A Genetic Polymorphism Maintained by Natural Selection in a Temporally Varying Environment

Environments that are crowded with larvae of the fruit fly, Drosophila melanogaster, exhibit a temporal deterioration in quality as waste products accumulate and food is depleted. We show that natural selection in these environments can maintain a genetic polymorphism with one group of genotypes specializing on the early part of the environment and a second group specializing on the late part. These specializations involve trade‐offs in fitness components. The early types emerge first from crowded cultures and have high larval feeding rates, which are positively correlated with competitive ability but exhibit lower absolute viability than the late phenotype, especially in food contaminated with the nitrogenous waste product, ammonia. The late emerging types have reduced feeding rates but higher absolute survival under conditions of severe crowding and high levels of ammonia. Organisms that experience temporal variation within a single generation are not uncommon, and this model system provides some of the first insights into the evolutionary forces at work in these environments.

Another important innovation in the theory of evolution was the examination of the outcome of natural selection when fitness varied over generations. This theory abstract: Environments that are crowded with larvae of the fruit has included models in which the environment passed fly, Drosophila melanogaster, exhibit a temporal deterioration in through fixed cycles, varied at random, or possessed quality as waste products accumulate and food is depleted. We show that natural selection in these environments can maintain a some autocorrelated variation (Wright 1948; Kimura genetic polymorphism with one group of genotypes specializing on 1954; Dempster 1955; Haldane and Jayakar 1963; Gillesthe early part of the environment and a second group specializing pie 1973, 1991Hartl and Cook 1973;Jensen 1973; on the late part. These specializations involve trade-offs in fitness Felsenstein 1976). However, all of these theoretical modcomponents. The early types emerge first from crowded cultures els are similar in their assumption that within a generaand have high larval feeding rates, which are positively correlated tion the environment assumed a fixed state and that gewith competitive ability but exhibit lower absolute viability than notypes could be characterized by a set of constant the late phenotype, especially in food contaminated with the nitrogenous waste product, ammonia. The late emerging types have fitness values. reduced feeding rates but higher absolute survival under condi-An unexplored but feasible extension of this theory tions of severe crowding and high levels of ammonia. Organisms would examine environments that go through a temporal that experience temporal variation within a single generation are sequence of deterioration within a generation, for innot uncommon, and this model system provides some of the first stance, habitats that are ephemeral and change rapidly insights into the evolutionary forces at work in these environover time. This deterioration of the environment could ments. be a function of ecological conditions like population Keywords: Drosophila melanogaster, density-dependent selection, density. One example may be excrement from large nitrogen wastes, ammonia, urea. mammals that serves as a habitat for many insects and microorganisms but dries out and decays over time. For Drosophila, similar conditions may occur in fresh fruit An important goal of evolutionary biology is to develop that falls to the ground and starts to decay. Drosophila an understanding of the role the natural environment has larvae in these habitats may often be at suboptimal denin molding adaptations and affecting allele frequency sities (Grimaldi and Jaenike 1984). The concentration of change (Partridge and Harvey 1988;Roff 1992; Stearns organic compounds also changes over time in Drosophila 1992). In the classical models of natural selection, the cultures. In particular, Drosophila food that initially has fitness of a genotype was assumed to be constant and high levels of ethanol shows a marked decline in ethanol thus unresponsive to changes in the environment. The levels and an increase in acetic acid levels as the cultures theory of density-dependent natural selection was one of age (Hageman et al. 1990). the first attempts to alter this view of evolution by devel-Many plant species may find themselves in environments in which the quality declines over time. For in-*Present address: Animal Behaviour Unit, Jawaharlal Nehru Centre for Advanced stance, goldenrod usually occupies recently cleared, early Scientific Research, Bangalore 560 064, India. successional habitats. As additional species settle nearby, †To whom correspondence should be addressed; E-mail: LDMUELLE@UCI.EDU. the shading and competitive environment are altered Am. Nat. 1998. Vol. 151, pp. 148-156. © 1998  significantly (Abrahamson and Weiss 1997). A similar sort of successional series can be created by fires; selec-adults. Eggs were then collected from each of the four population types (CU-early, CU-late, UU-early, and UU-tion pressures may change over time in these environments (Scheiner 1989). Any character, like dispersal abil-late) and passed through two generations of common conditions, which consisted of low larval and adult den-ity or development time, that exposes individuals to different slices of these sorts of temporal decays makes it sities. This type of standardization insures that any observed phenotypic differences between the four popula-possible for selection to act on traits that differentially adapt organisms to these changing aspects of the envi-tion types cannot be attributed to different environments the individuals were raised in (acclimation) or the differ-ronment.
This study develops a model system in which to study ent environments the mothers of the tested individuals were raised in (maternal effects). Consequently, these adaptation to these types of heterogeneous environments.
Here we examine populations of Drosophila melanogaster phenotypic differences ought to be due to genetic differences between the populations (Clausen et al. 1941). used to study density-dependent natural selection, specifically through crowding in the larval stage (Mueller et Since each population was replicated fivefold there were a total of 20 experimental populations on which the ex-al. 1993). While the populations are cultured on a fully discrete regime of reproduction, the larval environment perimental assays were done. shows a gradual deterioration over time as food is depleted, nitrogen wastes accumulate, and dead larvae de-Feeding Rates cay. This temporal variation permits some genotypes to specialize on the early part of the environmental se-Eggs were collected from adults that had been through the two-generation standardization procedure described quence and others to specialize on the late part of the sequence.
in figure 1. Newly hatched larvae from these eggs were raised on petri dishes with agar and live yeast paste. At 48 h of larval development, feeding rates of 20 larvae per population were measured by methods described Methods elsewhere (Joshi and Mueller 1988), with the following Populations modifications. At least 1 min of feeding behavior was videotaped with a camera attached to the dissecting mi-The two selected populations are both derived from a long-standing laboratory-adapted population called the croscope. Feeding rates were then counted from videotape records by two different people. If any feeding rates B's (Rose 1984). One population called the UU population has evolved in the laboratory under uncrowded lar-for a single larva differed by more than 10 retractions per minute the results were rechecked by each investigator. val (50-80 larvae/8-dram vial) and adult (50 adults/8dram vial) densities (Joshi and Mueller 1996). Each UU All feeding rates were completed during 1 wk. Since all populations could not be finished on a single day, popu-population consists of 40 vials. The second population, called CU, is maintained the same as the UU's except lations were broken into blocks. On a single day, all the populations with the same subscript were tested (e.g., that larvae are crowded (Ͼ1,000 larvae/6-dram vial). Each CU population consists of 20 vials. The UU and CU CU 1 -early, CU 1 -late, UU 1 -early, UU 1 -late). Thus, the possibility exists that for tests conducted on different days populations had evolved for approximately 117 and 145 generations, respectively, before the onset of the experi-uncontrolled experimental variables could produce differences in feeding rates. This experimental design is ments. Each of these populations is replicated fivefold so that differences between the CU and UU populations due handled by the block design ANOVA discussed later. A second feeding rate assay was conduced without the UU-to natural selection (a deterministic process) can be separated from differences arising due to genetic drift (a sto-derived populations. This experiment involved an independent derivation of the early and late populations as chastic process; Rose et al. 1996). All populations have breeding adult numbers of more than 1,500 adults each described previously. We present these results only to illustrate that the large differences between the CU-early generation. Eggs from each of these populations (the five CU and the five UU ) were collected and raised under the and -late population feeding rates are repeatable phenomena. same high larval densities ( fig. 1). Early adults are those that emerge during the first 72 h of adult eclosion. Previous work suggested that this sample will include about Viability 15%-20% of all eclosed adults. During the next 9-10 d, eclosing adults were removed daily from their crowded Adults that had undergone the standardization procedure were used to collect 60 eggs on nonnutritive agar. These cultures and not used. Flies emerging after this period were collected for about 48-72 h and classified as late eggs were placed in 8-dram vials with 5 mL of food, which contained the following: standard banana-molasses standard banana-molasses food. Each of these vials was replicated 20 times. On each experimental day, four vials food, standard food with 0.25 M ammonium chloride, or standard food with 0.3 M urea added. Each treatment from a given population were removed, and ϳ1 mL of food was stored at Ϫ20°C. Larvae were excluded from was replicated eight times for each of the 20 populations. The high density experiment was conducted on standard the samples as much as possible. Food homogenates were prepared by grinding 100 mg of food in 4 mL of water. food with 1,000 eggs per vial and replicated five times for each population. These experiments were performed on Homogenates were stored at Ϫ70°C until assayed. Ammonia concentrations were determined using a nicotin-all populations simultaneously. Prior to performing an ANOVA the viability data were subject to an arcsin amide adenine dinucleotide-linked assay (Mondzac et al. 1965). The reaction mixture (3 mL final volume) con-square-root transformation.
If one allele is fixed (A i , i ϭ 1, 2) the condition for the increase of the rare alternative allele is v 11 e 12 ϩ (1 Ϫ v 12 )l 12 Ͼ v ii e ii ϩ (1 Ϫ v ii )l ii . If mating is completely Statistics random, then the initial increase condition is the same, The evaluation of significant effects was made with the although the internal equilibrium frequencies have not aid of ANOVA implemented on SAS for Windows (SAS been determined. With the aid of this relationship, we Institute 1991). Population (CU vs. UU ), period (early can predict the conditions necessary for a protected polyvs. late), and food type (standard vs. ammonia, etc.) were morphism (Hartl and Clark 1989) or the point at which treated as fixed effects. Population replicate was treated natural selection will not fix either the A 1 or the A 2 as a block effect because of the common origin of CU i allele. For example, consider the point illustrated in and UU i populations, and in the case of the feeding rate figure 2. Here all three genotypes have the same early viexperiments these populations shared a common day of ability: the ''early genotype,'' A 1 A 1 , has a high fraction of analysis (Joshi and Mueller 1996). Survivorship data were adults emerging early (0.5) but low viability in the late transformed using the arcsin square-root transformation. environment (0.1). Conversely, the ''late genotype,'' Multiple comparisons were done using the Tukey-A 2 A 2 , has fewer adults emerging early (0.1) but higher Kramer method. viability in the late environment (0.5). The heterozygotes have an intermediate viability and emergence fractions, demonstrating that a polymorphism is possible Results without overdominance in each component of fitness.

Evolution in Theory
The important point is that there are broad conditions under which both alleles may be stably maintained by We start by developing a simple population genetic evolution due to genotypes with differing abilities to do model that illustrates how evolution might work in a well in either the early or late portion of the environtemporally variable environment. Assume a single locus ment. with two alleles, A 1 and A 2 . For genotype A i A j , the fraction that emerges early in the environmental profile is v ij . The viability of this early emerging group is e ij . The re-Evolution in the Laboratory maining portion of the genotypes (1 Ϫ v ij ) emerges dur-Several aspects of the larval environment change over the ing the late portion and their viability is l ij . If we assume time that larvae develop in crowded cultures. Of course there is complete assortative mating, for example, early food is depleted, and in our own laboratory the volume emerging types only mate with other early types, then we of food is often reduced by 50%-80%. Since the food must keep track of genotype frequencies since eggs will contains a growing population of yeast there is also an not be in Hardy-Weinberg proportions. Let the freaccumulation of acetic acid (Hageman et al. 1990). We quency of genotype A i A j among zygotes be x ij . Then the have sampled the food from crowded cultures and meafrequency of the A 1 allele in the early emerging adult sured levels of urea and ammonia ( fig. 3). These data population is show that there are almost no detectable levels of urea p ′ 1 ϭ (x 11 v 11 e 11 ϩ 1/2 x 12 v 12 e 12 ) w′ , (contrary to previous reports; Botella et al. 1985) but a significant and steadily increasing amount of ammonia. Thus, larvae that are more slowly developing in crowded where the mean fitness is w′ ϭ x 11 v 11 e 11 ϩ x 12 v 12 e 12 ϩ x 22 v 22 e 22 . The frequency of the A 1 allele in the late por-cultures are more likely to be exposed to high levels of ammonia through ingestion of polluted food. Over this tion of the adult populations is To test whether these environments could harbor a polymorphism similar to the one described by the previous model, we studied two types of laboratory populations of Drosophila melanogaster. We have isolated two subpopulations from the UU and CU populations that we call early and late, as illustrated in figure 1. There are two important features of the protocol outlined in figure  1. The use of two generations of common environmental conditions just prior to the assays insures that any differences observed between the four population groups (CUearly, CU-late, UU-early, UU-late) will be due to genetic differences among the populations, not environmentally induced differences. Also, if there are differences between the CU-early and the CU-late but not between the UUearly and the UU-late, then we can reasonably infer that natural selection due to larval crowding is the cause of the observed genetic differences. One phenotype that evolves in response to larval crowded cultures of the CU (adapted to crowded larval condicrowding is competitive ability (Mueller 1988a(Mueller , 1988b. tions) and UU (adapted to uncrowded larval conditions) populations.
In environments with limited food, increased competitive  Figure 5: Larval feeding rates for four different populations, of the CU (adapted to crowded larval conditions) and UU with * indicating the population with the significantly increased (adapted to uncrowded larval conditions) populations. Freshly feeding rate. An ANOVA of these data indicates that there is a made food contains 3% ethanol, equal to 650 µmoles per gram significant population (CU vs. UU ) by period (early vs. late) inof food.
teraction (P Ͻ .005). This result is due almost entirely to the significantly greater feeding rates of the CU-early larvae compared with the CU-late larvae (one-tailed test, P Ͻ .0004). The UU-late versus the UU-early, in contrast, show no significant ability affects viability and male mating success and fe- difference. The small insert shows the additional test of feeding male fecundity-through changes in adult size-in a frerates on just the CU populations, which also reveal large and quency dependent manner; for example, the fitness benesignificant differences (P Ͻ .001).
fits enjoyed by good competitors are greatest when they are rare (Mueller 1988a(Mueller , 1988b. It has been shown several times that competitive ability in Drosophila larvae is test, P Ͻ .03). This suggests that the differentiation of the CU-early and -late subpopulations is due to both the highly correlated with larval feeding rate (Burnet et al. 1977;Joshi and Mueller 1988). The larval feeding rates of CU-late larvae becoming more tolerant of ammonia and the CU-early larvae becoming less tolerant. Larvae were the four populations derived in figure 1 were measured and compared (fig. 5). These results show that the CU-also raised in standard food at high larval densities.
Again the CU-late larvae had a significantly higher rate of early larvae feed at a significantly higher rate than the CU-late larvae but that there is no difference between the survival than the CU-early population and no difference was observed between UU-early and late ( fig. 6). If these UU-early and UU-late larvae.
Egg-to-adult viability was also examined under four absolute viabilities are expressed as fractions relative to the CU-late viability, then the relative fitness of the CU-different conditions. Larvae from the four populations were raised at low densities but in three different food early larvae at high density (0.84) and in ammonia (0.84) is much less than their relative fitness at low density environments: standard food, standard food with ammonia (0.25 M), and standard food with urea (0.3 M). In all (0.96), even though there are significant differences between CU-early and -late populations in all three cases. three cases the CU-late larvae show a significantly higher viability than the CU-early larvae, while there are no sig-Nevertheless, the fitness advantage of the CU-late types is most pronounced under conditions of crowding and nificant differences between UU-early and UU-late ( fig.  6). On ammonia the CU-early larvae also have a signifi-high ammonia concentrations.
The experiments in figure 6 measured absolute viabil-cantly lower viability than the UU-early larvae (one-tailed crowded conditions. A crucial component of this polymorphism is the trade-off between feeding rates and absolute viability. Additional support for this trade-off comes from recent experiments in which the CU populations were cultured at reduced larval densities and experienced a significant decline in feeding rates relative to similar populations kept at high larval densities (Joshi and Mueller 1996). These observations further justify the important role that trade-offs play in the theory of lifehistory evolution (Stearns 1992).
While we have emphasized the temporal aspects of the environmental variation, there are similarities between the environmental decay in the CU populations and temporal variation. In the CU environments, only a portion of the total population experiences the early environment (by this we mean the completion of development in this time interval) and only a portion experiences the late environment. In standard models of temporal variation, the entire population would be assumed to experience each new environmental state.  Botella et al. (1985) found that there is a significant population (CU vs. UU ) by period (early urea and uric acid levels changed over time in relatively vs. late) interaction (P Ͻ .025). For each food treatment, the uncrowded cultures. Urea is an uncommon nitrogenous viability of the CU-late subpopulation was significantly greater that the CU-early subpopulation, and there were no significant waste product in insects, and uric acid production is usudifferences between the two UU subpopulations. At high denally associated with terrestriality (Cochran 1985). We sity there was also a significant population by period interaction could not detect uric acid, and urea levels were much (P Ͻ .025) with the CU-late subpopulation having a signifilower than ammonia ( fig. 3). The levels of urea and uric cantly greater viability than the CU-early subpopulation (oneacid measured by Botella et al. (1985) were no higher tailed test, P Ͻ .0008). than a few mmol/kg food. These values are similar in magnitude to the urea concentrations we measured under much more crowded conditions but were probably ity differences of early or late phenotypes only. Although competitive ability affects fitness through changes in via-not high enough to be toxic. In our populations, ammonia appears to be the primary nitrogenous waste product, bility, these viability effects can only be seen when good competitors are placed in competition with poor com-although we cannot exclude the possibility that it is generated microbially. petitors. The results in figure 6 show that the CU-early phenotype suffers a reduction in absolute viability rela-Ethanol in Drosophila cultures may evaporate or be converted to acetic acid by microbes (Hageman et al. tive to the CU-late phenotypes. 1990). In our crowded conditions, over 90% of the ethanol initially present in fresh foods disappears within 4 d,

Discussion
and insignificant quantities are present after 8 d ( fig. 4). The rate of disappearance is much slower in food held It appears that natural selection in crowded Drosophila cultures has led to a polymorphism that can be dissected without larvae or with uncrowded larvae (A. G. Gibbs, unpublished observations). The larvae may use ethanol along an axis of developmental times. Fast developing larvae have high feeding rates and reduced exposure to as a major energy source during early development (Geer et al. 1993), although microbial degradation may also the late part of the larval environment that is characterized by low levels of food and high levels of ammonia. occur.
Even under uncrowded conditions, other chemical Conversely, the more slowly developing larvae have higher absolute viability, especially under conditions of changes are likely. For example, changes in ammonia and acetic acid levels will affect the pH of the medium. The high levels of waste products and also under very