Correlated responses to artificial body size selection in growth, development, phenotypic plasticity and juvenile viability in yellow dung flies

Authors


Wolf U. Blanckenhorn, Zoologisches Museum, Universität Zürich, Winterthurerstrasse 190, CH-8057 Zürich, Switzerland.
Tel.: +41 44 635.4755; fax: +41 1 44 635.4780;
e-mail: wolfman@zoolmus.unizh.ch

Abstract

Most life history traits are positively influenced by body size, whereas disadvantages of large body size are poorly documented. To investigate presumed intrinsic costs of large size in the yellow dung fly (Scathophaga stercoraria; Diptera: Scathophagidae), we established two replicates each of three body size laboratory selection lines (small, control and large; selection on males only), and subjected flies of the resulting extended body size range to various abiotic stresses. Response to selection was symmetrical in the small and large lines (realized h2 = 0.16–0.18). After 24 generations of selection body size had changed by roughly 10%. Female size showed a correlated response to selection on male size, whereas sexual size dimorphism did not change. Development time also showed a correlated response as, similar to food limited flies, small line flies emerged earlier at smaller body size. At the lowest larval food limit possible, flies of all lines emerged at the same small body size after roughly the same development time; so overall phenotypic plasticity in body size and development time strongly increased following selection. Juvenile mortality increased markedly when food was extremely limited, large line flies showing highest mortality. Winter frost disproportionately killed large (line) flies because of their longer development times. Mortality at high temperatures was high but size-selective effects were inconsistent. In all environments the larger males suffered more. Initial growth rate was higher for males and at unlimited food. Small line individuals of both sexes grew slowest at unlimited larval food but fastest at limited larval food, suggesting a physiological cost of fast growth. Overall, extension of the natural body size range by artificial selection revealed some otherwise cryptic intrinsic juvenile viability costs of large size, mediated by longer development or faster growth, but only in stressful environments.

Introduction

It is generally agreed that fecundity and sexual selection are the main forces favouring large body size in many species (Darwin, 1859; Roff, 1992). In most organisms (female) fecundity correlates positively with body size (Shine, 1988; Honek, 1993), and sexual selection typically favours large male size, either by male–male competition or female choice (Andersson, 1994), although small males are sometimes also favoured in species where agility plays a role (Ghiselin, 1974; e.g. Steele & Partridge, 1988; Blanckenhorn et al., 1995). Whereas these selective advantages of large body size are often obvious and strong, the counterbalancing selective factors, most often relating to viability selection, remain cryptic and far less well documented in general (reviewed by Blanckenhorn, 2000, 2005).

There are a number of reasons why life history costs of large body size are difficult to detect in nature. One is that presumed trade-offs only become visible in stressful environments because individuals can invest maximally in all traits when environmental conditions are good (Ojanen et al., 1979; van Noordwijk & de Jong, 1986; Alatalo et al., 1990; Schluter et al., 1991; Rowe & Houle, 1996). It is therefore imperative to experimentally investigate individuals under stress. Another reason is that in many species large individuals may die early during development, i.e. before they can be investigated, and hence are not part of the natural size distribution (Blanckenhorn, 2000). A third reason is that if the function linking fitness to body size is mainly flat with steep decreases in fitness occurring only at the very ends of the size distribution, then the disadvantages of large body size can remain largely undetected. Expanding the natural body size range of flies by artificial selection can overcome the latter two fundamental problems.

Body size (at maturity) is a highly plastic trait that strongly depends on environmental conditions such as temperature, food availability, photoperiod or predators encountered during development. Typically organisms become larger at lower temperatures and unlimited food (Atkinson, 1994; Atkinson & Sibly, 1997; de Moed et al., 1999; Gillooly et al., 2001; Davidowitz et al., 2003). To become larger, animals must either grow faster or develop for a longer time, and these trade-offs have been central to life history theory in the context of body size evolution (Roff, 1980, 1986, 1992; Sibly et al., 1985; Stearns & Koella, 1986; Kozlowski, 1992; Stearns, 1992; Abrams et al., 1996; Arendt, 1997). If development is prolonged, juveniles necessarily suffer a higher cumulative risk of death because of predation (e.g. Loader & Damman, 1991; Werner & Anholt, 1993) or parasitism (Benrey & Denno, 1997). Organisms developing in ephemeral habitats such as temporary ponds, water holes, dung, carcasses or rotting fruits may risk dying through starvation or habitat drying (Newman, 1992; Juliano & Stoffregen, 1994). Growing faster, on the other hand, is energetically, ecologically or physiologically costly. Faster growing individuals may have to forage and therefore move more, making them more conspicuous to predators or parasites (Werner & Anholt, 1993; Bernays, 1997; Munch & Conover, 2003). A greater amount of energy used for growth, especially when food is limited, reduces energy allocated to reproduction, maintenance, storage or repair mechanisms (Cody, 1966; Partridge & Sibly, 1991; Arendt, 1997; Wikelski & Ricklefs, 2001; Zera & Harshman, 2001; Koella & Boëte, 2002). Additionally, faster growth may carry inherent costs of rapid mass accumulation or inefficient energy use (Higgins & Rankin, 2001), resulting in higher mortality during or after food shortage (Dingle, 1992; Gotthard et al., 1994; Blanckenhorn, 1998). All these trade-offs have been invoked in the past to play a key role in limiting body size (Blanckenhorn, 2000).

The yellow dung fly Scathophaga stercoraria (Diptera: Scathophagidae) is a well-studied model organism in terms of behaviour, ecology and evolution. It abundantly inhabits the cooler north-temperate regions of the Old and the New Worlds (Gorodkov, 1984), and is stressed by high temperatures (Ward & Simmons, 1990; Blanckenhorn, 1998; Blanckenhorn et al., 2001). The flies mate and lay their eggs on fresh cow dung (Parker, 1970a). Larvae feed on the dung, which they thereby deplete, pupate there or in the ground underneath, and emerge depending on temperature and food availability approximately 2.5–5 weeks later. The frost-sensitive larvae have to reach the pupal stage to survive the winter (Blanckenhorn, 1998). Sexual size dimorphism (SSD) is male biased (Borgia, 1982; Simmons & Ward, 1991), the atypical situation for insects, probably because of strong sexual selection on male size (Jann et al., 2000). Body size is heritable (Simmons & Ward, 1991; Blanckenhorn, 2002; Kraushaar & Blanckenhorn, 2002) and strongly correlates with clutch size in females (Parker, 1970b; Jann et al., 2000; Kraushaar & Blanckenhorn, 2002) and mating success in males (Borgia, 1982; Parker & Simmons, 1994; Sigurjónsdóttir & Snorrason, 1995; Otronen, 1996; Jann et al., 2000; Kraushaar & Blanckenhorn, 2002, Blanckenhorn et al., 2003). Thus, extensive previous studies have revealed strong net selection for large body size with very little countervailing selection (summarized in Table 1). These numbers reflect long-term averages over several seasons and populations, thereby incorporating the possibility of oscillating or fluctuating stabilizing selection (Istock, 1981; Gibbs & Grant, 1987). This situation is typical for many animal species (Blanckenhorn, 2000; Kingsolver & Pfennig, 2004) and puzzling given the assumption of predominantly balancing selection in nature (Schluter et al., 1991).

Table 1.   Previous mean ± 95 % CI field estimates of linear sexual (pairing success), fecundity and viability selection differentials or gradients on body size (hind tibia length) for the yellow dung fly over several seasons in multiple Swiss populations, plus corresponding juvenile viability estimates from this study (averaged over the two dung treatments in all cases).
 FemalesMalesReference (number of estimates)
  1. Previous selection estimates were hardly ever negative, whereas the ones from the current study almost always are.

Pairing success+0.372 ± 0.130Jann et al., 2000 (2)
Kraushaar & Blanckenhorn, 2002 (1)
Blanckenhorn et al., 2003 (4)
Fecundity+0.205 ± 0.035+0.009 ± 0.020Jann et al., 2000 (1)
Kraushaar & Blanckenhorn, 2002 (1)
Juvenile viability+0.225 ± 0.177−0.234 ± 0.349Unpubl. data fromBlanckenhorn, 1998 (3)
+0.014 ± 0.028−0.007 ± 0.002Larval food limitation experiment 1 (2)
−0.041 ± 0.063−0.088 ± 0.020Extreme larval food limitation experiment 2 (2)
−0.128 ± 0.219−0.202 ± 0.236Hot temperature experiment 3 (4)
−0.311 ± 0.317−0.237 ± 0.128Winter frost experiment 4 (2)
Adult viability+0.122 ± 0.014+0.005 ± 0.065Burkhard et al., 2002 (2)

The central aim of our study was to reveal intrinsic but apparently cryptic factors selecting against large body size in yellow dung flies by expanding the natural size range via artificial selection. We here document the basic results of our artificial selection, as well as correlated responses in development time, growth rate and phenotypic plasticity in growth and development. We consequently investigate potential underlying genetic body size limitations such as pleiotropic effects, genetic trade-offs or intrinsic mortality costs, which are otherwise difficult or impossible to detect in nature. As we selected only on males, we expected a correlated response in females because of a strong genetic correlation between the sexes (Blanckenhorn, 1998, 2002), and potentially a correlated change in SSD (Reeve & Fairbairn, 1996). We also expected a correlated genetic response in development time, thus directly demonstrating a trade-off (Roff, 1980, 1986), although previous investigations suggest the size/development time correlation is low in yellow dung flies (Blanckenhorn, 1998, 1999). Alternatively or additionally, we expected a correlated change in growth rate, at least in some environments, as flies can achieve larger sizes also by growing faster. To directly investigate viability costs of growing large, we subjected flies of the resulting extended genetic (the selection lines) and phenotypic (environmentally induced) body size range to various abiotic stresses augmenting juvenile (i.e. egg, larval and pupal) mortality: (1) limited and (2) extremely limited food, (3) hot (25 °C) temperature, and (4) winter frost. Although we have some previous (phenotypic) evidence for several of these mortality factors in yellow dung flies, we here attempted to demonstrate stronger, genetic and intrinsic life history trade-offs by using selection lines with an expanded body size range, thus circumventing some of the above-mentioned fundamental problems in detecting counter-selection on body size.

Methods

Selection procedures

To start the selection lines, 120 copulating pairs of S. stercoraria were randomly collected on four different days in April and May 2000 from cow pats in Fehraltorf, near Zürich, Switzerland. Females were brought to the laboratory and provided with a smear of dung into which they could lay their eggs. These eggs were used to start two replicate lines, each from at least 50 different females. Twenty-five eggs from each clutch were transferred in 50 ml plastic containers containing 80 g of cow dung (= larval food), and the offspring were raised at 20 °C and 60 % humidity. For each replicate, 50 males and 50 females from as many families as possible were randomly assigned to three selection lines: a small (S), a large (L) and a control (C) body size line. Within each line, 50 pairs were formed randomly, and selection started on the offspring of the F2 generation. Beginning with the F2, every generation head widths of 150 males – if possible three from each family – were measured, and the 50 smallest (S) or largest (L) or 50 randomly chosen (C) males were allowed to reproduce with 50 randomly chosen females of the same line (head width is a practical surrogate of overall size, as all body components correlate strongly and no particular trait was found to be the target of selection in the field; Blanckenhorn et al., 2003). If females did not lay after copulation, they had another opportunity to lay 4 days later or the corresponding males were given the possibility to copulate with another female, so that always 50 clutches produced the next generation for each line. To reduce the number of flies generated per generation and to keep the selection coefficient constant, from generation 12 onwards only 75 instead of 150 males were measured for each line, and 25 instead of 50 families were used to produce the following generation. To produce the next generation, 20–25 eggs of each clutch were transferred into 50-ml plastic containers with 80 g of dung (more than 2 g dung/larva can be considered unlimited larval food conditions: Amano, 1983). The containers were always kept at a constant 20 °C and 60 % humidity until offspring emerged after 19–22 days. Adult flies were kept singly in 100-ml vials until sexually mature, at 20 °C, 60 % humidity and at 12-h photoperiod with water, sugar and Drosophila melanogaster ad libitum as prey. Head widths of mothers of the next generation were measured after egg laying.

Selection was performed for 21 generations. At generation 21, flies from the two replicates within each selection line were crossed to control for potential inbreeding effects. Selection was further continued with these crossed S, L and C lines until generation 24; thereafter, the flies were propagated without selection for two more generations.

Realized heritabilities were estimated for the replicate small and large lines as the slope of the linear regression of the cumulative selective response on the cumulative selection differential (Falconer, 1989). The regression was forced through the origin as all three lines stemmed from the same population. The selection differential was assessed in each generation for each line as the difference between the mean head width of the selected males and the mean head width of all measured males of the respective line. To control for temporal variation caused by dung quality (which could not be avoided), the cumulative response was calculated for each generation as the difference in mean head width between the control line and the small or large line. SSD was estimated each generation as one minus the mean size of males divided by the mean size of females (Lovich & Gibbons, 1992).

Juvenile mortality and phenotypic plasticity in body size and development time in response to food limitation

Experiment 1: larval food limitation

At generation 11, in each of the two replicates and for each of the three selection lines (S, C and L), clutches of 30 females were split and assigned to two different larval treatments: 15 eggs were transferred into 80 g of cow dung (unlimited larval food) and 15 eggs of the same clutch into 5 g of dung (limited food: Amano, 1983). The containers with the larvae were kept in the laboratory at 20 °C and 60 % humidity. When emergence started, containers were checked daily, thus scoring egg-to-adult development time, and the head widths of the emerging offspring were measured under a binocular microscope at 16× magnification.

Experiment 2: extreme larval food limitation

The results of experiment 1 suggested that dung flies can survive food conditions even lower than 5 g of dung per 15 eggs. To assess mortality at the lowest limit (i.e. the highest larval food stress) possible, we repeated experiment 1, at generation 16 of selection, with even less dung: instead of 15 eggs, 20 larvae were transferred into 5 g of dung, and 15 larvae into 80 g of dung. We used larvae this time to avoid losses because of egg mortality. To assess egg mortality, egg numbers were counted and the number of unhatched eggs scored 36 h later.

To exclude potential viability reductions because of inbreeding depression of the selected lines, we repeated this experiment with flies stemming from a cross between the two replicates of each selection line. Generation 16 flies of one replicate were mated with flies of the other replicate within selection lines. The offspring were raised and some were used as parents of the larvae used in experiment 2.

Experiment 3: hot temperature and limited larval food

This experiment was only performed with the crossed replicates (cf. experiment 2) at generation 17. Clutches of females of all three lines were split among the two temperatures 20 and 25 °C. For half the females, 20 larvae were transferred into each of two containers (one for each temperature) with limited dung (5 g); for the other half of the females, 15 larvae were transferred into each of two containers with unlimited (80 g) dung. For each selection line, 25 of these replicate pairs were produced, resulting in a total of 75 limited and 75 unlimited dung containers at 20 and 25 °C for all three lines. Emergence rate and body size were recorded as before. To distinguish between larval and pupal mortality, limited dung containers additionally were checked for dead pupae after all flies had emerged. It was, however, not possible to find the pupae in the 80 g containers.

Experiment 4: winter frost and limited larval food

Again, only crossed replicates were used for this experiment. In early December 2002, at generation 20, flies of both replicates were crossed and their clutches split: 20 larvae were transferred into 5 g of dung and 15 larvae into 80 g of dung. Dung containers with the larvae were first kept for 2 days at 12 °C and then half submerged into big boxes filled with soil to give the larvae and pupae some protection against the frost. S. stercoraria pupae overwinter in the soil under or next to the dung pats. The boxes were moved outside, under a roof together with a temperature recorder, and left there over winter until emergence was scored in spring. Females did not lay synchronously, as small females laid first and large females laid later. Further, the onset of winter frost is not precisely predictable. Therefore, larvae from all lines were moved outside on several days, and this date was added as a covariate in the analysis. Thirty small line containers were moved outdoors on 6 December and six more on 16 December; 9, 11 and 10 control line containers were moved outdoors on 7, 8 and 13 December, and 4, 7 and 9 large line containers on 7, 8 and 14 December respectively. We ended up with 20 pairs of containers for the large line, 30 pairs for the control line and 36 pairs for the small line. Containers were checked occasionally during winter and more regularly in spring when emergence started. Again, emergence rate and body size were recorded.

Data analysis

In general, families with no emerged flies under good and bad larval conditions were excluded from the analyses. Emergence rate and mean body size were calculated for each of the full sib families (i.e. per family container), separately for the treatments (temperature or amount of dung) and sexes (assuming equal sex ratios at laying). Emergence rates were arc-sine square-root transformed. Data were analysed using repeated measures anovas with sex and larval treatment of the emerged flies as repeated and the selection lines and replicates as crossed factors. In the extreme food limitation experiment 2, we included the crossed replicate as a third replicate in the analysis, following the logic that this produces another, partially independent genotype (analogous to offspring from parents); however, the results were qualitatively similar when analysing only the original two replicates (as for experiment 1). In some treatment groups in the winter frost experiment 4, only single flies of one sex or none at all emerged; therefore, body size could not be analysed in this fashion.

To compare the juvenile viability results of this study to our previous results presented in Table 1, we calculated univariate selection differentials using standard methods (Arnold & Wade, 1984a, b). In all cases, we regressed the standardized proportion of emerged adults of a given clutch or family (family emergence/overall emergence) on the z-scored mean body size of that family, separately for the sexes and the two dung (food) treatments (as well as the two temperature treatments in the temperature experiment 3), but over all three body size selection lines combined (the genetic body size gradient). In case of 100 % mortality for a particular family, we used the overall mean body size for that particular treatment combination and selection line as a best substitute estimate.

Larval growth trajectories

At generation 24, larval growth rate of the three (crossed) selection lines was measured at unlimited and very limited dung (i.e. at the lowest limit of phenotypic plasticity as in the extremely limited food experiment 2 above). Twenty-four hours after laying the hatching larvae were measured as described below and moved into dung. From each line larvae of 20 different families were moved either singly (unlimited larval food) or together with larvae of the other 19 families (limited larval food) into small plastic containers (2 × 2 × 2 cm3) filled with 5 g of cow dung. Each day one larva of each of the 20 different families was measured at limited and unlimited larval food, so each day 20 larvae of each line and treatment were measured. Larval length and width were measured under a binocular microscope. Larval volume was then estimated using the formula for a cylinder. Additionally, larvae were weighed from day 3 onwards. Larvae that crawled out of their dung container and pupated there were excluded. After measurement, all larvae were moved singly into new 5-g dung containers, and the sex of the emerging adults was noted and adult head width measured.

To compare the growth trajectories among treatment combinations, we subdivided them in two parts: the initial exponential growth phase from days 1 to 5, and the asymptotic phase beginning with day 7 (purposely omitting day 6 marking a break point; see Results). To investigate differences in the initial growth phase using ancova, we analysed cube-root-transformed (i.e. linearized) larval body volume as a function of age with selection line, sex and larval food level as fixed factors. To investigate treatment differences in the asymptote using anova, we analogously analysed cube-root-transformed larval body volume with selection line, sex and larval food level as fixed factors by simply using all data from days 7 to 11 (thus ignoring age here).

Results

Selection response, realized heritability and correlated response in size dimorphism

Body size responded to artificial selection equally in both directions (Fig. 1). Mean head width remained relatively constant in both replicates of the control lines. In comparison with the original population, body size had increased by 11 % in the large and decreased by 9.5 % in the small line at generation 24.

Figure 1.

 Mean head width of males (black) and females (white) of the three selection lines small (circles), control (squares), large (triangles) for both replicates (a) and (b) over the artificial selection period.

The realized heritabilities, estimated by the slope of the linear regression of the cumulative selective response on the cumulative selection differential, were h2 = 0.171 ± 0.007 (SE) and h2 = 0.155 ± 0.007 for the two small line and h2 = 0.185 ± 0.008 and 0.162 ± 0.010 for the two large line replicates. The response to selection was symmetrical as the slopes of the small and the large lines did not differ significantly (t-test; replicate 1: t40 = 0.468, P = 0.642; replicate 2: t40 = 0.314, P = 0.755).

Even though we only selected on male body size, female size showed a correlated response, verifying a realized genetic correlation between male and female size (cf. Blanckenhorn, 2002; Fig. 1). From generation 16 onwards, females of the large line became on average larger than the males of the small line (Fig. 1), and at generation 24 large line females were nearly as large (mean head width 2.51 ± 0.013 mm, SE) as control (i.e. field) males (2.55 ± 0.012 mm). SSD did not change significantly in response to selection in the various lines (ancova with line and replicate as fixed factors and generation as covariate: line × generation interaction F1,84 = 2.16, P = 0.146), i.e. did not decrease in the small lines (r = 0.154, P = 0.504 for replicate 1; r = −0.179, P = 0.438 for replicate 2) or increase in the large lines (r = −0.251, P = 0.272 for replicate 1; r = −0.365, P = 0.104 for replicate 2).

Juvenile mortality and phenotypic plasticity in body size and development time in response to food limitation

Experiments 1 & 2: (extreme) larval food limitation

Flies of the three selection lines differed in body size in the food limitation experiments 1 and 2 (Fig. 2; line effect in Tables 2 and 3), as could be expected because artificial selection worked (Fig. 1). As female size showed a correlated response this was also true for females (sex effect in Tables 2 and 3 demonstrating the normal size dimorphism). Development time also showed a correlated response to selection, as large line flies emerged from unlimited dung 1–2 days later than small line flies (line effect in Tables 2 and 3; Fig. 2).

Figure 2.

 Bivariate plot of the overall mean (based on family means ± SE; if SE does not show it is smaller than the symbol) body size (head width) and development time for males (black) and females (white) of the large (triangles), control (squares) and small selection lines (circles). (a) and (b) show the two replicates at limited and unlimited (upper points) larval food (=dung) at generation 11 (larval food limitation experiment 1); (c), (d) and (e) show the two replicates plus the crossed replicate at extremely limited and unlimited (upper points) larval food at generation 16 (extreme larval food limitation experiment 2).

Table 2.   Repeated measures anova for family mean body size (head width), egg-to-adult development time and juvenile mortality (emergence rate) with sex and dung (=food) as repeated and selection line and replicate as crossed factors (larval food limitation experiment 1).
 Head widthDevelopment timeEmergence rate
d.f.MSFPMSFPMSFP
  1. Significant effects are in bold.

Dung118.34905.17<0.001180.3899.24<0.00167.0916.47<0.001
Dung × replicate10.115.370.0221.901.040.3096.841.680.197
Dung × line20.209.98<0.0014.792.640.0763.370.830.440
Dung × replicate × line20.062.840.0625.172.840.0620.070.020.983
Within family error (dung)1220.02  1.82  4.07  
Sex15.651331.8<0.001336.23525.59<0.0010.000.000.999
Sex × replicate10.000.850.3590.210.330.5690.470.090.761
Sex × line20.024.340.0151.111.740.1804.840.950.388
Sex × replicate × line20.000.460.6350.230.350.7031.080.210.809
Within family error (sex)1220.00  0.64  5.08  
Dung × sex11.04248.51<0.00148.93101.82<0.0012.640.540.464
Dung × sex × replicate10.001.050.3080.140.290.59410.492.140.146
Dung × sex × line20.012.020.1373.747.780.0011.390.280.754
Dung × sex × replicate × line20.011.110.3323.066.360.0022.960.600.548
Within family error (dung × sex)1220.00  0.48  4.90  
Replicate10.3419.92<0.00197.5539.33<0.00117.161.910.170
Line21.6595.93<0.00164.7926.12<0.0013.150.350.706
Replicate × line20.052.790.0666.512.630.07631.143.460.034
Between family error1220.02  2.48  9.00  
Table 3.   Repeated measures anova for family mean body size (head width), egg-to-adult development time and juvenile survival (adult emergence rate) with sex and dung (=food) as repeated and selection line and replicate (including replicate cross) as crossed factors (extreme larval food limitation experiment 2).
 Head widthDevelopment timeEmergence rate
d.f.MSFPMSFPMSFP
  1. Significant effects are in bold.

Dung1105.514888.9<0.0011090.01652.9<0.0010.5535.42<0.001
Dung × replicate20.188.44<0.00116.0324.32<0.0010.3220.34<0.001
Dung × line21.0850.14<0.0016.409.71<0.0010.010.750.475
Dung × replicate × line40.135.93<0.0015.428.22<0.0010.010.600.666
Within family error (dung)2010.02  0.66  0.02  
Sex16.61583.77<0.001459.712153.4<0.0010.5216.19<0.001
Sex × replicate20.011.020.3625.6026.24<0.0010.061.980.140
Sex × line20.021.810.1671.798.39<0.0010.144.220.016
Sex × replicate × line40.011.160.3290.311.450.2180.030.920.453
Within family error (sex)2010.01  0.21  0.03  
Dung × sex14.33370.59<0.00146.71230.64<0.0010.4715.26<0.001
Dung × sex × replicate20.000.240.7872.5112.41<0.0010.092.990.052
Dung × sex × line20.021.650.1940.331.620.2000.010.360.700
Dung × sex × replicate × line40.032.200.0700.492.440.0480.020.540.706
Within family error (dung × sex)2010.01  0.20  0.03  
Replicate20.115.110.007133.61101.69<0.0010.7126.44<0.001
Line23.79175.20<0.00162.7247.74<0.0010.2910.89<0.001
Replicate × line40.010.580.6794.193.190.0140.072.660.034
Between family error2010.02  1.31  0.03  

As expected at limited larval food (Blanckenhorn, 1999), flies emerged earlier at smaller size (Fig. 2; dung effect in Tables 2 and 3). In the larval food limitation experiment 1 at generation 11, flies were not at their lowest limit of phenotypic plasticity because at limited food males were still larger than females, and females of the large line were larger than control and small line females (dung × line and dung × sex interactions in Table 2; Fig. 2a and b). At generation 16 with even less larval food (experiment 2), however, males and females of all three lines emerged smaller and all at roughly the same body size (Table 3: significant dung × replicate, dung × replicate × line and dung × sex × replicate × line interactions). So at least in replicate 1, flies were at their lowest limit of phenotypic plasticity (Fig. 2c). Development time corresponded to body size and did not differ between the three lines at the lowest limit of phenotypic plasticity (Fig. 2c). But interestingly, males (the larger sex) of all selection lines needed at least 1 day longer to develop than females (Fig. 2c and d) despite the fact that under very restricted larval food conditions they emerged at same size as females (significant higher order interactions between dung, sex and line in Table 3). In the larval food limitation experiment 1 at generation 11, large line males at unlimited food were 40–55 % larger than small line females at limited food (spanning the maximal extent of body size plasticity), with a correlated increase in development time of 18–23 % (Fig. 2a and b). The corresponding numbers for the extreme larval food limitation experiment 2 at generation 16 were 69–81 % for head width and 20–32 % for development time (Fig. 2c–e). This correlated response in development time agrees roughly with that expected based on previous estimates of the genetic correlation between the two traits (r = 0.2–0.5; Blanckenhorn, 1998).

Larval food limitation reduced the proportion of emerged flies in the larval food limitation experiment 1 (dung effect in Table 2), but in the two replicates the three lines were affected in different ways (replicate × line interaction in Table 2; Fig. 3a). In replicate 1, flies of all lines survived equally at unlimited food, whereas at limited food mortality of flies of the small line was highest. In replicate 2, in contrast, flies of the small line survived best at limited and unlimited food (Fig. 3a). The number of emerged females and males did not differ significantly at either food treatment (Fig. 4a; sex effect and sex × dung interaction, NS) or among the three lines (sex × line interaction, NS in Table 2), demonstrating the ability of larvae to avoid death by extensive growth plasticity. These results are thus inconclusive.

Figure 3.

 Mean juvenile survival (adult emergence proportions ±SE) for the three selection lines (S: small, C: control and L: large) at limited (white symbols) and unlimited larval food (black symbols) for all four experiments: (a) larval food limitation experiment 1, (b) extreme larval food limitation experiment 2, (c) hot temperature experiment 3 (circles 20 °C, squares 25 °C) and (d) winter frost experiment 4.

Figure 4.

 Proportion ±SE of emerged males in the three selection lines (L: large, C: control and S: small) at limited (white symbols) and unlimited larval food (black symbols) in all four experiments: (a) larval food limitation experiment 1, (b) extreme larval food limitation experiment 2, (c) hot temperature experiment 3 (circles 20 °C, squares 25 °C) and (d) winter frost experiment 4 (no female and only one male of the large line survived the winter at unlimited food).

In the extreme larval food limitation experiment 2 at generation 16, larval survival differed between the three replicates (i.e. the two replicates plus the cross between the two; replicate effect in Table 3), survival of the crossed lines being highest (Fig. 3b). This suggests some (weak) inbreeding effects caused by artificial selection. In all replicates the large line survived worst and in two replicates the smallest line best (Fig. 3b; line effect in Table 3). As in the larval food limitation experiment 1 five generations earlier (Fig. 3a), food limitation reduced emergence rate (dung effect in Table 3), but this effect was stronger in replicate 1 (dung × replicate interaction in Table 3; Fig. 3b). In contrast, in the extreme larval food limitation experiment 2 the number of emerged males (the larger sex with longer development time) was overall lower than the expected 50 % (sex effect in Table 3; Fig. 4b): more females than males emerged when food was limited (dung × sex interaction). This effect again differed between the replicates (dung × sex × replicate interaction), and the sex ratio was most female biased in the large line (sex × line interaction in Table 3).

Excluding totally infertile families, egg mortality in the extreme larval food limitation experiment 2 did not differ between the lines (F2,244 = 0.49; P = 0.612), but differed between the replicates (F2,244 = 4.44; P = 0.013), being lowest in the crossed treatment. Together with higher adult emergence rates in the crossed lines (see below), this again suggests some (weak) inbreeding depression.

Experiment 3: hot temperature and limited larval food

As in the extreme larval food limitation experiment 2, males and females of all three lines at limited food emerged at both temperatures at nearly the same body size (Fig. 5; Table 4: dung × line, sex × dung × line and sex × dung interactions). As expected, flies emerged overall smaller at 25 °C (Fig. 5; temperature effect in Table 4), but temperature only showed an effect on body size in the unlimited larval food treatment (Fig. 5; temperature × dung interaction in Table 4). Also as expected, stressfully high temperature (25 °C) during development strongly reduced juvenile survival (Fig. 3c; temperature effect in Table 4). Larval food limitation again reduced adult emergence (Fig. 3c; dung effect in Table 4), but contrary to the food limitation experiments flies emerged slightly better at limited food. Temperature and food stress overall did not interact significantly (temperature × dung interaction, NS in Table 4), although this differed between the lines (temperature × dung × line interaction in Table 4). Moreover, overall survival did not differ among the three lines (Fig. 3c; line effect, NS in Table 4). However, at 20 °C juvenile mortality of small line flies was lowest at both food levels, whereas at 25 °C it was highest at limited food (Fig. 3c; dung × line, temperature × line and temperature × dung × line interactions in Table 4), suggesting that small line (and not large line) flies were disproportionately affected by the double stress.

Figure 5.

 Mean body size ±SE for males (triangles) and females (circles) of the three selection lines (S: small, C: control and L: large) at limited (white symbols) and unlimited larval food (black symbols) in response to hot temperature (experiment 3) and winter frost (experiment 4; no female and only one male of the large line survived the winter at unlimited food).

Table 4.   Repeated measures anova for family mean body size (head width) and juvenile survival (adult emergence rate) with sex and temperature as repeated and dung (=food) and selection line as crossed factors (hot temperature experiment 3).
 Head widthEmergence rate
d.f.MSFPd.f.MSFP
  1. Significant effects are in bold.

Temperature10.4240.97<0.001113.67547.26<0.001
Temperature × dung10.4139.44<0.00110.020.820.368
Temperature × line20.032.760.06920.166.520.002
Temperature × dung × line20.011.230.29620.135.240.006
Within family error (temperature)830.01  1400.02  
Sex12.06439.49<0.00110.7730.68<0.001
Sex × dung11.27271.65<0.00110.166.490.012
Sex × line20.000.190.82620.093.700.027
Sex × dung × line20.036.530.00220.010.540.582
Within family error (sex)830.00  1400.03  
Temperature × sex10.034.570.03511.2254.97<0.001
Temperature × sex × dung10.023.840.05310.010.440.508
Temperature × sex × line20.000.590.55620.010.450.639
Temperature × sex × dung × line20.010.990.37520.020.920.400
Within family error (temperature × sex)830.01  1400.02  
Dung126.311453.66<0.00110.174.530.035
Line20.9150.28<0.00120.030.770.466
Dung × line20.4223.07<0.00120.318.27<0.001
Between family error830.02  1400.04  

The sex ratio of the emerging flies was, overall, again female biased (sex effect in Table 4), but this effect was entirely because of greater male mortality at 25 °C (Fig. 4c; temperature × sex interaction in Table 4). As in the extreme food limitation experiment 2, the sex ratio was more female biased at limited food (Fig. 4c; sex × dung interaction in Table 4). The sex ratio bias also differed between lines (sex × line interaction in Table 4), as fewer large line males emerged (Fig. 4c).

Excluding four families with total failure, egg mortality differed marginally between the lines in the hot temperature experiment (F2,119 = 2.69; P = 0.072), being highest in the control line. Pupal mortality was only assessed at limited larval food. The majority of dead pupae were fully metamorphosed. Nearly, all dead pupae were found at 25 °C (temperature effect: F1,75 = 457.13; P < 0.001); more dead pupae were found in the small (mean ± SE: 8.96 ± 0.46) than the control (6.96 ± 0.62) line and fewest in the large line (5.35 ± 0.55; line effect: F1,75 = 8.14, P < 0.001; mean ± SE number of dead pupae at 20 °C for the small, control and large lines: 0.12 ± 0.09, 0.42 ± 0.14 and 0.37 ± 0.10 respectively). Consequently, there was a strong temperature × line interaction (F2,75 = 12.18, P < 0.001). For 380 of the 608 metamorphosed dead pupae the sex could be identified: 68 % of 360 dead pupae at 25 °C and 90 % of 20 dead pupae at 20 °C were males (both P < 0.001, binomial test).

Experiment 4: winter frost and limited larval food

In this experiment body size distributions again agreed with those of the other experiments, but could not be properly analysed because only one fly of the large line emerged from unlimited food (Fig. 5). The date when eggs were laid had a significant influence on juvenile mortality (date effect in Table 5): offspring produced and moved outdoors later emerged at lower rates. Severe frost did not occur during the first days of the larval phase, when larvae feed and grow (Blanckenhorn, 1999), but did occur for about 10 consecutive days in early January. Interestingly, more flies died at unlimited larval food (Fig. 3d; dung effect in Table 5), especially in the large line (only four individuals survived the winter). Juvenile survival differed between the three selection lines, being overall lowest for the large line and highest for the small line (Fig. 3d; line effect in Table 5). Disregarding the very few large line emergences at unlimited food, the sex ratio was again female biased in both food treatments (sex effect in Table 5; Fig. 4d). Overall, these results clearly indicate that large line and male flies, which develop for longer, were disproportionately killed by the frost in early January during the larval or early pupal stage.

Table 5.   Repeated measures ancova for family mean juvenile survival (adult emergence rate) with sex and dung (=food) as repeated and selection line as crossed factors and laying date as covariate (winter frost experiment 4).
 d.f.MSFP
Dung10.4913.40<0.001
Dung × line20.0020.060.945
Dung × date10.0020.010.938
Dung × line × date20.0040.110.896
Error (dung)750.04  
Sex10.147.530.008
Sex × line20.020.960.389
Sex × date10.021.160.285
Sex × line × date20.031.630.202
Error (sex)750.02  
Dung × sex10.031.050.308
Dung × sex × line20.176.920.002
Dung × sex × date10.031.430.235
Dung × sex × line × date20.114.490.014
Error (dung × sex)750.02  
Line20.263.940.024
Date10.345.240.025
Line × date20.081.250.292
Between family error750.07  

Larval growth trajectories

As usual, body mass gain of larvae was initially exponential, and consequently close to linear when plotting cube-root-transformed body volume as in Fig. 6, and then levelled off in preparation for pupation. Graphical break point analysis (Chappell, 1989) revealed that the break point (i.e. the levelling off) generally occurred between days 5 and 7 (Fig. 6), for most treatment combinations on day 6 (this could not be resolved any further because data were taken in daily intervals).

Figure 6.

 Growth trajectories (mean cube-root-transformed body volume ±SE; N = 3–13 per treatment combination and day) of (a) male and (b) female larvae of the large (triangles), control (squares) and small (circles) lines at limited (open symbols) and unlimited larval food (closed symbols) (0 = day of egg laying).

Analysis of the initial growth phase revealed that cube-root-transformed larval body volume increased with age (F1,408 > 7000, P < 0.001), was greater for males (F1,408 = 5.37, P = 0.021), varied only marginally with larval food (F1,408 = 2.89, P = 0.090) but not among the three selection lines (F2,408 = 1.65, P = 0.194). Furthermore, the slope of the body volume increase with age (estimating growth rate) was steeper for males (sex × age interaction: F1,408 = 17.75, P < 0.001) and at unlimited food (dung × age interaction: F1,408 = 35.01, P < 0.001), the difference in growth rate between the two larval food treatments being for all lines greater in males (dung × sex × age interaction: F1,408 = 5.03, P = 0.025). Crucially, the growth rate varied among the lines and larval food treatments (dung × line × age interaction: F2,408 = 6.91, P = 0.002) such that small line individuals of both sexes grew slowest at unlimited larval food (slope of the initial growth phase ±SE for small line females 0.550 ± 0.017 and males 0.627 ± 0.020 vs. large line females 0.607 ± 0.016 and males 0.687 ± 0.022) but fastest at limited larval food (small line females 0.630 ± 0.018 and males 0.666 ± 0.022 vs. large line females 0.567 ± 0.027 and males 0.586 ± 0.023; cf. Fig. 6). All other effects were not significant.

Analysis of the asymptotic cube-root-transformed larval body volume showed a picture corresponding to adult size in Fig. 2. Male larvae and those developing at unlimited larval food were larger (both F1,382 > 290, P < 0.001), large selection line larvae being largest (F2,382 = 57.2, P < 0.001). Furthermore, unlimited larval food exaggerated both the sex and selection line differences (significant dung × sex and dung × line effects: F1,382 = 99.0 and F2,382 = 73.5, respectively, both P < 0.001), size differences among the lines were stronger in males (sex × line interaction: F2,382 = 4.30, P < 0.05), and there was even a three-way sex × line × dung interaction (F2,382 = 3.55, P < 0.05; Fig. 6).

Discussion

Artificial selection on male body size (head width) in S. stercoraria resulted in a significant and symmetrical genetic size change of about 10 % in both directions over 24 generations, a corresponding correlated response in female body size, but no change in SSD. Development time and initial growth rate also changed with body size such that large flies developed for longer and grew initially faster. However, at limited larval food (i.e. dung), growth rate was highest for small line individuals, indicating some growth compensation and suggesting a physiological cost of fast growth. The overall extent of body size and development time plasticity increased in large line flies, as flies from all lines emerged at the same minimum size under limited larval food. Some intrinsic mortality costs of growing large became apparent at extreme larval food limitation. Furthermore, winter frost disproportionately killed large line flies and flies growing up in abundant dung, an effect mediated by their longer development time, whereas size-dependent mortality at high-temperature stress was inconsistent. In all environments the larger males suffered more than the females, as evident in the selection differentials calculated from our data in Table 1. Below we discuss these main results in turn.

Response to artificial body size selection

Although we found some indications of inbreeding depression after 24 generations of selection, as evident in higher preadult mortality, genetic variability in body size was still present after 24 generations of selection (ca. 2.5 years), and body size seemed not yet to have reached its upper limit. Realized heritabilities of h2 = 0.16–0.18 for body size were lower than heritabilities previously estimated for males by parent–offspring regression (range: h2 = 0.28–0.48; Simmons & Ward, 1991; Blanckenhorn, 2002) or half-sib analysis (range: h2 = 0.37–0.85; Blanckenhorn, 2002). This is not uncommon (Falconer, 1989), and similar to results found by Reeve & Fairbairn (1996) for body size in D. melanogaster.

The typically high genetic correlations between the sexes previously estimated for body size in S. stercoraria (rg(sex) = 0.7–0.8; Blanckenhorn, 2002) is reflected in the strong correlated response in female body size to selection on males only. For species with larger males, Rensch's rule predicts an increase in SSD with increasing body size (Reeve & Fairbairn, 1996; Fairbairn, 1997). Fairbairn & Preziosi (1994) hypothesized that this may be caused by sexual selection on male size in combination with an imperfect genetically correlated response in females, which was simulated here, resulting in females not growing as fast as males (see also Reeve & Fairbairn, 1996; Kraushaar & Blanckenhorn, 2002). Here we found no significant change in SSD over 24 generations in response to selection on males. Perhaps SSD would have changed significantly in our selection lines given more time (Lande, 1980; Reeve & Fairbairn, 1996), although this appears unlikely as none of the large and only one of the small lines showed a response in the predicted direction. The equilibrium model for the evolution of SSD posits that SSD results if the selection pressures favouring (primarily fecundity selection in females and sexual selection in males) and disfavouring (primarily viability selection) large size equilibrate differently in the sexes (Price, 1984; Arak, 1988; Schluter et al., 1991; Fairbairn, 1997; Blanckenhorn, 2000). Alternatively, therefore, the SSD equilibrium may change in unexpected ways because in the laboratory selection pressures encountered in the field are no longer present (Matos et al., 2000).

Correlated responses in development time and growth rate

To change body size, an individual may alter its growth rate or its development time. In previous studies development time was found to be heritable (narrow sense heritability for males estimated from half-sibs, h2 = 0.02–0.11; and from parent–offspring, h2 = 0.09–0.52; Blanckenhorn, 2002), and its genetic correlation with body size was found to be low (rg = 0.2–0.5: Blanckenhorn, 1998). Here, selection on body size in yellow dung flies produced a correlated response in development time largely predictable from these previous estimates. At generation 16, there was an average increase relative to the control line of 8 % in body size and of 4 % in development time in the large line, and a decrease of 8 % in size and of 2 % in development time in the small line. Large line flies developed for 1–2 days longer than small line flies. In addition, at unlimited larval food the initial growth rate of flies of both sexes increased from the small to the large selection line. Thus both faster growth and longer development contributed to the larger final size of large line flies. However, males (the larger sex) increased initial growth rate at unlimited compared with limited larval food more than females. Furthermore, at limited larval food small line flies of both sexes grew fastest. This suggests a trade-off and a viability cost of growing fast in large (line) flies. And indeed, large line males suffered the highest juvenile mortality at extremely limited larval food (discussed further below). Alternatively, large line flies may grow more slowly at limited larval food because they grow for longer and are thus able to meet a presumed target body size, whereas small line flies, in contrast, might have to increase growth rate to meet the same target size. In any case, selection on body size produced a correlated genetic response in growth rate and in growth rate plasticity.

When larval food (i.e. dung) was limited, yellow dung fly larvae of all lines grew only slightly slower to emerge earlier and smaller. This response is rare in general (Stearns & Koella, 1986) but common in species that are regularly threatened by habitat depletion (Møller et al., 1989; Newman, 1992; Juliano & Stoffregen, 1994; Blanckenhorn, 1998, 1999). Indeed, at the lowest limit of phenotypic plasticity, presumably marking the minimum amount of dung required to survive, males and females of all lines emerged roughly after the same development time at roughly the same size. Except for slightly elevated larval mortality at extremely limited larval food in large line flies (Table 1), probably because of their higher energy demand (see below), there was no difference in the ability of the genetically different line flies to escape mortality via phenotypic plasticity in growth and development.

Correlated response in phenotypic plasticity

The amount of phenotypic plasticity a genotype can express for a specific trait is usually quantified by the difference in trait value of related individuals reared in different environments (Scheiner, 2002). Applying this definition to the body size plasticity of yellow dung flies here means that large line flies generally are more plastic than small line flies. This is because the minimum size attained at the lowest limit of phenotypic plasticity was the same for all lines, whereas the maximum size expressed under good larval conditions was greater for genetically large flies. From this, we may conclude that the evolution of larger body size in yellow dung flies does not lead to, i.e. trade-off with a decrease in phenotypic plasticity in body size (cf. van Tienderen, 1991, 1997; Newman, 1992; Arendt, 1997). On the contrary, here it resulted in greater plasticity. This could be interpreted as showing that trait (here body size) and plasticity in trait size are positively genetically correlated (Via & Lande, 1985; Via, 1993a, b; Via et al., 1995). On the other hand, it may be argued that selection for larger size has left the lowest attainable body size unaffected, perhaps because it is determined by a different proximate mechanism than the maximum size attainable, implying the increase in plasticity is an epiphenomenon (Scheiner, 1993a, b). The latter seems likely, as two different and independent physiological mechanisms seem to proximately control body size plasticity: the minimum weight that has to be reached to survive to pupation, and the critical weight at which larvae pupate depending on food availability (de Moed et al., 1999; Stern, 2001; Davidowitz et al., 2003; Nijhout, 2003). Note that all this equally holds for development time, which was correlated with body size (Fig. 2), albeit to a lesser extent.

Correlated responses in juvenile viability

Only under extreme larval food limitation, when body size plasticity reached its lowest limit and males and females of all selection lines emerged at roughly the same size (Fig. 2c–e), did genetically large flies (Fig. 3b) and the larger males of all lines (Fig. 4b) show higher mortality. This is reflected in generally negative selection differentials (Table 1) and demonstrates a physiological cost of growing large as well the adaptive nature of this phenotypic plasticity, as in their unpredictable ephemeral habitat it is better to emerge small than to die (Blanckenhorn, 1998, 1999). Dingle (1992) also found that milkweed bugs selected for large size showed higher larval mortality when raised under food restriction. It could be that large (line) flies have greater mass-specific metabolic requirements or a less efficient energy use (Blanckenhorn, 2000, 2005; Reim et al., 2006a, b), potentially mediating their higher intrinsic mortality.

We found consistent evidence that males suffered higher juvenile mortality when conditions became stressful (see their more negative selection differentials in Table 1). This is unlikely to be because of our artificial selection having been performed on males only, as similar sex-biased effects also occurred in earlier experiments using field-derived phenotypes (Blanckenhorn, 1998). Under extreme larval food limitation, males emerged at the same body size as females but nevertheless required 1 day longer for development (Fig. 2). This difference does not merely reflect the larger genetic body size of males, as males of the small and control lines after selection were similar in size to females of the control and large lines respectively (Fig. 1). Males also suffered more than females at hot temperatures, when growth rate is higher. Furthermore, particularly in winter, when real-time growth is slow, males suffered from their longer development time. In principle, such sex-specific effects can relate to fundamental differences between the sexes per se. Thus it has been argued that strong and prevailing sexual selection on males (but not females) produces higher costs in terms of mortality even when males are the smaller sex (Promislow, 1992; Andersson, 1994). Furthermore, male gonads might be more costly to produce than female gonads, incurring some sex-specific life history costs (Pitnick et al., 1995). Alternatively, sex-specific effects can relate to the expression of SSD, the larger sex having to pay a higher cost, e.g. in terms of increased development time or mortality (Promislow, 1992; Blanckenhorn, 2000; Badyaev, 2002). We cannot differentiate between these alternatives here, as yellow dung fly males are also the larger sex.

Winter frost was the factor most strongly selecting against large flies (Table 1), both in this study involving a gradient of genetically different selection lines of artificially extended body size range and in a previous phenotypic study by Blanckenhorn (1998). Extensive frost kills larvae that have not yet pupated, which here were large line flies, flies growing up in unlimited food featuring longer development times, particularly males that develop for longer, and larvae from eggs laid later in the season. We thus obtained direct evidence that end-of-season time limitations can select against large body size when size and development time are positively genetically correlated (cf. Roff, 1980, 2000), which is the case in yellow dung flies, albeit not strongly (r = 0.2–0.5: Blanckenhorn, 1998).

Higher temperatures reduce body size in most organisms (Atkinson, 1994). Yellow dung flies show their fastest growth at 25 °C, but this is associated with high mortality (Blanckenhorn, 1997, 1998, 1999). In our experiment, only 24 % of all individuals emerged at this temperature, but large (line) flies were not necessarily affected more (Fig. 3c). Interestingly, the main mortality occurred during the pupal phase. Dead pupae were fully developed, suggesting that despite fast growth larvae had obtained sufficient energy to metamorphose. Possible reasons for higher mortality at high temperatures are starvation, desiccation or insufficient heat dissipation. The first two mechanisms should favour large individuals that have a relatively smaller surfaces exposed to evaporation and/or a more efficient energy use (Atkinson, 1994; Lighton et al., 1994). For the same reason, the last mechanism should favour small animals with a greater surface to better dissipate heat (Atkinson, 1994). Desiccation is unlikely in our climate chambers because of humidity control. Moreover, differences in survival between phenotypically large (unlimited food) and small (limited food) flies at 25 °C were only marginal, and if so in the wrong direction. Mortality of the selection lines in response to heat stress was inconsistent. At unlimited food small line flies survived best, but this was also the case for their siblings at 20 °C and therefore cannot be attributed to temperature alone. In contrast, at limited food large line flies survived best, exactly opposite to their siblings at 20 °C, even though all flies attained about the same (small) final size. Perhaps at limited food hot temperatures stressed small flies more because of inefficient energy use in physiological processes involved in cooling. Despite the evidence lacking for starvation disadvantages of large flies, males, which potentially have a higher energy demand during development (cf. above: e.g. Pitnick et al., 1995), suffered higher mortality. Overall, size-selective mortality effects of hot temperatures were equivocal here and hence prove inconclusive.

Conclusions

Selection experiments are a useful tool for studying life history trade-offs or constraints, although they can be associated with problems of inbreeding depression and adaptation to the laboratory (Harshman & Hoffmann, 2000; Matos et al., 2000; Zera & Harshman, 2001). Our study revealed some, albeit very weak inbreeding effects in terms of lower juvenile survival when comparing the pure selection lines with the replicate crosses (Fig. 3b). Together with the fact that body size of the large line continued to increase when artificial selection was terminated at generation 24, however, this suggests few negative pleiotropic mortality effects potentially constraining the evolution of larger yellow dung flies. This is not uncommon. For example, the tobacco hornworm Manduca sexta has been shown to double its body weight over a 30-year period in laboratory culture (D'Amico et al., 2001).

Overall, in correspondence with an earlier study (Blanckenhorn, 1998), we found three connected, genetically correlated responses that apparently trade-off against large body size at the juvenile level: a development time, a growth rate and a juvenile mortality increase. These costs of large body size were, however, only evident at extreme food shortage at the lower limit of phenotypic plasticity and during winter frost, predominantly in the larger males. This testifies to the strong buffering effect of growth plasticity in this species (Blanckenhorn, 1998). Given nonzero mortality probability at all times, higher cumulative mortality caused by prolonged juvenile periods clearly can select against large body size to some degree (Roff, 1980, 1992, 2000; Newman, 1992; Stearns, 1992). Although a number of egg, larval and pupal predators and parasites have been identified in and around dung (Laurence, 1954; Skidmore, 1991), quantitative assessment of juvenile or adult size-dependent mortality caused by predation or parasitism in yellow dung flies is difficult in the field (Blanckenhorn, 1998; Burkhard et al., 2002), however necessary to obtain a full picture. Our gut feeling is that size-selective predation and parasitism, if present at all, is not strong, as our admittedly limited field estimates of adult viability selection against large body size (including predation) are indeed small in magnitude (Burkhard et al., 2002; Table 1). Taken together, previous plus the newly generated viability selection estimates using our artificially enlarged selection lines may well be sufficient to counteract the strong size advantages generated by sexual and fecundity selection and thus to halt body size evolution in yellow dung flies (Table 1); however, ultimately parameterized models will be necessary to investigate this question. We are aware that some of our experiments are unreplicated because of crossing the replicate lines. However, at minimum, we investigated a genetic body size gradient with three points (small, control and large) plus a phenotypic gradient (caused by food limitation), which yielded robust results.

Our results are in line with the notion that presumed trade-offs underlying the evolution of adaptive life histories often become visible only in stressful environments (van Noordwijk & de Jong, 1986; Alatalo et al., 1990; Schluter et al., 1991). They also suggest that the function linking fitness to body size is largely flat so that (strong) fitness decrements only occur at the fringes (here at the high end) of the naturally existing size distribution. Extension of the available body size range by artificial selection, as done here, revealed some fitness effects that otherwise would have remained cryptic.

Acknowledgments

We thank various local colleagues for help and encouragement, and the Swiss National Fund for financial support of the PhD project.

Ancillary