Direct and correlated responses to selection on age at physiological maturity in Drosophila simulans

Authors


DanielPromislow Department of Genetics, University of Georgia, Athens, GA 30602-7223, USA. Tel.: +1 706 542 1715; fax: +1 706 542 3910; e-mail: promislow@arches.uga.edu

Abstract

Biologists who study the timing of development in insects have focused on variation in duration of pre-adult stages almost without exception. However, development is not complete until adults are not only morphologically mature, but also reproductively mature. Here we describe an experiment in the fruit fly, Drosophila simulans, in which we used artificial selection to create lines with shortened and lengthened duration from eclosion to the age when the first egg was laid. We found significant genetic variation for this trait. The response to selection on age when the first egg was laid was due to variation among females. Delayed adult development was correlated with rapid pre-adult development and longer life span in females. The approach we use here resolves some difficulties with previous approaches used to study the genetics of senescence, and provides an opportunity to study the hitherto unexamined predictions derived from classic models for the evolution of senescence.

Introduction

In most of the previous studies of development in the fruit fly, Drosophila spp., it has been assumed that development is complete at eclosion (but see Pitnick et al., 1995 ). In fact, development continues until the eclosed adult is ‘reproductively’ mature and is able to produce viable sperm or eggs. In fruit flies, full reproductive maturity is not reached for 1–4 days after eclosion in D. melanogaster and D. simulans, and may take as long as 2–3 weeks in the males of giant sperm species, such as D. pachea and D. bifurca ( Pitnick et al., 1995 ). Pitnick et al. (1995) discovered enormous variations in post-eclosion development time within the genus, especially in males. Whereas we know a great deal about genetic variation for larval development time ( Sang & Clayton, 1957; Marien, 1958; Clarke et al., 1961 ), and how larval development is correlated with other traits, for example, size at maturity (reviewed in Roff, 2000), fertility ( Hiraizumi, 1961), egg hatchability ( Dawson, 1966), body weight ( Nunney, 1996), larval survival ( Chippindale et al., 1997 ) and ageing (e.g. Lints & Lints, 1969; Partridge & Fowler, 1992; Roper et al., 1993 ; Chippindale et al., 1994 ; Zwaan et al., 1995a ), we know very little about genetic variation for adult development time in insects species, or about how this trait may correlate genetically with other life history traits.

Such variation may be important for several reasons. Firstly, the developmental cycle is not complete until adults reach sexual maturity. If we are to understand the overall genetic structure of development, we need to examine all stages of this process. Secondly, as Pitnick et al. (1995) have pointed out, rates of maturation may be inextricably linked to patterns of sperm competition. Thirdly, previous work has suggested that patterns of ageing may be genetically correlated with rates of early development, which is the primary focus of this study. In the late 1960s, Lints & Lints (1969) suggested that development time may determine the rate of adult survival. However, subsequent work by Chippindale et al. (1994 , 1997) and Zwaan et al. (1995a) has refuted this developmental theory of ageing.

Years before the work of Lints and Lints (1969), Williams (1957), in his antagonistic pleiotropy theory for the evolution of senescence, predicted a specific relationship between development and ageing. Williams’ antagonistic pleiotropy theory posits that genes that decrease fitness at late ages (e.g. through a reduction in late-life survival or fecundity), can be favoured by selection if these same genes lead to beneficial effects on early age fitness traits. Williams (1957) derived nine predictions from his theory, at least three of which can serve as the basis of selection experiments to delay ageing. Specifically, he argued that (1) the age at which mortality rates begin to increase as a result of senescence should coincide with the age at maturity; (2) individuals that mature early should show more rapid senescence (and vice versa); and (3) selection for increased longevity should lead to the correlated response of decreased ‘vigour’ at young ages.

In light of Williams’ third prediction, Edney & Gill (1968) suggested that one could alter life span by changing the average age at which individuals are allowed to breed. This was successfully demonstrated by a series of experiments in which researchers allowed females to mate throughout life, but prevented them from contributing offspring to the next generation until they were relatively old ( Rose & Charlesworth, 1981; Luckinbill et al., 1984 ; Rose, 1984a; Partridge & Fowler, 1992; Zwaan et al., 1995b ; Partridge et al., 1999 ). These studies showed that a demographic delay in reproduction was, indeed, genetically correlated with longer life span. However, the reproductive delay affected by this selection regime was not necessarily correlated with any physiological delay in maturity. Rather, females that survived until the later breeding age appeared to have a genetic predisposition for higher survival rates throughout life, and passed these genes on to their offspring. Thus, these experiments did not address the first two of the three predictions listed above.

Unfortunately, much less attention has been paid to these two predictions. In a comparative test of the relationship between physiological age at maturity and onset of ageing, Promislow (1991) showed that the onset of ageing in natural populations of mammals occurred much later than was predicted by theory. However, as Promislow et al. (1999) later pointed out, the apparent lack of concordance between the onset of senescence and the onset of reproduction could be as a result of a statistical sampling artefact.

Among closely related species of Drosophila, Pitnick et al. (1995) found substantial variation for male age at maturity. If this interspecific variation is a reflection of underlying genetic variation within species, we may be able to use artificial selection to delay age at physiological maturity, and so test Williams’ physiological predictions directly.

With this in mind, we examined whether age at physiological reproduction is genetically correlated with larval development time and patterns of ageing in adults. We used artificial selection in D. simulans to create replicate lines with relatively early and late age at physiological maturity, defined here as the time between eclosion and when a female lays her first egg. We then examined correlations between this trait and other life history traits, including body size, age-specific survival and fecundity. This approach allowed us to determine whether development in adults is largely decoupled from pre-adult development, as molecular evidence suggests ( Blake et al., 1996 ) or whether selection on delayed adult development leads to extended life span, as predicted by life history theory ( Williams, 1957; see also Harvey & Zammuto, 1985; Promislow & Harvey, 1990).

Materials and methods

Flies

Female fruit flies (D. simulans) were obtained from a peach orchard at the USDA South-eastern Fruit and Tree Nut Research Laboratory in Byron, GA, USA in August 1997. Approximately 200 mated females were brought into the lab, and placed singly in 8-dram vials with 10 mL of standard molasses-agar-cornmeal-yeast medium. Flies were reared at 24 °C for two generations, allowing time to reduce maternal effects in the novel lab environment. Throughout the experiment, flies were maintained under these conditions, with the proviso that the developmental process from egg to eclosion occurred at 26 °C (see below).

Selection regime

The starting population was derived from the second generation of flies obtained from the wild. This population consisted of 250 randomly mated pairs of flies, which were collected in three separate blocks over 2 days, each within 4 h of emergence. Each of the same-aged pair was placed in a vial at 24 °C at approximately 60% RH under a 12/12 h light–dark cycle, and monitored every 6 h for the appearance of eggs. After 5 days, we isolated females from the 50 pairs that had taken the longest to lay the first egg, and the 50 pairs in which the females had laid the first egg within the shortest period of time. Each group of 50 ‘early laying’ females and each group of 50 ‘late laying’ females was divided randomly into two groups of 25, from which we created four selection lines: early A, early B, late A and late B (hereafter referred to as Ea, Eb, La and Lb, respectively). The selected pairs were transferred into freshly yeasted 8-dram vials, allowed to lay approximately 40 eggs and discarded. These vials were kept at 26 °C until eclosion.

After this initial generation of selection, we carried out the following protocol. On the day of peak emergence, vials were cleared at 0600 hrs and newly eclosed flies not older than 4 h were collected at 1000 hrs and sexed under light CO2 anaesthesia. For each replicate selection line, we set up 125 male–female pairs, chosen at random, and placed each pair in an 8-dram vial with 5 mL of freshly yeasted standard medium at 24 °C. We monitored each pair at 6 h intervals, and recorded the time at which the first egg appeared. After 5 days, the 40 fastest pairs to lay eggs (within each of the Ea and Eb lines) or the 40 slowest pairs to lay eggs (within each of the La and Lb lines) were chosen to be the parents of the next generation. In the late lines, females that did not lay any eggs within 5 days were discarded.

We placed the selected flies in 1/2-pint plastic bottles (Applied Scientific, San Francisco, USA), with five females per bottle, and allowed them to lay eggs for 24–36 h at 26 °C, at which point we removed the adults. At this point, each bottle contained approximately 250 eggs. We deliberately limited the numbers of eggs to minimize the effects of density on larval competition and the time of emergence. The offspring were allowed to develop in these bottles at 26 °C until eclosion, at which point we repeated the selection process. We carried out selection for a total of 12 generations, with the exception of generation 11, when selection was relaxed for one cycle.

To test for significant differences in age at maturity among lines each generation, we used a mixed-model nested analysis of variance with a fixed treatment effect, and a random replicate within treatment effect, as shown in Eqn 1: inline image The experiment did not include a set of unselected, control lines. However, in a divergent selection regime such as ours, the downward selected lines act as a control for the upward selected lines and vice-versa ( Falconer & Mackay, 1996). Although unselected controls would allow us to determine the degree to which environmental changes might have influenced secular trends in the data, they are not strictly necessary to answer the specific questions addressed by this work.

Correlated responses

After 12 generations of selection and one further generation of relaxed selection, we carried out a series of assays to determine correlated responses. In particular, we determined whether selection response was specific to one sex and tested for correlated responses in development time, body mass, age-specific fecundity and age-specific mortality to selection on age at maturity.

Male and female response to selection

We set up a large assay of age at maturity that allowed us not only to obtain precise estimates of maturation time, but also to determine whether any response to selection that we observed was as a result of the changes in female age at maturity, male age at maturity or both. To perform this, we set up a total of 700 pairs of males and females in a series of crosses. These crosses included pairs made up of newly eclosed males and females, mature females with newly eclosed males and mature males with newly eclosed females. In addition to this, we set up males from Ei or Li treatments with females from the opposite treatment. In total, there were 20 different types of crosses, with 35 flies per cross.

Reproductively mature flies, which had been collected from each line during the previous generation of selection, were stored as virgins at 24 °C until the day of the assay. We collected the newly eclosed flies during a 4-h period on the morning of the day that we set up the assay. The 700 pairs of flies were placed haphazardly in trays to minimize block effects, and kept at 24 °C. Every vial was checked on a 3-h interval until the appearance of the first egg.

Development time

At generation 13, we compared egg-to-pupa and egg-to-adult development time among lines. Within each line, we placed virgin 1-week-old males and females together for 3 days. At the end of this period, we placed 40–50 females from each line in a lucite 0.32 L egg-laying chamber (see Promislow et al., 1998 ) for 3 h. Eggs were then collected from the chamber and placed in 8-dram vials with 10 mL standard medium at a density of 30 eggs per vial, with 10 vials per line. These vials were kept at 26 °C, which was the same temperature used for pre-adult development throughout our selection procedure. Vials were checked for pupae twice daily, with new pupae noted on the outside of the vial with a marking pen. Eclosed flies were collected and sexed 2 to 3 times daily. These data provided estimates of mean and variance for egg–pupa and egg–adult development time. We analysed the data using nested analysis of variance, with a random effect of replicate within treatment, and a fixed effect of treatment (Eq 1), using JMP ( SAS Institute, 1995).

Adult mortality rates

To estimate age-specific mortality rates, after 10 generations of selection and one of relaxed selection, we set up 8 single-sex cages (four with males, four with females) for each replicate line, with approximately 250 flies in each cage. The mortality cage is made of a 32-oz. plastic container (Reynolds Co., Louisville, USA), with a 75-mm-diameter screen cut in the lid for airflow, and a standard 8-dram vial with 5 mL fly medium attached by a plastic tube to the bottom of the cage. Vials were replaced daily to ensure a fresh food supply.

Every day, we removed and recorded the number of dead flies in each cage. After all flies had died, we determined the number of flies in the cage on the first day and each subsequent day. From this information, we were able to estimate the daily mortality, μx=−ln(px), where px is the probability of survival from age x to age x + 1.

We used the log-rank test in the Kaplan–Meier survival analysis package in JMP ( SAS Institute, 1995) to test for significant differences in survival rates between lines ( Lee, 1992).

Fecundity

After 12 generations of selection, we collected 50 newly eclosed females in each line and placed them singly in vials with 1-week-old males from the same line. After 3 days, we removed the males and allowed the females to lay eggs in a fresh vial for a 24 h period. We then transferred the females to fresh vials, and before counting the eggs from the first vial, we poured liquid nitrogen on them to stop development. We repeated 24-h fecundity estimates twice a week for 4 weeks, with 1-week-old males added one day prior to each 24-h egg laying period to ensure an adequate supply of sperm.

Body size

For generations two and three, we measured live body weight in 10 single etherized flies of each sex from each line. After that, to increase the number of flies weighed, we measured mass in five groups of five flies for each sex in each line.

After generation 12, we measured the wing size in 20 flies of each sex from each line (except LB, in which we measured 10 flies of each sex). We obtained digitized images for left and right wings of each fly with a Leica MZ8 dissecting scope attached to a Macintosh computer with NIH IMAGE software (obtainable from the National Institutes of Health, at http://rsb.info.nih.gov/nih-image). We digitized 12 landmarks on each wing, and then compared wing centroid size using GRF-ND morphometric software (developed by D. Slice, SUNY Stony Brook, and available at no charge from http://life.bio.sunysb.edu/morph/software.html). GRF-ND (for generalized rotational fitting of N-dimensional landmark data) uses standard morphometric approaches to calculate the size and shape of objects based on a pre-established set of landmarks ( Bookstein, 1991). The program provided us with centroid size for each wing for each fly. We then used analysis of variance to compare size among lines.

Results

Direct selection response

All four lines – early A and early B, late A and late B – responded to selection for age at maturity in the expected direction. The mean age at maturity across all lines did not change significantly over the course of the experiment, although from the fourth generation onwards, Ea and Eb had considerably reduced age at maturity compared with La and Lb ( Fig. 1; one-tailed nested analysis of variance, P < 0.05 for each of generations 2, 5, 6, 9, 10 and 12 after correcting for multiple comparisons; Fisher’s combined probability across all generations: χ222=66.4, P < 0.0001). At generation seven, there was a one-generation reversal in the La line, which had a shorter age at maturity than Eb. This was likely because of an unidentified environmental effect, as La was slower than either of the early lines in subsequent generations.

Figure . 1.

Selection response. Open and closed symbols represent lines selected for short and long duration, respectively, from eclosion to first egg. For each generation, the values are calculated (in hours) as the difference between each line and the mean of the four lines.

The divergence between lines reached a maximum at nine generations of selection, after which the total difference declines. Given the relatively short period of selection, we cannot be sure whether this decline was a result of an environmental or genetic effect.

Previous studies suggest that lines selected for rapid pre-adult development show a weaker response than lines selected for delayed development ( Clarke et al., 1961 ). Given the absence of unselected control lines, we cannot compare the total response between L and E lines. However, we can compare the heritability for different selection regimes, based on the relationship between the response to selection and the cumulative selection differential for each line ( Falconer & Mackay, 1996). Taking all data into account, heritability was slightly higher for the early lines than for the late lines (h2: Ea=0.048, Eb=0.031, La=0.019, Lb=0.020). Only heritability for Ea was significantly greater than 0 (F1, 11=16.1, P=0.002). If we exclude the last four generations of selection, after which the total response declined, heritabilities appear to be substantially higher (h2: Ea=0.084, P=0.0009; Eb=0.071, P=0.054; La=0.006, P=0.8; Lb=0.095, P=0.0071; n=8 for all cases). These heritabilities are somewhat lower than for pre-adult development (e.g. Zwaan et al., 1995a ), but in line with typical heritabilities for fitness-related traits.

Differences between males and females

In our comparison of age at maturity between E- and L-line individuals, we paired newly emerged females with mature males, and newly emerged males with mature females. In the first case, L females laid their first egg at a significantly later age than E females ( Fig. 2, Table 1). In contrast, in the second case (new males with mature females) there was no difference in the age at which the first egg was laid for mature females with E vs. L males. Thus, it appears that the response to selection on physiological age at maturity was primarily a result of response in females. Furthermore, when immature females from one treatment were crossed with mature males from a different treatment, there was no effect of time to first egg (F1,126=0.27, P=0.61), although there was still an effect of time to first larva (F1,113=7.82, P=0.0061).

Figure . 2.

Age to first egg for different crosses within lines, including newly eclosed males and females, newly eclosed females with mature males, and mature females with newly eclosed males. Note that the differences between treatments occur only when females are newly eclosed, suggesting that the response in selection was due to females and not males.

Table 1.  Differences between treatments for time when first egg was laid, at when first larva appeared, and duration from egg to first larva. Thumbnail image of

Correlated selection response

Body size

Females were consistently larger than males both in body mass and wing size ( Fig. 3). Body size was measured in generations 2–9. After table-wide correction for multiple comparisons, there were no significant differences between treatments for body size. Similarly, selection regime did not have an effect on wing centroid size ( Fig. 3).

Figure . 3.

Wing size, measured as the centroid based on 12 landmarks on each wing (see text for details).

Pre-adult development

Lines selected for earlier age at maturity had significantly longer egg-adult developmental periods than lines selected for delayed age at maturity (nested analysis of variance comparing mean development times per vial –F1,36=9.46, P=0.004; Fig. 4). In the vials set up to measure development time, egg densities varied from 10 to 39 eggs. There was no effect of density in the nested analysis of variance, and development times remained significantly different between treatments even after controlling for density (F1, 35=8.844, P=0.0053). The E lines also had significantly longer duration to pupal formation (F1, 36=7.34, P=0.010) and from pupa to eclosion (F1,36=6.7, P=0.014).

Figure . 4.

Relationship between hours elapsed from egg to eclosion (Y-axis) and from eclosion to first egg laid (X-axis), as measured at generation 13. Early maturity and late maturity lines are denoted by open and closed squares, respectively. Bars represent 1 standard error.

The experiment was designed to minimize correlated effects on development time. For each generation of selection, we obtained newly eclosed flies simultaneously from each line. Despite this experimentally imposed constraint, there was still a strong correlated effect on development time. In fact, towards the end of the experiment, we had to use flies that were among the latest to eclose from the L lines and the first to eclose from the E lines to ensure comparable development times for the two treatments ( Fig. 4).

Generation time

We define minimum generation time as the time elapsed from the first egg laid to the next, physiological mature female emerging and laying her first egg. To determine minimum generation time, we combined data for egg-adult development time with data for age at first egg laid. When we combine these two datasets, we find that there is no effect of selection regime on generation time (though early B has a longer generation time than the other three lines). This pattern arises as a result of the negative correlation between development time and age at first egg ( Fig. 4).

Adult demography and fecundity

According to standard evolutionary theories of ageing ( Williams, 1957), selection on delayed maturity should lead to longer life span. Previous studies have selected on the age at first allowable reproduction, without attempting to alter physiological age at maturity directly ( Rose & Charlesworth, 1980, 1981; Luckinbill et al., 1984 ; Partridge & Fowler, 1992; Zwaan et al., 1995b ; Partridge et al., 1999 ).

In this experiment, we found that after 10 generations of selection, longevity was higher in females from delayed maturity lines than in females from early maturity lines ( Fig. 5, Table 2), although differences were only 1–2 days. The pattern in males was less clear. Males from line Lb had longer life expectancy than the other three lines, but one of the two early lines (Eb) was longer lived than La ( Table 2).

Figure . 5.

Survival curves for females (top) and males (bottom). The solid lines refer to the delayed maturity treatments, and the dashed lines refer to the early maturity treatments. Late-maturing females live significantly longer than early-maturing females, but the effect is relatively weak (see text). Survival rates in males do not differ significantly.

Table 2.  Comparison of life expectancy in L and E lines. Thumbnail image of

There were no consistent differences between treatments in fecundity ( Fig. 6). The delayed maturity lines showed both highest (Lb) and lowest (La) fecundity through much of the life span.

Figure . 6.

Age-specific fecundity, measured as average number of eggs laid per female in a 24-h period, in lines selected for early (dashed lines) and late (solid lines) age at maturity. There was no significant relationship between fecundity and selection on age at maturity.

Discussion

Developmental traits

Previous studies of developmental rates in flies have focussed explicitly on pre-adult development. The results presented here demonstrate that there is significant genetic variation for the timing from eclosion to the age at which a female lays her first egg, that this duration shows a strong negative correlation with pre-adult developmental rate even when pre-adult development time is held constant between treatments during the selection process, and that the time from eclosion to first egg is positively correlated with adult life span.

Previous experiments on pre-adult development time in Drosophila have shown a relatively high amount of genetic variation, with realized heritabilities as high as 58% ( Sang & Clayton, 1957; Marien, 1958; Clarke et al., 1961 ; Marinkovic & Ayala, 1986; Zwaan et al., 1995a ). Our estimates of genetic variation for post-eclosion development are substantially lower, although continued selection may have uncovered more of a response. McMillan et al. (1970) found significant genotypic variance for age at first egg. However, that study did not measure additive genetic effects, and the actual trait was only measured indirectly. Life history theory suggests that age at maturity should be under very high selective pressure ( Cole, 1954; Dobzhansky et al., 1964 ; Lewontin, 1965). Thus, it is perhaps not surprising that we found very little genetic variation for time to egg laying.

In addition to strong selection, several other factors may have influenced the realized heritability in the study, including maternal effects, novel environment effects, inbreeding, drift and genetic trade-offs. In this study, we observed an initially high response to selection followed by a levelling off. This pattern is consistent with a strong maternal effect component influencing selection response ( Falconer, 1965; Kirkpatrick & Lande, 1989), with little or no heritable variation. A previous selection experiment on pre-adult development time in Drosophila found that after an initially rapid response, divergence between lines declined by the eighth generation of selection ( Zwaan et al., 1995a ). However, after approximately 12 generations of selection, the response began to increase again in Zwaan’s study, suggesting that both maternal and additive genetic effects may be important. In our own study, given the relatively small number of breeding pairs per line per generation (40 males and 40 females), we may have seen relatively low heritability because of loss of genetic variation from inbreeding or drift. Finally, novel environmental factors acting on wild-caught flies may have selected on traits other than those that we were directly selecting, leading to a reduced rate of response (see, for example, Van Tienderen & De Jong, 1994). We shall examine this last issue in greater detail later in the discussion.

At the same time that several factors may have limited genetic variation, in the longer term, the trade-off that we observed between pre- and post-eclosion development may help to maintain genetic variation. Our findings suggest that flies may be genetically and physiologically constrained such that they can alter the time from egg to eclosion, but not the minimum generation time (the time that it takes for a fertilized egg to develop into a mature female and produce a fertilized egg) (but see Chippindale et al., 1997 ). One evolutionarily important factor that this study does not address is the genetic variation not simply for the age at which the first egg is laid, but for the number of eggs laid early in life. We know that there is substantial genetic variation for fecundity at very early ages ( Tatar et al., 1996 ). In fact, Tatar et al. (1996) argue that this variation may be maintained by a trade-off between optimal time at eclosion and optimal age at maturity, consistent with the results that we found here.

The negative correlation between pre- and post-eclosion development time that we observed was consistent with much earlier work by Bonnier (1926) and Li (1927), who suggested that if one component of development were accelerated (in the present case, by artificial selection), the other component would ‘compensate’ by slowing down. In contrast, Chippindale et al. (1997) found a strong response to selection for decreased egg-to-egg development time, reducing the combined duration of pre- and post-eclosion development from 12 to 8 days over 125 generations. Although their study did not examine the correlation between larval and adult development, the strong response they found suggests that the negative genetic correlations could eventually be broken down if both traits are selected on simultaneously. However, attempts to deliberately disrupt genetic correlations through selection have met with mixed success, at least in plants ( Camara & Pigliucci, 1999; Camara et al., 2000 ). Of course, the negative genetic correlations that we observed for a recently caught wild strain may not have existed in the lab-adapted flies used by Chippindale et al.

Genetic constraints clearly play an important role in the short term evolution of life history strategies in flies. What might account for the apparently tight constraint that regulates the total ‘egg-to-egg’ development time in the population of Drosophila examined in this study? One possibility is that the response to selection on age at first egg is not a result of differences in rates of physiological maturity, but simply of the duration that a female holds fertilized eggs before laying them. If females in the L lines held eggs longer than females in the E lines, once the eggs were laid, we may expect them to hatch and mature sooner. However, the total development time from the time of fertilization would be no different. This trade-off may be relevant for flies in natural conditions, where the cost of delaying the time to lay eggs may be offset by the benefit of increased time available to find an ideal site for larval development. Future experiments will need to examine the physiological basis of this trade-off in greater detail.

The response to selection that we observed was confined to females. This is not surprising, given that the selection in this experiment focused primarily on female age at maturity. However, comparative evidence suggests that the greatest interspecific variation for age at maturity within the genus Drosophila is found in males ( Pitnick et al., 1995 ). Future studies might benefit from direct selection on male age at maturity to determine if genetic variance for male age at maturity is commensurate with high among-species variation.

Lab vs. wild flies

This study differs from most other selection studies on ageing in its use of flies newly caught from the wild. In a highly influential paper, Service & Rose (1985) argued that in order to obtain unbiased estimates for genetic correlations, one needs to use organisms that have adapted to the environment in which the selection will take place. The risk with wild-caught organisms is two-fold. First, those individuals with genotypes that are well-adapted to the new lab environment could show relatively high values for most fitness traits. Thus, genetic correlations will appear to be positive, solely as a result of the effects of the novel environment. Second, during the course of the selection experiment, the response to direct selection by the experimenter may be limited by selection on other, unexamined traits imposed by the novel laboratory regime (e.g. Van Tienderen & De Jong, 1994).

The flies we used had been brought into the lab very recently, and had therefore not yet adapted themselves lab culture. In light of recent studies on the genetic structure of laboratory populations ( Promislow et al., 1999 ; Sgrò & Partridge, 2000), we were specifically interested in finding evidence for standing genetic variation in natural populations. By using divergent selection, we were able to identify natural genetic variation and, at the same time, to minimize the confounding effects of lab adaptation. Presumably, the lab environment would have led to selection in the same direction for both E and L treatments. In this case, the effect of inadvertent selection would have been to increase type II error rates – a conservative response. Similarly, if the influence of the novel lab environment had affected observed correlations, we would have been more likely to observe positive correlations between age to first egg and both egg-eclosion duration and adult mortality. In both cases, we observed negative correlations. Whether these fitness trade-offs would be maintained in the lab environment over the long term remains to be seen ( Leroi et al., 1994a , b).

Ageing

Over the past 20 years, a series of selection experiments has addressed the relationship between age at maturity and longevity in Drosophila ( Rose & Charlesworth, 1981; Luckinbill et al., 1984 ; Rose, 1984a; Partridge & Fowler, 1992; Zwaan et al., 1995b ; Partridge et al., 1999 ). In each of these studies, age at maturity has been defined as the age at which flies are allowed to breed. We refer to this approach as selection on demographic maturity. By contrast, in this case we have used artificial selection to alter the age at physiological maturity. This latter approach has enabled us to look directly at the relationship between adult development and ageing. The approach is relevant for testing two models of ageing.

The first model, the developmental theory of ageing ( Lints & Lints, 1969), predicts a positive genetic correlation between pre-adult development time and longevity. A series of studies in the past 5 years has rekindled interest in this theory ( Zwaan et al., 1991 , 1992; Chippindale et al., 1994 , 1995a), although in general, this work fails to support the developmental theory. In the present study, we found that after a relatively short period of selection, flies selected for slow post-eclosion development had longer life span, although the effect was found only in females and was relatively weak. In fact, the negative genetic correlation that we found between larval and adult development implies a negative correlation between larval development and longevity – the opposite of that predicted by Lints & Lints (1969).

Second, the approach that we have used here offers a novel and potentially fruitful way to address predictions derived from Williams’ antagonistic pleiotropy theory of senescence. Based on his theory, Williams predicted that ‘rapid individual development should be correlated with rapid senescence’ ( Williams, 1957; p. 409). To test these theories of aging ( Medawar, 1952; Williams, 1957), Edney & Gill (1968) proposed breeding only from young flies. Trade-offs between early vigor and late survival would then lead to higher late-age mortality. They dismissed as too difficult attempts to select on increased longevity by reducing late-age mortality rates. But it was this latter prediction that led to a series of experiments where researchers selected on demographic maturity by breeding from flies that survived to a relatively late age. In one sense, these experiments were enormously successful ( Rose & Charlesworth, 1981; Luckinbill et al., 1984 ; Rose, 1984a; Partridge & Fowler, 1992; Zwaan et al., 1995b ; Partridge et al., 1999 ). By enforcing a regime of delayed breeding, researchers were able to dramatically increase longevity in selected lines.

However, the degree to which these experiments actually support Williams’ hypotheses has been questioned ( Clark, 1987; Partridge & Fowler 1992; Curtsinger et al., 1995 ). In most cases the selection regime creates an effectively semelparous population, where generations do not overlap and individuals have only a few days during which breeding is possible, spaced apart by as much as 10 weeks (but see Stearns et al., 2000 ). The most successful individuals in this regime will be those that can survive to relatively late age, and having survived, are the most fecund. Thus, as Partridge & Fowler (1992) noted, in an attempt to determine whether there is a correlation between age-specific fecundity and survival, most experimenters have inadvertently selected on both traits simultaneously, thereby creating the very correlation one was hoping to find. Whereas many of the potentially confounding effects of earlier selection experiments have been rectified in recent work ( Partridge et al., 1999 ), to our knowledge only one study has fully eliminated the coupled selection on survival and fecundity ( Zwaan et al., 1995b ).

Notwithstanding the problems with these ‘delayed breeding’ experiments, the evidence is incontrovertible that demographic selection experiments can extend longevity. Our more immediate concern with these studies, and one which is directly relevant to the study we have carried out, is our belief that at least some of these experiments have been based on a misinterpretation of Williams’ original theory.

By delaying the onset of reproduction, selection will be maintained at a high and constant level until reproduction starts ( Hamilton, 1966). Thus, a demographic delay in reproduction is equivalent to a demographic delay in the onset of the decline in selection intensity ( Rose, 1991).

Unfortunately, this focus on demographic maturity has led us to ignore much of Williams’ original focus. In his discussion on reproductive maturity, Williams is clearly referring not to demographic maturity (i.e. age at breeding), but to actual physiological maturity. Williams defines the life course as ‘a fixed developmental sequence’, and says that ‘if development were greatly accelerated…senescence would be accelerated and appear at the usual developmental stage’ ( Williams, 1957; pp. 409–410). By ‘usual developmental stage’, he means age at physiological maturity. From this and the subsequent discussion in his paper, it is clear that Williams is thinking of maturity as a part of the developmental process. Delayed breeding experiments alter the age at which adults are allowed to breed, but do not directly select on their developmental trajectory. Williams predicted that delayed maturity should lead to extended life span, and conversely, individuals that mature earlier should show more rapid senescence and that the onset of ageing should coincide with the onset of reproduction. In each case, we think he could only have been referring to age at physiological maturity. Each of these predictions relating to age at physiological maturity has hitherto been ignored.

In this light, demographic selection experiments that delay breeding, but which do not select on reproductive maturity, do not directly test the predictions about development as outlined by Williams. Although the demographic approach is certainly useful in understanding some elements of the ecological and genetic basis of senescence, it does not address predictions about the effects of variation in physiological maturity. One well-known proponent of demographic approaches to studying ageing has defined senescence as ‘a persistent decline in the age-specific fitness components of an organism because of internal physiological deterioration’ ( Rose, 1984b; p. 20). For evolutionary biologists interested in senescence, ageing is typically measured in demographic terms, but as Rose’s definition illustrates, the phenomenon itself is inextricably linked with underlying physiological processes. While some workers have attempted to select on physiological traits and determine if these traits are genetically correlated with longevity ( Rose et al., 1992 ; Zwaan et al., 1995a ), no one to our knowledge has tried to examine the correlation between physiological maturity and longevity in Drosophila. This is particularly surprising, given the central role that age at maturity plays in the fitness of an organism ( Lewontin, 1965).

This distinction between demographic and physiological selection is further complicated by the fact that Drosophila is a mitotically limited organism. In Drosophila, almost all cells are post-mitotic soon after eclosion. This clearly limits the degree to which the developmental trajectory can be altered, at least in terms of continued growth into adult life. In many organisms, continued growth is a critical component of the maturation process, and may play an important part in senescence ( Finch, 1990). It may turn out that at least the more developmental aspects of Williams’ model are best tested with organisms that are not mitotically limited.

One might argue that the distinction between physiological and demographic selection that we make here is misleading, as the demographic selection approach used by Rose and colleagues has led to a delay in pre-adult development time ( Chippindale et al., 1994 ) – surely a physiological parameter – as predicted by Williams (1957). However, there is an alternative explanation for the result of Chippindale et al. (1994) . In their study, which found a correlation between longevity and pre-adult development, the nonselected, short-lived lines had been maintained in standard 2-week culture. In this culture regime, flies are placed in bottles, allowed to lay eggs, and then discarded. Two weeks after the bottles are set up, newly emerged flies from the next generation are transferred to fresh bottles, and the cycle starts anew. This culture regime leads to extremely strong selection for rapid development. Flies that do not eclose and develop a large number of mature oocytes before being transferred to fresh bottles will be unlikely to contribute offspring to subsequent generations ( Clark, 1987; Promislow & Tatar, 1998; Sgrò & Partridge, 2000). In contrast, in the ‘delayed breeding’ selection regime, flies can delay development substantially (caused by mutation accumulation and genetic drift) without negative consequences on fitness. Thus, the delay in development could be a result of a relaxation of directional selection on rapid development seen in control lines, rather than direct selection favouring delayed development in selection lines.

The trade-off that we observed between different components of development suggests that the relationship between development and ageing in insects may be more complicated than was previously thought. If different elements of development show negative pleiotropy, there can be no simple relationship between rates of ageing and rates of development, as put forth by Williams (1957) and Lints & Lints (1969). Further work is needed to determine the generality of the trade-off that we observed (which was not seen in Chippindale et al., 1997 ), and which particular elements of development are most likely to influence adult life history traits.

In conclusion, we have drawn an explicit distinction between demographic selection and physiological selection on age at maturity. Demographic selection changes the ecology or external environment of the organism, whereas physiological selection changes the physiology, or internal environment of the organism. We think that both can contribute to our understanding of the genetics and evolution of senescence. But selection on physiological age at maturity provides new opportunities for analysing the genetics of ageing by effectively reversing the classical approach carried out by so many previous workers. Rather than selecting on longevity and asking how early adult reproductive traits respond, we advocate an approach where one selects instead on early adult reproductive traits and measures the response in survival. This approach avoids the confounding effects of simultaneously selecting on fecundity and longevity ( Partridge & Fowler, 1992). It takes the phenotypic approach whereby costs of reproduction are measured by manipulating reproductive effort and measuring survival (e.g. Partridge & Andrews, 1985; Partridge et al., 1987 ; Sgrò & Partridge, 1999) and uses artificial selection to determine the genetic basis of the relationship (see Reznick, 1985). Our approach addresses an essential component of Williams’ theory that has long been neglected. It is our hope that through continued selection experiments on early adult reproductive physiology, we may be able to understand more fully the physiological and genetic basis of ageing and life history strategies.

Acknowledgments

We would like to thank Graham Bell, Dale Hoyt, Amber Keyser, Paul Mack, Scott Pitnick, Nick Priest, Marc Tatar and two anonymous referees for critical comments on an earlier draft of this manuscript, Judy Willis for discussions on developmental issues, and the USDA South-eastern Fruit and Tree Nut Research Laboratory in Byron, GA, USA for letting us out of the compound after hours. This work was supported by National Institute on Ageing grant AG14027 to D. Promislow.

Ancillary