Changes in genetic architecture during relaxation in Drosophila melanogaster selected on divergent virgin life span


Corneel Vermeulen, Department of Ecology and Genetics, Aarhus University, Ny Munkegade Buildg. 540, DK-8000 Aarhus C, Denmark.
Tel.: +45-8942-3135; fax: +45-8612-7191;


Artificial selection experiments often confer important information on the genetic correlations constraining the evolution of life history. After artificial selection has ceased however, selection pressures in the culture environment can change the correlation matrix again. Here, we reinvestigate direct and correlated responses in a set of lines of Drosophila melanogaster that were selected on virgin life span and for which selection has been relaxed for 10 years. The decrease in progeny production in long-lived lines, a strong indication of antagonistic pleiotropy, had disappeared during relaxation. This was associated with a higher cost of reproduction to long-lived flies in mated, but not in virgin life span. These data strongly suggest that genetic mechanisms of mated and virgin life span determination are partly independent. Furthermore, data on body weight, developmental time and viability indicated deleterious effects of longevity selection in either direction, giving rise to a nonlinear relationship with life span for these characters. In order to reclaim original patterns, we founded a new set of derived lines by resuming selection in mixed replicate lines of the original set. Although selection was successful, most patterns in correlated characters remained, showing that these new patterns are resistant to new episodes of selection.


Research on the genetic determination of life span and aging processes is driven by a broad academic interest in the evolutionary processes shaping aging in metazoan organisms. In addition, there is medical interest in application of this kind of research in the extension of human longevity and the underlying causes of diseases at old age. Research on genetic variation for life span in model organisms will outline genetic possibilities and constraints that determine the evolution of life span and generate candidate genes and processes that facilitate medical research into the aging process in humans (Guarente & Kenyon, 2000; Tatar et al., 2003). According to the disposable soma theory (Kirkwood & Holliday, 1979), a life history trade-off exists between early reproduction and late age survival. Several studies in Drosophila have demonstrated that artificial selection on increased longevity indeed is accompanied by a correlated decrease in early age reproduction (Rose & Charlesworth, 1981; Luckinbill et al., 1984; Zwaan et al., 1995; Partridge et al., 1999). In addition, several other genetic correlations have been reported between life span and life history characters such as body weight, developmental time and egg-to-adult viability (e.g. Chippindale et al., 1994; Zwaan et al., 1995; Stearns et al., 2000). These changes in fitness-related characters often are deleterious in the laboratory culture environment at which the selection lines are maintained and become adapted to (population cages or bottle cultures with a two-week cycle). Therefore, natural selection aims to restore fitness by driving populations back into their ancestral optimal state, when artificial selection is no longer sustained. Whether this will be accomplished depends, among other things, on the evolutionary history of a population (Teotonio & Rose, 2000). Selection pressure is especially high on female early fecundity as this character is a highly relevant fitness trait in laboratory bottle culture (Matos et al., 2000; Sgro & Partridge, 2000; Teotonio & Rose, 2000; Teotonio et al., 2002). Given the negative genetic correlation between life span and early fecundity, life span is expected to decline again once artificial selection for increased longevity is ceased. Similar declines have been observed for stress resistance, which is negatively correlated with early female fecundity, in natural populations of Drosophila that were introduced to the laboratory (Hoffmann et al., 2001). Other important life-history characters, such as developmental rate, viability and body weight seem to be subject to strong evolutionary constraints and it is difficult to predict whether these will be restored to ancestral values during reverse evolution, as there appear to be multiple alternative routes to restore fitness (Teotonio & Rose, 2000; Teotonio et al., 2002).

Our model consists of a set of lines of Drosophila melanogaster selected for divergent virgin life span. Selection on virgin life span may be ecologically less relevant, but the important aspect is that inadvertent selection on the reproductive schedule is avoided, allowing more robust conclusions concerning genetic correlations (Zwaan et al., 1995). In addition, the required design of family selection allows the establishment of both lines selected for long- and short-life span. Instead of having accumulated rare deleterious mutations as is often assumed (e.g. Lints et al., 1979), short-lived lines have been shown to include relevant variation for lifespan that normally is not addressed. Zwaan et al. (1995) established two replicates of each of a short living, a long living and a control line and demonstrated several intriguing correlated responses. Among other findings, flies selected for long life were shown to have decreased (early) progeny production, which was concordant with other findings (Rose & Charlesworth, 1981; Luckinbill et al., 1984; Partridge et al., 1999). It is noteworthy that mated life span displayed few significant differences between the contrasting selection lines, whereas lines selected on mated life span (age at reproduction) were shown to display a positive correlation between virgin and mated life span (Service, 1989). Thus, the correlated response in early female fecundity was due to its negative genetic correlation with virgin life span, rather than with mated life span, in these lines. Long-lived lines did not appear to have correlated changes in any other characters, but in the short-lived lines increased developmental time and decreased female body weight were demonstrated (Zwaan et al., 1995).

In this study, we assess the robustness of the initial response and the effect of natural selection during laboratory culture on correlated responses in several life history characters, after initial selection had ceased and had been relaxed for 10 years. Note that selection pressure did not reverse, as (virgin) life span is not a character under selection during laboratory culture whereas, conversely, selection on early reproduction was not a feature of the artificial selection procedure, because inadvertent selection on the reproductive schedule was deliberately avoided. We show that, many years after selection was relaxed, significant differentiation in life span still was present. Given the strong selection for increased early reproduction, egg-to-adult viability and developmental rate, compensatory evolution is to be expected to modify initial character changes. Therefore we also assessed several relevant life-history characters and demonstrated that these indeed experienced changes during relaxation. Since we had also resumed selection in the original lines (primary set), we could assess if new rounds of selection would restore the original pattern as found by Zwaan et al. (1995). The new selection lines (derived set) were founded by mixing the original replicate lines to possibly restore some level of variation if replicates had initially diverged and adding five generations of life span selection to the six already present. This resulted in an increased response, which restored the direct response to the levels achieved by Zwaan et al. (1995) just after the initial selection. With these new lines we could show that genetic correlations with other characters had also changed during relaxation.

Materials and methods


Two replicates of lines selected for short- and long-virgin life span and two control lines (respectively S1 and S2, L1 and L2 and C1 and C2) were used. Details on the selection procedure and initial responses of those lines can be found in Zwaan et al. (1995). Summarizing, the selection procedure consisted of a design of family selection. For six generations, those families of flies, stored at 15 °C, were selected for the next round of selection, whose full sibs exhibited the most extreme (short or long) longevity phenotype at 29 °C. After establishment, the lines were cultured in uncrowded quarter-pint bottles (30 mL standard medium: 26 g dead yeast, 54 g sugar, 17 g agar and 13 mL nipagine solution per litre).

Procedure for resumed selection

During approximately 220 generations of relaxation, the original lines were kept in three to five culture bottles with regular mixing, resulting in a large population size of at least 500 individuals per generation. Inbreeding is not expected to become a severe problem with this census size. The replicates were used to found mixed stocks of each of every selection regime (short, control and long-life span). This was done for every mixed stock by setting up five bottles for each reciprocal cross with 20 female flies from one replicate crossed to 20 males of the other replicate (e.g. S1 × S2 and S2 × S1). Subsequently, five bottles were started with 20 fertilized F1 females from each cross. These stocks were allowed five rounds of recombination before selection. From each of three mixed stocks two new replicate lines were established named A and B. Thus, in addition to the original lines we obtained six new lines: SA, SB, CA, CB, LA and LB. The selection procedure was performed as described in Zwaan et al. (1995) with slight modifications. To prevent inbreeding in the set-up, we allowed no brother–sister mating. Instead, we constructed all possible family combinations, by making all reciprocal crosses among the offspring of six pairs (excluding brother–sister mating), to ensure the maximum amount of possible genotypic variation. Life span measurement was performed using vials in the first generation of selection. From the second generation onwards, life span measurements were obtained using trays supplied with food, in which flies were able to feed from their vials through a gauze in the bottom (modified from Fukui et al., 1992), which simplifies the logistics of this elaborate experiment. Although longevity in the trays is shorter than in vials (Fig. 1), the correlation in life span between these procedures was inspected and proved to be significantly high (females r = 0.802, P < 0.01; males r = 0.919, P < 0.001). Therefore, no problems due to gene-by-environment interactions for the different assay methods were expected. Selection was carried out during five additional generations of selection. To distinguish the original set of selection lines that were established by Zwaan et al. (1995) from the newly established set, these are denoted as the primary set and the derived set respectively. Note that we used a design of resumed selection, where the primary set of selection lines was used to found new base stocks for each of the selection regimes. The advantage of this method is that new variation was generated by recombination of favourable alleles that were lost in one of the replicates, but as a consequence the derived set does not consist of true replicates and therefore joint analysis with the primary set is required.

Figure 1.

Selection progress and inspection data for the derived set. The mean lifespan at 29 °C for females (a) and males (b) in days are given for all selection lines. Also, the response to selection for the S- and L-lines as absolute deviations from the mean C-line values (solid black lines) are given for females (c) and males (d).

Longevity before and after completion of selection

To assess repeatability of life span measurements, longevity of the original set of lines was determined at two time points, with approximately 3 months time between both experiments. Flies were grown at 25 °C in quarter-pint bottles [standard medium + ampicillin (100 mg L−1)] in noncrowding conditions and collected as virgins. Fifty to sixty flies for each line and sex were distributed to vials containing 9 mL of food (five flies per vial). During experiment 1 only 20 C1 females could be obtained. Vials were transferred to 29 °C (±1 °C). Note that longevity at 29 °C gives a measure of the direct response. Dead flies were scored three times per week and vials were refreshed twice per week.

Also, we investigated the final life span characters during the inspection that was performed 3 months after selection had ceased. Both the primary and the derived set were assessed to ascertain the response had increased. The longevity assay at 29 °C was performed simultaneously with an assay at 25 °C to confirm that 29 °C longevity, which is the character under selection, closely correlates with life span under standard breeding conditions. Longevity was scored three times per week and vials were refreshed once per week.

Correlated characters

Progeny production and female mated life span

For each line, 25 females were kept individually in vials with two scarlet (st) males at 25 °C. Females were 1–2 days and st males were 1–3 days of age at the start of the experiment. Males were replaced every 14 days and in case of occasional death. Vials were refreshed three times per week. At these times mortality of the females was scored for assay of mated life span, until all females had died. Used vials were kept at 25 °C and all eclosing progeny was counted. Two egg counts were performed in week 2 of the experiment to estimate egg-to-adult viability, which proved to be high (90%). Therefore, differences in progeny production can be taken as an accurate reflection of differences in egg production. anova was performed on mated life span and lifetime progeny production (LPP). To allow anova on LPP, sterile females and outliers (points over three standard deviations removed from the mean) were omitted (three data points or 1%). Weekly progeny production was statistically analysed using a Kruskal–Wallis one-way analysis of variance by ranks.

Developmental time and egg-to-adult viability

Parental flies were allowed to lay eggs on standard food supplied with live yeast within a 4-hour window, preceded by 30 min to remove eggs that might be withheld in the body of the females. For each line, ten vials of 100 eggs were set up at 25 °C (±0.5 °C). When pupation was complete, all pupae were counted. This allowed egg-to-adult viability to be partitioned into egg-to-pupa and pupa-to-adult viability. Finally, all eclosing flies were collected, counted and sexed two times per day until all flies had emerged. Viability was angular transformed and analysed by anova. Developmental time was calculated separately for each sex. Eclosion percentages (after probit transformation) were regressed on time (logarithmically transformed) and median developmental time was calculated for each vial as performed by Zwaan et al. (1991). These data also were analysed by anova.

Body weight at eclosion

Five vials containing standard medium with 100 eggs were set up per line and transferred to 25 °C. Larvae were allowed to pupate on transparent film that was inserted into the vials. After pupation was complete, pupae were transferred to vials containing only agar, so no feeding could occur after eclosion. Finally, flies were collected as virgins at eclosion (0–12 h) and killed with ether vapour. Body weight was assessed by weighing five groups of five flies per line for each sex.

Data analysis

Longevity of the flies was taken as the midpoint of the interval between two successive scorings during which the fly died. Few flies escaped or died because of accidents and were censored for life table analysis [e.g. calculation of median survival time (MST)] and dropped from analysis of mean life span (i.e. anova). Data for all characters were analysed by anova using general linear model (GLM) procedure. For life span analysis, first-week mortality was not included in anova for normalization purposes and for conceptual reasons, because this mortality is unlikely to be aging related. Selection regime was set as fixed factor and replicate was nested within selection regime. Sex was included as a fixed factor if appropriate. All interactions were included in the model. A priori contrasts with respect to the control line were tested for both selection lines. Data from the primary and the derived set were analysed separately. If the sex-by-selection regime interaction turned out to be significant, separate analysis of male and female data was performed. Finally, correlations between characters were performed on mean values including lines of both the primary and the derived set. Since the derived set does not consist of independent replicates, one should be cautious with the results from such correlations, but these may reveal linear relationships indicating causal or closely linked mechanisms.


Robustness and repeatability of the direct response

Prior to the founding of the derived set, we obtained data on the magnitude and robustness of the selection response in the primary set after such a considerable period of relaxation. In order to assess the repeatability of the longevity assay, we performed correlation analysis of MST between two longevity assays, spaced 3 months apart. We observed a highly significant positive correlation (r = 0.817, P = 0.001), indicating that the differentiation in life span between lines had a genetic basis (mean line values shown in Table 1). Both mean downward and upward response over both experiments still were significant (t1200 = 6.8, P < 0.001 and t1200 = 8.1, P < 0.001, respectively). The long-term persisting differences between S- and L-lines summed up to approximately 7 days for females and 5 days for males. After founding the derived set, almost three years later, the difference between S- and L-lines in the primary set was as high as 9.4 days for females and 9.3 days for males (inspection after selection, mean line values shown in Table 2). These experiments show that the original set of lines still display a large amount of genetic differentiation.

Table 1.  Mean virgin life span in days [±standard error (SE)] for selection lines of the primary set in both longevity assays of the repeatability experiment prior to selection (see text).
ExperimentT (°C)SexS1S2C1C2L1L2
  1. All means are corrected for first-week mortality.

Experiment 129Female25.3 (0.75)32.2 (0.59)32.6 (1.98)31.3 (0.88)36.7 (1.10)33.3 (0.77)
Male21.7 (1.05)24.8 (1.03)25.4 (0.98)22.2 (0.87)30.3 (1.27)34.2 (0.82)
Experiment 229Female30.4 (0.68)29.4 (0.65)36.4 (0.95)35.0 (0.68)39.2 (0.58)36.7 (0.53)
Male26.9 (0.71)26.9 (0.62)32.7 (1.21)25.3 (0.92)31.3 (1.00)31.8 (0.77)
Table 2.  Mean life span in days [±standard error (SE)] for all selection lines of the primary and the derived set in longevity assays of virgin and mated life span during inspection experiments after selection.
ExperimentT (°C)SexPrimary set
Virgin life span25Female35.3 (2.07)35.7 (0.90)39.7 (1.42)43.2 (1.36)51.6 (1.46)54.4 (1.40)
Male27.8 (1.60)37.2 (1.23)48.3 (1.60)40.2 (1.65)44.9 (1.79)51.6 (1.24)
29Female23.7 (1.15)27.3 (0.67)30.4 (0.67)29.0 (0.82)36.7 (0.74)33.2 (0.62)
Male20.1 (0.85)28.2 (0.56)29.1 (0.95)25.3 (0.72)33.2 (1.09)34.0 (0.75)
Mated life span25Female39.1 (2.19)40.6 (1.84)36.8 (2.67)39.2 (1.78)29.1 (2.28)39.0 (2.53)
   Derived set
  1. All means are corrected for first-week mortality.

Virgin life span25Female33.2 (0.60)33.3 (0.92)43.4 (1.22)39.2 (1.65)60.0 (1.62)54.4 (1.50)
Male36.2 (1.19)30.4 (1.11)49.1 (1.51)46.6 (1.27)57.6 (1.77)58.4 (1.26)
29Female22.1 (0.49)23.4 (0.59)32.3 (0.56)29.4 (0.70)39.3 (0.86)34.9 (0.77)
Male25.3 (0.62)23.7 (0.57)28.9 (0.80)28.3 (0.84)40.8 (0.84)38.2 (0.92)
Mated life span25Female32.6 (1.96)32.9 (2.02)39.8 (2.29)42.6 (1.82)32.5 (1.90)36.4 (2.12)

Response to selection

Large life span differences were already present in the base stocks, but during selection additional progress was being made. Due to the fact that we used vials in generation 0 and trays later on during the selection procedure (see materials and methods) and because fluctuations in time are considerable, the response to renewed selection is not immediately evident from Fig. 1. However, the data presented in Table 2 clearly show that resumed selection was successful. This was mainly shown by the increased direct response in longevity at 29 °C in the derived set as compared to the primary set, since the mean difference between S- and L-lines had increased from 9.4 days in the primary set to 14.8 days in the derived set (sexes pooled). Differences between each selection regime and the control treatment were highly significant in the anova for the derived set (t625 = 12.17, P < 0.001 and t625 = 16.47, P < 0.001 for downward and upward response respectively). For virgin longevity at 25 °C similar increases were found (see Fig. 2; Table 2), with the mean difference between S- and L-lines changing from 16.4 days in the primary set to 24.4 days in the derived set (sexes pooled). These values closely correspond to the genetic differentiation obtained by Zwaan et al. (1995), thus it can be concluded that selection restored the original magnitude of the response, although not necessarily by similar genetic changes. The increased difference and the fact that correlation between MST at 25 and 29 °C is excellent (r = 0.914, P < 0.001), suggests that the differences in life span at both temperatures are caused by the same underlying mechanism, presumably the aging process and not other potentially confounding factors, such as resistance to high temperatures or conditionally expressed inbreeding effects (Vermeulen & Bijlsma, 2004a,b). It can thus be concluded that the resumed selection has been highly successful.

Figure 2.

Cumulative survival curves at 25 °C in the inspection experiment after completion of selection for both sexes of the primary set and the derived set.

Correlated characters

Mated longevity

Although the selection factor was significant in the anova on life span data of the primary set (F2,136 = 3.43, P < 0.05), neither the S- nor the L-lines were significantly different from the control lines (mean line values shown in Table 2). In the derived set however, both selection regimes had significantly decreased mated life span as compared to the control lines (S-lines: t139 = 4.19, P < 0.001 and L-lines: t139 = 3.33, P < 0.01). The absence of any repeatable patterns in mated life span causes the correlation between MST of female virgin life span and mated life span at 25 °C to be nonsignificant (r = −0.371, n.s.). To address this effect, we performed an anova on female life span combining the data of both virgin and mated life span at 25 ° in one analysis. By including mating status as a fixed factor, we could test if the sensitivity to mating was significantly different between selection regimes. Such differences would be indicated by a significant selection regime-by-mating status interaction. The resulting anova indeed shows highly significant interactions (Table 3, F2,443 = 40.11, P < 0.001 and F2,446 = 60.79, P < 0.001 for primary and derived set, respectively). These interactions are visualized in Fig. 3. Apparently, there is a larger deleterious effect of mating on life span in the L-lines than present in the S- and C-lines. This is consistent with Zwaan et al. (1995), who did not detect significant differences in mated life span between females of the respective selection regimes, although they had diverged in virgin life span. However, in their survey, females of the L-lines had, although not significantly, longer mated life span still than females of both C- and S-lines and the order of life span differences between lines in their experiment is the same for virgin and mated life span. This is clearly opposite to our results.

Table 3. anovas using general linear model (GLM) procedure with female life span as the dependent variable and selection regime, replicate within selection regime and mating status as fixed factors for both sets.
  1. The selection regime-by-mating status interaction indicates differences in sensitivity to mating between selection regimes. SS, sums of squares; MS, mean squares.

  2. **P < 0.01,***P < 0.001.

Primary set
 Selection regime2306.21021153.10510.141***
 Replicate (selection regime)1593.2953531.0984.671**
 Mating status3537.16013537.16031.107***
 Selection regime-by-mating status9122.31124561.15540.113***
 Replicate (selection regime)- by-mating status428.1743142.7251.255
R2 = 0.313
Derived set
 Selection regime11219.29825609.64961.974***
 Replicate (selection regime)41.193313.7310.152
 Mating status6009.12016009.12066.388***
 Selection regime-by-mating status11005.25425502.62760.792***
 Replicate (selection regime)- by-mating status1113.0503371.0174.099**
R2 = 0.500
Figure 3.

Interaction between selection regime and mating status in the primary set and the derived set. Lines depict the mean of the replicate lines, whereas symbols show the individual values of each replicate. Values are relative to the mean control value, which is set to 1. Lifespan was decreased in L-lines upon mating, relative to S- or C-lines.

Progeny production

anova of LPP showed no significant differences between selection regimes for both the primary set (F2,133 = 0.65, n.s.) and the derived set (F2,138 = 2.66, P = 0.073). Age-specific progeny production per week is shown in Fig. 4. Daily progeny production was highly skewed for some weeks and contained increasing proportions of sterile flies, so a nonparametric test was applied to test differences between selection regimes in every week. For the primary set, only significant patterns in age-specific progeny production of weeks 1 and 2 were obtained (H = 13.21, P < 0.01 and H = 6.21, P < 0.05, respectively). Multiple comparisons at the 0.05 significance level showed that in week 1, S-lines had decreased progeny production as compared to the control lines, whereas L-lines were not statistically different from either selection regime. In week 2, the S-lines had in fact a higher progeny production than the L-lines, but neither of these lines was statistically different from the control lines. This leads us to conclude that in the primary set the large and significant decrease in early progeny production, which was prevalent just after selection (Zwaan et al., 1995), had disappeared over time. In the derived set, one might expect that this pattern could be reclaimed. This was not the case. Only in week 1 did the age-specific progeny production differ among selection regimes and this was caused by a superior progeny production of the L-lines over the other lines (H = 25.41, P < 0.001, see Fig. 4, derived set), which is in fact the opposite of what Zwaan et al. (1995) observed. However, early progeny production did not show a significant correlation to either virgin (r = 0.556, n.s.) or mated (r = −0.105, n.s.) life span at 25 °C. Also, there was no significant correlation between body weight and LPP (r = 0.198, n.s.), but the correlation between early (first week) reproduction and body weight was marginally significant (r = 0.589, P = 0.044).

Figure 4.

Age-specific progeny production per week (in eggs per day per female) for each selection regime of the primary set and the derived set. Lines are the mean of the median values from two replicates. Median values were used because progeny production data become skewed to high values in the course of the experiment.

Egg-to-adult viability

Mean egg-to-adult viability in the total experiment was 74%. In the anova significant effects of selection regime were apparent in both sets (mean values shown in Fig. 5). Both in the primary and the derived set, egg-to-adult viability was significantly decreased in the S- and L-lines as opposed to the control lines (t54 = 2.78, P < 0.01 and t54 = 2.80, P < 0.01, respectively, in the primary set and t54 = 6.13, P < 0.001 and t54 = 3.60, P < 0.001 in the derived set). Partitioning into egg-to-pupa and pupa-to-adult viability showed that this effect was mainly due to differences in the egg-to-pupa viability, since there were significant differences in this character at P = 0.001 in all comparisons (for short- and long-lived flies t54 = 3.85 and t54 = 4.31, respectively, in the primary set and t54 = 4.67 and t54 = 3.99 in the derived set), whereas for pupa-to-adult viability only the S-lines of the derived set experienced decreased viability (t54 = 2.62, P < 0.05). This indicates that selection on life span, in either direction, has affected either larval viability or egg hatching success. It is noteworthy that this pattern was not yet apparent directly after establishment of the primary set (Zwaan et al., 1995).

Figure 5.

Developmental viability data for the primary set and the derived set. Egg-to-adult viability has been partitioned in egg-to-pupa and pupa-to-adult viability. Asterisks denote a significant difference with the control lines (*P < 0.05,**P < 0.01, ***P < 0.001). Error bars indicate standard errors, based on residual error of anova.

Body weight at eclosion

Both males and females of the S-lines experience a significant decrease in body weight in both the primary and the derived set (see Fig. 6). This decrease was of equal magnitude in both sexes, as indicated by the nonsignificance of the selection-by-sex interaction and was −0.094 mg for the primary set (t48 = 7.73, P < 0.001) and −0.066 mg for the derived set (t48 = 5.63, P < 0.001). In the L-lines also a weight loss was observed in the primary set (t48 = 3.51, P < 0.001), but this had disappeared in the derived set (t48 = 0.06, n.s.). Weight loss was also observed by Zwaan et al. (1995) in the S-lines after selection in the primary set was completed, but only in females. Due to this correlated response in the S-line, there is a positive correlation both for males and for females between 25 °C life span and body weight at eclosion, but neither of these is significant (for females r = 0.454 n.s. and for males r = 0.326 n.s.). This is because the weight loss in the L-lines did not conform to the positive correlation.

Figure 6.

Wet body weight (mg) at eclosion for both sexes of the primary set and the derived set. Asterisks denote a significant difference with the control lines (***P < 0.001). Error bars indicate standard errors, based on residual error of anova.

Developmental time

In the primary set the L-lines had increased developmental time (time to 50% eclosion) with respect to the C-lines (+7.7 h, t108 = 5.04, P < 0.001) and this pattern persisted in the derived set (see Fig. 7). However, since the sex-by-selection interaction was significant in the derived set, male and female data required separate analysis. This showed that increased developmental time in the L-line was significant only for females (+3.4 h, t108 = 3.49, P < 0.001). In the derived set, developmental time was also increased in the S-lines, both for males and females. Such an increase for S-lines was also found after establishment of the primary set, but was absent in the L-lines at that time (Zwaan et al., 1995). Developmental time is deemed an important life history character in insects, mainly because of its relation to body size (Stearns et al., 2000). However, no positive correlation was observed between developmental time and body weight both for females (r = −0.039 n.s.) and males (r = 0.138 n.s.). This might be caused by the fact that increases in developmental time have occurred during the pupal stage, rather than in the feeding larval stage. Since developmental time to pupation was not measured, we cannot test this hypothesis.

Figure 7.

Developmental time (h) as mean 50% eclosion time for both sexes of the primary set and the derived set. Asterisks denote a significant difference with the control lines (***P < 0.001). Error bars indicate standard errors, based on residual error of anova.


The robustness of the selection response after ten years of relaxation, established during few rounds of initial selection (six) suggests rapid fixation of a few major genes. Furthermore, the fact that the response to resumed selection in the lines of the derived set, derived from mixing the replicates of the primary set, did not surpass the initial response achieved by Zwaan et al. (1995) indicates that either identical or incompatible mechanisms of life span determination are present in the replicates of the selection lines of the primary set. Alternatively, the available genetic variation for life span was already exhausted during the first selection procedure or had disappeared during relaxation as a result of genetic drift. The first explanation is unlikely, since Zwaan et al. (1995) mention no slackening of the response at the time that selection was ceased. Furthermore, since flies have been kept in large bottle populations the total loss of available genetic variation also is not plausible, because random genetic drift would not have depleted genetic variation at the same loci in both replicates.

Compensatory evolution during relaxation

The negative correlation between early reproduction and longevity is usually thought to be due to antagonistic pleiotropic loci, which are central to theories of the evolution of ageing (Williams, 1957; Kirkwood & Holliday, 1979). However, although the direct response was still present, the initial difference in progeny production in the long-lived lines, which was a key feature of the initial correlated responses, had disappeared. During normal laboratory breeding culture in bottle populations with two-week cycles, there is a large selection pressure on early reproduction (Matos et al., 2000; Sgro & Partridge, 2000). It is therefore to be expected that during relaxation, selection for restoration of high early reproduction in the L-lines will occur (Teotonio & Rose, 2000; Teotonio et al., 2002). In several studies it has been demonstrated that, in the course of selection, the decrease in early fecundity diminishes and long-lived lines may eventually even surpass the base stock for this character (Leroi et al., 1994; Arking et al., 2002). Leroi et al. (1994) have argued that this was due to genotype-by-environment interactions caused by differences in breeding conditions in those experiments. Since comparable differences in breeding conditions did not exist in our experimental set-up, we conclude that the compensation of the decreased early progeny production shortly after selection is no interaction effect. Importantly, the genetic constitution responsible for this compensatory effect in the L-lines has proven to be resistant against further selection, demonstrating that early reproduction had become uncoupled of virgin life span. An important finding in this respect is the lack of correlation between mated and virgin life span in our lines. Zwaan et al. (1995) also found that there were no significant differences in female-mated life span between selection lines. In contrast, in lines selected on mated life span, virgin and mated life spans have been shown to be highly correlated (Service, 1989). The lack of correlation appears to be caused by a higher cost to adult survival in long-lived flies than in control and short-lived flies when mated. We can but speculate why long-lived females are more sensitive to mating, but this either involves a higher cost of egg production, or an increased sensitivity to the presence of the males themselves. The latter may be due to increased sensitivity to seminal fluids (Chapman et al., 1995) or increased activity levels because of male harassing. Preliminary data indeed suggest differentiation in locomotor activity between our lines (unpublished results). It is presumably the lack of correlation in our lines that allowed female fecundity levels to be restored in the L-lines, without affecting the direct response. This is remarkable, since the decrease in progeny production when mated was a correlated response to selection on virgin life span (Zwaan et al., 1995). This suggests that the genetic basis of mated and virgin life span are partly independent, and that mated life span is genetically linked to reproduction whereas virgin life span can become genetically uncoupled, i.e. determined by a (partly) different set of genes, although the initial response shows a genetic relationship. In corroboration of our results, Leips & Mackay (2002) have shown that mating status has a profound effect on the genetic architecture of life span differences between two lines of Drosophila, which also suggests different sets of genes are involved in virgin vs. mated life span.

How could this possibly be explained? Let us assume that the response to selection for increased age at reproduction is due to two separate responses. On the one hand there is up-regulation of maintenance genes and on the other hand a decrease of egg production and mating behaviour, decreasing mortality in long-lived lines (Sgro & Partridge, 1999; Sgro et al., 2000). However, in lines selected on virgin lifespan, only the former mechanism operates, as egg-production normally is greatly reduced in virgin flies and mating is absent. In both types of selection methods a genetic change in the balance between reproduction and maintenance will initially lead to a change in life span at the cost of fecundity. Flies selected for increased age at reproduction have a positively correlated response in the alternative (virgin) mating status (Service, 1989). For females selected for divergent virgin lifespan, mated lifespan did not show significant differences (Zwaan et al., 1995). This may be partly related to the fact that the food conditions in our laboratory without live yeast lead to lower rates of egg production than in many other laboratories (Ackermann et al., 2001). Since virgin lifespan is not directly affected by deleterious effects of egg production, compensatory evolution on early reproduction can occur in this type of selection line, without affecting the initial direct response. In lines selected on age at reproduction this is not possible, since the deleterious effects of egg production are directly affecting the response in mated life span. This was demonstrated in the lines of Teotonio & Rose (2000), which were previously selected on age at reproduction. When the long-lived lines were returned to the original culture environment, early fecundity reversed to the ancestral state in females and this was accompanied by a concurrent decrease in longevity (Passananti et al., 2004). Note that, as opposed to their experiment, selection was not reversed in our design, as there is no selection on virgin life span in bottle culture and since selection on the reproductive schedule was avoided in our selection procedure.

Deleterious effects of selection on life span

It seems that, regardless of the direction of selection, there is a deleterious effect of selection on virgin life span that is manifest from the changes in egg-to-adult viability, developmental time and body weight. Decreased developmental viability of selection lines may be caused by close linkage of life span genes with viability genes, but is more likely to be a direct negative consequence of the changes in physiology provoked by the selection. Arking et al. (2002) also mentioned deleterious egg-to-adult viability effects in long-lived lines. As opposed to these results, Chippindale et al. (1994) reported increased egg-to-adult viability in their long-lived O-lines with respect to the control B-lines. However, they also mentioned concurrent increases in developmental time for these lines, which resemble the increases found in our study for the females of the long-lived L-lines. The retarded developmental time of the S-lines in the derived set may have a similar explanation, since none of the increases in developmental time in our study led to increases in body weight. Neither egg-to-adult viability, developmental time nor body weight correlates significantly with longevity. It is therefore unlikely that the correlated changes in these characters are adaptive or causal determinants of virgin life span.

Although it is tempting to associate the decrease in several fitness-related characters in the S-lines with a decrease in overall fitness, either caused by accumulation of deleterious mutant alleles or inbreeding effects, this is not a plausible explanation for the findings in these lines. The inbreeding risk during selection is minimized by a deliberate out-breeding design and controlled for by subjecting all lines to the same procedure during selection, which makes the C-lines appropriate controls for these effects. Path analysis of pedigrees during selection showed S-lines did not have higher inbreeding coefficients than control or L-lines (data not shown). Also, female fecundity, male fertility, developmental rate and egg-to-adult viability need to be sufficiently high during the selection procedure to sustain enough individuals for the next generation, leading to purging of deleterious alleles. In support of this, S-line females indeed show normal levels of progeny production, indicating they are not crippled by the selection procedure and display normal levels of fitness. Finally, Zwaan et al. (1995) did not find large differences in correlated responses by crossing the replicates of the primary set, showing that inbreeding depression did not significantly interfere with the observed patterns. We therefore argue that long-term changes in life span in either direction most likely are accompanied by complex changes in metabolism and genetic constitution that negatively affect body weight, developmental time and egg-to-adult viability.

Given the low level of replication of our lines, there is a risk that some patterns were affected by genetic drift. Nevertheless, our results suggest caution in the extrapolation of correlated responses and inferences on antagonistic pleiotropic patterns concerning these characters, since many may have nonlinear relationships with life span, resulting in genetic correlations shaped like convex curves with control lines at the optimum. Our results are supported by the findings of Phelan et al. (2003) and Archer et al. (2003), who demonstrated that the positive genetic correlation between longevity and stress resistance tends to break down at extreme values of stress resistance, resulting in a similar complex nonlinear relationship.

Concluding remarks

The decreased progeny production in the L-lines that originally accompanied the increase in virgin longevity in the primary set has disappeared during relaxation. Presumably, compensatory evolution on early reproduction during laboratory culture did not affect the direct response, because the impact of reproduction only affects mated life span. Therefore, the genetic correlation between early female fecundity and virgin life span has disappeared in these lines. Perhaps fixation of alleles during artificial selection forced long-lived flies to employ alternative routes to restore early fecundity that were unrelated to virgin life span. Such processes may also affect other life history traits. Both in the lines selected for long life and for short life, we observed deleterious effects in several key life-history characters. These may reflect changes in the underlying physiology in response to artificial selection and ensuing natural selection in the culture environment, which disrupted the genetic background of the laboratory adapted base population. The new correlation matrix has proven to be resilient to new episodes of selection, suggesting it represents a new fitness optimum for these lines. Therefore, irrespective of the direction of selection, negative pleiotropic fitness effects may often be observed upon selection on (virgin) life span.


We are very grateful to Anneke Boerema, Désiree Joubert, Marielle van Rijswijk and Marco van der Velde for taking care of the longevity assays and selection procedure during several occasions in the course of this research. Also, we thank Wilke van Delden and two anonymous reviewers for constructive comments on the manuscript.