Density-dependent natural selection in Drosophila: Adaptation to adult crowding

The effects of adult crowding on two components of fitness were studied in three sets of Drosophila melanogaster populations, subjected to life-stage-specific, density-dependent natural selection in the laboratory for over 50 generations. Three days of crowding, early in adult life, were observed to increase mortality significantly during the episode of crowding, as well as decrease subsequent fecundity. Populations selected for adaptation to high adult densities suffered significantly lower mortality during episodes of adult crowding, as compared to populations selected specifically for adaptation to larval crowding, as well as control populations typically maintained at low larval and adult densities. Moreover, populations adapted to larval crowding seemed to be adversely affected by adult crowding to a greater extent than the controls, raising the possibility of trade-offs between adaptations to larval and adult crowding, respectively. Preliminary evidence suggests that the populations adapted to adult crowding may have evolved a propensity to stay away from the food medium, which is where most deaths occur when adults are crowded in culture vials.


Introduction
The signi®cant role of density-dependent natural selection in moulding the evolution of life histories and adaptive strategies is now widely recognized (MacArthur and Wilson, 1967;Pianka, 1970;Boyce, 1984;Elgar and Catterall, 1989;Travis and Mueller, 1989). Moreover, explicit mathematical models of density-dependent and age-speci®c selection have underscored the importance of the timing and precise mechanisms of density-dependent regulation of ®tness components for the evolution of particular life-history strategies in populations (Charlesworth, 1971(Charlesworth, , 1980King and Anderson, 1971;Clarke, 1972;Roughgarden, 1979;Iwasa and Teramoto, 1980;Asmussen, 1983;Nunney, 1983;Mueller, 1988a). Nevertheless, there have been very few systematic investigations of patterns and processes in the evolution of speci®c adaptations to crowding in carefully controlled, and relatively well characterized, laboratory populations (reviewed in Ro, 1992;Joshi, 1997). Most of these studies have used laboratory populations of Drosophila (e.g. Taylor and Condra, 1980;Mueller, 1988bMueller, , 1990Joshi and Mueller, 1988Bierbaum et al., 1989;Mueller et al., 1993), and have typically focused on the impact of larval crowding on the evolution of larval and adult ®tness components.
Interestingly, despite numerous laboratory and ®eld studies documenting the impact of adult density on a variety of ®tness correlates in many species (Pearl et al., 1927;Park, 1932;Utida, 1941;Davis, 1945;Chiang and Hodson, 1950;Frank et al., 1957;Tanner, 1966;Mueller and Ayala, 1981;Graves and Mueller, 1993;Condit et al., 1994;Tonn et al., 1994;Ostfeld and Canham, 1995), there has never been, to our knowledge, any empirical study of the evolution of speci®c adaptations to high levels of adult crowding. There have been attempts to document dierences in components of ®tness among extant wild populations, and ascribe them to inferred density dierences in the past (Pianka, 1970;Gadgil and Solbrig, 1972;Abrahamson and Gadgil, 1973;McNaughton, 1975). In such studies, it is not possible to assert unambiguously that any observed dierences are due to density-dependent selection. Moreover, it is also impossible to link crowding experienced during a speci®c life stage to any observed dierences among populations.
The inability to tease apart the evolutionary eects of larval versus adult crowding has also been a problem in interpreting results from the very few studies comparing adult ®tness correlates in Drosophila populations subjected to varying density-dependent selection regimes (Taylor and Condra, 1980;Mueller and Ayala, 1981;Mueller et al., 1993). Taylor and Condra (1980) measured body size, longevity and age-speci®c fecundity on two pairs of r-and K-selected populations of Drosophila pseudoobscura. They observed signi®cantly greater longevity in the K-selected females, and also found some evidence suggesting that K-selected females may have greater late-life fecundity than r-selected females. Their results are, however, dicult to interpret in terms of adaptations to larval or adult crowding because their K-selected lines were subjected to crowding both as larvae and as adults, and their r-selected lines were under direct selection for decreased egg± adult development time, which could potentially aect the evolution of ®tness components quite independently of density.
So far, the clearest evidence for adaptation to adult crowding has come from studies on three pairs of Drosophila melanogaster populations subjected to selection at low and high densities, respectively (the r-and K-populations of Mueller and Ayala, 1981). The K-populations were observed to have greater rates of population growth at high adult densities (Mueller and Ayala, 1981), as well as greater tolerance to the detrimental eects of adult crowding on longevity . Although these results were relatively unambiguous, the K-populations, nevertheless, diered from the r-populations in several features of their maintenance regime other than adult density . Larval density in the K-populations was typically much higher than in the r-populations. More importantly, the K-populations were maintained with overlapping generations, permitting reproduction throughout life, whereas the r-populations were maintained on a discrete generation cycle that restricted reproduction to the ®rst few days of adult life. These features of the r-and K-selection regimes, therefore, confounded the potential eects of age-speci®c selection and density-dependent selection on the larval and adult stages, making it dicult to ascribe unequivocally any observed changes in these populations to speci®c causes .
This drawback with the r-and K-system motivated the creation of a set of 15 populations subjected to crowding speci®cally during either the larval or the adult stage: the CU, UC and UU populations (described in Mueller et al., 1993, Joshi and. The ®rst letter in these population designations refers to larval density and the second letter to adult density (e.g. CU: crowded as larvae, uncrowded as adults). These populations constitute a powerful system for studying the evolutionary eects of extreme crowding at dierent life stages, because they allow observed dierences among populations subjected to varying selection regimes to be ascribed unambiguously to density-dependent selection acting speci®cally on the larval or adult stage of the life cycle.
In this paper, we report results from two experiments in which we investigated the eects of adult crowding on key ®tness components in the CU, UC and UU populations, to ascertain whether populations reared at high adult densities exhibit increased tolerance to adult crowding. We also examined the issue of whether populations selected for adaptations to high larval densities had undergone any correlated evolutionary changes in their ability to withstand adult crowding.

Experimental populations
This study used three sets of ®ve replicate populations of D. melanogaster that had each been subjected to diering levels of larval or adult density for over 50 generations . All populations were maintained on banana-molasses food at 25°C and continuous light, and had a generation time of about 3 weeks. Population sizes each generation were about 2000±4000 breeding adults. The ®ve populations crowded as larvae (CU 1 ¼CU 5 ) were reared at densities of 1000 or more larvae per 6 dram vial. Eclosing adults were collected daily from these vials, and kept at a low density of about 60±80 adults per 8 dram vial. The ®ve uncrowded populations (UU 1 ¼UU 5 ) were reared at low larval densities of 60±80 larvae per 8 dram vial; eclosing adults were subjected to the same density as the CU populations. The ®ve populations crowded as adults (UC 1 ¼UC 5 ) were reared at low larval densities of 60±80 larvae per 8 dram vial; eclosed adults were collected from these vials on the 13th day after egg-lay and kept in 8 dram vials at densities of about 160±200 adults per vial. Thus, the three sets of populations diered in the degree of larval or adult crowding to which they were exposed, with the UU populations acting as controls to both the UC and CU populations. Prior to initiating a new generation, all the eclosed adults from a population were dumped into a plexiglass cage (25.5´20´14.4 cm 3 ) and supplied with liberal amounts of live yeast paste for 2 days before egg collection. All three sets of populations were derived from the ®ve B populations of Rose (1984), each B population being used as the progenitor of one CU, one UC and one UU population. Consequently, CU, UC and UU populations bearing the same numerical subscript are more closely related to each other, as compared to other populations subjected to the same density regime.

First experiment: 3 days of adult conditioning
Collection of adult¯ies for assays. Prior to initiating the assays described below, all test populations were passed through one complete generation of identical rearing conditions, so as to eliminate any dierences among selected lines due to environmental or maternal eects. Eggs were collected from the adults of each population and placed in 8 dram vials at low densities of 60±80 eggs per vial. Eclosing adults from these vials were then collected into cages; eggs laid by these adults were collected into 8 dram vials at low densities of 60±80 eggs per vial. Adult¯ies eclosing in these vials were collected one day after eclosion, and put into one of two conditioning treatments: crowded (75 males and 75 females per vial) or uncrowded (25 males and 25 females per vial). All conditioning vials contained exactly 5 ml food medium. The two conditioning densities were chosen to approximate the adult densities experienced by the UC (crowded) and the CU/UU (uncrowded) populations in their respective selection regimes. For each population, seven vials were set up at each conditioning density, resulting in a total of 210 vials (3 selection regimes´5 replicate pop-ulations´2 conditioning densities´7 vials). The¯ies remained in these vials for 3 days, after which they were used for setting up the fecundity and longevity assays described below. Any¯ies that died during the 3 days of conditioning were sexed, and the number of male and female dead in each vial was recorded.
Fecundity assay. After the 3 day conditioning period, the fecundity of females was assayed at two dierent densities: high (25 males and 25 females per vial) and low (1 male and 1 female per vial). This was done to determine whether an episode of adult crowding aects the subsequent sensitivity of female fecundity to adult density at the time of egg laying. For each population´conditioning combination, 15 vials were set up at the low assay density; the number of vials set up at the high assay density varied from 4 to 8, based on the availability of¯ies that survived the conditioning treatments. Flies to be assayed were placed in 8 dram vials containing about 3 ml of charcoalsucrose medium (Rose and Charlesworth, 1981), and a dab of live yeast paste on the side of the vial to ensure an abundant supply of food (Mueller and Huynh, 1994). The¯ies were given exactly 24 h to lay eggs in these vials, after which the adults were discarded and the number of eggs in each vial counted.

Second experiment: 5 days of adult conditioning
The second experiment was conducted after the results of the ®rst experiment had strongly suggested that the UC¯ies were better adapted to tolerate adult crowding. It also appeared that UC ies, in both crowded and uncrowded vials, tended to spend more time away from the food surface, close to the sponge plug at the top of the vial (A. Joshi, personal observation). In the second experiment, we increased both the degree and duration of the adult crowding, in an attempt to dierentiate more clearly among the response to adult crowding exhibited by the dierent selection lines. We also measured the gain in weight of¯ies during the ®rst 5 days of adult life; if UC¯ies indeed tend to avoid the food in the vials, they might be expected to gain less weight during this time, as compared to their UU and CU counterparts.
As in the ®rst experiment, test populations were passed through one complete generation of identical rearing conditions prior to being assayed. Virgin females were collected within 3 h of eclosion and frozen for subsequent weighing. Flies were collected in batches of eight females; six or seven such batches were collected for each replicate population. Prior to weighing, the¯ies were dried in an oven at 80°C for 24 h. In addition, two adult conditioning treatments were set up: crowded (100 males and 100 females per vial) and uncrowded (20 males and 20 females per vial). All conditioning vials contained exactly 5 ml food medium. For each population, 10 vials were set up at each conditioning density, and the¯ies remained in these vials for 5 days. Any¯ies that died during the 5 days of conditioning were sexed, and the number of male and female dead in each vial was recorded. For each population, six to seven batches of eight females each were collected on the sixth day after eclosion, from among the survivors of the uncrowded conditioning treatment. Thesē ies were frozen, dried and weighed as described earlier for the virgin¯ies. Due to the extremely high mortality in the crowded conditioning vials (see Results), we were unable to collect a reasonable sample of¯ies from the crowded treatment for weighing.

Statistical analysis
All analyses of variance (ANOVA) were performed using the procedure GLM of SAS for Windows version 6.08. Due to the pattern of relatedness among the CU, UC and UU populations (CU i , UC i and UU i are more closely related to each other than any of them is to other populations with which they share the same selection regime, i 1 F F F 5, sets of CU, UC and UU populations, matched by subscripted indices, were treated as random blocks in the analyses. Selection regime and conditioning density were treated as ®xed eects crossed within each block. Mortality data from both experiments were subjected to the arcsin square-root transformation prior to analysis (Freeman and Tukey, 1950). For these data, sex was treated as a crossed, ®xed factor, along with selection and conditioning, and the units of analysis were individual estimates of mortality from each vial.
For the data from the fecundity assay, selection, conditioning and density of measurement were treated as crossed, ®xed factors. The units of analysis were estimates of mean fecundity across all females in a vial. Consequently, individual data points from the high assay density were mean fecundities (averaged across 25 females in a vial), whereas those from the low assay density were fecundities of individual females in a vial. For the dry weights of female¯ies at days 0 and 6 after eclosion, the units of analysis were the weights, in grams, of individual batches of eight¯ies; the ANOVA model included selection regime and day of measurement as ®xed main eects crossed with the random blocks.

First experiment: 3 days of adult conditioning
Mortality during conditioning. The ANOVA results for mortality during the 3 days of adult conditioning showed signi®cant eects of sex and conditioning density, as well as a signi®cant selec-tion´sex´conditioning interaction (Table 1). In general, males suered greater mortality than females, and¯ies in the crowded conditioning treatment suered greater mortality than those subjected to uncrowded conditioning (Fig. 1). In the uncrowded conditioning treatment, CU males suered signi®cantly greater mortality (`0X01) than UC males; all other dierences were non-signi®cant (Fig. 1a). In contrast, in the crowded conditioning treatment, CU¯ies suered the highest mortality, followed by¯ies from the UU populations. The UC¯ies suered the lowest levels of mortality. This pattern was consistent across both sexes, and all dierences among populations were signi®cant at the 0.01 level (Fig. 1b). The overall levels of mortality over 3 days of crowding, however, were fairly low (Fig. 1). Eect of adult crowding on fecundity. The ANOVA results for fecundity after 3 days of adult conditioning showed signi®cant eects of conditioning density and the density at which fecundity was measured (Table 2). In general, fecundity decreased with increasing density of conditioning and/or measurement, and these eects were more or less additive (Fig. 2). The pattern of mean fecundity after crowded and uncrowded conditioning also suggested that the detrimental eect of adult crowding on fecundity may be greatest on the CU females and least on the UC females. After the uncrowded conditioning treatment (Fig. 2a), the mean fecundity of UU females, averaged across the two measuring densities, was signi®cantly lower (`0X02) than that of females from both the UC and CU populations (UU < CU UC). However, after the crowded conditioning treatment (Fig. 2b), UU and CU females did not dier from each other in mean fecundity, although both populations showed signi®cantly lower fecundity (`0X005) compared to the UC populations (UU CU < UC). Strictly speaking, if the appropriate interaction eect is not signi®cant, then signi®cant dierences seen in multiple comparisons should, at best, be considered suggestive of, rather than evidence for, the existence of meaningful dierences among cell means in an ANOVA. We have included the results from these multiple comparisons for two reasons. The pattern of dierences in fecundity among the UU, UC and CU populations is the same as that seen for the eect of adult crowding on mortality in this study, and on longevity (Joshi and Mueller, 1997) in these populations; the likelihood of the same pattern arising by chance in three separate assays of dierent ®tness components is rather remote. Moreover, the lack of a signi®cant selec-tion´conditioning interaction in the ANOVA is entirely due to the anomalous response of one population (CU 5 ) to adult crowding. In our laboratory, we have previously noted that this population diers dramatically from the other four CU populations for several ®tness traits related to fecundity in Drosophila; for example, dry weight and lipid content (D.J. Borash and L.D. Mueller, unpublished data). Indeed, removing block 5 (UU 5 , UC 5 , CU 5 ) from the dataset has the eect of rendering the selection´conditioning interaction highly signi®cant (p 10X02, `0X02); the magnitude of the other ANOVA eects is not signi®cantly altered.

Second experiment: 5 days of adult conditioning
Mortality during conditioning. Increasing the duration of adult conditioning from 3 days, in the ®rst experiment, to 5 days, in the second experiment, resulted in higher mortality overall (Fig. 3), especially in the crowded treatment (compare Fig. 3b to Fig. 1b). In the more severely crowded conditioning treatment of the second experiment, females, in general, suered higher mortality than males (`0X01), although the dierence between the sexes was signi®cant only in the crowded conditioning treatment ( 0X001). Overall, the eect of selection regime was signi®cant, as were the selection´sex, selection´conditioning and conditioning´sex interactions ( Table 3). The rank order of mortality suered by the selection lines during the 5 days of crowded conditioning (CU > UU > UC) was the same as in the ®rst experiment. Under the more severe crowding of the second experiment, however, only the dierences between the UC and CU Note: Fecundities were measured at low (1 male and 1 female per vial) and high (25 males and 25 females per vial) densities for each combination of population and conditioning. Signi®cant ®xed main eects and interactions are indicated in bold type (terms enclosed in parentheses denote abbreviations, not nested eects).
( 0X0001) and UC and UU ( 0X0001) populations were signi®cant; the populations did not dier signi®cantly in mortality during the 5 days of uncrowded conditioning.
Dry weight at days 0 and 6 of adult life. Overall, the eect of selection regime on dry weight of females was only marginally signi®cant ( 0X045); there were, however, signi®cant eects due to day (virgins at day 0 post-eclosion vs mated¯ies after 5 days of uncrowded conditioning) and the selection´day interaction (Table 4). Both the UU and CU populations showed a signi®cant (`0X001) increase in weight over the 5 days of adult conditioning at low density; the increase in weight exhibited by the UC populations, on the other hand, was negligible ( 0X16) (Fig. 4).

Discussion
The pattern of mortality experienced by the UU, UC and CU populations during adult conditioning, in both the ®rst and second experiments, clearly shows that adult¯ies from the UC The error bars depict 95% con®dence intervals about the mean of the ®ve replicate populations of each selection regime, and were calculated using least-squares estimates of the standard errors of cell means in the randomized block ANOVA.
populations were the least aected by crowding, followed by those from the UU and CU populations, in that order (Figs 1 and 3). The same pattern of sensitivity to adult crowding was also seen for fecundity in the ®rst experiment, as well as longevity (Joshi and Mueller, 1997), clearly indicating that the UC populations have evolved increased tolerance to the detrimental eects of adult crowding in response to being maintained at high adult densities. Although not previously documented in an unambiguous manner, this result is exactly what would be intuitively expected, given that adult crowding is known to have detrimental eects on ®tness in Drosophila.
The decline of female fecundity with increasing adult density in Drosophila is a well-documented phenomenon, and one that is known to be at least partly independent of the decrease in per capita food availability caused by increasing density (Mueller, 1985). Typically, studies of the eect of density on fecundity have involved measuring fecundity on female¯ies at dierent densities, as opposed to examining the eect of a discrete period of crowding on subsequent egg production (reviewed in Mueller, 1985). Our results clearly demonstrate that brief episodes of crowding can have a detrimental eect on subsequent fecundity, regardless of the density at the time of measurement (Fig. 2). The lack of a signi®cant conditioning density´measuring density interaction in Figure 3. Mean mortality of¯ies during the 5 days of uncrowded (A) or crowded (B) conditioning as adults in the second experiment. The error bars depict 95% con®dence intervals about the mean of the ®ve replicate populations of each selection regime, and were calculated using least-squares estimates of the standard errors of cell means in a randomized block ANOVA on untransformed mortality data. the ANOVA (Table 2) indicates that previous exposure to high density does not markedly aect the sensitivity of female fecundity to the density of adults at the time of egg-laying. Since we assayed fecundity only once, at the end of the conditioning period, our data do not, however, permit an assessment of how episodes of crowding may aect the long-term fecundity pro®le of females.
The data on the dry weight of females from the UU, UC and CU populations, as freshly eclosed virgins and after 5 days of adult life at low density (Fig. 4, Table 4), lend some credence to the speculation that the decreased mortality of UC¯ies in crowded cultures may in part be due to their having evolved a tendency to stay away from the food medium, which is where most deaths occur in crowded Drosophila cultures. In stark contrast to the UU and CU populations, females from the UC populations underwent almost no increase in weight during the ®rst 5 days of adult life at a  very moderate density of 20 males and 20 females per vial (Fig. 4). The UC¯ies, especially in crowded vials, tend to congregate in the upper part of the vial, near the sponge plug; UU and CŪ ies, on the other hand, tend to spend a lot of time on the food surface (A. Joshi, personal observation). Taken together, these two observations suggest that crowded adult conditions may possibly impose a selective advantage to¯ies with a tendency, perhaps due to negative geotaxis, to spend more time in the upper part of the vial, away from the food surface. Though by no means conclusive, the current results suggest that further studies on this behavioural aspect of adult life may help us to understand the mechanisms underlying adaptations to adult crowding in laboratory populations of Drosophila.
The consistently greater susceptibility to adult crowding of the CU populations, compared to the UU controls, is very interesting, inasmuch as it suggests the possibility of a hitherto unsuspected trade-o between adaptations to larval and adult crowding, indicating that studies of the performance of UC larvae under crowded larval conditions may be worth pursuing to determine whether the trade-o between adaptation to larval and adult crowding is symmetrical. This ®nding also underscores the importance of being able to separate the eects of crowding during dierent life stages in any attempt to study how density-dependent selection may shape the evolution of life histories.
In other experiments in our laboratory, we have documented the adaptation of the CU populations to their high-density larval rearing conditions. The CU populations have evolved higher larval feeding rates  and greater tolerance to metabolic waste, relative to the UU populations (Shiotsugu et al., 1997), as a consequence of being reared at high larval densities. It is, therefore, clear that both the UC and CU populations have diverged from the UU control lines as a result of adapting to their respective density-dependent selection regimes. The mechanisms of adaptation to larval crowding in these populations, and indeed in Drosophila in general, are much better studied and understood than are the mechanisms conferring enhanced tolerance to adult crowding. The possibility of a trade-o between adaptation to larval and adult crowding suggested by this study highlights the importance of elucidating exactly how the UC populations are able to withstand the detrimental eects of adult crowding, and why the CU populations are so sensitive to high adult densities. Such an understanding will be important for a clearer appreciation of the physiological and genetic factors that may shape the evolutionary response of populations to extreme density experienced at various stages of the life cycle.