Environmental dependence of inbreeding depression and purging in Drosophila melanogaster


Bijlsma Department of Genetics, University of Groningen, Kerklaan 30, NL-9751 NN Haren, The Netherlands. Tel: +31 50 3632117; fax: +31 50 3632348; e-mail: r.bijlsma@biol.rug.nl


Elimination or reduction of inbreeding depression by natural selection at the contributing loci (purging) has been hypothesized to effectively mitigate the negative effects of inbreeding in small isolated populations. This may, however, only be valid when the environmental conditions are relatively constant. We tested this assumption using Drosophila melanogaster as a model organism. By means of chromosome balancers, chromosomes were sampled from a wild population and their viability was estimated in both homozygous and heterozygous conditions in a favourable environment. Around 50% of the chromosomes were found to carry a lethal or sublethal mutation, which upon inbreeding would cause a considerable amount of inbreeding depression. These detrimentals were artificially purged by selecting only chromosomes that in homozygous condition had a viability comparable to that of the heterozygotes (quasi-normals), thereby removing most deleterious recessive alleles. Next, these quasi-normals were tested both for egg-to-adult viability and for total fitness under different environmental stress conditions: high-temperature stress, DDT stress, ethanol stress, and crowding. Under these altered stressful conditions, particularly for high temperature and DDT, novel recessive deleterious effects were expressed that were not apparent under control conditions. Some of these chromosomes were even found to carry lethal or near-lethal mutations under stress. Compared with heterozygotes, homozygotes showed on average 25% additional reduction in total fitness. Our results show that, except for mutations that affect fitness under all environmental conditions, inbreeding depression may be due to different loci in different environments. Hence purging of deleterious recessive alleles can be effective only for the particular environment in which the purging occurred, because additional load will become expressed under changing environmental conditions. These results not only indicate that inbreeding depression is environment dependent, but also that inbreeding depression may become more severe under changing stressful conditions. These observations have significant consequences for conservation biology.


During the last two decades, the potential loss of genetic variability for adaptive evolution and the loss of fitness faced by small and isolated populations, due to genetic drift and inbreeding, has been a major issue in conservation biology ( Frankel & Soulé, 1981; Soulé, 1986, 1987; Lande, 1988; Van Treuren et al., 1991 , 1993; Hedrick & Miller, 1992; Bijlsma et al., 1994 , 1997; Loeschcke et al., 1994 ; Frankham, 1995). The detrimental effects of inbreeding have been observed for many centuries and have been subject to extensive discussions from a number of perspectives ( Darwin, 1876, 1885; Lerner, 1954; Charlesworth & Charlesworth, 1987). The loss of fitness that often accompanies the increase in homozygosity due to inbreeding is generally referred to as inbreeding depression, which has become a topic of interest in recent years for conservation biology.

Although the precise genetic basis of inbreeding depression is still debated and overdominant genes might be involved in the causation of inbreeding depression, most of the fitness effects are thought to be the result of increased homozygosity for (partially) recessive deleterious alleles ( Charlesworth & Charlesworth, 1987). Based on studies of Drosophila using chromosome balancer techniques ( Dobzhansky & Queal, 1938; Wallace, 1956; Sved & Ayala, 1970), it appears that approximately half of the inbreeding depression is due to rare recessive lethal or sublethal mutations, while the remainder can be attributed to a large number of mildly deleterious mutations ( Simmons & Crow, 1977; Charlesworth & Charlesworth, 1987). Others have suggested that part of the inbreeding depression may also result from synergistic or epistatic interactions ( Templeton & Read, 1994, and references therein; Charlesworth, 1998). The level of inbreeding depression expressed for a given population strongly depends on the rate of inbreeding. If inbreeding proceeds gradually there appears to be ample opportunity for natural selection to purge recessive deleterious mutations ( Ehiobu et al., 1989 ; Hedrick, 1994; Lande, 1995). Except for very mildly deleterious mutations, with only a slight effect on fitness in homozygous condition, which can easily become fixed at even low rates of inbreeding ( Barrett & Charlesworth, 1991; Hedrick, 1994; Lande, 1995), most detrimentals and certainly all lethals and sublethals are expected to become purged from populations for even moderate rates of inbreeding unless the number of these deleterious alleles in the populations is extremely high ( Hedrick, 1994).

Though the phenomenon of inbreeding depression has been studied extensively in itself, only a few studies have documented the occurrence of inbreeding depression in natural populations ( Van Noordwijk & Scharloo, 1981; Jiménez et al., 1994 ; Keller et al., 1994 ; Hedrick et al., 1996 ; Keller, 1998) and empirical data on this subject are often conflicting. The importance of inbreeding for the persistence of endangered species has therefore become increasingly questioned ( Caro & Laurenson, 1994; Caughley, 1994). One of the most frequently used arguments to reason that inbreeding problems for small natural populations are far less than expected is that populations, in principle, can reduce or eliminate inbreeding depression by evolutionary changes at the contributing loci (purging) ( Templeton & Read, 1984; Barrett & Charlesworth, 1991; Hedrick, 1994; Ballou, 1997). This view is supported by the observation that populations of different species that have gone through severe bottlenecks in the past and thereby lost most of their genetic diversity nevertheless prosper currently ( Ellegren et al., 1993 ; Hoelzel et al., 1993 ). This suggests that these populations indeed purged most of their genetic load during the early stages of the inbreeding process and preserved a relatively unaffected long-term fitness. Similarly, Lacy & Ballou (1998) suggest that the differences in inbreeding load expressed upon inbreeding for different Peromyscus polionotus populations can possibly be explained by different histories of inbreeding and purging.

However, several studies have suggested that inbreeding depression strongly depends on the environmental conditions, and that the inbreeding load becomes increasingly expressed under more stressful conditions ( Coman & Wallace, 1973; Miller, 1994; Dahlgaard et al., 1995 ; Bijlsma et al., 1997 ; Loeschcke et al., 1997 ). Miller (1994), for instance, demonstrated that Drosophila melanogaster flies homozygous for the second chromosome showed a significant increase in inbreeding depression under lead stress compared with the same homozygotes in an ‘unleaded’ environment. Therefore, the purging process might only be effective under specific environmental conditions but inbreeding depression could still be expressed (again) under changing and deteriorating conditions. Thus, the consequences of inbreeding will not only depend on the previous history of inbreeding and selection, but will also strongly depend on the (then) prevailing environmental conditions. As many endangered species actually have to cope with catastrophic deterioration of and rapid changes in their environment, it is important to evaluate the nature of purging and its consequences for the persistence of endangered species.

Because of the problems inherent to conducting research on natural and endangered populations, we investigated the effects of inbreeding on fitness in an experimental situation by using Drosophila melanogaster as a model system. Using the classical balancer technique ( Muller, 1930; Dobzhansky & Queal, 1938; Wallace, 1956) we sampled second chromosomes from a wild population and estimated their viability both in homozygous and in heterozygous condition. By selecting quasi-normals, i.e. chromosomes that in homozygous condition had a viability comparable with that of the balancer heterozygote, we artificially purged those chromosomes that carried deleterious recessive alleles from the population. Next we tested these selected quasi-normal chromosomes for fitness as homozygotes under both optimal (control) and different environmental stress conditions. We show that those homozygotes that initially had a normal relative fitness under optimal conditions often show significantly decreased fitness under stress conditions, indicating that purging of deleterious recessive alleles depends significantly on the environment.

Materials and methods


For the experiments flies from the Groningen 83 (G83) wild type strain were used. This strain was founded in 1983 with 403 females captured at the fruit market in Groningen (The Netherlands), and maintained as a large population since ( Zwaan et al., 1991 ). Unstressed flies were raised under standard conditions at 25 °C, 40–60% RH, and normal food containing 32 g dead yeast, 54 g sugar, 17 g agar and 13 mL nipagine solution (10 g nipagine in 10 mL 96% alcohol) per litre.

Chromosome extraction procedure

By means of the well-known balancer technique ( Muller, 1930; Dobzhansky & Queal, 1938; Wallace, 1956), second chromosomes were isolated from the G83 population. This procedure results in flies that are either homozygous (isogenic) for one entire second chromosome (almost 45% of the total genome) or heterozygous for this same chromosome over a standard balancer chromosome ( Fig. 1, cross IV). We modified the procedure slightly by using two different balancer stocks in the initial crosses (IA and IB), and we used 2–3 females in cross II. This procedure ensured that the resulting wild type flies from cross III, although isogenic for the second chromosome, were highly variable with respect to the rest of the genetic background. As the CyO/CyO homozygotes die in the embryonic stage due to a recessive lethal, cross III results only in two genotypes with respect to the 2nd chromosome: either homozygous for one specific extracted wildtype chromosome or heterozygous for the same wildtype chromosome in combination with a standard CyO chromosome. Assuming normal Mendelian segregation, the expected ratio of homozygotes (+/+) to heterozygotes (+/CyO) is 1:2. Therefore, the viability (egg to adult survival) of these chromosome homozygotes can be expressed relative to the viability of its accompanying heterozygotes, based on the numbers for each genotype, as twice the ratio (+i/+i)/(+i/CyO). By this procedure the relative viability of homozygotes (RV) for a specific wildtype chromosome was determined for 453 independently extracted 2nd chromosomes. Those chromosomes that showed a RV between 0.80 and 1.20 (quasi-normals) in homozygous condition were selected for further experimentation, thereby artificially purging most of the chromosomes carrying lethal or deleterious recessive alleles. These lines were maintained by selecting heterozygous females every second or third generation, in order to keep them segregating for both the wildtype and the balancer chromosome.

Figure 1.

 Schematic representation of the chromosome extraction procedure. Note that in the crosses IB and II single males are used, and consequently all wild-type homozygotes (+i/+i) emerging from cross IV are isogenic for one entire second chromosome. For explanation of the mutant symbols see Lindsley & Zimm (1992).

Viability experiments

Of a set of 53 quasi-normal lines the RV of the homozygotes was again tested five generations after the initial extraction. For each line, the RV was determined from the cross between +i/CyO females and +i/+i males, and calculated as the number of homozygous wildtype offspring divided by the number of heterozygous +i/CyO offspring. From 200 eggs cultured under uncrowded conditions, the RV was determined both under standard control conditions (CON) and under two stressful conditions: high temperature (HT), where the breeding temperature was increased to 30 °C, and ethanol stress (ETH), in case of which the standard food was supplemented with 12% ethanol. For logistic reasons the latter was only done for a subset of 38 of the quasi-normal lines.

Equilibrium experiments

The basic theory behind these experiments is outlined by Sved & Ayala (1970). CyO/CyO homozygotes are lethal and +/+ homozygotes are in most cases at a disadvantage compared to +/CyO heterozygotes. Then the overall fitness of heterozygotes is greater than both homozygotes, the classical overdominance model, and a heterotic equilibrium is expected with a frequency of the wildtype chromosome of 1/(1 + s), s being the selection coefficient against the wildtype homozygote. If, thus, homozygotes and heterozygotes are allowed to compete in a population for several generations, the fitness of the homozygotes can readily be estimated from the equilibrium value that eventually will be reached in the population. Only when the fitness of the chromosome homozygotes is the same or larger than that of the balancer heterozygotes, which was seldom the case (see Results section), the balancer will be lost from the population and the relative fitness of the chromosome homozygotes cannot be adequately estimated.

This procedure was used to estimate the relative fitness of 50 quasi-normal wildtype chromosome homozygotes, selected out of the 53 used in the viability experiment. For each of the lines a cross between +i/CyO females and +i/+i males was made (10 pairs of flies in a half pint bottle). After hatching, the progeny was transferred to a new bottle with fresh medium and the females were allowed to lay eggs for the next generation. After these females had produced 300–400 eggs, all parents were removed. Every odd generation these parents were scored phenotypically to determine the frequency of the chromosome homozygotes and heterozygotes. This was done for the following environmental conditions:

1 Control (CON): on standard food and at 25 °C.

2 High temperature (HT): as control but with a breeding temperature of 28.5 °C. (Compared with the viability experiment the temperature had to be lowered as it was impossible to complete the entire life cycle at 30 °C).

3 Ethanol (ETH): as control but 10% ethanol added to the food.

4 Alternating (ALT): an alternation between environmental conditions 2 (every even generation) and 3 (every odd generation).

5 DDT: as control but 5 p.p.m. DDT was added to the food. The concentration was increased to 7.5 p.p.m. in generation 3, to 8.5 p.p.m. in generations 4–7, and finally to 9 p.p.m. DDT in generations 8 and 9.

6 Crowding (CRW): as control, but the females were allowed to lay eggs for a considerably longer period resulting in, on average, the production of more than 600 eggs, and thus severe crowding, in the bottles.

In all environments, except of course CRW, females were removed after they had laid between 300 and 400 eggs, which means that these populations were maintained under uncrowded conditions. Generally, offspring were transferred to new bottles after all flies had emerged from the pupae. For CRW, as a consequence of crowding, developmental time was greatly increased and rather skewed, with many individuals developing very slowly. Therefore, in this environment the offspring from these bottles were already transferred to new bottles and scored phenotypically when only 50–70% of the progeny had emerged. If homozygotes and heterozygotes did differ in developmental time, this procedure might have affected the fitness estimates. From this progeny a random sample of approximately 250 flies was scored phenotypically.


Viability experiment

Out of a total of 540 independent males ( Fig. 1, cross IB), the RV of 453 chromosome homozygotes could be determined. In the homozygotes, recessive or partly recessive detrimental alleles become expressed, resulting in a significantly decreased viability of its carriers ( Fig. 2, top). About 15% of the chromosomes were found to be lethal or nearly lethal (RV < 0.10), while another 35% of the chromosomes carried detrimentals (0.10 < RV < 0.80). On the other hand, approximately 12.5% of the chromosomes revealed supervitals (RV > 1.20) when made homozygous. The picture emerging is not unlike that generally observed for natural populations ( Wallace, 1956; Tracey & Ayala, 1974).

Figure 2.

 The effect of high temperature and ethanol stress on the relative viability (egg to adult survival) of 2nd chromosome homozygotes. The top figures (identical) show the frequency distribution of the relative viabilities of all independently extracted 2nd chromosomes under control conditions. The quasi-normals with a relative viability between 0.80 and 1.20 are indicated by the horizontal bar in these top figures. The middle figures show the viability of a subset of quasi-normals under unstressed control conditions five generations later. (The 38 chromosomes in the right figure are a subset of the 53 to the left). The bottom figures show the viability of the same subsets but under stressed conditions: high temperature stress (left) and ethanol stress (right). N denotes the number of chromosomes tested.

The process of purging by natural selection was simulated by selecting only quasi-normals (0.80 < RV < 1.20). The 53 selected quasi-normals had, of course, a significantly increased mean RV (Mann–Whitney U-test, < 0.0001) and a greatly reduced variance (F-test, F452,52 = 7.32, < 0.0001) compared to the whole group of 453 lines from which they were selected (Table 1, A vs. B). When retested, five generations after the chromosome lines were isolated, the effect of artificial purging is still clearly noticeable ( Fig. 2, middle): most low-viability chromosomes have been purged from the distribution. Although the variance in RV is still significantly lower than observed for the whole set of extracted lines, the variance has increased significantly compared with the situation directly after isolation (Table 1, B vs. C, F452,52 = 2.66, < 0.0003). Because directly after isolation we only obtained a point estimate of the RVs, such an increase in variance is expected in these parent–offspring combinations due to environmental deviations. Moreover, the chromosome extraction results, with respect to the 1st and 3rd chromosome of D. melanogaster, in strong departures from linkage equilibrium. Fitness interactions between the 2nd chromosome and its genetic background and changes in the genetic background therefore could also partly account for the observed increase in variance in generation 5. Together with the fact that we selected chromosome lines within a narrow range (0.80 < RV < 1.20), this does explain the apparent lack of correlation between the RVs of lines directly after isolation and five generations later ( Fig. 3; Spearman rank correlation rs = 0.0343, P = 0.806).

Table 1.   Mean relative viability and variance of the second chromosome homozygotes for the different sets tested in the stress viability experiment. Thumbnail image of
Figure 3.

 The relation between the relative viability (ratio) of the chromosome homozygotes of the selected quasi-normals at the end of the extraction procedure and their relative viability (ratio) when retested after five generations. The diagonal represents the values of the ratio being the same for both generations.

The applied stresses, HT and ETH, affected the egg-to-adult survival considerably, and total absolute viability, irrespective of the genotype, was decreased 44.6% and 36.8%, respectively, compared with control conditions (data not shown). For HT this severe stress had, moreover, a significant effect on the RV of the chromosome homozygotes ( Fig. 2, bottom left; Table 1; Fig. 4). For 41 out of 53 extracted chromosomes the RV at 30 °C was lower than under control conditions ( Fig. 4, sign test two-tailed, < 0.0001), resulting in a significantly lower mean RV (Table 1, Wilcoxon signed-rank test: < 0.0001). Actually, 20% of the chromosomes again became lethal or nearly lethal at 30 °C, and the left top and bottom figures ( Fig. 2) show qualitatively the same distribution. This is also indicated by the strong increase in variance at 30 °C compared with the same lines at 25 °C (F52,52 = 1.94, P = 0.0093), which approaches the value observed for the original set of chromosomes (Table 1, A vs. B). These findings demonstrate that, although we had purged most deleterious alleles from the original set artificially, under the altered environmental conditions new deleterious effects became manifest that were not expressed under control conditions. Because we used samples of the same batch of eggs for each of the environments, one would expect, everything else being equal, the RVs to be highly correlated. Therefore, the absence of a significant correlation between the RV of specific chromosomes under the two experimental conditions ( Fig. 4, top: Spearman rank correlation rs = 0.16, P > 0.25) strongly supports the conclusion that new detrimentals became expressed at high temperature.

Figure 4.

 The relation between the relative viability of individual lines under control condition (x axes) and under stress condition. Top: high-temperature stress, rs=0.160, n=53,  0.25; bottom: ethanol stress, rs=0.548, n=38, < 0.01. The diagonal represents the values where the relative viabilities are identical under both environmental conditions.

For ETH the effects of stress on RV are much less pronounced ( Figs 2 and 4, Table 1). Nevertheless, 25 out of 38 chromosome homozygotes showed a lower RV when tested at ethanol than at control conditions ( Fig. 4, two-tailed sign test, P = 0.073). Although the whole distribution seems to be shifted somewhat downwards and the mean RV at ETH is decreased, the difference from control conditions is only significant at the 10% level (Table 1, Wilcoxon signed-rank test: < 0.0983). For ETH the correlation between the RV under stress and control conditions turned out to be significant ( Fig. 4, bottom: Spearman Rank Correlation rs = 0.55, < 0.01).

In conclusion, the data show that, at least for HT stress, the viability of chromosome homozygotes is more negatively affected by environmental stress than their accompanying heterozygotes.

Equilibrium experiments

From the 50 bottle populations started in each environment, eight, four and nine became extinct at HT, ETH and ALT, respectively, during the nine generations. Two of these lines became extinct in all three environments, while HT and ALT shared five extinct lines. This indicates a possible correlation between the performance of lines over environments. In the other three environments all lines persisted during the experiments. Population sizes varied considerably both between generations and between lines. On average, the mean number of adult flies per bottle population was between 225 and 275 for the environments CON, HT and DDT. These numbers were considerably lower for ETH and ALT, on average 150 and 180 flies, respectively. For CRW the number of offspring used to start the next generation was around 400 individuals, while the number of adults that could have emerged, as estimated from the number of pupae, was over 550.

The changes in mean frequency of wild-type chromosomes in the bottle populations are shown in Fig. 5. The starting frequency is at about 50% homozygotes, as expected from the cross +/CyO × +/+ giving a wild-type chromosome frequency of 0.75, and this frequency increases gradually for all environments before seemingly approaching an equilibrium frequency in most cases. The rate of increase differed clearly between environments, with HT and DDT increasing more slowly than CON and ETH. The ALT populations show during the first generation an increase more similar to the HT than to the ETH environment. We observed an unexpected increase in the frequency of homozygotes in generation 9 for the environments ETH and ALT. After counting also generation 8 for these environments, it became clear that the sudden increase happened between generations 8 and 9 for ETH, and between generations 7 and 8 for ALT. For the latter environment this was the generation in which the populations were also given ethanol-supplemented food. Although the real cause of the sudden increase is still unknown, we infer that something went wrong with the preparation of the ethanol-supplemented food. Because, for ETH, the equilibrium had seemingly already been reached in generation 8 (see also Table 2), further inferences for this environment are based on the frequencies observed in generation 8 instead of generation 9. Because for ALT an equilibrium apparently had not yet been reached in generation 7, we have omitted this environment from further calculations.

Figure 5.

 Mean frequencies of the chromosome homozygotes averaged over all extant lines during nine generations at different environments.

Table 2.   Summary of the results of the equilibrium experiment. Thumbnail image of

Compared with the control environment, some stresses apparently caused a decrease in the relative fitness of chromosome homozygotes, resulting in a significant lower mean frequency averaged over all lines in those environments (Table 2). HT and, to a lesser extent, DDT had a significantly decreased mean frequency as compared with CON, while CRW and ETH were not significantly different from CON. This is more clearly shown when for each line the frequency of wild-type homozygotes reached at a particular stress environment is compared with the frequency reached under control conditions. On average the difference, calculated as freq.stress minus freq.control, was negative for both HT and DDT and for both the 95% confidence interval did not include zero. In order to be able to calculate the selection coefficient for the chromosome homozygotes, on the basis of the frequency reached in each environment, we have to assume that each line has attained its equilibrium frequency. To test this assumption we calculated the change in homozygote frequency from generations 7 to 9 for each line. Table 2 shows that on average there has been little change during this period, and that for each environment the 95% confidence interval includes zero. This indicates that the lines have indeed more or less reached an equilibrium value. The estimated selection coefficients are shown in the last column of Table 2, and they demonstrate that on average chromosome homozygotes under control, ethanol and crowding conditions show an approximately 25% decrease in fitness compared with heterozygotes. It is clear that, compared with control conditions, HT and DDT cause a significant further reduction in relative fitness of chromosome homozygotes and have some 33% lower fitness. This indicates that under these latter stress conditions the fitness of chromosome homozygotes is on average decreased compared with the heterozygotes, and thus that homozygotes are more negatively affected by stress than are heterozygotes.

In the previous section we have ignored the variation between individual lines. Figure 6 shows the effect of stress on the equilibrium frequency of chromosome homozygotes for individual lines compared with their frequency reached under control conditions. It is clear that the effects differ considerably between environments. For CRW most values fall along the diagonal, indicating once more that crowding stress did not affect the equilibrium frequencies significantly. For HT most values fall below the diagonal (31 out of 42; two-tailed sign test, P = 0.029), but the distribution is more or less parallel to the diagonal. This is also reflected by the fact that the observed variance in frequency of the homozygotes is not different for HT compared with CON (F41,49 = 1.13, P = 0.343). This suggests that HT stress affects most lines in a similar way. For DDT and ETH stress the situation is different. Both environments revealed an increase in variance in chromosome homo- zygote frequency between lines compared with the variance observed under control conditions (see also Table 2). This increase was not significant for ETH (F45,49 = 1.40, P = 0.127), but clearly significant for DDT (F49,49 = 2.07, P = 0.006). However, the way the variance is brought about is different for the two environments. For ETH the data points are more or less symmetrically distributed on both sides of the diagonal, indicating that for some lines the relative fitness of homozygotes increases while in others it decreases. The picture for DDT is highly asymmetrical, and, although the number of points above the diagonal (26) is nearly the same as below (24), the values below the diagonal in particular are responsible for the increase in variance. This suggests that, whereas in some lines DDT stress hardly affects the relative fitness of homozygotes, it decreases the relative fitness very strongly in others.

Figure 6.

 The relation between the (equilibrium) frequencies after nine generations of the chromosome homozygotes for individual lines reached under control conditions (x axes) and the frequency of the same lines under the four stress conditions. Crowding: top left; high-temperature: top right; ethanol: bottom left; and DDT: bottom right. The diagonal represents the values of the frequency being the same under control and stress condition.

In general, we observed a significant correlation between the equilibrium frequencies reached under control conditions and those reached in the four different stress environments (pairwise correlation coefficients range from 0.38 to 0.65, < 0.05 for all). This suggests that some fitness effects are general and environment independent. To a lesser extent, significant correlations were observed over stress environments (range 0.17–0.49) but these were significant at the 5% level only for three (ETH-CRW, ETH-DDT and CRW-DDT) out of the six pairwise comparisons. All these correlations over stress environments, however, became nonsignificant when partial coefficients between these environments were calculated while adjusting for the frequencies found under control conditions. This shows that an important part of the fitness effects observed under stress are in fact due to genotype–environment interactions specific for a particular environment.


The fitness profiles of the extracted chromosome homozygotes in this study are consistent with the results from previous studies ( Wallace, 1956; Tracey & Ayala, 1974). The relative viabilities of the chromosome homozygotes show a bimodal distribution. A large fraction of the chromosomes, around 50%, carry recessive lethal or sublethal mutations and the fraction of chromosomes carrying a lethal was estimated to be about 15% in this population. This value falls well within the range observed for natural D. melanogaster populations ( Oshima & Watanabe, 1973), but much higher values have also been recorded for natural populations ( Greenberg & Crow, 1960; Mackay, 1985). The other peak of the distribution lies somewhat below unity, showing that most of the chromosome homozygotes are less fit than the corresponding heterozygotes. The average relative viability of the chromosome homozygotes, including lethals, found in this study (0.737) is also comparable with older studies on D. melanogaster ( Greenberg & Crow, 1960; Tobari, 1966). The fitness decrease of homozygotes is in fact larger as the balancer heterozygotes on average show a 20–30% lower fitness than natural wild-type heterozygotes ( Tracey & Ayala, 1974; Mackay, 1985), indicating that the homozygotes have about a 40–50% reduction in viability compared with wild-type heterozygotes. This shows that the inbreeding load of D. melanogaster populations can be very large, the more so because the third chromosome and to a lesser extent the X-chromosome also carry substantial levels of genetic load.

Given this high level of genetic load observed for Drosophila populations, and also for many mammal species ( Ralls et al., 1988 ), it is not surprising that they show considerable levels of inbreeding depression upon inbreeding. However, the range in viability observed for the different individual chromosome homozygotes is large, with a substantial fraction of the chromosomes revealing the same or even higher viability than the accompanying heterozygotes. As mentioned above, it has to be realized that ‘wild-type’ heterozygotes, i.e. individuals heterozygous for two independent wild-type chromosomes, in general have a significantly higher viability (1.20–1.30; Tracey & Ayala, 1974; MacKay, 1985) than the balancer hetorozygotes we used to estimate the relative viability of homozygotes. Hence, homozygotes, as measured by our method, on average should show a viability of at least 1.20–1.30 in order to have a viability comparable with wild-type heterozygotes. Therefore, there is ample opportunity for natural selection to select against these lethals and sublethals, thereby purging a population for most of its genetic load, as is supported by the distribution of the relative viabilities of the artificially selected quasi-normals (Table 1). However, this purging appears not necessarily to be effective under all environmental (stress) conditions. When tested at 30 °C many of these quasi-normals show a decreased relative viability. The amount of load expressed by the selected quasi-normals under high-temperature stress is of the same order of magnitude as observed for the unselected base population, and also its distribution looks more or less the same. In fact, the average loss in relative viability of the chromosome homozygotes is almost twice as large for these quasi-normals under high-temperature stress than observed for the unselected base population under control condition. Together with the observation that there is no correlation between the viability of these homozygotes under benign control conditions and high-temperature stress ( Fig. 3), this indicates that new deleterious alleles become expressed under the altered environmental conditions. These (conditional) detrimentals obviously could not be purged from the population because they were either not expressed or even beneficial under control conditions. Some of these chromosomes even seem to carry mutations that are lethal or near lethal at 30 °C and in fact behave as conditional lethals. Temperature-sensitive mutations are well known in many organisms and Suzuki et al. (1967 ) estimated that for D. melanogaster about 10% of all newly EMS-induced lethal mutations were possibly temperature-sensitive mutations that became lethal at 29 °C. Here we found 10%, 5 out of 53, of the extracted lines to produce no wild-type offspring at 30 °C. A decrease, but to a lesser extent, in the viability of chromosome homozygotes at extreme temperatures has also been observed by Tobari (1966) for D. melanogaster and by Dobzhansky et al. (1955 ) and Dobzhansky & Levene (1955) for D. pseudoobscura. In the latter case it appeared that the low-viability chromosomes, in particular, were sensitive to a change in temperature.

Not all novel stress environments do evoke the same amount of concealed genetic load. On ethanol, an environment that caused approximately the same level of stress as HT, some decrease in the average relative viability of homozygotes was observed, but this loss of fitness was not statistically significant. This difference between ethanol and temperature stress was also observed for the equilibrium experiment, showing that there possibly is less variation for fitness genes sensitive to ethanol stress than to thermal stress segregating in this population. This is somewhat surprising as the G83 population shows genetic variation at the Adh locus, located at the second chromosome, and some of the tested quasi-normals carried the more ethanol-sensitive AdhS allele while the balancer chromosome carries the AdhF allele. On the other hand, it has been shown that homozygous AdhS/AdhS eggs are able to phenotypically adapt to ethanol concentrations of 10–12% ( Bijlsma-Meeles, 1979). Moreover, the frequency of the AdhS allele is low in the base population ( Oudman et al., 1991 ), and only six of the 53 chromosomes carried this allele.

That the effectiveness of purging is dependent on the environment is also corroborated by the equilibrium experiments. In these experiments the consequence of homozygosity is not only tested for viability, but also other fitness components, such as fecundity, sterility and male mating ability, may be involved. Mackay (1985) and Miller & Hedrick (1993) have shown that these fitness components also contribute significantly to the level of inbreeding depression. Under control conditions the average fitness of the homozygotes compared with the balancer heterozygotes was found to be 0.763, which is considerably lower than the value observed for viability alone (0.893, Table 1), indicating that other fitness components do indeed contribute significantly to the level of inbreeding depression.

Our fitness estimate for the homozygotes of 0.763 is significantly higher than previous estimates of 0.12–0.32 for second and third chromosome homozygotes in D. melanogaster by Tracey & Ayala (1974) and Mackay (1985), but these authors used all extracted chromosomes having a viability greater than 0.10 in homozygous condition and not only quasi-normals, and moreover they corrected for the lower fitness of balancer heterozygotes compared to wild-type heterozygotes.

Different environmental conditions affect the fitness of the chromosome homozygotes differently. Averaged over all lines, crowding and ethanol stress do not affect the relative fitness of the homozygotes significantly compared with control conditions (Table 2). For ethanol, however, there was a (nonsignificant) increase in the variation between the different extracted chromosomes, suggesting that in some lines the homozygotes were more resistant to ethanol stress and in others less resistant than the heterozygotes. This difference between lines, however, was not correlated with variation at the Adh locus, indicating the observed differences possibly should be attributed to other loci.

Both high-temperature and DDT stress increased the inbreeding depression significantly (Table 2), though in different ways ( Fig. 6), which probably reflects the genetic determination of the resistance/sensitivity to these stresses. For DDT stress it appears ( Fig. 6) that half of the extracted chromosome homozygotes are relatively insensitive to the added DDT, while others seem to be highly sensitive. As a consequence, the variance in fitness between chromosome lines is significantly increased, and some of the extracted chromosome lines seem even to be (near) lethal at DDT when homozygous. Depending on the intensity of selection, resistance to chemical stresses such as DDT often involves a simple genetic basis ( Rousch & McKenzie, 1978; Hoffmann & Parsons, 1991). In D. melanogaster, resistance to DDT is most often found to be governed by one or a few specific major genes on the second chromosome ( Morton, 1993, and references therein). If the G83 population is segregating for tolerance to DDT, the dichotomous reaction of the chromosome lines could be explained.

The results for HT are strikingly different. In this case, most extracted chromosomes show a similar decrease in fitness at HT when homozygous. This is more consistent with thermosensitivity being determined by many genes with small effects than by one or a few major genes, where many of the extracted chromosomes carry on average the same number of sensitive alleles, although often for different loci. This would result in the observed change: a decrease in the mean fitness of the homozygotes at high temperature, but no increase in variance between the lines. This view agrees with recent observations that thermotolerance is a quantitative character and that D. melanogaster populations possess considerable genetic variation for thermal sensitivity ( Cavicchi et al., 1995 ; Krebs et al., 1996 ; Loeschcke et al., 1997 ). At first sight, this seems inconsistent with the results of the viability experiment on high temperature, as in that case we observed the expression of new alleles with large effects on fitness, and these viability lethals are not observed in the equilibrium experiment. However, we used different temperatures for the viability and equilibrium experiment, 30 °C and 28.5 °C, respectively, as it was not possible to breed the extracted lines at 30 °C because of fitness components other than viability, like male sterility. It is well known that D. melanogaster males become sterile when cultured continuously at temperatures of 30 °C and higher ( David et al., 1983 ). Thermal sterilization of males corresponds to a very narrow threshold and flies either carrying the balancer chromosome or being homozygous for the second chromosome seemingly were more thermally sensitive than wild-type flies. Evidently, 28.5 °C is also more permissive for some of the temperature-sensitive detrimental mutations that were observed for viability at 30 °C. Moreover, three out of the eight lines that went extinct during the equilibrium experiment belonged to the class of (near) lethals for viability at 30 °C.

The results of both the viability and equilibrium experiment clearly show that inbreeding depression is environment dependent because of a strong genotype–environment interaction, and also indicate that inbreeding depression will often become more severe under environmental stress conditions. This agrees well with previous observations that the fitness of inbred individuals is further reduced as the environment becomes less favourable ( Parsons, 1959; Coman & Wallace, 1973; Jiménez et al., 1994 ; Keller et al., 1994 ; Miller, 1994; Dahlgaard et al., 1995 ; Loeschcke et al., 1997 ). Our data show, however, that the enhancement of inbreeding depression by stress is not simply decreasing the relative fitness difference of homozygotes compared with heterozygotes. First, although high-temperature stress and ethanol have almost the same effect on absolute viability, and thus exert a comparable severeness, they have a completely different effect on the amount of genetic load expressed in the viability experiment. Second, we have shown that under certain stress conditions new deleterious alleles became expressed that were not apparent under more favourable conditions. Third, when corrected for the difference in fitness of wild-type chromosome homozygotes under favourable conditions, there is no correlation between the relative fitness of homozygotes under the different stresses. This indicates that, except for those mutations that affect fitness under all environmental conditions, inbreeding depression will be due, at least partly, to different loci in different environments. This shows that purging of deleterious alleles can be effective only for the environment in which the purging occurred. Thus, even if purging of all genetic load could be realized during gradual inbreeding, which can be doubted for mildly deleterious mutations ( Barrett & Charlesworth, 1991; Hedrick, 1994; Lande, 1995), additional load will become expressed under changing environmental conditions. This environmental dependence of the expression of inbreeding load may explain, at least partly, the contrasting results observed between different studies concerning the occurrence and importance of purging ( McCall et al., 1994 ; Hauser & Loeschcke, 1996; Lacy & Ballou, 1998). If previously unexpressed load becomes expressed under changing environmental conditions, inbred populations are expected to show a decreased fitness under most novel environmental conditions and, consequently, their extinction probability may be seriously affected (Bijlsma et al., in preparation).

Evidently, these findings have significant consequences for the conservation of small populations of endangered species. An increase in homozygosity due to inbreeding and genetic drift in small populations may, due to purging of environment-specific deleterious alleles, have no immediate fitness consequences, as is suggested by the success of the reintroduction of the beaver in Sweden ( Ellegren et al., 1993 ), but this is no guarantee for survival in a changing and deteriorating environment. The populations may have become homozygous for alleles that act neutral or near-neutral under the given conditions, but that could be deleterious or even lethal under other stress conditions. Possible examples of these situation are homozygosity for conditional lethals and fixation for alleles that increase the susceptibility to a virus infection. This also indicates that deliberate inbreeding of captive populations as a potential management measure to rapidly purge these populations of most of their genetic load ( Templeton & Read, 1994) might in the long run be counter effective. Of course, the impact of environmental stress may be less significant if the environment changes much more gradually. However, many endangered species actually have to cope with human-induced catastrophic and rapid changes in their environment.

Our findings may also in another way have considerable consequences for captive breeding and reintroduction programmes. Although it might seem that relatively optimal captivity conditions have bred a healthy population for reintroduction, there may be a high chance of failure when under the new natural, and often stressful and variable conditions, ‘silent’ detrimentals become expressed. Breeding programmes therefore not only should avoid inbreeding but breeding should also preferably be done under conditions that ‘mimic’ future natural situations.


We thank E. Bijlsma-Meeles and P. W. Hedrick for critically reading the manuscript, and R. Lande, V. Loeschcke, and F. J. Weissing for stimulating discussions on earlier drafts. R.B. was in part supported by a grant from the University of Aarhus.