The genetic architecture of variation in adult life span was examined for a population of recombinant inbred lines, each of which had been crossed to both inbred parental strains from which the lines were derived, after emergence from both high and low larval density. QTL affecting life span were mapped within each sex and larval density treatment by linkage to highly polymorphic roo-transposable element markers, using a composite interval mapping method. We detected a total of six QTL affecting life span; the additive effects and degrees of dominance for all were highly sex- and larval environment-specific. There were significant epistatic interactions between five of the life span QTL, the effects of which also differed according to genetic background, sex, and larval density. Five additional QTL were identified that contributed to differences among lines in their sensitivity to variation in larval density. Further fine-scale mapping is necessary to determine whether candidate genes within the regions to which the QTL map are actually responsible for the observed variation in life span.
MOST eukaryotic organisms have limited life spans. Despite this near universal property, the potential life span of individuals varies a great deal among species (Finch 1990; Austad 1997; Gagnyet al. 1997; Lasset al. 1997; Martínez 1998; Vaupelet al. 1998) and populations (Austad 1996; Ricklefs 1998). A large number of studies have verified that much of this variation results from the interaction of genetic and environmental influences (Finch 1990; Rose 1991); however, the details of this interaction remain largely unknown. Understanding the genetic basis of life span variation and how the genetic components interact with environmental influences to limit life span is not only of practical importance to the medical community, it is also fundamental for understanding how the process of aging itself has evolved.
Two major, but not mutually exclusive, hypotheses have been proposed to explain the evolution of aging that ultimately determines the maximum attainable life span: mutation accumulation and antagonistic pleiotropy. Both theories are based on the assumption that the strength of selection declines with increasing age following the onset of reproduction (Hamilton 1966; Charlesworth 1994). The mutation accumulation hypothesis first proposed by P. J. Medawar in 1952 (as cited by Rose 1991; Charlesworth 1994) proposes that aging evolves from the build up of mutations in a population, which have no effect on fitness early in life, but have deleterious effects at later ages. These mutations remain in a population because the force of selection is weak or nonexistent on alleles expressed after reproduction and parental care has ceased. The antagonistic pleiotropy hypothesis (Williams 1957; Rose 1985) proposes that aging has evolved from the maintenance of alleles that enhance fitness at early ages, but have deleterious effects at later ages. Such alleles remain in a population because of their positive effects on fitness at early ages when selection is strongest. Evaluation of these hypotheses requires identifying the genes that contribute to variation in aging, determining their age- and environment-specific effects, and the pleiotropic effects of these genes on traits directly related to fitness (e.g., age-specific fecundity, age at maturity).
Recent studies have identified many genes involved in the aging process that all act to limit life span in many taxa. Candidate genes have been identified in Saccharomyces cerevisiae (Jazwinski 1996; Sinclairet al. 1998; Kirchmanet al. 1999), Caenorhabditis elegans (Friedman and Johnson 1988; Larsenet al. 1995; Gemset al. 1998; Hekimiet al. 1998; Tissenbaum and Ruvkun 1998; Taubet al. 1999), and Drosophila melanogaster (Luckinbillet al. 1990; Stearnset al. 1993; Orr and Sohal 1994; Dudas and Arking 1995; Hariet al. 1998; Linet al. 1998; Parkeset al. 1998). Most of these genes code for metabolic enzymes or are involved in stress and antioxidant defense responses.
While studies of candidate genes have begun to reveal many of the physiological mechanisms of aging and life span limitation, complete understanding of the evolution and variation in life span requires that we also study these correlated traits in a quantitative genetic context. One such approach involves studying genetic regions throughout the genome that act in concert to determine life span and evaluating their effects in a range of environments and different genetic backgrounds. This is necessary for two reasons. First, the effects of specific genes on aging and life span limitation are mediated by environmental variation (Kenyonet al. 1993; Gemset al. 1998). Second, there is increasing evidence that many genes that contribute to variation in life span have pleiotropic effects on other traits and interact epistatically with other genes to influence life span (Gemset al. 1998; Tissenbaum and Ruvkun 1998).
A promising approach to address this complex question is to use polymorphic molecular markers to map quantitative trait loci (QTL) that contribute to the variation in life span among genetically distinct lines. This procedure can be used to confirm and evaluate the effects of genetic variation at previously identified candidate genes, as well as identify new genetic regions for further study. In addition, this procedure can be used to identify sex- and environment-specific effects of different QTL in a range of environments and genetic backgrounds.
Recent studies in our laboratory using 98 recombinant inbred (RI) lines of D. melanogaster identified a number of QTL that contribute to the variation in life span. Interestingly, many of these QTL have sex- (Nuzhdinet al. 1997) and environment-specific effects (Vieiraet al. 2000). In addition, a subset of the QTL identified in both studies have antagonistic effects on life span in the different sexes and across different environments.
In this study, we extend our previous work on these lines to address three important questions. First, how are the effects of QTL that contribute to variation in life span influenced by the genetic background in which they occur? This question was addressed by monitoring the life span of virgin offspring from backcrosses of the 98 RI lines to the two parental strains used to initiate them. This design allows an estimation of the effects of QTL on life span in lines that are heterozygous over a proportion of the genome, a more “natural” genetic state for Drosophila. It also allows a comparison of heterozygous and homozygous effects, and hence degrees of dominance, of QTL affecting life span. Second, how are the effects of life span QTL influenced by variation in larval density? Larval density varies a great deal in natural populations and is known to affect life span (Miller and Thomas 1958; see review of Graves and Mueller 1993) as well as a number of traits directly related to fitness (e.g., age and size at eclosion; Miller and Thomas 1958; Barker and Podger 1970; Prout and McChesney 1985). Differences in larval density also affect the amount of heritable variation in life span that is expressed (e.g., Clare and Luckinbill 1985; Bucket al. 1993a). Thus, larval density is an important ecological variable with potentially important effects on the evolution of life span. Third, what is the genetic basis of the plastic response of life span to variation in larval density? Specifically, can we identify QTL that affect the sensitivity of life span to variation in larval density, and if so, are they the same as those that explain the variation in life span among the lines? Two general mechanisms have been proposed to explain the plastic response of quantitative traits to environmental variation (Viaet al. 1995), which can be examined using the QTL approach. The first is that alleles have varying effects on the phenotype in different environments (the “allelic sensitivity” hypothesis). Extending this hypothesis to the current study, we would predict that the QTL affecting the sensitivity of adult life span to variation in larval density should map to the same regions as those that explain the variation in the average life span among lines. Differential expression of alleles in these QTL regions across densities would explain the covariation between life span and larval density. The second possibility is that special regulatory genes respond to specific environmental cues to turn on or adjust the expression of structural genes that directly influence the phenotype (the “gene regulation” hypothesis). In this case, we would predict that the QTL regions that explain the variation among lines in the sensitivity of life span to larval density will be distinct from QTL that contribute to the variation in life span among the lines within a given density.
MATERIALS AND METHODS
D. melanogaster stocks: Two unrelated isogenic parental strains (Oregon, Lindsley and Zimm 1992; and 2b, Pasyukova and Nuzhdin 1993) and 98 RI lines derived from them were used in this experiment. The lines were initially constructed by S. Nuzhdin and E. Pasyukova (Nuzhdinet al. 1997) and a brief summary of their construction follows. The RI lines were initiated by first crossing the Oregon and 2b parental strains. F1 offspring from this cross were backcrossed to 2b, and their progeny were randomly mated for four generations to increase the amount of recombination between genetic regions. The backcross to 2b was carried out because the 2b strain is generally less vigorous than the Oregon strain and the backcross helped ensure that the proportion of 2b and Oregon alleles were equally represented within each RI line. The 98 RI lines were subsequently created by carrying out single pair matings of full sibs for 25 generations. RI lines were maintained by mass transfer of large numbers of individuals each generation to minimize the rate of fixation of new mutations.
Focal individuals for this experiment were the offspring of crosses of females of each of the 98 RI lines to males of each of the two inbred parental strains, Oregon (O) and 2b (B). RI females were mated to males of each parental strain so that QTL for life span on the X chromosome could be identified. In the resulting lines, all three genotypes (OO, OB, and BB) are expected at life span QTL, with OO and OB genotypes occurring in the offspring of the cross to Oregon, and BB and OB occurring in the offspring of the cross to 2b. This crossing scheme is, in essence, a modified North Carolina Design III (NC III; Comstock and Robinson 1948, 1952), with the exception that the RI lines in this experiment represent the F2 generation of the NC III design.
Experimental treatments and life span assay: To minimize the influence of nongenetic maternal effects on life span, the density of all lines was controlled for two generations prior to the start of the experiment by restricting egg laying of 20 pairs of flies per vial for a 7-day period. It should be noted here that this does not completely remove the possibility that some maternal effects may have contributed to life span differences among lines. Genetic differences in unmeasured traits (e.g., fecundity or larval viability) among lines may have produced differences in larval density among stocks in the parental and grandparental generations even though adult density was controlled. We are currently examining other traits in these lines to examine this possibility. Also, nongenetic paternal effects may have been present because the density of the Oregon and 2b lines was not controlled prior to allowing males of each stock to mate with females of the RI lines. However, in this case the paternal effects would not contribute to the genetic variance in life span among lines, and any contribution of the paternal effects to life span would contribute to the error variance of the analyses.
High and low larval densities in each cross were created by allowing either 2 or 20 mated females to oviposit for 24 hr in vials containing standard cornmeal-agar-molasses medium. In addition, a small amount of live yeast was added to each laying vial to stimulate oviposition. Two replicate breeding vials per cross and density were used, and virgin males and females were collected from each replicate over a 24-hr period from the day at which the first individuals from a vial began to emerge. Individuals of each sex from each replicate were kept separate during this procedure. Virgin males and females from each replicate breeding vial were housed separately in two replicate vials containing 5 ml standard medium for monitoring the life span (five single-sexed individuals per vial). No additional yeast was added to the vials used for determination of life span. Flies were maintained in a constant temperature room at 25° and transferred to new vials once a week. Although we had two replicate vials for each cross, sex, and density treatment combination, a few flies escaped during the course of the experiment, resulting in an unbalanced design. A total of 7733 flies were used in this experiment. Adult life span was monitored every other day. For logistic reasons, the density assays were initiated sequentially, 3 mo apart.
Additional crosses within and between the parental strains were carried out at each density to compare the effects of larval density on the life span of each parental strain and their F1 hybrids (Oregon♂♂ × 2b♂♂, 2b♂♂ × Oregon♂♂). Two replicate vials containing five single-sex individuals for each genotype, sex, and density treatment combination were used. The life spans of individuals within each treatment were monitored concurrently with individuals from the RI line crosses described above.
Statistical analyses: In the first analysis, life span data from the crosses within and between the isogenic parental strains were analyzed by analysis of variance (ANOVA). We modeled the fixed effects of genotype (the two homozygote classes, Oregon and 2b, and two classes of F1 heterozygotes, Oregon ♂♂ × 2b♂♂ and 2b♂♂ × Oregon♂♂), sex, and density on life span. Inspection of the residuals from the analysis on untransformed data indicated that no transformation was necessary to satisfy the assumptions of ANOVA. Post hoc comparisons of mean values were carried out using Tukey's “honestly significant difference test” (Zar 1984).
The second set of analyses was carried out on the life span of the offpsring from crosses between females of each RI line to males of the two isogenic lines. The experimental design allowed us to examine the genetic and environmental influences on life span variation of the RI line crosses at several levels. Data were analyzed hierarchically in five different ANOVAs, so that the variance attributable to the main effects and their interactions could be examined in subsets of the data as well as in the full data set.
A random effects model was first used to identify variance attributable to differences among lines within each cross (to Oregon or 2b), sex (male or female), and larval density (high or low) treatment. In the first of four mixed-model ANOVAs, the data were separated by cross and density treatments and the variance in life span due to line, sex, and their interaction was determined. Variance components from this analysis were used to evaluate the significance of any genotype (i.e., line within a parental cross)-by-sex interaction (GSI) and to calculate rGS, the correlation of the mean male and female life span of each line when crossed to the different parental strains and in the two densities. rGS was calculated as cov12/(σL1 σL2) (Robertson 1959), where cov12 is the covariance among line means between males and females within a cross and larval density treatment, and σL1 and σL2 are the square roots of the variance components for the RI line term from the reducedmodel analysis within cross, density, and sex. The second mixed-model analysis estimated the effects of parental cross, RI line, and the cross-by-line interaction within each sex and density treatment. The interaction term from this analysis reveals how the genes that contribute to differences among RI lines are influenced by the genetic background in which they occur. The third mixed model included the effects of cross, line, density, and their interactions on life span within each sex. This analysis provided a measure of how life span was affected by the interaction between the genetic background and larval density. The fourth model investigated all main effects and their interactions in the entire data set. In each mixed-model analysis, the parental cross, sex, and density were treated as fixed effects and all others as random effects.
Because life span was measured on individuals sharing a vial within each replicate, vial effects were removed by including a term for replicate vials in each ANOVA model. Unlike the ANOVA, which compared the life span of the two parental lines and their F1 hybrids, all life span data from the RI line crosses to the parental lines were transformed to natural logs prior to analysis to satisfy assumptions of ANOVA (Sokal and Rohlf 1981). ANOVAs were carried out using the SAS GLM and VARCOMP procedures (Version 6.12; Sas Institute 1988).
To compare the amount of genetic variance expressed among lines in the different crosses and larval density treatments, the coefficient of genetic variation was calculated separately for each sex within each cross and density treatment. The coefficient of variation was calculated as , where is the average life span among lines (Houle 1992; Vieiraet al. 2000) and VL,U is the variance component attributable to line effects from ANOVA on untransformed life span values.
Molecular marker map: Cytological insertion sites (Lindsley and Zimm 1992) of the roo-transposable element were used as molecular markers to determine the genotype of the Oregon, 2b, and RI lines (as described by Nuzhdinet al. 1997). Five additional informative markers were added to the 76 markers used by Nuzhdin et al. (1997) and Vieira et al. (2000) for the QTL mapping analyses. The positions of these five additional sites and their estimated cytological positions (map positions) are as follows: 16D (1-128.10), 46A (2-154.82), 77E (4-164.91), 78D (4-166.06), and 85A (4-170.66). In total, 81 informative markers were used, which have an average spacing between markers of 3.2 cM (Nuzhdinet al. 1997). A complete list of the markers and their estimated positions is available from the authors upon request. Map positions (d in centimorgans) of the markers on the map were estimated from the observed recombination frequencies (r) between pairs of markers using the Kosambi map function: 100dM = 0.25 ln[(1 + 2r)/(1 − 2r)], where dM is the distance between markers in Morgans. Cytological positions 1–20, 21–60, and 61–100 are located on chromosomes 1, 2, and 3, respectively, on the standard D. melanogaster genome map (Lindsley and Zimm 1992). Chromosome II was divided into two linkage groups because the recombination frequency between markers 50F and 57C was >0.5 (Nuzhdinet al. 1997).
QTL mapping: QTL contributing to the variation among lines in mean life span and in the sensitivity of mean life span to larval density were identified by composite interval mapping (Zeng 1994) using the randomly mated second generation intercross (RF2) option for analysis in QTL Cartographer (version 1.13). This procedure tests the hypothesis that an interval between adjacent markers contains a QTL affecting the trait, while simultaneously controlling for effects of linked QTL on the same chromosome but outside the test interval. Markers on which the QTL analyses were conditioned were chosen in each separate analysis by a forward selection-backward elimination stepwise regression. Because the results of the analyses can depend on the size of the “conditioning window” used around the tested interval (Nuzhdinet al. 1997), we used a range of window sizes (2, 5, 10, 15, 20, and 30 cM) to evaluate this effect in each analysis. On the basis of the results of this screening procedure, a window size of 10 cM was chosen because QTL identified with this window size were common to all analyses using other window sizes and so is the most conservative choice. The likelihood-ratio (LR) test statistic is −2 ln(L0/L1), where L0/L1 is the ratio of the likelihood under the null hypothesis (there is no QTL in the test interval) to the alternative (there is a QTL in the test interval). The test statistic at a genomic location is distributed as χ2 with 2 d.f. under the null hypothesis and was evaluated every centimorgan.
The significance level for each analysis was determined by permutation. Empirical distributions of LR test statistics under the null hypothesis of no association between test intervals and trait values were obtained for each analysis by randomly permuting the trait data and calculating the maximum LR statistic across all intervals for each permutation. LR statistics calculated from the original data that were exceeded by the permutation maximum LR statistics <50 times are significant at α = 0.05 under the null hypothesis (Churchill and Doerge 1994; Doerge and Churchill 1996).
The life span data used in the QTL analyses were the line means of males and females from each cross (to Oregon or 2b). Line means were not transformed to natural logs prior to QTL analyses because the untransformed data approximated a normal distribution. Four QTL analyses were carried out so that differences among lines crossed to each parental strain were compared. Analyses were carried out separately for each sex within each density treatment because software is not currently available to perform QTL analyses across different environments.
To identify QTL contributing to the variance among lines in the sensitivity of life span to high and low larval density, a sensitivity score (S) was calculated separately for males and females of each line within a given cross as (Falconer 1990). Here, S is the sensitivity phenotype, and and are the average life spans of line i (where i ranges from 1 to 98) within a cross after exposure to high and low larval densities, respectively. D is the difference between the average life span of all lines when reared in high and low larval density (within each cross and sex). QTL analyses were subsequently carried out on the sensitivity scores separately for males and females of each line as described above.
QTL effects on life span: Estimates of the additive and dominance effects of each life span QTL within each sex and density are provided by the QTL mapping analyses. However, these analyses do not estimate the genotypic effects of these QTL on life span across densities and sexes (Fryet al. 1998). To estimate the interaction effects of QTL across sex and density, the genotypes of markers closest to each significant QTL peak were used as categorical variables in ANOVA. A separate analysis was carried out for each significant marker. Each model included the main effects of all significant marker genotypes, sex, larval density, and all possible interactions between the marker of interest and sex and density. Examination of the residuals from these analyses indicated that transformation was not necessary to satisfy the assumptions of ANOVA.
Epistatic effects of life span QTL: Pairwise epistatic interactions were tested in ANOVA. We tested for epistasis only between nonadjacent life span QTL that were identified as significant on the basis of the permutation tests. Although a more sophisticated method has been developed to identify epistasis among QTL (Kaoet al. 1999), software for this analysis is not readily available. Tests were carried out within each cross separately because, although each life span QTL could be either homozygous or heterozygous, both classes of homozygotes (OO and BB) did not occur together within a line because of the breeding design used. For each interaction of interest, the genotype of each marker (OO, OB, or BB) closest to each significant QTL peak was used to evaluate the significance of the marker interactions on life span. For each test, a model was fitted that included the main effects of density, sex, and all life span QTL identified by the QTL mapping analyses, the two-way interaction term for the two focal marker genotypes, and all possible interactions between the two focal markers, sex, and density. These three- and four-way interactions were used to investigate sex- and environment-specific interactions between QTL. The least-squared mean life span of each line was used as the dependent variable. Examination of the residuals from these analyses indicated that transformation was not necessary to satisfy the assumptions of ANOVA.
For all ANOVAs in which we tested for epistasis, we also report the results after application of the sequential Bonferroni procedure (Holm 1979; Rice 1989). The initial Bonferroni correction value used for testing the significance of each interaction was based on the number of interactions tested within each cross.
Effects of breeding density on larval density: Although the density treatment was analyzed as a categorical variable, the actual number of larvae that were competing in each vial is unknown and so the effects of larval density on life span should be interpreted with caution. To estimate the relative differences in larval densities created by the different parental densities, we counted all individuals that eclosed during the first 5 days of emergence from a subset of the lines (109 lines from the low density cross to Oregon and 2b, and 181 lines from the high density cross to Oregon and 2b). We also obtained dry weights on the first five males and females that emerged from these vials to determine if the density treatments affected body mass at eclosion. On average, the high adult breeding density treatment produced 30% more individuals during this period than low breeding density treatments (Low density, eclosing individuals; high density, eclosing individuals). Visual inspection of the vials indicated that the majority of individuals from the low density treatment had eclosed by the 5th day. This was not true for the high density treatment. On average, newly eclosed male and female flies from the low density vials were 15 and 16% heavier, respectively, than those that emerged from the high density vials. On the basis of these data, it is clear that our treatments were effective in producing different larval densities. The consequence of the uncontrolled density is that the effects of the two treatments are less distinct than they would have been if we had been able to count eggs or larvae. Analytically, variation in larval density within treatments creates a bias against detecting life span differences between treatments, making any conclusion regarding the density effects conservative.
Life span phenotypes of the parental strains and F1 parental hybrids: The average life span of the parental strains (Oregon♂ × Oregon♀ = 37.0 ± 2 days, 2b♂ × 2b♀ = 37.5 ± 2 days) was significantly shorter than that of their F1 hybrids (Oregon♂ × 2b♀ = 60.6 ± 2 days, 2b♂ × Oregon♀ = 55.3 ± 2 days; F3127 = 37.53, P < 0.0001). No other main effects (sex and density) or their interactions significantly affected life span. There was a nearly significant sex-by-density interaction (F1127 = 3.37, P = 0.07), with the trend being for increased larval density to increase the average male life span but have little if any effect on female life span (Table 1). It should be noted that the average male and female life spans of the Oregon and 2b lines in this experiment are in general agreement with those reported by Vieira et al. (2000), which provides good evidence of the repeatability of life span measurements under these conditions.
Life span phenotypes and genetic variation in life span in RI line crosses: In the cross to Oregon, the average male life span ranged from 29.8 to 73.7 days among lines, and the average female life span ranged from 8.6 to 99.8 days. In the cross to 2b, the average male life span ranged from 24.7 to 73.6 days, and the average female life span ranged from 26.6 to 79.1 days.
Analyses within each cross, sex, and density using random effects models identified significant differences in life span among lines, but differences were cross-, density-, and sex-specific (Table 2). In the cross to Oregon, genetic variation for male and female life span was only significant in the high larval density treatment. In the cross to 2b, significant genetic variation for male and female life span was also found in high density; in contrast to results of the cross to Oregon, genetic variation for male life span was also found in the low density treatment. The proportion of the phenotypic variation explained by genetic differences among lines [see the VL/(VL + VR) column of Table 2] varied with cross, sex, and density and ranged from ~0 (RI × Oregon cross, females from low density) to 21% (RI × Oregon cross, females from high density). This is also reflected in the coefficient of genetic variation, CVG, which was generally larger in the high density treatments.
Genotype-by-sex and -environment interactions: The first set of mixed-model analyses (within each cross and density) identified significant effects of sex, line, and line-by-sex interactions that were cross- and density-specific (Table 3). In the cross to Oregon, none of the main effects were significant when considered alone. In the cross to 2b, the main effect of sex was significant, and females lived ~20% longer than males in both densities. In this cross, line effects were also significant but only in high density. Line-by-sex interactions were significant in both crosses, but only in the high density environment (Table 3, Figure 1). In high density, the correlation between the sexes within lines, rGS, differed in sign and magnitude depending on the cross (Table 2). In the cross to Oregon, the correlation was slightly, but significantly, negative; in the cross to 2b the correlation was significant and positive.
In the second set of mixed-model analyses (within each sex and density treatment), the main effect of cross and the cross-by-line effects were sex- and density-specific (Table 4). For males, the only significant effect on life span was due to the cross-by-line interaction, which occurred in both density treatments. This interaction explained 9 and 11% of the total variation in life span in the low and high density treatments, respectively. In females, the direction of the cross significantly affected life span in both densities. The female offspring of lines crossed to 2b lived 14–16% longer than those crossed to Oregon. There was also a significant cross-by-line interaction for female life span in high density that explained 15% of the variation. This interaction was not significant in low density.
In the third set of mixed-model analyses (within each sex), most of the variation in life span explained by the model was attributable to interactions between the main effects (Table 5). In males, the only significant effects on life span were due to the cross-by-line and the line-by-density interactions. For females, the main effect of cross was significant, as was the cross-by-line-by-density interaction.
In the full model, most of the variation in life span explained by the model resulted from differences between sexes and the cross-by-sex interaction terms (Table 6). The cross-by-line interaction was also significant.
QTL analyses for life span variation among lines: Six QTL contributed to the variation in life span among lines, but different QTL were identified as important in each sex and larval density (Table 7, Figures 2 and 3). On chromosome II, one QTL (35B–38E, 43A) was found in two treatment combinations, males and females from low larval density. The remaining two QTL on chromosome II (46C–49D and 50D) were found only in males from the high density treatment. On chromosome III, one QTL (67D–68B, C) was detected in two of the four analyses (females in both densities). The remaining QTL on chromosome III (71E, 72A–77A, and 69D–87B) were found in each of two treatment combinations, males and females in low density. These QTL are close to, or share, regions of overlap in the different analyses and so it is not entirely clear whether these represent the same or different QTL. It should be noted that two other regions that did not quite exceed the significance threshold might harbor QTL affecting male life span. One region is on chromosome I, near cytological position 7D, and the other on chromosome II, near 34EF (Figure 2).
Within each sex and density treatment, the QTL exhibited a range of additive and dominance effects (Table 7). Strong overdominance was observed at one QTL (35B–43A), but this effect occurred only in males from low density. Three other QTL exhibited weaker overdominance and these effects were also sex- and environment-specific (67D–68C, females from high density; 69D–87B, females from low density; and 71E, males from low density). The effects of the remaining QTL ranged from partial to complete dominance, depending on the sex and density in which they were found.
QTL for sensitivity of life span to larval density: Five QTL affected the sensitivity of life span to variation in larval density, but none were common to both sexes (Table 8, Figure 4). Male sensitivity to density was affected by two QTL on chromosome III. Interestingly, these were not the same QTL that contributed to the among-line variation in males within a density treatment. Three sensitivity QTL were found for females, only one of which (67D–68C) had previously been identified as a region contributing to variation in female life span among lines. The other two were on chromosome I (4F–5D and 6E–7E), where no significant life span QTL were identified.
The genotypic effects at these sensitivity QTL ranged from dominant to overdominant (Table 8). Overdominance was observed at one QTL region (6E–7E), and the remaining QTL exhibited varying degrees of partial dominance.
Density and sex dependence of effects of QTL genotype on life span: The genotypic effects on life span at many life span QTL depended on sex and larval density, particularly when lines were heterozygous or homozygous for the 2b strain at a particular QTL. QTL genotype-by-sex interactions occurred at markers 48D (F2671 = 3.12, P = 0.04; Figure 5A), 68B (F2671 = 3.18, P = 0.04; Figure 5B), and 76B (F2671 = 4.62, P = 0.03; Figure 5C). QTL genotype-by-larval-density interactions were significant for markers 38E (F2700 = 18.31, P < 0.0001; Figure 6A), 48D (F2740 = 21.03, P < 0.0001; Figure 6B), 50D (F2764 = 32.40, P < 0.0001; Figure 6C), 68B (F2756 = 44.80, P < 0.0001; Figure 6D), and 76B (F2740 = 31.93, P < 0.0001; Figure 6E). In no case was there a significant three-way interaction between marker genotype, sex, and density.
In general, most of the interactions with density resulted from changes in the rank order of the life span of marker genotypes across densities. This occurred in only one marker-by-sex interaction (Figure 5A). Of particular interest, genotypes that exhibited overdominance in one sex or density often exhibited additive or partially dominant effects on life span in the contrasting sex or environment. This intriguing result suggests that the degree of dominance of a particular QTL genotype is density- and also possibly sex-dependent.
Epistasis between life span QTL: Five life span QTL exhibited significant epistatic interactions (Table 9) but their effects on life span varied, depending on sex (Figure 7A) and larval density (Figure 7, B and C). In the single significant marker-by-marker-by-sex interaction (Figure 7A), the genotypic effect of marker 48D on male life span depended on the genotype at marker 76B. No such interaction was seen in females. Figure 7, B and C, illustrates the range of effects that other marker interactions had on life span across densities. It is important to note that in some cases the lines that were heterozygous in both interacting regions were not the longest lived, a situation that might be expected under associative overdominance (Lynch and Walsh 1998). By far, most epistatic interactions occurred in the cross to 2b. Only three analyses identified significant epistatic interactions in the cross to Oregon, and these involved only three QTL. In only one case was there a significant marker-by-marker-by-sex interaction, suggesting that the effects of most genetic interactions on life span are consistent across sexes. After the sequential Bonferroni procedure, three of the five two-way interactions were still significant, but only two of the seven three-way interactions remained significant. These all occurred in the cross to 2b and involve interactions between the two locus marker genotypes and larval density (Table 9).
Genetic and environmental effects on life span: Many genetic and environmental influences can act independently to influence the age at which an organism dies. Our results suggest that the interaction of these influences can also be especially important in determining the adult life span of Drosophila. Using offspring from backcrosses between a panel of 98 RI lines and their two isogenic parental strains, we identified significant genetic variation for adult life span, but this variation was largely influenced by the direction of the genetic cross, sex, and the larval density experienced. In general, genetic differences among RI lines were evident for both males and females after exposure to high larval densities; after exposure to low larval density, genetic differences were only apparent in the cross to the 2b parental genotype and only in males. The genetic correlation between male and female life span was highly sensitive to the genetic background and larval density experienced. This sex- and environment-dependent expression of genetic variation in life span is consistent with previous work on these lines (Nuzhdinet al. 1997; Vieiraet al. 2000) and those of other studies (e.g., Buck et al. 1993a,b). In addition, our finding that genetic variation for life span is more prevalent in high than in low density is consistent with the observation that selection for increased life span is more effective when flies are reared in high larval densities (e.g., Clare and Luckinbill 1985; but see Zwaanet al. 1995 for a contrasting result). The total variation in life span was comparable in the two density treatments, suggesting that this association between larval density and genetic variance is a genetic property, not a statistical one. The genetic basis of this pattern is unknown, but an intriguing possibility is that the conditions of high density invoke a stress response, and genetic differentiation among lines in their general stress response may be reflected by differences in life span. Mild thermal stress has been shown to increase adult life span in D. melanogaster (Khazaeliet al. 1997; Vieiraet al. 2000) and yeast (Shamaet al. 1998). Higher larval density generally leads to longer life span in D. melanogaster (Miller and Thomas 1958; Zwaanet al. 1991; Bucket al. 1993a), although in our experiment the response to density depended on sex and genotype (Table 5). If a stress is imposed by high larval density (e.g., from high concentration of waste products or reduced oxygen levels) and genetic covariation between the stress response and life span exists, then genetically based differences in life span might be revealed as a consequence of their different stress response capabilities.
Additive effects of life span QTL: Six QTL were found to contribute to differences in life span among lines but no QTL were detected in all sex and larval density combinations. The QTL identified within each sex and larval density exhibited a range of additive and dominance effects on life span, and the relative magnitude of these effects depended on the sex in which they were detected and the larval density experienced. Even when the same QTL were identified in high and low larval densities or both sexes, their effects on life span were sex- and environment-dependent (Table 7). The additive effects of the 2b genes at the QTL in region 67D–68C increased female life span compared to the effects of Oregon in low density, but reduced female life span relative to Oregon in high density. This indication of antagonistic pleiotropy was also seen between sexes for the QTL at 69D–87B, in which the additive effects of the 2b genes increased male life span relative to the effects of Oregon, but had negative relative effects on female life span. The additive genetic effects were not always antagonistic, however. The genetic effects of 2b genes at the QTL at region 35B–43A reduced the life span of both sexes relative to Oregon in low density, but the negative effects of the 2b genes affected female life span more than that of males.
Two previous QTL studies of life span using the same RI lines, but in a homozygous state, also found many of the same QTL identified in our study. Nuzhdin et al. (1997) identified five QTL affecting life span in a single environment, four of which were also identified in our study (QTL 1, 4, 5, and 6 as designated in Table 7). In Nuzhdin et al. (1997), QTL 1 affected only female life span, while 4, 5, and 6 were specific for males. Vieira et al. (2000) screened for QTL affecting life span in five environments: starvation, after mild heat shock, and in 14°, 25°, and 29°. Their study identified five of the six QTL found in this study (QTL 1, 2, 4, 5, and 6). QTL 1 of our study overlapped two QTL identified by Vieira et al.'s study. These QTL had significant effects on the life span of at least one sex in each environment. QTL 2 was specific for males in the starvation environment. QTL 4, 5, and 6 had significant effects on the life span of each sex in at least one environment. In all cases, the effects of individual QTL on life span depended on sex and environment. Only one region, QTL 3, was not identified by Nuzhdin et al. (1997) or Vieira et al. (2000) and so is unique to this study. Taken together, these independent studies have consistently verified the importance of particular regions in explaining genetic variation in life span in these lines. A striking feature of these QTL is the inconsistency of their effects on life span between sexes and among environments.
Degrees of dominance of life span QTL: An important result revealed by the use of the backcross design is that the degree of dominance of life span QTL is not a fixed property. Changes in the degree of dominance across environments (which may include the disparate physiological environments of males and females) have a sound theoretical basis (Wright 1977) and have also been documented empirically (Wright 1977; Bourguetet al. 1996). While our QTL study cannot accurately predict allelic effects on life span within loci (our QTL regions undoubtedly contain many genes), it is plausible that the genotypic effects we observed represent the true allelic effects. Wright's theory predicts that the degree of dominance is a function of the relationship between gene activity and the physiological response of the reaction of its product with those of other genes and with the developmental and environmental conditions (Wright 1977, pg. 500). Environmental differences that affect any step in the process of converting genotype to phenotype, such as the different physiological environment of males and females or differences in larval environments induced by variation in density, should in theory alter the relative dominance relationship between alleles.
Epistasis between life span QTL: A critical assumption of Wright's theory is that the phenotype of the trait in question is the product of a number of interacting loci. Our analyses of pairwise interactions among the life span QTL clearly demonstrate a large degree of epistasis with dramatic effects on life span. That these interactions are often sensitive to larval density, sex, and the genetic background is consistent with the basic premise of Wright's theory on dominance discussed above. Other studies have also demonstrated the importance of gene interactions affecting life span. Buck et al. (1993b) found that genes on chromosomes I and II interact to affect the expression of genes on chromosome III to produce an extended life span phenotype in a line of D. melanogaster selected for increased life span. A large number of studies have also revealed strong epistatic and pleiotropic effects of genes involved in an insulin-like signaling pathway, which affect life span, metabolism, and fertility in C. elegans (see Tissenbaum and Ruvkun 1998 and references therein; Gemset al. 1998; Hsin and Kenyon 1999).
QTL for the differential sensitivity of life span to larval density: We found five QTL that affected the sensitivity of life span to variation in larval density. Only one of these also mapped to a region that contributed to the variation in life span among the lines (67D–68C). While none of the other sensitivity QTL coincided with significant life span QTL, careful examination of the position of the likelihood peaks from all sets of analyses indicated congruence between the position of the sensitivity QTL and potential life span QTL that did not exceed the permutation threshold for significance. Thus our results provide weak support for the “allelic sensitivity” hypothesis of the genetic basis of plasticity (Viaet al. 1995). This conclusion remains tentative, however, and it may well be that these regions harbor some of the regulatory loci that control the expression of genes that directly influence life span.
Life span QTL and candidate genes: The QTL identified in our study contain many genes that have been previously identified as candidate genes affecting life span (Table 6). Adh and Pgm are important metabolic enzymes (Davidet al. 1976; Clark 1990; Heinstraet al. 1991) that may play a role in aging and life span limitation. Adh has been shown to have higher expression levels in lines of D. melanogaster selected for long life (Arkinget al. 1993). In a separate experiment, Deckert-Cruz et al. (1997) examined allozymic differentiation at the Pgm locus between lines of D. melanogaster selected for postponed senescence and unselected control lines derived from the same base population. They found that control and selected lines were fixed for different allozymes at the Pgm locus and that the longevity of laboratory fly populations could be directly predicted by the allelic frequency at this locus.
Insulin degrading metalloproteinase (Ide) is the primary enzyme responsible for the degradation of insulin (Garciaet al. 1988; Duckworthet al. 1994; Bennettet al. 1997). Because insulin is a critical hormone controlling glucose metabolism (Wolfe 1995) and the insulin-like receptor pathway regulates adult life span and reproduction in C. elegans (Tissenbaum and Ruvkun 1998), variation at Ide could presumably affect intracellular insulin levels and so indirectly influence life span through its effect on glucose metabolism.
Sod and Cat are two genes involved in the elimination of reactive oxygen species (ROS), toxic byproducts of metabolism that are primarily generated from the mitochondrial respiratory chain (Wallace and Melov 1998). Oxidative damage from ROS is thought to be a major factor in aging (Harman 1956) and life span limitation, and several studies support this hypothesis (e.g., Orr and Sohal 1994; Dudas and Arking 1995; Deckert-Cruzet al. 1997; Hariet al. 1998; Parkeset al. 1998).
The remaining two candidate genes are EF1α and Hsp70. EF1α is an essential protein for protein synthesis. Stearns et al. (1993) demonstrated that mated females with enhanced expression of EF1α lived longer but laid fewer eggs early in life than did control females. However, enhanced expression of this gene had no effect on virgin male and female life span, which argues against its role in contributing to the variation in life span in our study. Hsp70 is a heat shock protein that has a variety of functions including autoregulation of the heat shock response and signal transduction (Mirimotoet al. 1997). Hsp70 has also been implicated as directly influencing life span on the basis of experiments in which Hsp70 was overexpressed in flies containing extra copies of this gene (Tataret al. 1997).
Future fine-mapping and quantitative complementation studies (e.g., Longet al. 1996; Mackay and Fry 1996) are necessary to determine what genetic loci correspond to the life span QTL in the Oregon and 2b strains. The next step will be to assess whether allelic variation at these loci is associated with phenotypic variation in life span in natural populations, using linkage disequilibrium mapping (Mackay and Langley 1990; Laiet al. 1994; Longet al. 1998). Until these studies are complete, it is not possible to assess whether the observed variation between the Oregon and 2b strains is attributable to inbreeding depression caused by rare deleterious alleles that contribute little to naturally occurring variation or by alleles at intermediate frequency maintained in nature by a balancing selection mechanism.
Conclusions: Our results have documented genetically based differences in life span among heterozygous lines, but these differences depended on the genetic background, sex, and larval density. A QTL mapping analysis indicated that much of this variation results from sex- and density-dependent effects of QTL genotypes on life span and the nature of genetic interactions among life span QTL. Many of the genetic regions identified in our QTL analysis contain candidate genes with known influences on life span and whose relative effects are modulated by interactions with modifier or regulatory loci in other parts of the genome (e.g., Graf and Ayala 1986; Arkinget al. 1993).
Our finding that the effects of QTL genotypes on life span depend on genetic background, sex, and larval environment is surprising; if this is a general property of genes affecting all life-history traits, this has important implications for the maintenance of genetic variation in these traits. Life-history traits are the primary components of fitness and so (with the exception of life span) are thought to be under strong directional selection (Roff 1992; Stearns 1992). Life-history theory predicts that little, if any, genetic variation should persist for these traits because natural selection should fix the genes that produce the optimal phenotype in a given environment. However, a large amount of genetic variation exists for these traits in natural populations and several hypotheses have been proposed to explain this variation (Roff 1992; Stearns 1992). None of these hypotheses explicitly incorporate the types of genetic effects on life-history traits that we observed. If such genetic behavior is common, it is easy to see how a large amount of genetic variation could be maintained. Except when mutations are unconditionally deleterious, natural selection will have difficulty culling genetic variation at quantitative trait loci. This is because the fitness contribution of different alleles is context-dependent, and so their effects will be small when averaged between sexes and over all environments and genotypes. Whether or not the alleles that contribute to the variation in life-history traits really represent moving targets for natural selection remains to be seen.
If the extreme environmental sensitivity of the QTL effects on life span can be extrapolated to true genetic effects, this may prove problematic for those studies attempting to identify single nucleotide polymorphisms (SNPs) associated with complex traits, e.g., multifactorial human diseases (e.g., Cargillet al. 1999; Halushkaet al. 1999). The detection of such associations assumes that the allelic effects on a trait are constant (Risch and Merikangas 1996). If the environmental and genetic contexts are important for the expression of such alleles (i.e., the main effects of these genes are small), the detection of such loci may be more difficult than is currently appreciated.
We thank C. Dilda, T. D'Souza, B. Hackett, S. Heinsohn, F. Lawrence, and R. Anholt for help with the flies. Thanks to M. De Luca, R. Fuller, D. Houle, J. Travis, B. Weir, and Z-B. Zeng for discussion of various aspects of this experiment. We also thank A. Clarke, C. Dilda, L. Horth, and two anonymous reviewers for comments on the manuscript. This work was supported by National Institutes of Health NRSA grant GM18818-03 to J.L., and GM 45146 and GM 45344 to T.F.C.M. This is a publication of the W. M. Keck Program for Behavioral Biology.
Communicating editor: A. G. Clark
- Received December 22, 1999.
- Accepted April 17, 2000.
- Copyright © 2000 by the Genetics Society of America