and mean fecundity for the 5-day period and this relationship tended to be stronger in the stress treatments. No association was found between FA and mean fecundity for any of the traits measured. Similarly, a significant positive relationship was detected between wing trait size and development time but no association was detected between trait FA and development time. There were no differences between mean fecundity or development time of extreme asymmetry phenotypes compared with modal phenotypes. These results are discussed with reference to suggestions in the literature that FA can be used to estimate individual fitness.

Introduction

Fluctuating asymmetry (FA) of bilateral morphological traits has been used in a wide range of studies as a measure of environmental or genetic disturbance during an organism's development or as an indicator of individual and population fitness. FA is particularly appealing as a measure of fitness, because unlike other fitness traits it is possible to define an optimum phenotype (perfect symmetry) (Palmer & Strobeck, 1986) from which deviations are thought to be nonadaptive (Watson & Thornhill, 1994). This has led some authors to suggest that FA can be used as an indicator of individual fitness (Møller, 1997; Møller & Swaddle, 1998). Other authors have suggested that developmental stability might provide an important tool to conservation biology for fitness estimations of endangered species populations (Clarke, 1995) and as a quality control indicator in insect mass rearing colonies (Clarke & McKenzie, 1992).

The present study considers several Drosophila experiments designed to test for an association between fitness traits and FA at the individual level. Two evolutionary mechanisms could maintain a relationship between FA and fitness (Markow, 1995). First, sexual selection could lead to an association between fitness and FA if levels of asymmetry influenced mate choice. The majority of experiments that have examined FA-fitness relationships have focused on sexually selected traits. Levels of FA are generally thought to be negatively correlated with fitness as some studies have demonstrated mate preference for and higher mating success in symmetrical individuals (see reviews of Møller, 1993; Watson & Thornhill, 1994). However, a number of studies have failed to detect such an association (Markow & Ricker, 1992; Jennions, 1998) and it is possible that studies reporting negative results have more difficulty reaching publication (Markow, 1995).

Secondly, natural selection could maintain an FA-fitness relationship if increased levels of FA are associated with reduced performance for life history traits. To date this proposal has received relatively little direct experimental testing, however, despite the lack of empirical evidence, a relationship between fitness traits and FA is often presumed to exist (Pomiankowski, 1997). Several authors have reviewed the published literature on studies examining the relationship between FA and fitness and have arrived at quite different conclusions about the strength of the relationship and the potential of FA to indicate fitness. Leung & Forbes (1996) conducted a meta-analysis on studies of FA-fitness relations and concluded that there was little evidence that increased levels of FA impacted negatively on fitness traits and that in general FA was a poor predictor of fitness. In contrast, a review by Møller (1997) claims there is clear evidence of a negative relationship between FA and components of fitness such as growth, survival, fecundity and longevity. This has been questioned by Clarke (1999) who suggested that of the studies cited by Møller (1997) as supporting a positive relationship between FA and fitness components, over half did not directly address the topic.

When investigating FA–fitness associations, it is essential that a clear distinction be made between studies that have tested for this association at the population level from those that have tested it at the individual level. Population level studies on different strains of Lucilia cuprina (Clarke & McKenzie, 1992; McKenzie & O'Farrell, 1993) have found correlations between FA and development time, percentage egg hatch and levels of larval densities. Although FA was originally intended as a population level measure it has recently been suggested that FA can also be used as an estimate of individual fitness (Møller, 1997; Møller & Swaddle, 1998). However, few empirical studies have directly tested for a relationship between individual FA and a fitness trait. Whereas some studies have found evidence of such a relationship (Thornhill, 1992; Naugler & Leech, 1994; Ueno, 1994), they must be interpreted with caution as they measured only a single trait for FA and were based on very small sample sizes. Moreover other studies have found no association between FA and fitness (e.g. Windig, 1998).

The experiments described in this paper examine the relationship between individual FA and two fitness traits in Drosophila: fecundity and egg to adult development time. Two papers (Parsons, 1962; Agnew & Koella, 1997) have been cited as providing evidence of a negative relationship between FA and fecundity (Møller, 1997). However, close inspection of these studies reveals that this conclusion is misleading as the FA-fecundity relationship was not measured directly in either study. Our experiments are multiple replications with control and stress treatments and address the following questions:

1. What is the relationship between morphological trait values and these fitness components?

2. Is the relationship considered in question 1 different in favourable and stress environments?

3. What is the relationship between FA of morphological traits and fitness components?

4. Is the relationship considered in question 3 different in favourable and stress environments?

Multiple replicated experiments may be required to detect FA associations as a previous study found inconsistent FA levels between replicates of the same treatment for some traits (Woods et al., 1999). There are two reasons for including stress treatments. First, stress might increase the variability of traits and thereby extend the range over which we can test for a relationship between trait values and FA. Secondly, patterns of correlations between traits may be affected by changes in the environment (Service et al., 1985; Holloway et al., 1990) and particular correlations may be environment dependent. Van Dongen & Lens (2000) have emphasized the need for studies that examine patterns of developmental stability under a range of stress environments. In the present study multiple traits were measured on both meristic (bristle) and metric (wing) characters, each with low levels of measurement error for estimates of trait size and FA (Woods et al., 1998, 1999). The FA of all individual traits was first tested for an association with the fitness traits. A composite index of FA from all morphological traits was then computed. It has been suggested that composites of multiple FA traits are more likely to detect significant correlations with fitness components as a composite increases the amount of FA variance accounted for by shared individual differences (Gangestad & Thornhill, 1999) and this has some empirical support (Hewa-Kapuge & Hoffmann, 2001).

Materials and methods

Stocks

The fecundity and development time experiments were conducted with different Drosophila stocks founded from wild collections of adult females from rotting fruit piles used as breeding sites. Flies for the fecundity experiment were obtained from a mass bred laboratory population of Drosophila melanogaster initiated in March 1995 from 100 field-inseminated females collected from an orchard at Wandin North, approximately 80 km north-east of Melbourne. Maintenance of the population is described in Sgrò & Hoffmann (1998). The population had been cultured in the laboratory for approximately 20 generations when the fecundity experiment was conducted. Development time was examined with three separate experiments using two populations of D. melanogaster and one population of D. simulans. For the first experiment, we used the offspring of 94 adult female D. simulans collected from a site (Silvan, 37°45′S, 145°25′ E) near Melbourne in March 1994. For the second experiment more than 100 adult female D. melanogaster were collected from a site near Cobram, Victoria (35°55′S, 145°40′E) and cultured as a mass bred population at a population size of around 2000 for 4 months prior to the experiment. Finally, more than 100 adult female D. melanogaster were collected at Flower Pot in Tasmania (43°12′S, 147°14′E) in February 1997 and cultured in the laboratory (n > 2500) for 12 months prior to the experiment.

Experiments

Fecundity was examined in a single multiple replicated experiment whereas development time was examined in three smaller independent experiments. In all experiments flies were cultured and maintained under a 24-h light regime and flies were held at 25 °C unless otherwise stated.

Fecundity

Fecundity data were obtained from an experiment examining heritability of fecundity in optimal and extreme environments (Sgrò, 1997; Sgrò & Hoffmann, 1998; Woods et al., 1999). Briefly, the experiment was conducted by following 100–150 families over three consecutive generations under optimal (25 °C) or stressful conditions that decreased fecundity and increased the variability of this trait (Sgrò & Hoffmann, 1998). Stress was imposed at the egg, larvae and adult stage of the life cycle. Stressful growth conditions for the eggs and larvae involved 8.5% ethanol added to a sucrose, agar, yeast medium which also contained 50% of the dried Torula yeast normally used in our medium [3% (v/v) instead of 6%]. In addition, eggs and first instar larvae were each exposed to a cold stress at 1 °C for 45 min by placing vials with medium into a refrigerated bath. The adult stress involved exposing flies in empty vials to a cold shock of −1 °C for 30 min prior to collecting eggs for the next generation. Fecundity of females from the optimal and stressed treatments was scored for the first 5 days after eclosion. A virgin female was paired with a virgin male from a different family of the same treatment and placed into an empty 30 mL vial containing a spoon filled with 2 mL of medium. Pairs were left for 24 h before spoons were replaced. The spoons from the previous 24 h were frozen at –15 °C for a day and then stored at 4 °C until eggs could be counted.

Development time

Egg-to-adult development time was scored in three experiments by collecting eclosing adults at intervals until all flies had emerged. Flies were frozen in plastic Eppendorf tubes and kept at –4 °C until they were scored for asymmetry.

Experiment 1. This was undertaken with the D. simulans population collected from Silvan. Field flies were kept in vials containing a sucrose-agar-dead yeast medium at a density of about 10 females per vial and fed live yeast for several days immediately after collection. They were then placed individually into vials with plastic spoons that contained a small amount of medium covered with a layer of live yeast. These vials were kept in darkness under high humidity for 10 h to increase egg laying. Four vials each with 10 eggs on 10 mL of medium were set up per female. Vials were placed at 25 °C and scored for emergence every 8 h. The experiment was repeated with vials placed at 28 °C; because development was more rapid at the higher temperature, emergence was scored every 6 h.

Experiment 2. This was conducted with the D. melanogaster population from Cobram. Five bottles each containing around 300 adults were attached to watch glasses containing 5 mL of medium covered with a layer of live yeast paste. Bottles were kept in darkness under high humidity for 10 h to stimulate egg laying. Eggs were transferred to vials with 10 mL of medium. Thirty eggs were set up per vial. This experiment was repeated four times. For two of the repeats, 50 vials were set up using the optimal conditions described for the fecundity experiment above. For the other two repeats, 100 vials were set up under stressful conditions identical to those described above for the fecundity experiment above, including the stress medium (low yeast-high ethanol) as well as cold shocks at the egg and larval stages. Development time was scored at 6-h intervals in the optimal treatment and 8-h intervals in the stressed treatment.

Experiment 3. This experiment used the D. melanogaster population collected from Flower Pot. Thirty eggs were placed into each of 200 vials containing the stress medium (low yeast-high ethanol). Both eggs and larvae were exposed to a cold shock and a third cold shock of 1 °C for 45 min was applied four days after vials were set up. Development time was scored at 8-h intervals.

Data analysis

Traits were measured following Woods et al. (1998, 1999). FA was estimated as the absolute difference between sides, and a total asymmetry value was calculated for each individual by summing a standardized value of the FA of each trait. Standardization involved subtracting the population mean and dividing through by the population standard deviation (Zar, 1996). Measurement error and tests for directional asymmetry and antisymmetry were carried out as described in Woods et al. (1998). The results of these analyses indicate that no directional asymmetry or antisymmetry was present in the samples and consequently FA was the only form of asymmetry analysed. A relationship between FA and trait size was tested by regression of absolute asymmetry on trait values following Palmer (1994). Regression coefficients were not significantly different from zero, indicating that FA does not vary with trait size in these samples.

To test for a relationship between morphological trait values and a fitness trait, a linear regression of an individual's fitness on morphological trait values was performed. Because several traits were scored for the same individuals, we corrected probability values for multiple comparisons using the Dunn–Sidak method (Sokal & Rohlf, 1995). For experiments that included control and stressed treatments, we also tested if the regression coefficients were significantly different between these treatments using a Tukey–Kramer test (Sokal & Rohlf, 1995). To estimate the distribution of fitness values for different traits and the type and intensity of selection on traits, we computed selection coefficients (Lande & Arnold, 1983; Endler, 1986). Trait and fitness values were first standardized by subtracting the population mean and dividing through by the population standard deviation (Zar, 1996). To estimate directional and variance selection (stabilizing or disruptive) independent of the effects of each other, a multiple regression was performed. For each morphological trait, the fitness character was regressed onto the trait value and a square of the trait value. This process was repeated for the asymmetry values of each trait.

We also examined genetic correlations between fecundity and the morphological traits. As data from individuals within families over three generations were available, it was possible to calculate additive genetic correlations between fecundity and the wing characters based on parent offspring comparisons involving generations 1/2 and generations 3/4. Standard errors for these estimates were computed following Falconer & Mackay (1996). Genetic correlations for asymmetry were not computed because of the absence of significant phenotypic associations between fecundity and asymmetry (see below) as well as the lack of heritable variation for asymmetry (Woods et al., 1999).

To test for a relationship between extreme phenotypes and FA (cf. Soulé & Cuzin-Roudy, 1982; Hoffmann et al., 1999), the mean fecundity and development time of extreme asymmetry phenotypes were compared with modal phenotypes for each morphological trait. Individuals from each treatment were sorted into two groups: extreme asymmetry phenotypes were determined as those that were more than two standard deviations from the population asymmetry mean, and modal phenotypes as those whose asymmetry values were within two standard deviations. The fecundity and development time values of these two groups were then compared by single factor analyses of variance.

Results

Trait values and fitness components

Fecundity

Mean fecundities for the two treatments and different generations are summarized in Table 1. Sample sizes and means for morphological traits are also given. The fecundity of females reared under the stress conditions was lower than that of females from the control treatments.

Table 1.   Morphological and fecundity data for three generations of D. melanogaster in optimal and stressed environments. Morphological data shown are trait means (average of left and right sides), coefficients of variation (CV) and absolute fluctuating asymmetry (FA) of the two bristle and three wing characters. Fecundity data are the means and standard deviations of the eggs laid over a five day period. Sample sizes for control treatments were N = 125 (generation 1), N = 179 (generation 2) and N = 236 (generation 3). For stress treatments, samples sizes were N = 98 (generation 1), N = 212 (generation 2) and N = 123 (generation 3). Thumbnail image of

The results of regressions of fecundity on trait means for all morphological characters in each treatment are presented in Table 2. After adjustment of probability values for the number of traits scored, significant regressions were evident in five out of eighteen wing comparisons, all of them in the stress treatment. Following the methods outlined in Lande & Arnold (1983), a quadratic term was used in a multiple regression of fecundity on all morphological traits to test for stabilizing and disruptive selection. After correction for multiple comparisons, no quadratic terms were significant (data not presented) indicating that selection was only directional. Comparisons of regression coefficients (results not shown) indicated significant differences between stress and favourable conditions for wing width and wing length in generations 1 and 2.

Table 2.   Coefficients (b) and standard errors (SE) from regression of fecundity on trait values for three generations of control and stress treatments. Probability values are for one-tailed t tests to determine if the regression coefficients are significantly different from zero. Regression coefficients were calculated on standardized data. Thumbnail image of

Genetic correlations between fecundity and morphological traits could not be computed when there were negative regression coefficients for parent–offpspring comparisons of traits: this involved four of the twenty comparisons (Table 3). The remaining correlations between fecundity and wing characters tended to be negative whereas those involving wing traits tended to be positive (Table 3). Several of these correlations were more than three standard errors from 0 suggesting that the association between fecundity and wing traits has a genetic basis.

Table 3.   Genetic correlations (rA) and standard errors between fecundity and morphological traits based on comparisons of parents and offspring. Because measurements were made over three generations, two parent- offspring comparisons could be made. Thumbnail image of

Development time

In the first experiment undertaken with D. simulans, development time was shorter in the 28 °C treatment (mean=175.80 h, SD=6.54, n=300) compared to the 25 °C treatment (mean 200.75 h, SD=8.89, n=325) as expected. The trait means of all three bristle traits were decreased in the 28 °C treatment compared with the 25 °C treatment (Fig. 1). Two of these differences (for the sternopleural and orbital traits) were significantly different (t229=3.05, P=0.002, and t229=2.95, P=0.003, respectively) (Fig. 2). Trait variability was higher for all three traits in the 28 °C treatment, although none of these differences was significant (Fig. 2).

Figure 1.

 Mean and SD (shown by error bars) of traits in development time experiments.

Figure 2.

 Mean and SE (shown by error bars) of fluctuating asymmetry of morphological traits in development time experiments.

In the second experiment involving D. melanogaster, development time was shorter in the control (mean =201.94 h, SD=9.90 h, n=111 and mean=200.41, SD=9.25, n=102 for replicates 1 and 2, respectively) than stressed (mean=268.94 h, SD=18.64 h, n=132 and mean=272.60, SD=18.38, n=158 for replicates 1 and 2, respectively) treatments. The difference between the replicates probably reflects minor differences in culture temperatures because replicates were tested at different times. Means of wing traits were found to be significantly larger in the stressed treatment in five of the six comparisons (Fig. 2). Only the mean of wing length in replicate 2 was not different between treatments. This contrasts with a previous study in which a significant and consistent reduction in the means of wing characters was found in the stressed treatments (Woods et al., 1999). The means of bristle characters were only significantly different for the sternopleural trait in the second replicate which showed a reduction in the stressed treatment (Fig. 2). Trait variability (CV) was not significantly different between stressed and control treatments in either replicate 1 or 2 for any trait (data not presented).

In the third experiment with D. melanogaster from Tasmania tested only under a stress treatment, the mean development time was 302.49 h (SD=24.07 h, n=313). Flies from this experiment were smaller than those from experiment 2 although this population came from a lower latitude (Fig. 2). This may reflect the additional cold shock to which the larvae in this experiment were exposed.

To test for a relationship between development time and trait values, linear regressions were conducted on all traits for each of the treatments in the three populations (Table 4). Following correction for multiple comparisons, significant negative regression coefficients were detected for the majority (10 of 15) of the wing traits (length, width and cross vein length). When the stressed treatments of experiments 2 and 3 are considered, eight of nine regressions for wing characters are highly significant. Wing characters are known to be positively correlated with body size, which suggests that larger flies developed faster than smaller flies under the experimental conditions used here. No significant results were obtained from the regression of development time onto bristle trait values. Tests for stabilizing and disruptive selection by multiple regression with a quadratic term were also not significant for any trait (data not presented) indicating that only directional selection was present. To test whether regression coefficients were significantly different between control and stressed environments in the second experiment, a Tukey–Kramer test (Sokal & Rohlf, 1995) was performed with each trait between environments. Significant differences were detected in comparisons of wing width and wing length between control and stress treatments in replicate 2 (results not shown).

Table 4.   Coefficients and standard errors from regression of development time on trait values for experiment 1 (control and stress treatments), experiment 2 (control and stress treatments, replicates 1 and 2) and experiment 3 (stress treatment). Probability values are for one-tailed t tests to determine if the regression coefficients are significantly different from zero. The stress conditions in experiment 1 was a treatment developed at 28°C, while for experiments 2 and 3 a combined stress was used. Regression coefficients were calculated on standardized data. Thumbnail image of

Asymmetry and fitness

Fecundity

To test whether asymmetry could predict fecundity levels, regressions of mean fecundity on the fluctuating asymmetry of the five morphological traits and the composite total asymmetry value were performed. Coefficients and their standard errors are presented in Fig. 3. Following correction for multiple comparisons, no regression coefficient was found to be significantly different from zero. The sign of the regression coefficient was inconsistent with almost half (17 of 36) being negative.

Figure 3.

 Linear regression coefficients and their standard errors of regressions of fitness measures (fecundity and development time) on fluctuating asymmetry of morphological traits in stress and control environments.

The fecundity of extreme asymmetry phenotypes (outgroup) was compared with mean fecundity levels of individuals whose asymmetry was within two standard deviations of the population mean (ingroup). After correction for multiple comparisons, none of the comparisons were significant (the lowest P-value without correction (0.0124) was for the comparison of cross veins in generation (3). In both the control and stress treatments, fecundities of ingroups were relatively higher in nine of fifteen comparisons. Thus there is no evidence that the fecundity of highly asymmetrical individuals is lower than individuals with a modal asymmetry phenotype.

Development time

Fluctuating asymmetry was not significantly different between treatments in experiment one for any of the individual traits (Fig. 2) or for the total asymmetry comparison (Z=0.052, P= 0.958, d.f.=299). In experiment 2 a significant difference in FA levels was detected between treatments for the cross vein character in replicate two which showed lower levels of FA in the stressed treatment than the control treatment (Fig. 2). Comparisons of total asymmetry values between treatments in experiment two did not reveal any differences (Z=–0.284, P=0.776, d.f.=110 for replicate one; Z=–0.177, P=0.859, d.f.=101 for replicate two).

To test for a relationship between development time and the asymmetry of each trait and the total asymmetry value, linear regressions were performed (Fig. 3). A significant negative association was only detected for the orbital bristle character in replicate one of the control treatment in experiment two. The direction of the regression coefficients was not consistent with almost half (17 of 36) having a negative sign. These results indicate that development time is not consistently related to the asymmetry of individual traits or to the total asymmetry value in the control or stress environment.

The development time of extreme asymmetry phenotypes was compared by the method previously described. No significant differences were found between the mean development time of outgroups and ingroups for any of the traits (results not shown). Means for ingroups exceeded those for outgroups in eight of thirteen comparisons for the control treatments and nine of sixteen comparisons of the stress treatments. Hence, the development time of highly asymmetrical individuals is not different to that of individuals with a modal asymmetry phenotype.

Discussion

Fitness components and trait asymmetry

The experiments described in this paper were primarily designed to investigate FA-fecundity and FA-development time relationships at the individual level in Drosophila under optimal and stress environments. Tests for a relationship between FA and mean fecundity were not significant, indicating that there is no association between individual FA of any of the characters and mean reproductive output. The relationship between individual FA and fecundity in invertebrates has not been empirically tested prior to the experiments described here; the results indicate that FA is not an accurate measure of individual fecundity and should not be used as a surrogate measure of reproductive potential in Drosophila.

Despite the fact that we used stressful conditions, as evident from the considerable reductions in egg-to-adult viability and significant decreases in body size between control and stressed treatments, there was no evidence that an association between FA and fecundity became evident under these conditions. This supposition follows from the hypothesis that stressful conditions exacerbate the expression of individual differences in developmental instability. Even in the case of FA in orbital bristles, which showed significant and consistent increases in FA levels under the conditions tested here (Woods et al., 1999), there was no evidence for an association between FA and fitness.

We also found no relationship between FA and development time for any trait in either the control or stress treatments. One concern is that we had little power to test for this association because development time was only scored every 6–8 h. To test the power of our design to detect such a relationship we simulated random data sets from a bivariate normal distribution. The data sets were the same size as those of the experiment from distributions using observed trait means and variances with different levels of correlation between the traits. Distributions for development time and trait asymmetry were broken down into discrete intervals and the simulated data were then analysed in the same manner as for the experiment. These simulations indicated that with our sample sizes and design we could have detected a significant correlation for FA as low as 0.2 for the bristle traits and 0.15 for the wing traits (at P < 0.05).

Another concern with our experiments is that the stresses we used may have resulted in the death of the most asymmetrical individuals from a sample. Thus no relationship between FA and fitness may have detected under stress conditions because only the most symmetrical individuals were measured. To further examine this possibility, experiments could be undertaken under less stressful conditions that do not result in mortality of immature stages.

In attempting to examine the relationship between FA and fitness it is important to select appropriate morphological traits for estimates of FA, as only specific traits are likely to reliably detect the relationship (Leung & Forbes, 1996). It has been suggested that traits whose symmetry directly influences fitness will be the most reliable indicators of an FA–fitness association (Eggert & Sakaluk, 1994; Palmer, 1994). One study that may have identified such a relationship was a field experiment with a parasitoid wasp (Bennett & Hoffmann, 1998), which found that released wasps with lower asymmetry for a wing length trait were more likely to reach oviposition sites. In this instance it seems plausible that symmetry of wing length would directly influence flying capabilities and hence fitness. In the present study, we have shown that particular traits are under directional selection. However, this does not mean that the asymmetry in the traits is under selection: for instance while wing length traits are positively correlated with fecundity, most likely because large flies have a relatively higher egg production, it is unlikely that wing trait asymmetry will directly influence egg production. It remains to be seen if there are consistent associations between FA and fitness when the asymmetry in specific traits is likely to be directly selected.

Trait values and fitness components

We anticipated a positive relationship between the size of wing characters and fecundity as wing characters are known to be highly correlated with body size and previous studies have demonstrated a relationship between body size and fecundity (Chiang & Hodson, 1950; Robertson, 1957). In agreement, we found that large flies tended to have a higher fecundity. Selection on wing characters was predominantly directional with only one significant quadratic term in the multiple regression analysis. The strength of selection appeared to be influenced by the environment; in the stress environment there was stronger selection for larger flies as indicated by the significantly higher selection coefficients of directional selection for wing width and wing length characters. Genetic correlations between wing characters and fecundity also tended to be higher in the stressed treatments.

In contrast to the fecundity results, size associations with development time did not fit findings from other studies. Previous studies have usually found a positive relationship between development time and body size (Robertson, 1960; Roff, 1992), whereas the negative coefficients from the linear regression of development time on wing characters in experiment two suggest that larger flies developed faster. In previous Drosophila studies, body size or development time was scored as correlated responses to selection rather than directly at the phenotypic level. For example, D. melanogaster lines selected for shorter larval development period have shown a correlated decrease in body size (Zwaan et al., 1995; Nunney, 1996; but see Robertson, 1963). Similarly, D. melanogaster lines selected for increased body size have an associated increased development period (Partridge et al., 1999).

These different patterns may relate to the stressful conditions under which some of the flies were reared. Because the stress conditions led to mortality, the emerging individuals represent a biased sample of the initial population and it is possible that the emerging individuals were the largest ones. In the absence of small individuals, any association between size and development time may have become obscured. The fact that the stress conditions were associated with an increase in wing size in experiment 2 is consistent with this hypothesis. In contrast, stress conditions in the fecundity experiment undertaken with a different stock of D. melanogaster resulted in a reduction in the size of the wing traits. However, this hypothesis does not explain the negative correlation between development time and wing traits observed under the control conditions when mortality levels were low.

Concluding remarks

The experiments described in this paper are one of the few empirical tests for an association between individual FA and both fecundity and development time in an invertebrate. The results suggest that under optimal and stress environments the FA of several meristic and metric characters when analysed individually or in composite do not reflect an individual's fitness for either life history trait. A third trait, body size, considered by some authors as closely related to fitness in Drosophila (Partridge & Farquhar, 1981; Wilkinson, 1987), was also uncorrelated with FA in the experiments. In the literature, studies reporting a negative correlation between FA and other life history traits in invertebrates have either considered the relationship at the population level, or are characterized by small sample sizes and single trait estimates of FA. Individual FA may be a poor indicator of fecundity and development time even under stress conditions although the issue of whether stress results in individuals with a high level of FA remains to be resolved.

Ancillary