Standardized measures of the strength of selection on a character allow quantitative comparisons across populations in time and space. Spatiotemporal variation in selection influences patterns of adaptation and the evolution of characters and must therefore be documented. For the dung-breeding fly Sepsis cynipsea, we document patterns of variation in sexual, fecundity and larval and adult viability selection on body size at several spatiotemporal scales: between-populations, over the season, over the day and between dung pats. Adult viability selection based on residual physiological survivorship in the laboratory was nil or weakly negative. In contrast, larval viability selection in two laboratory environments was weakly positive for males at low competition and females at high competition. Fecundity selection was positive and strong at all times and in all populations. Sexual selection reflecting pairing success was overall strongly positive (about three times stronger than fecundity selection), while selection reflecting male reproductive success via the clutch size of his mate (i.e. assortative mating) was essentially nil. Only sexual selection varied significantly at coarse (between populations and seasonally) but not at fine (within a day or between pats on a pasture) spatial and temporal scales. Quadratic and correlational selection differentials were low and inconsistent in all episodes except for fecundity selection, where there was some evidence that clutch size reaches an asymptote at large body sizes, implying weaker selection for large size as females get bigger. Implications of these results for the evolution of body size and body size dimorphism are discussed.
Spatiotemporal variation in environmental conditions is well known to affect a variety of morphological, life history and behavioural traits. Such variation will consequently also affect the relationship of fitness to, and therefore selection on, and ultimately the evolution of those traits. For example, between-population variation in a given species in sexual selection favouring large males is hypothesized to result in corresponding variation in the average body size and sexual size dimorphism between those populations (Fairbairn & Preziosi, 1994). This hypothesis requires that between-population differences in selection regimes are, on average, consistent (i.e. repeatable) over evolutionary time (Preziosi & Fairbairn, 1996). If this is not so, fluctuating selection may produce different evolutionary patterns (e.g. Istock, 1981); if this is so, within-population temporal variation in selection will merely prolong the expected evolutionary response. However, such variation is always of practical importance as it decreases the reliability of an estimate of selection at any point in time. Therefore, for accurate interpretation and prediction of patterns of adaptation and the evolution of characters, particularly in terms of mechanisms, spatiotemporal variation in selection needs to be assessed (Lande & Arnold, 1983; Endler, 1986; Wade & Kalisz, 1990). In recent years, some studies have documented such variation in selection at some scales in a variety of species, characters and scientific journals (Kalisz, 1986; Gibbs & Grant, 1987; Conner, 1989; Arnqvist, 1992a; McLain etal., 1993; Nishida, 1994; Carrol & Salamon, 1995; Conner etal., 1996; Dudley & Schmitt, 1996; Preziosi & Fairbairn, 1996, 1997). However, such studies are not as frequent as they should be; they are necessary for any study species for which evolutionary inferences are to be drawn from field measurements of selection.
This paper investigates selection on body size in Sepsis cynipsea, a small (≈4–5 mm in body length) and very abundant dung fly of ant-like appearance (Hennig, 1949; Parker, 1972a,b; Pont, 1979; Meier, 1996). We document variation in selection on body size in this species in Switzerland at multiple spatiotemporal scales. An initial study in 1994 investigated between-population (long-distance spatial) variation in selection among three disjunct populations and additionally considered temporal (seasonal) variation by sampling the same populations repeatedly over the season. In a further study in 1995 we investigated seasonal, daily and between-pat (i.e. short-distance spatial) variation in sexual selection in one of the three populations more thoroughly. We attempted to be as complete as possible by considering all major selection episodes reflecting the primary components of individual reproductive success: larval and adult viability selection on both sexes, fecundity selection on females and sexual selection on males. This has been achieved in only very few studies to date.
Materials and methods
The study species
Dung flies of the genus Sepsis are common in Eurasia and Africa (Zuska & Pont, 1984). Many similar-looking species coexist in dung of various animals, some of them being dung specialists and some generalists (Hammer, 1941; Pont, 1979; Meier, 1996). S. cynipsea is the most common and abundant European species of this group and specialized on cow dung (Hennig, 1949; Parker, 1972a,b; Pont, 1979; Schulz, 1989; Meier, 1996). Females oviposit into the dung, on which the developing larvae feed and thereby deplete, together with larvae of many other fly and beetle species (Hammer, 1941). Intra- and interspecific competition thus typically limits resource availability and imposes a time constraint on development (Blanckenhorn etal., 1998).
The mating system of S. cynipsea has been well described by Parker (1972a,b). Large numbers of males await females ready to lay eggs at the oviposition site, a fresh cow pat (Hammer, 1941; Parker, 1972a). Operational sex ratios are typically highly male-biased. As soon as a female arrives, males try to seize her, guard and defend her against other males during oviposition, and attempt to copulate with her off the dung thereafter (Parker, 1972a,b). In response to male mating attempts, females show characteristic shaking behaviour during all phases of pairing indicating reluctance to mate and/or some sort of male assessment; in any case, there is sexual conflict over mating (Ward etal., 1992; Rowe etal., 1994; Blanckenhorn etal., unpublished). Only about 40% of the pairs formed eventually copulate (Parker, 1972b; Ward, 1983; Ward etal., 1992). Ward (1983) showed in one English population of S. cynipsea that larger males are more likely to be paired, to be subsequently mated and to acquire a larger female, but he did not quantify selection. Because the smaller males cannot force copulations upon the females in this species, the influence of the female on the outcome of the pairing, i.e. some form of female choice, must be ranked higher than that of male–male competition (Ward etal., 1992; Blanckenhorn etal., unpublished). It is not known when and where females copulate the first time or how often they copulate during their lifetime. It is also unclear where exactly females spend the rest of their time in the field. From laboratory rearing we know that individuals acquire the protein needed for the production of eggs and sperm by feeding on dung and that individuals require sugar (Schulz, 1989; personal observation), which they seem to acquire in the field by foraging on nectar (Bährmann, 1993; Warncke etal., 1993; Meier, 1996).
Field samples and laboratory treatment
In 1994, we worked with three disjunct populations around Zürich, Switzerland: Luzern (N47°02′, E8°19′), Obfelden (N47°16′, E8°26′) and Fehraltorf (N47°23′, E8°44′). Starting with the last week in May, about 2 weeks into the season, we sampled one fresh cow pat per week for 5 (Obfelden: 6) weeks in a row. S. cynipsea larvae develop very rapidly at high temperatures (egg to adult in 7 days at 30 °C; Schulz, 1989; Blanckenhorn, 1997). We have estimated that this species features at least six generations per season in our lowland (≈500 m), Fehraltorf population (Morf, 1997), so 5 weeks corresponds to about the first third of the season and one to two field generations. All populations were sampled on dry days between 10:00 and 15:00 h, but not necessarily on the same weekday. At each sampling, we first estimated local male density on and ≈30 cm around the pat by counting the number of single individuals (i.e. unpaired males) and pairs. This estimate of the proportion of paired males on and around a pat p* served as our measure of the relevant local competitive environment later used in calculating sexual selection coefficients. We then proceeded to collect at least 20 pairs and 20 single males into small glass bottles using an aspirator. We did not differentiate between pairs in copula or not (cf. Ward, 1983). We additionally noted a variety of climatic data such as temperature, wind conditions, cloud cover and the freshness of the dropping as potential covariates.
In the laboratory, we measured head width (including eyes) of all individuals under CO2 anaesthetic using a binocular microscope at 40× magnification (to the nearest 1/80 mm). In S. cynipsea, phenotypic and genetic correlations between head width and other morphological measures like thorax, wing or hind tibia length are high (r > 0.75; Reusch & Blanckenhorn, 1998). Furthermore, Ward (1983) showed a good correlation (r > 0.75) between wet weight and hind tibia length in females of this species, and Zerbe (1993) showed an even better correlation (r > 0.95) between dry weight and hind tibia length in Sepsis punctum. Therefore, head width can serve as a practical surrogate measure reflecting direct and indirect selection on overall body size.
After measurement, the pairs were maintained together for 24 h in 100-mL glass bottles containing miniature dishes of sugar and pollen (for food), a wet piece of cheese cloth (for water), and fresh dung (for food and oviposition). Thereafter, the males were removed and maintained until their death as a group in large 3.5-L plastic containers with ad libitum quantities of the above nutrients. All single males were held from the beginning in an identical manner in another group container. For the females (who can store sperm for weeks) we recorded the size of their first two clutches after capture, which typically would be laid within 5–9 days; if they did not die before, they were then also transferred to a third, identical group container. The death date for all individuals was recorded, and all individuals were re-measured upon their death. This was necessary because the grouped individuals were not individually marked. The laboratory environment was a constant 30–40% relative humidity, 27–30 °C (range), and 15 h photoperiod.
In our 1994 study, we considered four selection episodes on body size reflecting the primary components of individual reproductive success: (i) adult viability selection on both sexes; (ii) fecundity selection on females; (iii) sexual selection on males reflecting pairing success (SexS1) and (iv) assortative mating (SexS2). SexS1 compared the sizes of paired and unpaired males. SexS2, which some authors (e.g. Ward, 1988) call ‘natural’ selection (an unspecific term that we try to avoid), refers to the reproductive success accruing to larger males due to their ability to secure larger, more fecund females, given pairing. There are actually three selection episodes detectable in the sexual context in S. cynipsea (Parker, 1972a; Ward, 1983; Ward etal., 1992; Allen & Simmons, 1996): (a) some males do not get paired at all (SexS1), (b) of those that get paired only about 40% achieve copulation and (c) a larger male may copulate with a larger female, her generally larger clutch size (Blanckenhorn, 1997) conferring higher reproductive success to him as she presumably uses his sperm the next time(s) she oviposits (SexS2). For practical reasons, we ignored episode (b) in our study. As selection for large males tends to occur in all three episodes (Ward, 1983; Ward etal., 1992), this leads to underestimates of the overall sexual selection intensity. Fecundity selection was simply the mean of her first two clutches in captivity as a function of body size. Assessing adult survival in the field, for example via mark–recapture techniques, is practically impossible in this small and numerous species, so we used the residual physiological longevity of the individuals after capture as an estimate. Assuming that the individuals captured at any point in time are a representative cross-section of the mating population with regard to age, i.e. that there is no consistent association between size and time of season, this measure allows some inferences about the effects of sex, pairing status and size on (residual) survivorship. Viability selection on body size cannot necessarily expected to be the same for males and females (cf. Arak, 1988; e.g. Blanckenhorn etal., 1995), so it should be estimated separately for the sexes.
In 1995, we intensively studied only the Fehraltorf population, essentially following the same methodology. This time we concentrated on sexual selection as it had proven most variable in 1994, but we took a bivariate approach by measuring head width and hind tibia length. Compared to univariate analyses, multivariate analyses allow separation of direct and indirect selection on several correlated components of body size, which may help identify the actual targets, and hence the potential mechanism of selection (Lande & Arnold, 1983). Once a week, two pats were sampled within a short time period (<30 min, i.e. almost simultaneously) on any weekday between early June and mid September. In addition, on five days of the season (11 and 25 July, 11 and 23 August, and 6 September), five (on 6 September only four) pats were sampled at 2-h intervals between 09:00 and 17:00 h to estimate daily variation in sexual selection. As only morphological measures were taken, the individuals could be frozen upon arrival in the laboratory for later measurement.
Univariate (1994) and bivariate (1995) selection coefficients were estimated using regression following Lande & Arnold (1983) and Arnold & Wade (1984a,b). For each individual sample (i.e. pat), we first produced standardized z-scores for body size x by subtracting the sample mean from each value and dividing by the standard deviation, zi=(xi−x¯)/SDx. The absolute fitness estimators Wi corresponding to the four selection episodes in 1994 (square root(residual longevity), mean clutch size, mated or not mated, and mean clutch size of mating partner) were transformed to relative fitnesses, wi= Wi/w¯. We used the univariate models of relative fitness on standardized body size w=c+β1z to estimate the linear (β1) and w=c+β′1z+(γ1/2)z2 to estimate the quadratic (γ1) coefficients. The resulting coefficients β1 and γ1 are the linear and nonlinear (quadratic) selection differentials, reflecting the combined effects of direct and indirect selection on body size (Endler, 1986; Brodie etal., 1995). In the 1995 study, we used the corresponding bivariate models w=c+β2,hzh+β2,tzt for the linear (β2,j) and w=c+β′2,hzh+(γ2,h/2)z2h+β′2,tzt+(γ2,t/2)z2t + γh,tzhzt for the quadratic (γ2,j) and correlational (γh,t) coefficients, where the subscripts h and t refer to head width and tibia length. These coefficients are the bivariate linear and nonlinear selection gradients (Brodie etal., 1995). In both studies, we produced mean coefficients ± SE for each population from the weighted (by sample size) weekly or per-pat coefficients.
The difference of all regression coefficients from a slope of zero (the null hypothesis of no selection) was tested for each sample (i.e. pat). We tested for overall selection, as well as for spatial and temporal variation in selection, by using a multiple regression (ANCOVA) approach of relative fitness on body size with population and week (plus all interactions) as fixed factors in 1994, and week and pat (for seasonal and between-pat variation) and day and time of day (for daily variation) as fixed factors in 1995. This allowed testing for the effects on relative fitness of all main factors plus their interactions (all terms not including standardized size), for overall selection on body size (effect of standardized size), and for spatiotemporal variation in selection (all other interaction terms including standardized size) at the same time, avoiding multiple analyses of the same data set.
In case of the binary mating success, we weighted body size means and variances by the proportion of paired males p* estimated just prior to sampling the pat (see above) before producing z-scores, as our proportions p of pairs and (1 – p) of single males sampled did not necessarily reflect their actual occurrence (cf. Arnold & Wade, 1984b): x¯total=p*·x¯paired+(1−p*)·x¯unpaired and Vartotal=p*·Varpaired+(1−p*)·Varunpaired+p*· (1−p*) · (x¯paired−x¯unpaired)2. When pairing success (i.e. absolute fitness) is one or zero, relative fitness is the proportion of paired males in the sample (Brodie & Janzen, 1996). As unpaired females rarely occur at the pat this number is equivalent to the operational sex ratio (OSR = number of females/number of males), which defines the local competitive environment and thus the intensity of sexual selection (Emlen & Oring, 1977; Arnold & Duvall, 1994). By using p* instead of the actually sampled proportion of paired males p our weighting procedure corrects for sampling bias while significance testing remains unaffected (Blanckenhorn etal., 1999). Significances were derived as described in the previous paragraph with p* as the dependent variable. As pairing success is not normally distributed, we additionally assessed significant deviations from zero slopes using logistic regression for each sample (i.e. pat) as well as the full model. Unless otherwise stated, the significances of SexS1 given refer to the logistic regression results. However, as is common practice, we present the coefficients derived from linear regression and not those derived from logistic regression because the latter ‘are not yet interpretable in the context of equations for evolutionary change’ (Brodie etal., 1995).
Larval survivorship experiment
We estimated larval viability selection in a separate laboratory experiment. Unless the character under selection (here body size) can be reliably estimated from the egg or the young larva, estimation of body size independent of the estimation of larval mortality is required. This is because those individuals which die during development cannot later be measured, thus biasing the results. Furthermore, as in most ectotherms, adult body size of S. cynipsea depends on the environmental conditions during development, most notably temperature and resource availability, the latter largely being a function of inter- and intraspecific competition. These factors are highly variable in nature (Blanckenhorn, 1997) and can be manipulated in the laboratory (Blanckenhorn etal., 1998). One way to circumvent these problems is to estimate the larval survivorship of families of individuals at a variety of environmental conditions and relate it to the body size estimated from relatives (parents, full or half sibs) at ideal conditions, at which mortality should be minimal and presumably not biased with regard to the trait investigated (here body size). The mean phenotype of a family at ideal conditions is probably the best possible estimate of a genotype because the environmental variance component is minimized. Estimating larval survivorship in more than one environment is advised because it better reflects the natural situation and because the expression, and selection on, a trait in various environments need not necessarily be correlated.
Roughly 100 S. cynipsea pairs were collected in September 1995 from Fehraltorf. In the laboratory, the pairs were distributed evenly but randomly into six 3.5-L plastic containers, in which the individuals had continuous access to ad libitum amounts of the above nutrients. They were kept in a climate chamber at 27–30 °C (range), ≈30–40% relative humidity and 15 h photoperiod. Subsequent generations were bred in the laboratory using standard methods described in Blanckenhorn etal. (1998).
For the experiment, F2 laboratory-generation individuals were separated by sex as soon as they eclosed in the large rearing containers. After a few days, mixed-sex groups of various operational sex ratios and densities were randomly assembled into new containers with the above nutrients, subject to the constraint that males and females stemmed from different F2 rearing containers to avoid brother–sister matings. This treatment was for reasons extraneous to this experiment. Of importance here is merely that females could mate with one or more males over a period of 3 days, implying that their offspring may be half rather than full sibs.
After 3 days, females from the various containers were isolated into 100-mL bottles containing the above nutrients. There they were held until they had laid one clutch of 40–70 eggs into a miniature dish with about 3.5 g of dung. The eggs in the clutch were counted and subsequently split into two environments, one with overabundant dung (> 0.5 g per larva) and one with limited dung (< 0.1 g per larva; cf. Blanckenhorn, 1997). Typically, a fraction of the dung containing about half the eggs (which were again counted) was carefully spooned out of the miniature dish and transferred into a 50-mL plastic bottle containing at least 20 g of dung (the no- or low-competition environment), which was then capped with a toilet paper lid. The eggs remaining in the miniature dish were also counted again and the dish was weighed (the tare was known) so we could estimate the amount of dung available per larva in this food-limited treatment (the high-competition environment). The dish was then put into a 100-mL bottle also containing miniature dishes of sugar, pollen and water, which was capped with a paper stopper. The split broods of a total of 75 half or full sib families could then develop side-by-side in a climate chamber at 22–23 °C (range), ≈60% relative humidity, and 13 h photoperiod.
We counted the number of male and female offspring of each family that emerged from each larval environment, whereupon all individuals were frozen for later measurement. We measured the left or right hind tibia length of five individuals per sex, larval environment and family. For analysis, we produced mean body sizes separately for each sex and larval environment of each family. Larval (preadult) survivorship was estimated as (adults emerged/eggs entered) for each environment, yielding one estimate per family for each sex. To estimate larval viability selection differentials, we regressed relative, angular-transformed survivorship (i.e. fitness) on z-standardized mean family hind tibia length at low competition, separately for both sexes and both larval environments. Significances were derived from these regressions as well as from the full repeated-measures model with sex and larval environment as fixed factors.
1994 data set
To exclude handling effects on mortality, the longevity analysis was based only on individuals that had lived for at least 3 days after capture. The full linear model revealed that residual longevity in the laboratory was greater for males than for females (overall mean ± SE: 19.3 ± 0.37 vs. 15.5 ± 0.56 days; F1,935=17.11, P < 0.001; cf. Table 1) and that it varied in space and time (significant effects of population, week, population by week, and population by week by sex: all F > 4.3 and P < 0.001). There was marginally significant overall adult viability selection against large size (negative effect of standardized size on relative residual longevity: F1,935= 3.75, P=0.053), implying that unpaired males lived longer than paired males (22.0 ± 0.44 vs. 15.3 ± 0.59 days). There was also significant temporal (i.e. seasonal) variation in selection, as evident in effects on relative residual longevity of the week by body size (F4,935 = 4.04, P = 0.003) and the sex by week by body size (F4,935 = 3.25, P = 0.012) interactions, implying differences in the weekly slopes. All other terms including standardized size were not significant (P > 0.1), particularly all those also including population, implying no spatial variation in selection. Table 1 gives the weekly data on size and untransformed residual longevity with the corresponding linear adult viability selection differentials β. The latter were negative whenever significant. Quadratic selection differentials γ showed no particular pattern in either sex, so they are not presented in Table 1: nine were positive and seven negative (one significantly so) for males, and 11 were positive (one significantly so) and five negative for females. Correspondingly, the full nonlinear model revealed no significant effects at all.
Table 1. Mean (±SE) residual physiological longevity and body size of males and females with the corresponding adult viability selection differentials β for three Swiss populations of Sepsis cynipsea at various times of the season 1994. **P < 0.05; **P < 0.01; ***P < 0.001.
Linear fecundity selection differentials were positive throughout and mostly significant (Table 2). The clutch size data showed a bimodal distribution separating partial from full clutches, with the minimum between the two maxima at about 25 eggs. Blanckenhorn's (1997) laboratory data also showed that even the smallest females’ clutch sizes are no less than 30. Therefore, only those clutches comprising more than 25 eggs were included in the analysis. For the females that laid two full clutches (the majority) we analysed their mean clutch size. Analysis of the full model revealed strong fecundity selection for large female size (F1,261=73.28, P < 0.001). There was no significant spatiotemporal variation in selection and no spatiotemporal variation in clutch sizes (all other terms P > 0.1). Quadratic selection differentials γ were low throughout but mostly negative (Table 2). In combination with positive linear selection differentials β, this suggests that the increase in clutch size levels off at the largest body sizes. The full nonlinear model indeed revealed a small but significant overall negative quadratic coefficient (F1,260=3.92, P=0.049, mean γ=−0.035).
Table 2. Mean (±SE) clutch and body size of paired females with the corresponding linear, β, and quadratic, γ, fecundity selection differentials for three Swiss populations of Sepsis cynipsea at various times of the season 1994. *P < 0.05; **P < 0.01; ***P < 0.001.
Sexual selection differentials based on mating success (SexS1) tended to be positive and were occasionally significant (Table 3). In contrast, selection reflecting assortative pairing (SexS2) was weak and nonsignificant throughout (Table 3). The full linear model revealed a significant large male mating advantage (overall mean ± SE head width of paired and unpaired males: 0.911 ± 0.002 vs. 0.897 ± 0.002 mm; positive effect of standardized size on relative pairing success: F1,1119=21.06, P < 0.001; logistic regression: χ2=15.12, P < 0.001). There was also variation in SexS1 across populations and time, as evident in significant population by size (F2,1119=3.83, P=0.022; χ2=10.09, P=0.003) and population by week by size interaction terms (F8,1119=2.16, P=0.028; χ2=3.37, P=0.071). The week by population interaction was also significant, indicating spatiotemporal variation in pairing success p* (Table 3; F8,1119=15.36, P < 0.001). All other terms were P > 0.1. The corresponding model for SexS2 yielded no overall effect of male size on his reproductive success (F1,261=3.34, P=0.069) and no spatiotemporal variation in selection (all other terms P > 0.1). Again, quadratic sexual selection differentials γ showed no particular pattern and so are not presented in Table 1: seven were positive (two significantly so) and nine negative for SexS1, and seven were positive and nine negative for SexS2. The full nonlinear model revealed no significant effects at all. Note that the sixth Obfelden sample was dropped in all the above analyses to balance the design.
Table 3. Mean (±SE) body size of paired and unpaired males with the corresponding estimated proportion of being paired p*, the selection differentials β for mating success (SexS1) of all males and those of reproductive success (SexS2) of paired males for three Swiss populations of Sepsis cynipsea at various times of the season 1994. Reproductive success of males are the clutch sizes of their mates (with corresponding sample sizes) given in Table 2. *P < 0.05; **P < 0.01; ***P < 0.001.
Table 4 compares the mean parameters of the three populations. The populations varied significantly in body size and longevity but not clutch size (for the latter two variables, the main effects from the models described above are given). Body size additionally varied from week to week (i.e. seasonally), as indicated by a significant population by week interaction term (F8,1742= 5.61, P < 0.001), but not in a systematic way (see Tables 12–3 for the weekly size data). Sexual and viability selection (at least for females) tended to be strongest in the population with the greatest mean body size (Fehraltorf) and weakest in the population with the smallest mean body size (Luzern). The significance of the per-population selection differentials given in Table 4 was determined by dropping week from the models described in the previous paragraphs. None of the environmental variables reflecting climate or the quality of the dung significantly covaried with the temporal change in body size or any of the selection differentials, so they were excluded from all analyses. The only environmental factor that influenced selection was local competitor density, probably reflecting population density; this is treated elsewhere (Blanckenhorn etal., unpublished).
Table 4. Population comparison of the mean (±SE) parameters of size, residual longevity, clutch size and the mean (±SE) linear, β, and nonlinear, γ, selection differentials for the season 1994 with sample sizes in parentheses. *P < 0.05; **P < 0.01; ***P < 0.001.
1995 data set
Analyses of the 1995 data set revealed that sexual selection based on pairing success (SexS1) did not significantly vary over a given day. This was indicated by nonsignificant time of day by size and time of day by week by size interactions for either morphological measure (head width or hind tibia length). Sexual selection did also not vary among the two pats sampled about the same time on any given day throughout the season 1995, which is actually what ideal-free theory would predict (Milinski & Parker, 1991). We could consequently remove pat and time of day from the full model when analysing seasonal variation. The resulting analysis revealed seasonal variation in sexual selection over the 13 weeks sampled, for both morphological measures (significant week by size effects on pairing success, i.e. differences in the weekly slopes; Table 5). The significant effect in Table 5 of terms not including body size simply means that p*, the estimated operational sex ratio, varied over the day and the season. Only the least-squares (but not the logistic) regression results are relevant in this context because only they were run using the corrected estimate p* (see Methods). The results of the bivariate analyses are highly concordant with those of the univariate analyses in Table 5 and consequently not presented separately.
Table 5. Univariate least-squares and logistic regression results for seasonal variation in sexual selection (SexS1) on head width and hind tibia length in 1995. Dependent variable: relative (p*) and absolute pairing success, respectively.
Table 6 shows the seasonal changes in the weekly linear, quadratic and correlational selection differentials in 1995; the overall means ± SE for the whole season are also given. When comparing the linear univariate coefficients it is evident that directional selection on head width was much stronger than on hind tibia length, a result also apparent in Table 5. Note that the mean univariate selection differential on head width (β1h) obtained in 1995 (Table 6) is very close to, and not significantly different from, that obtained the year before for the same, Fehraltorf population (Table 4). The bivariate analysis revealed that head width (or a correlate thereof) rather than hind tibia length is the trait under selection; in fact, selection on both morphological measures tended to be in opposite directions despite the generally positive correlation between the two (Table 6; Reusch & Blanckenhorn, 1998).
Table 6. Seasonal changes during 1995 in linear (β1j = univariate; β2j = bivariate), quadratic (γ1j = univariate; γ2j = bivariate) and correlational (γh,t) sexual selection (SexS1) gradients on head width (h) and hind tibia length (t) and the product–moment correlation of the two variables (rh, t), with the corresponding sample sizes. Dependent variable: estimated relative pairing success p*. *P < 0.05; **P < 0.01; ***P < 0.001.
Selection reflecting assortative pairing (SexS2), this time based on the body size (and not the clutch size) of the female partner, was again very weak in 1995 and only rarely significant. The data corresponding to those in Table 6 are consequently not presented individually. The mean linear univariate coefficients for head width and hind tibia length were β1h=+0.008 and β1t=+0.007, the mean bivariate linear coefficients were β2h=+0.008 and β2t=0, the mean univariate quadratic coefficients were γ1h=−0.002 and γ1t=−0.003, the mean bivariate quadratic coefficients were β2h=−0.012 and β2t=−0.005, and the mean correlational coefficient was γh,t=+0.009. Only the linear univariate coefficients β1h and β1t were significantly different from zero (P < 0.05). β1h was lower than the corresponding differential for the Fehraltorf population obtained in 1994 (Table 4), but they were produced with different fitness estimators.
In conjunction with the seasonal changes in sexual selection (Tables 5 and 6), body size (head width and tibia length) in the field also exhibited a strong seasonal pattern (Fig. 1). This can be primarily related to effects of temperature during development (extracted from meteorological data for the nearest available site, ≈5 km away) and population density, i.e. larval competition (Fig. 2). Interestingly, body size of the individuals sampled also decreased over the day (two-way ANOVA with week removed as a blocking factor; head width: F3,1345=7.13, P < 0.001; hind tibia length: F3,1422=6.44, P < 0.001), and more strongly so for females (sex by time of day interaction for head width: F3,1345=2.60, P=0.034; hind tibia length: F3,1422=1.95, P=0.100; Fig. 1). This must be a behavioural response.
Larval survivorship experiment
Table 7 shows the viability selection differentials resulting from our laboratory experiment. Linear coefficients were significantly positive for males at low (no) competition and for females at high competition. For both environments combined, repeated-measures ANOVA revealed that only the male linear coefficient (β=+0.053) remained significant at P < 0.01 while that for females (β=+0.017) was not. However, in the full (repeated-measures) model including both environments and both sexes all effects were nonsignificant at P > 0.1. Quadratic coefficients were never significant (Table 7).
Table 7. Fitness components (family mean ± SE) and linear, β, and quadratic, γ, larval viability selection differentials on body size for males and females at low and high larval competition. Relative family survivorship (angular transformed; n = 75 families) was regressed on mean family hind tibia length at low larval competition. *P < 0.05; **P < 0.01; ***P < 0.001.
More intense larval competition (i.e. lower per capita dung availability) indeed decreased larval survivorship (one-way ANOVA: F1,73=13.10, P < 0.001; Table 7). There were no systematic differences in mortality between the sexes, as evident in 1:1 offspring sex ratios in both larval environments. Adult body size also decreased when larval competition was high (two-way ANOVA with sex as the second factor: F1,71 = 637.8, P < 0.001) while development time did not (F1,71 = 1.57, P = 0.211).
Overall patterns of selection
Patterns of spatiotemporal variation in viability, fecundity and sexual selection on body size (head width) in three Swiss populations of the dung fly S. cynipsea differed markedly. To summarize briefly, adult viability selection, based on residual physiological survivorship in the laboratory, was nil or weakly negative. In contrast, larval viability selection was weakly positive for males at low and females at high competition. Fecundity selection was positive and strong at all times and in all populations. Sexual selection reflecting pairing success (SexS1) was overall positive, on average three to four times stronger than fecundity selection, but varied significantly at coarse (between populations and seasonally) but not at fine (within a day or between pats on one pasture) spatiotemporal scales. Selection reflecting male reproductive success via the body or clutch size of his mate (i.e. assortative mating; SexS2) was weak in comparison and only apparent in one population (Fehraltorf). Quadratic and correlational selection differentials were low and inconsistent for all episodes except for fecundity selection, which appeared to level off at large body sizes. We first discuss these findings in more detail and then their implications for body size evolution.
Even though we are aware of its many limitations, residual physiological longevity in the laboratory after capture may reflect at least some aspect of survivorship, provided that time of season (i.e. week) and size do not covary systematically. The 1995 study actually showed a clear seasonal pattern of body size variation: individuals were larger at the beginning and the end of the season. This can be largely explained by the differences in temperature and competition the six or more cohorts per season faced during development (Atkinson, 1994; Blanckenhorn, 1997; Morf, 1997). However, the summer decline in body size in 1995 really did not start before July, and a decline was also not evident early in the 1994 season (May and June). This is because all adults present during the first few weeks of the season are over-wintered individuals born in autumn of the previous year (Blanckenhorn, 1998a); the decline in size in early summer most likely coincides with the massive emergence of their first offspring which developed at higher temperatures. It is thus unlikely that our estimates of adult viability selection are strongly confounded by seasonal changes in size, though some offspring probably were present already in late June, thus possibly explaining the increasingly negative selection differentials toward the end of the 1994 sampling period (Table 1). However, we cannot dismiss the possibility that within a cohort the unpaired males were simply the young and inexperienced individuals and therefore lived longer in the laboratory after capture. Furthermore, our laboratory surrogate of adult viability selection does not include mortality due to predation or parasites and thus does not well represent the natural situation but, unfortunately, mark-recapture studies in the field are prohibitive in this small species. With all these caveats in mind, our data indicate that viability selection on body size as defined here was, if anything, only very weakly negative. Large males and females may thus be at a slight disadvantage, which can be interpreted as a consequence of greater environmental stress resulting from more pairing activity of large males and from higher reproductive investment of the larger females.
Larval viability selection was estimated with laboratory-reared F3 individuals to eliminate potential maternal effects, at two environments characterized by high and low intraspecific competition for food. Obviously, these two treatments do not encompass the variability in dung quantity and quality existing in nature. Nonetheless, based on our extensive rearing experience with this species, much lower per capita dung availability than that of the high competition treatment will rapidly reduce larval survivorship to zero. Our low (no) competition treatment estimated the base-line larval survivorship, and our high competition treatment led to a significant increase in larval mortality, as expected. In addition, body size was reduced by intraspecific larval competition but development time was not. Development times of males and females are the same despite the difference in body size, which may explain why there were no systematic differences in mortality between the sexes as a function of treatment (Blanckenhorn, 1997; this study). Somewhat surprisingly, larval viability selection on body size tended to be positive, at least for males at low and females at high competition. We expected the opposite, as greater body size is often associated with longer development time within individuals, which should entail a disadvantage in a rapidly depleting habitat such as dung (Roff, 1980; e.g. Partridge & Fowler, 1993; Blanckenhorn, 1998b). However, the presumed positive correlation between development time and body size apparently does not hold in S. cynipsea and, in retrospect, negates this argument (Blanckenhorn, 1997; this study). Instead, small, heritable differences in larval growth rate between individuals that become magnified during development may confer a competitive advantage to large genotypes and explain our results. This assumes that the base-line mortality that occurred at no competition did not selectively target large genotypes. Should this be so, increased mortality of large genotypes would be intrinsic (e.g. due to pleiotropic genetic effects) and difficult to assess experimentally. The crucial result is that, as was the case for our estimates of adult viability selection, larval viability selection as estimated here appeared weak in comparison to fecundity and sexual selection. We consider it unlikely that our results are strongly affected by two generations of laboratory rearing, which may alter or relax natural selection pressures, as individuals were reared and held in mass containers under seminatural conditions.
Fecundity selection was positive and strong at all times and in all populations, as could be expected for ectotherms (e.g. Wootton, 1979). Of course, clutch size is an imperfect estimator of lifetime reproductive success (Lande & Arnold, 1983; Arnold & Wade, 1984a,b; Endler, 1986). However, given that longevity in this species in the laboratory is highly variable and shows little pattern with regard to body size (Blanckenhorn, 1997; this study), using clutch size as an index of female fitness appears reasonable in this context. A negative nonlinear in combination with a positive linear regression coefficient suggests that clutch size reaches an asymptote at large body sizes, implying weaker selection for large size as females get bigger. Clutch size may be limited by mechanical or loading constraints that could reduce the flight performance and hence survivorship of large females, ultimately limiting female size (Fairbairn, 1990; Berrigan, 1991; Roff, 1992). However, this effect cannot be very strong, as we might expect it to be detectable even in the laboratory (compare above and below).
Based on head width as an index of body size, sexual selection for large males in S. cynipsea was seasonally variable but overall consistent. Throughout the year and in all populations it was positive or nil, but never negative, and our mean estimates for the two years in Fehraltorf were very similar. As already mentioned in the Methods, we did not differentiate between copulating and noncopulating pairs, thus ignoring one of the three selection episodes in the sexual context described in this species. Our sexual selection differentials are thus probably underestimates, as the large male advantage has been shown to occur in all three selection episodes (Ward, 1983; Ward etal., 1992) and the differentials of the three episodes are simply additive because the selection episodes are multiplicative (Arnold & Wade, 1984b). However, we do not know the magnitude of this supposed underestimation.
Implications of spatiotemporal variability in selection
Selection regimes, particularly those of sexual selection, are expected to differ in space and time, and this has important implications for evolutionary change (Lande & Arnold, 1983; Endler, 1986). For example, selection can change sign over the season and thus, on average, have a stabilizing effect (fluctuating-stabilizing selection; Istock, 1981). Various mechanisms can cause seasonal variation in sexual selection: it may be caused by seasonal changes in the availability of mating sites or host plants affecting local competitor density (McLain, 1992; McLain etal., 1993), by fluctuations in population density per se (Nishida, 1994) or by density-dependent female reluctance to mate (‘convenience polyandry’; Parker, 1972a; Thornhill & Alcock, 1983; Arnqvist, 1989; 1992a,b; Rowe, 1992; Rowe etal., 1994). The last of these mechanisms may operate in S. cynipsea (Blanckenhorn etal., unpublished). Of course, temporal changes in selection or in the conditions affecting selection within populations may be merely stochastic.
Our bivariate analysis suggests that head width is positively correlated with the real target of selection while for tibia length selection tends to be in the opposite direction, even though both (as well as other morphometric) measures are generally highly positively correlated in S. cynipsea (Reusch & Blanckenhorn, 1998). Why this is so is unclear at this point, but it implies that not all morphometric measurements can be taken as equally good indices for body size, as selection seems to target them differentially, thus affecting shape as well as size (Fairbairn, 1992). It also implies that the foregoing interpretation of our results has to be understood as evidence for selection on head width (or a correlate thereof) and not necessarily on body size per se. If females can measure male weight during pairing, a possible simple assessment mechanism, a measure directly reflecting overall body volume like head width may reflect this kind of selection better than tibia length, which additionally may be subject to different selection pressures (Fairbairn, 1990; Klingenberg & Zimmermann, 1992). Moreover, if females have an absolute weight threshold, the simplest selection mechanism possible, they should have difficulties choosing among a large number of invariably small males in summer, resulting in the lower sexual selection intensities observed during that time in our 1995 study. However, a multivariate selection study incorporating more morphometric traits (which is under way) is necessary to dissect the various selection pressures that affect size and shape in S. canipsea.
Temporal variation within populations notwithstanding, consistently more intense sexual selection favouring larger males should produce larger mean body sizes over evolutionary time, as fluctuating-stabilizing selection limiting the evolution of larger body sizes does not seem to occur in our populations (Istock, 1981). Sexual selection favouring larger males should also predictably affect the degree of sexual size dimorphism such that in species like S. cynipsea, where males are smaller, the size difference between males and females should decrease and, perhaps, ultimately reverse (Fairbairn & Preziosi, 1994). Our small population comparison indicates that sexual selection indeed tended to be strongest in the population with the greatest mean body size (Fehraltorf) and weakest in the population with the smallest mean body size (Luzern). We are currently extending our data set to rigorously test Fairbairn & Preziosi's (1994) hypothesis. The results of this study are of practical relevance in this context because they imply that sampling any pat at any time on a given day will not affect our population estimate of sexual selection, but when we sample in the season will. The latter may crucially hamper detecting variation in selection between populations if such variation is small compared to the seasonal variation detected here: in the extreme, sampling a population once in summer may yield a selection differential of zero even though, overall, sexual selection occurs in this population.
This work was supported by grant no. 31-40496.94 from the Swiss National Science Foundation. We thank B. Anholt, A. Barbour, D. Fairbairn, R. Preziosi, B. Sutherland, M. Wade, P. Ward and G. Wilkinson for comments and statistical advice. S. Waldmaier and S. Oehen helped gather some data.