It is a basic corollary of evolutionary theory that populations exposed to a novel environment are expected to undergo adaptation to that environment, where such adaptation can be defined as increasing mean fitness in the new environment due to genetic change. It is known that this effect is not a mathematical certainty; certain types of population genetic models can produce decreasing fitness in the course of selection (e.g. Pollak, 1978; Lewontin, 1979; Charlesworth, 1980, 1994; Pollak, 1995), and some empirical data support this possibility (e.g. Paquin & Adams, 1983). However, it is a common expectation that selection models will usually have the property of increasing mean fitness, in the absence of severe frequency-dependence, density- dependence, epistasis, etc. (cf. Nagylaki, 1977; Endler, 1986; Williams, 1992).
Associated with adaptation is an expectation that additive genetic variance for fitness will fall, following the lines of Fisher’s Fundamental Theorem (e.g. Fisher, 1930; Nagylaki, 1977, 1992; Ewens, 1979; Williams, 1992; Falconer & Mackay, 1996). In the simplest possible scenario, as selection proceeds the additive genetic variance for fitness is supposed to approach zero and adaptation then tends to slow or cease (cf. Maynard Smith, 1989). This inference has been generalized from fitness to its components (e.g. Mousseau & Roff, 1987; Roff & Mousseau, 1987; Falconer & Mackay, 1996), but this generalization is not theoretically certain (discussed further below).
It has also been proposed that the additive genetic correlations between fitness-components should tend to fall as adaptation proceeds, owing to convergence on equilibria of balancing selection involving antagonistic pleiotropy (e.g. Caspari, 1950; Wallace, 1968; Rose, 1982). This expectation is derived from the simple point that alleles having uniformly beneficial effects in the new environment should increase in frequency to a point close to fixation, while alleles that exhibit antagonistic effects in the new environment might meet the conditions for convergence on a genetically polymorphic equilibrium. This type of equilibrium may thereby preserve additive genetic variance for individual fitness-components ( Rose, 1982, 1985), contrary to the pattern discussed above. This does not mean that all additive genetic correlations between fitness components are expected to be negative at equilibrium, owing to the possible role of genes involved in multiple pleiotropic effects (some positive and some negative), or other genetic associations ( Charlesworth, 1990; Holloway et al., 1993 ).
In any case, the expectation of a directional trend of exhaustion of additive genetic variance of fitness as well as its components, and a shift of additive genetic correlations from positive towards negative values remains an open question. Empirical data on these issues are still unclear, whether concerning additive genetic variances of life-history traits (e.g. Holloway et al., 1993 ; Carriere & Roff, 1995; Crnokrak & Roff, 1995; Weber, 1996; Campbell, 1997; Holloway et al., 1997 ) or additive genetic correlations (e.g. Holloway et al., 1993 ; Carriere & Roff, 1995; Roff, 1996; Campbell, 1997). These studies are mostly based on either inferences from the analysis of selective responses or comparative studies between populations. A more adequate test should be a temporal analysis of the genetic architecture at the population level, as we propose to do here.
A possible outcome of the evolutionary process when mechanisms involving positive pleiotropic effects coexist with mechanisms involving negative pleiotropy is the occurrence of a biphasic evolutionary pattern for some traits, i.e. an increase in mean values followed by a depression in later generations since establishment in the novel environment. For example, antagonistic pleiotropy between early fecundity and survival might lead to a later drop in longevity after some generations of culture at early ages only.
It is essential to consider the source of genetic variance allowing the adaptation to a novel environment, i.e. genotype–environment interactions. One scenario is that, in a novel environment, new genetic variability is expressed owing to different effects of alleles that had the same effect in the previous environment (see de Jong, 1990a; Stearns, 1992). In this case the additive genetic variance of fitness is higher in the novel environment compared with the previous one, if the population was close to selective equilibrium in the latter, and some empirical studies fit this expectation (e.g. Holloway et al., 1990 ; Silva & Dykhuizen, 1993; Travisano et al., 1995 ; Guntrip et al., 1997 ). Some of the novel genetic variance is plausibly expressed in loci with positive pleiotropic effects. The most likely outcome is positive, or at least less negative, additive genetic correlations between main fitness components ( Rose & Service, 1985; Reznick et al., 1986 ; Clark, 1987). Some empirical studies support this possibility ( Service & Rose, 1985; Holloway et al., 1990 ; Guntrip et al., 1997 ). The consequence of this scenario is that, other things being equal, the additive genetic variance of fitness and its components, as well as the additive genetic correlations, will drop throughout the adaptation to the novel environment.
A different scenario, which leads to ‘unexpected’ evolutionary patterns, is a genetic trade-off between the novel environment and the previous one, specifically if the genotypes with higher fitness in one environment have the lowest in the other (see Via, 1987; de Jong, 1990b; Falconer, 1990; Stearns et al., 1991 ; Leroi et al., 1994a , b; Roff, 1997 ; but see Hawthorne, 1997). Under this scenario, the temporal change of both additive genetic variances and correlations throughout the adaptation to the novel environment may be highly unpredictable (see Charlesworth, 1990; Falconer & Mackay, 1996). The same will obviously apply to the evolutionary patterns of mean values of life history traits.
Nevertheless, the expectation of trends leading to an overall decrease in additive genetic variance and an increase in the expression of genetic trade-offs between life-history traits remains the simplest working hypothesis concerning evolution in a novel environment.
Adaptation to the laboratory has been mostly studied at the phenotypic level, in relation to the changes in mean values of life-history traits. Furthermore, few generations have been analysed, which does not allow a characterization of the trajectory (e.g. Souza et al., 1988 ; Harris & Okamoto, 1991; Economopoulos, 1992; Thomas, 1993; Cacoyianni et al., 1995 ).
Experimental evolution in Drosophila has been carried out by selection in a new culture regime. Such experiments show adaptation to novel environments involving high-density conditions (e.g. Mueller & Ayala, 1981; Mueller et al., 1993 ), altered age at reproduction (e.g. Luckinbill et al., 1984 ; Rose, 1984), starvation (e.g. Rose et al., 1992 ), temperature (e.g. Huey et al., 1991 ), etc. Most of these studies are also only at the phenotypic level (but see Hutchinson et al., 1991 ; Engstrom et al., 1992 ). Temporal changes of the genetic architecture of an evolving Drosophila population have been analysed more in terms of changes in gene frequencies than of quantitative genetic parameters (e.g. Buzzati-Traverso, 1955; McKenzie et al., 1992 , 1994; but see Tantawy & El-Helw, 1966; Tantawy & Tayel, 1970).
Adaptation to the laboratory environment has practically not been investigated in Drosophila. The few cases that have been published do not characterize the evolutionary trajectories of genetic parameters, being based on few generations or a comparative approach ( Dobzhansky et al., 1964 ; Tantawy & El-Helw, 1970; Pascual et al., 1990 ) or involving parallel selective regimes that hamper the interpretation ( Tantawy & El-Helw, 1966). More empirical studies about the tempo and mode of the evolution of the mean values and, most of all, genetic parameters of life-history traits as a population adapts to a novel environment are thus needed.
We endeavoured to test whether the predictions referred to above applied to the evolution of Drosophila subobscura Collin newly sampled from the wild. Two of these samples were studied, the first in more detail than the second. Our findings support the notion of adaptation during laboratory culture, but we are unable to test other expectations concerning the adaptive process.
Materials and methods
Source of laboratory samples
Both laboratory cultures of Drosophila subobscura were sampled from a pine wood near the village of Sintra, Portugal. Each sample was obtained from several collections, over 2 days, from early morning till late afternoon. About 140 females and 50 males were collected in the sample giving rise to the first laboratory culture, in January 1990, and about 230 females and 100 males were collected for the second, in March 1992.
The flies were maintained at about 18 °C, with a 12-h L : 12-h D photoperiod. Culture medium was similar to that described by David (1959), being composed of agar, corn meal, dead brewer’s yeast, charcoal colouring and nipagine. Flies were kept in 10 × 2-cm vials, with an adult density of around 50 individuals per vial. Larval densities were 60–80 per vial. The reproductive regime involved discrete generations, with egg collection at around 2 weeks of the imagoes’ life, close to peak fecundity. Maximum developmental time allowed was 21 days, some slight truncation selection for rate of development being involved. CO2 anaesthesia was used. The culture vials were maintained in the same incubators during the entire period of this study, except during manipulation, when they were handled at room temperature (normally around 22 °C). Adult population sizes were in general higher than 1000 individuals, never dropping below 400.
Flies were transferred to laying vials containing freshly prepared medium every day. Total egg-laying for periods of 1 week was used in the data analysis (female’s age being estimated using the day of emergence of the imago as reference). Female fecundity records were kept during their entire lives in the half-sib study (see below), which gave additional data on total and mean fecundity, as well as other traits (age of last egg laying, number of eggs laid during the last week of life, number of eggs of the last day of egg laying, number of days without egg laying before death). Females that did not lay a single egg during their entire lifespan were not used in the half-sib study (at most four females having been discarded from each sample).
Age of first reproduction
This trait was estimated as the number of days before first egg laying, counted since the emergence of the imago. In the comparison of new and established populations (from here on called W and B, respectively), we assigned the value 14 for age of first reproduction of the few females (maximum four, in generation 5) that did not lay eggs during the 2 weeks of life covered by the analysis. This way these females (which were mainly of population W) were not discarded from the analysis of this trait.
These data were collected in the half-sib study, from the same flies as those used in fecundity assays. Longevity was estimated as the number of days between the emergence of the female and the day in which the female was recorded as dead (this last day included). Females that escaped during the study or died from handling were excluded from all the analysis (never more than 4% of each sample). Females dying during the first 2 weeks of life were also excluded (always less than 5% of each sample, their longevities being deviant from the overall distribution of values).
Time of development
As the females of the half-sib samples did not emerge all in the same day, we estimated this trait as the day of formation of each pair, relative to the time of emergence of the first imagoes. For instance, if a female emerged on the second day after the start of emergence in the whole sample, that female was assigned the value 2 for this trait.
The thorax of the females of the half-sib study was measured as the distance between the most anterior point of the mesonotum and the most posterior point of the scutelum. An arbitrary scale was used, allowing the discrimination of around 15 size-classes.
In the second study, after the first 2 weeks of adult life (during which fecundity data were collected), mated pairs were placed in vials with a non-nutritive medium (plain agar) and deaths were recorded every 6 h.
The first laboratory population was subjected to a half-sib analysis every 4–6 generations, from the second generation to the 29th. To obtain each sample, we did an extra egg collection, applying the same procedures as for the maintenance of the population. From the first emergence of the imagoes, males and females were kept apart until the end of the fourth day of emergence. At that time, harems of one male and six females each were formed. These were kept for 2 weeks, after which the dams were placed in individual vials and an egg collection over 48 h took place. Finally, we removed, when required, some of the eggs from the vials to minimize differences in larval densities (around 15–25 larvae per vial, in general).
Development of the half-sib families took place in the same incubator. Flies from the first day of emergence were in general discarded, formation of pairs for assay starting on the following day and continuing for three consecutive days. CO2 anaesthesia was only used in the formation of mating pairs. Matings were made at random over the whole collection of families. Each sample was formed of 67 or 68 families (with family sizes ranging between two and six half-sibs each). As families were not randomized between racks, a higher block effect was defined to extract common environmental effects. Flies were transferred to fresh vials daily. Females remained with males until 30 days of life, by which time all males were removed (until that age dead males were replaced by other males of the same age). This procedure was necessary because males had a shorter life than females in these samples. A preliminary study of this population did not indicate a significant effect of removal of males on female egg laying.
Comparison of new and established samples
After 24 generations had elapsed from the formation of the first population, a new sample was obtained, as described above, in order to compare this new laboratory population with the established one. For generations 1–14 of the second population, a variety of characters were compared in mated pairs collected from the two laboratory cultures, especially fecundities during the first 2 weeks of life, as well as starvation resistance of males and females. In each assay, the samples of the two populations were randomized between racks in the incubator. Daily transfers and egg countings took place during the first 2 weeks of adult life, after which the mating pairs were placed in plain agar to assay starvation resistance. Finally, when the most recent culture was in its 47th generation, and the oldest was in its 71st generation, another comparison was made, using the same traits.
All data analysis was performed using IBM-compatible computers running SYSTAT, STATISTICA and EXCEL. No data interpolation was used in unbalanced ANOVA. Half-sib analysis involved a two-level nested ANOVA, with a higher block level (corresponding to the racks), a sire level (corresponding to the families within racks) and a dam (or error) level (corresponding to the half-sibs within each family). The formulas used to estimate additive genetic variances and correlations were those given by Kempthorne (1957). Sampling errors of additive genetic estimations were calculated using the Hammond & Nicholas (1972) approximate formulas. Estimations of genetic correlations involved only the females for which data were available for both traits. All regressions were Type I least-squares linear regressions.
Overview of the results
The results of the sequential sib analysis reveal no obvious pattern of means, genetic variances or genetic correlations during 29 generations of domestication to laboratory conditions. It is interesting to note that some striking patterns might have been inferred if our sib analyses had been stopped at generation 15 ( Fig. 2). Specifically, over just 15 generations, observed changes in genetic correlations meet the expectation of a progressive decrease in the frequency of alleles having positive pleiotropic effects (17 out of 18 linear regressions present a negative slope in terms of fitness, which is a highly significant departure from expectations of a binomial distribution).
Nevertheless, an analysis involving more generations reveals that this pattern was not sustained over the entire period of observation. It is an open question whether this loss of the generation 15 patterns is a result of genuine nonlinear effects of laboratory adaptation – see below – or whether it is merely an effect of statistical fluctuation. The latter would presumably have been revealed if there had been replication of evolving lines, as well as controls that had already adapted to the laboratory. It is nevertheless important to note that the average statistical power of the regressions was enough to detect a higher number of linear trends.
In any case, the results over 29 generations clearly do not fit the simplest expectations of the evolutionary theory discussed in the Introduction. That laboratory adaptation does nonetheless occur is revealed by the striking convergence of fecundity traits of B and W lines in the second group of experiments.
Quantitative genetics of adaptation
A laboratory is just another environment in the evolutionary history of a population, where a general pattern of adaptation might be expected.
In the particular example afforded by our experiments, it was possible to infer adaptation when a population already further along in the evolutionary process was compared with one just beginning the process. It was not, however, possible to detect any feature of adaptation, whether phenotypic or quantitative genetic, in a single population undergoing the process on its own. This was the case despite the repeated use of sib analysis, using multiple characters. These results suggest that the detection of adaptation will hinge critically on the availability of controls and replicates.
To our knowledge, almost no studies have tracked the evolutionary trajectories of the genetic architecture of life history traits during extensive adaptation (see Introduction). Had such a detailed analysis been developed in the selective experiments already done, we would suggest that it is open to question whether they would have been able to detect the expected patterns of change in genetic variation that we were unable to detect here.
In any case, population genetics theory already indicates that typical quantitative genetics expectations may not be met. Variances might increase under selection if beneficial alleles start at low frequencies and then take on intermediate frequencies during selection. As mentioned in the introduction, this may be expected if there is a genetic trade-off between the novel and the previous environment (e.g. see Leroi et al., 1994b ). This, together with the possible coexistence of genetic mechanisms affecting several life-history traits with a multitude of pleiotropic effects, may render any evolutionary pattern possible, and short-term analysis of little predictive power (see Charlesworth, 1990). For now, we offer no firm conclusions on the actual dynamics of genetic variation during adaptation.
We wish to express our gratitude to André Coelho and Henrique Teotónio, companion soldiers of this unfinished war. Also a word of thanks to Adam Chippindale, Ana Cruz, Armand Leroi, Ricardo Azevedo and Ted Nusbaum who, though not directly in the battlefield, helped with their inspiring (s)words. Henrique Teotónio and Ted Nusbaum helped also with their comments on the manuscript. This project was partially financed by JNICT no. PMCT/C/CEN/647/90 and C.R. had a 1-year grant from the same Institution. M.M. received financial support from FCG and from FLAD for some of the travels to U.C.I.