Sex-specific selection and sexual size dimorphism in the waterstrider Aquarius remigis


Ferguson Department of Biology, Concordia University, 1455 de Maisonneuve Blvd. W., Montreal, Quebec, Canada H3G 1M8. Tel.: +1 514 848 3397; fax: +1 514 848 2881; e-mail:


We estimated selection on adult body size for two generations in two populations of Aquarius remigis, as part of a long-term study of the adaptive significance of sexual size dimorphism (SSD). Net adult fitness was estimated from the following components: prereproductive survival, daily reproductive success (mating frequency or fecundity), and reproductive lifespan. Standardized selection gradients were estimated for total length and for thorax, abdomen, genital and mesofemur lengths. Although selection was generally weak and showed significant temporal and spatial heterogeneity, patterns were consistent with SSD. Prereproductive survival was strongly influenced by date of eclosion, but size (thorax and genital lengths in females; total and abdomen lengths in males) played a significant secondary role. Sexual selection favoured smaller males with longer external genitalia in one population. Net adult fitness was not significantly related to body size in females, but was negatively related to size (thorax and total length) in males.


Sexual size dimorphism (SSD) is common ( Reiss, 1989, pp. 91–128; Andersson, 1994, p. 252; Fairbairn, 1997) despite the generally high genetic correlation between male and female body size (e.g. Price, 1984; Cowley & Atchley, 1988; Preziosi & Roff, 1998), and therefore provides a useful system in which to study the evolution and maintenance of traits constrained by genetic correlations ( Lande, 1980). Male-biased SSD is often attrib- uted to sexual selection favouring large males ( Darwin, 1890; Selander, 1972, p. 187; Andersson, 1994), while female-biased SSD is widely attributed to fecundity selection favouring large females ( Darwin, 1890; Shine, 1988; Andersson, 1994). Such mechanisms, however, are insufficient to explain the existence of SSD since sexual selection for larger males has been detected in species with female-biased SSD (e.g. Howard, 1988; Fairbairn & Preziosi, 1994, 1996; Shine, 1994), while fecundity selection is a very general pattern ( Darwin, 1890; Clutton-Brock, 1988a, p. 7; Roff, 1992, p. 126; Honěk, 1993) and is neither necessary nor sufficient to explain female-biased SSD ( Arak, 1988; Hedrick & Temeles, 1989). Since SSD is just a difference in body size between the sexes, and body size is correlated with many life history traits ( Peters, 1983; Schmidt-Nielsen, 1984), selection acting at different life history stages (i.e. different episodes of selection) may influence SSD. By extension, we expect differences between the sexes in lifetime selection on body size to result in SSD at equilibrium ( Ralls, 1976; Price, 1984; Clutton-Brock, 1988b; Preziosi & Fairbairn, 1996), unless constrained by genetic correlations ( Charlesworth, 1990).

We report estimates of selection on body size in the waterstrider Aquarius remigis (Hemiptera, Gerridae), and relate contemporary patterns of selection on males and females to the pattern of SSD. Three components of lifetime fitness were measured: survival from eclosion as adult to the reproductive season, daily reproductive success (mating frequency or fecundity), and reproductive lifespan. These were then combined in an estimate of fitness through the entire adult lifespan. This study extends and expands the longitudinal study of selection on body size in A. remigis reported in Preziosi & Fairbairn (1996, 1997, in press). We present analyses of two more generations (1993–95), giving us estimates of selection through four consecutive generations (1991–95) in the same population. Analyses are also presented for two generations (1993–95) in a second population.

Study animal

Adult A. remigis are 10–17 mm long and live on the surfaces of streams and small rivers throughout most of temperate and subtropical North America ( Preziosi & Fairbairn, 1992). As in most insects, females are larger than males, but components of body size display divergent patterns of SSD ( Fairbairn, 1992; Preziosi & Fairbairn, 1996; Table 1).

Table 1.   Patterns of selection and sexual size dimorphism in Aquarius remigis. Dimorphism is reported as mean female size divided by mean male size (data from this study). Patterns of selection are taken from the literature. Thumbnail image of

In southern Quebec, females lay eggs from late April through early July on rocks under water. Young climb to the surface and develop through five nymphal instars. Adults do not become reproductively mature until the following spring, emerging from overwinter diapause as soon as the waters become free of ice. Mortality over the winter is 70–90% ( Fairbairn, 1985a; Blanckenhorn, 1994). Both sexes mate repeatedly with different partners ( Krupa & Sih, 1993; Preziosi & Fairbairn, 1996). During mating, males ride on the back of females, often remaining in copula for many hours ( Wilcox, 1984; Weigensberg & Fairbairn, 1994). The mean reproductive lifespan is about 4 weeks ( Preziosi & Fairbairn, 1997), and all overwintered adults die before midsummer ( Fairbairn, 1985a). Aquarius remigis move poorly overland, are rarely winged (≤3%) and, when winged, seldom fly ( Fairbairn, 1986; Fairbairn & Desranleau, 1987). Even on water, adult movements of greater than 100 m are rare in our study populations ( Fairbairn, 1985b, 1986). This low mobility results in genetic isolation of populations on streams separated by as little as a few hundred metres ( Preziosi & Fairbairn, 1992).

Previous evidence of selection on body size in Aquarius remigis

If SSD in A. remigis is adaptive, selection should differ between the sexes, favouring smaller males than females. Paradoxically, sexual selection favouring large body size in male A. remigis has been found in numerous studies (Table 1). However, such selection was lacking in some samples, and smaller males may occasionally be favoured through higher encounter rates with females ( Krupa & Sih, 1993; Blanckenhorn et al., 1995 ). Multivariate analysis of selection on body components suggests that the general pattern of sexual selection favouring larger total length may actually be due to selection for greater genital length (which makes up about 20% of male total length), with selection on the rest of total length being either neutral or negative ( Preziosi & Fairbairn, 1996). Such a pattern would be consistent with the pattern of SSD, with males having generally smaller bodies but much longer external genitalia than females.

Total length of females has been found to be positively associated with fecundity in a variety of experiments (Table 1). Multivariate analysis has revealed that fecundity selection actually targets abdomen length (Table 1), indicating that larger females lay more eggs because they have larger abdomens. Food availability also influences fecundity, and the influence of body size may be small or absent when food is limited ( Blanckenhorn, 1991a, 1995). Thus, fecundity selection favouring large female body size appears to be general in A. remigis, but is sensitive to food availability.

Body size may also influence SSD through differential prereproductive survival or reproductive longevity (Table 1). However, the lack of strong, general trends in either the pattern or the intensity of selection revealed by these data suggests that selection through differential survival is variable in A. remigis populations.

Preziosi & Fairbairn (in press) have attempted to measure selection through the adult lifetime (hereafter net adult selection) in this species. They found significant stabilizing selection on total length in both sexes (Table 1), which they attribute to negative relationships between total length and reproductive lifespan combined with positive relationships between total length and daily reproductive success. However, absence of these relationships in the second generation of males suggested temporal variance in selection regimes. To assess the generality of the fitness trade-offs suggested in Preziosi & Fairbairn (in press), we have measured selection on body size for two additional generations and in a second population. We have also altered the assay of sexual selection to determine if the results are robust to a change in methodology. Whereas Preziosi & Fairbairn (1996) estimated sexual selection in artificial pools (as did Sih & Krupa, 1992; Kaitala & Dingle, 1993), possibly reducing the influence of encounter rate on mate acquisition, we included encounter rate in our estimates of sexual selection by measuring mating frequency directly on the streams. Sexual selection may be divided into two stages: pre-encounter, when traits that maximize a male’s chances of encountering a female are favoured, and post-encounter, when traits that maximize a male’s chances of mating with the female encountered are favoured. Smaller males may be more efficient at finding females, and thus may be favoured in the first stage ( Blanckenhorn et al., 1995 ). Studies in which mating success is measured in an artificial enclosure or even a small naturally occurring pond, where encounter rates are high, are mainly measuring selection in the second stage. Sexual selection for longer external genitalia in male A. remigis occurs in the second stage, when it has been hypothesized that males with longer external genitalia may be better able to subdue females ( Preziosi & Fairbairn, 1996). If the first stage is important, then studying mating activity in a pool will provide an incomplete description of sexual selection.


We studied A. remigis on two streams at the McGill University Research Station on Mont St.-Hilaire, Quebec, Canada, about 35 km SE of Montreal ( Fig. 1), which support genetically distinct populations of A. remigis ( Preziosi & Fairbairn, 1992). The study area on South Creek was identical to that used by Preziosi & Fairbairn (1996), including an upper recapture-only area (30 m long), a central mark–recapture area (100 m) and a lower recapture-only area (100 m). On West Creek, the areas are 50, 200 and 100 m long, respectively. On both streams the study areas are bounded by barriers to dispersal such as waterfalls. Since A. remigis on these streams rarely move more than 100 m from where first captured ( Fairbairn, 1985b), any individuals marked in the central mark–recapture areas are unlikely to move out of the study areas. Nevertheless, 100–200 m beyond the study areas were searched monthly for marked individuals. Fewer than 1% of the marked individuals were ever observed outside of the study area, and thus death and dispersal are unlikely to be confounded in this study.

Figure 1.

 Study site on Mont St.-Hilaire, Quebec, Canada. Study areas included the extents of the creeks indicated, not including tributaries (which had few or no waterstriders).

The study areas were carefully searched weekly for A. remigis, and the number on each marked adult was recorded. All unmarked adults found in the mark–recapture areas were captured using hand nets, sexed, photographed in a ventral aspect, and marked with a unique number on the dorsal surface using enamel paint (see Preziosi & Fairbairn, 1996, 1997). These marks have no detectable effects on survival or mating success ( Wheelwright & Wilkinson, 1985; Butler, 1987; I. M. Ferguson, unpublished data). For each individual, total length, genital length, abdomen length, thorax length and (mean) mesofemur length were measured from the photographic negatives using a computerized digitizing system with MTV software ( Updegraff, 1990). The definitions of these components are detailed in Preziosi & Fairbairn (1996, 1997). Repeatabilities for these measurements are all at least 0.97 (R. F. Preziosi, unpublished data).

Adults were followed through two complete generations. Sampling continued as long as adults were active on the water surface: from 15 July to 19 October in 1993, from 27 April to 12 October in 1994, and from 12 April to 18 July in 1995. Catchabilities ( Krebs, 1989) in the reproductive season ranged from 67% to 91% with a mean of 79%, indicating that the mean probability of not capturing an individual known to be alive was 21% for 1 week, 4.4% for 2 weeks and 0.92% for 3 weeks.

Assessing components of fitness

Sexual dimorphism in total length does not occur in this species until the last nymphal instar (V. Simoneau, unpublished data), and since development time does not differ between the sexes ( Fairbairn, 1990), selection during the immature stages is unlikely to influence adult SSD. Thus, we have restricted our study to selection on adults.

Prereproductive survival (survival from eclosion to the spring reproductive season) was recorded as 0 ( =  did not survive) or 1 ( =  survived) for all individuals that were marked before the winter. Reproductive longevity of the surviving individuals was then calculated as:

inline image

where DL is the Julian date on which the individual was last captured and DF is the Julian date on which it was first captured after diapause.

To measure daily fecundity, buckets with screened lids, holes in the bottoms and a carefully cleaned rock for oviposition were placed in the streams so that water partially covered the rocks. Marked females were collected twice from each stream each year (on 15 June and 6 July 1994, and on 24 May and 14 June 1995) and held individually in the buckets for 2 days. Daily fecundity was estimated as the mean number of eggs laid per day for each individual over the 2-day trial.

The calculation of fecundity was straightforward in the first generation (1993–94), but in the second generation (1994–95) the mean fecundity differed between trials on both streams. To control for these differences, all fecundities in the second generation were converted to relative fecundity within each trial before the trials were combined for analysis.

To assess male mating success, we recorded the mating status (mating or single) of all males seen during the mark–recapture sampling. The study areas were also searched for mating males on one (1994) or two (1995) additional days each week. Mating frequency was then calculated for each marked male as the proportion of times that a male was found mating during the reproductive season.

Net adult fitness was calculated as prereproductive survival × reproductive lifespan × daily reproductive success (fecundity for females, mating frequency for males). While estimates of fitness for each episode were made for all individuals present for each episode, net adult fitness was only estimated for those individuals marked the previous fall. Although the net adult fitness of females is an estimate of lifetime fecundity, that of males can only be interpreted as an index of lifetime fitness. Mating status was assessed only once in any one field day, providing an ‘instantaneous’ estimate, rather than a true count of number of matings per day. Multiplying this by reproductive lifespan therefore gives a good index of relative mating success, but does not translate into total number of matings in a lifetime.

Female fecundity was determined for only a proportion of those surviving to the reproductive season. To avoid exaggerating the effects of prereproductive mortality, net adult fitness for females was calculated using the same proportion of the females that did not survive the winter. For example, on West Creek in 1994 fecundity was measured for 17 females out of 57 surviving from those marked in 1993 (29.8%). We therefore randomly selected 29.8% of the 152 females that did not survive the winter (i.e. 45) to include in the analysis of net adult selection, giving us a sample of 62. The random selection of nonsurviving females and subsequent analysis was repeated 10 times for each sample, and we report the mean coefficients and probabilities.

Statistical analysis

Date of eclosion (hereafter DOE: Julian date on which the individual ecloses from the last nymphal instar to the adult) is negatively correlated with total length in this species ( Blanckenhorn, 1994; this study, Table 2): waterstriders eclosing later in the year are significantly smaller. If DOE directly influences any component of fitness, its correlation with body size might result in a spurious correlation between body size and fitness. Therefore, DOE was included as an independent variable in the regression models to ensure that any selection on body size that was detected would be independent of DOE (see Mitchell-Olds & Shaw, 1987; Wade & Kalisz, 1990). For sexual and fecundity selection, DOE is not available for many of the individuals included in the data set. However, regression analysis using the subset of data for which DOE is available revealed no significant influence of DOE on the relationships between body size and male mating success or female fecundity. In the analysis of net adult fitness, DOE was available for all individuals and was included in the analyses.

Table 2.   Correlations between Julian date of eclosion and total length in Aquarius remigis at Mont St. Hilaire. Thumbnail image of

We assessed the relationships between our estimates of components of fitness and body size using multivariate regression techniques ( Lande & Arnold, 1983). Within each population and generation the estimates of fitness were converted to relative fitness (wi′ = wi/ where wi′ is relative fitness, wi is absolute fitness, and is mean absolute fitness), and each trait (body size component or date of eclosion) was standardized to a mean of 0 and a standard deviation of 1 {zi = (xi – )/sx where zi is the standardized trait value, xi is the (unstandardized) trait value, is the mean trait value and sx is the standard deviation of the trait values}. Relative fitness was regressed on standardized traits in four different models for each episode of selection. The linear model for total length included standardized total length and DOE, while the full model included the linear terms plus all possible quadratic terms: (standardized total length)2, (standardized DOE)2, and (standardized total length) × (standardized DOE). To identify possible selection on different components of body size, relative fitness was also regressed on the standardized lengths of the external genitalia (‘genital length’), abdomen, thorax and (mean) mesofemora; and DOE (both the linear and full models were estimated). The linear regression coefficients from the linear models (referred to as linear selection gradients, the ‘independent’β′ of Koenig et al., 1991 ) and the quadratic coefficients from the full models (referred to as quadratic selection gradients, γ) estimate selection on each trait, independent of selection on any other traits included in the model, for each episode of selection ( Lande & Arnold, 1983; Endler, 1986).

Both linear and quadratic selection gradients for DOE were estimated, but DOE was included only as a control variable, and this paper is concerned with sex-specific selection on body size. Therefore, selection gradients for DOE are not reported. Within each population, generation and sex we estimated five linear and 11 quadratic selection gradients on body size for each episode of selection and for net adult selection. The number of estimates leads to two problems: first, after Bonferroni correction for multiple tests there remains little power to detect significant selection; second, the large number of gradients generated makes interpretation difficult (5 + 11 = 16 gradients × 2 sexes × 2 populations × 2 generations × 4 episodes of selection [including net adult] equals 512 gradients). We therefore simplified the analysis by combining the standardized data from the different populations/generations before further analysis. This allowed us to test hypotheses about general patterns of selection, rather than estimating selection gradients within each population/generation. Only the linear selection gradients for net adult selection, for each sex in each population and generation, are reported in Table 3.

Table 3.   Standardized multivariate selection gradients for net adult selection on body size, with DOE in the model. Standard errors are in parentheses. n = sample size. Thumbnail image of

We tested for differences in selection gradients between samples (spatial and temporal heterogeneity in the fitness functions), so that we could avoid combining samples with different patterns of selection. This was accomplished by regressing relative fitness on standardized total length and DOE using the combined data set, then using partial F-tests ( Neter et al., 1985 , p. 281) to test whether the addition of the interactions between standardized total length and population or generation (included as indicator or ‘dummy’ variables as per Neter et al., 1985 , p. 328) improved the model (see Mitchell-Olds & Bergelson, 1990). The main effects of these dummy variables are always nil because traits were standardized within each population/generation. If an interaction was significant (indicating heterogeneity in the fitness function), the data were split as appropriate and the analysis repeated separately for each population/generation for all subsequent analyses in that episode of selection.

Patterns of selection on total length were estimated using the linear model (with standardized DOE and total length) and the full model (including all linear and quadratic terms) ( Lande & Arnold, 1983; Phillips & Arnold, 1989). If one of these models was found to be significant (F-test), then there must have been a significant relationship between fitness and at least one of the independent variables included in the model ( Neter et al., 1985 , p. 289). Stepwise regression (stepwise, forward selection, and backward selection; Neter et al., 1985 , p. 430) was used to reduce that model to the significant variable(s) (see Mitchell-Olds & Bergelson, 1990). Similarly, the linear and full models with standardized genital, abdomen, thorax, and mean mesofemur lengths, and DOE, were estimated and reduced by stepwise regression where significant. Because Preziosi & Fairbairn (1996) found that sexual selection on male genital and total lengths may be antagonistic, we also regressed relative mating frequency on standardized genital length and ‘pregenital body length’ (total length – genital length). In all cases, stepwise, forward selection, and backward selection techniques produced the same reduced models. The statistical software SPSS 8.0 ( SPSS Inc., 1997) was used to calculate all regression models.

The residuals from these regression models were not normally distributed, and therefore we confirmed significances using the program RT 1.02 ( Manly, 1992) to randomize the dependent variable (fitness) 9999 times. The randomization results were nearly identical to the parametric regression results (r2 = 0.99966 for probabilities generated by the two methods).


Through the entire study, a total of 1165 males and 1315 females were marked before the reproductive seasons. Prereproductive survival differed significantly between the sexes in the second generation on West Creek (males 10.7%, females 23.3%, χ2 = 7.18, d.f. = 1, P = 0.007), but this did not represent a general trend: overall 17% of males and 19% of females survived to the reproductive season (χ2 = 1.54, P = 0.12). Similarly, reproductive longevities estimated for the 194 males and 244 females that survived were not significantly different (male  = 26.12 days, female  = 26.40 days, U = 23522, P = 0.91). Mean daily fecundity estimated for 179 females was 4.98 eggs/day, while mean mating frequency estimated for 588 males was 23.7%. Factorial comparisons including population and generation indicated that males and females did not differ significantly in either DOE (date of eclosion to adult, F = 3.05, d.f. = 1, 2472, P = 0.081), or date of emergence from winter diapause (F = 0.002, d.f. = 1, 430, P = 0.96).

Male net adult fitness

Selection generally favoured smaller males in this study. Three of the four linear selection gradients for total length were negative, though none was significant (Table 3). The interactions between standardized total length and population or generation were not significant (F = 1.15, d.f. = 3, 1159, P = 0.33), indicating no significant temporal or spatial heterogeneity in selection through net adult fitness. The linear model for total length using the combined data set was significant (F =  4.83, d.f. = 2, 1162, P = 0.008), and stepwise regression reduced the model to [fitness  =  1 – 0.33(total)]. Therefore, there was a general pattern of selection favouring smaller males. Analysis of the components of body size also revealed significant selection for smaller size. The linear model was significant (F = 2.36, d.f. = 5, 1159, P = 0.038), and reduced by stepwise regression to [fitness  = 1 – 0.34(thorax)], suggesting that selection favouring smaller males may have been targeted at thorax length. The selection gradients in Table 3 are consistent with this result, the strongest pattern being negative gradients for thorax length in all four samples.

Male prereproductive survival

There was a significant interaction between standardized total length and generation (F = 4.48, d.f. = 1, 1161, P = 0.034), indicating temporal heterogeneity in the relationship between male prereproductive survival and total length. The two generations were therefore analysed separately. In the second generation the linear model for total length was highly significant (F = 12.00, d.f. = 2, 545, P << 0.001). The model reduced by stepwise regression was [fitness  =  1 + 0.30(DOE) – 0.24(total)], indicating that males with greater DOE (eclosed to adult later) and males with shorter total length had significantly higher survival ( Fig. 2). The same trend was observed in the first generation, but it was not significant {F = 0.73, d.f. = 2, 614, P = 0.48 [fitness = 1 + 0.063(DOE) – 0.072(total)]}. Therefore, selection favoured later eclosion and reduced total length, these trends being much stronger in the second generation.

Figure 2 Contour plot of estimated selection on male standardized total length (TOTAL) and Julian date of eclosion to adult (DOE) through prereproductive survival in Aquarius remigis at Mont St‐Hilaire, Quebec, in 1994–95. Fitness is estimated using the empirically derived equation: [prereproductive survival = 1 + 0.30(DOE) – 0.2.

Figure 2 Contour plot of estimated selection on male standardized total length (TOTAL) and Julian date of eclosion to adult (DOE) through prereproductive survival in Aquarius remigis at Mont St-Hilaire, Quebec, in 1994–95. Fitness is estimated using the empirically derived equation: [prereproductive survival = 1 + 0.30(DOE) – 0.2.

4(TOTAL)]. Darker shades of grey indicate higher survival.

The full model with components of body size was also significant in the second generation (F = 1.73, d.f. =  20, 527, P = 0.025), reducing to [fitness =  0.92 + 0.41(DOE) – 0.21(abdomen × DOE)] by stepwise regression. The interaction between abdomen length and DOE is difficult to interpret because of the strong direct effect of DOE. A contour plot based on the regression equation ( Fig. 3, right panel) indicates that fitness is maximized if males eclose late and have small abdomens. However, the effect of abdomen length depends on DOE such that long abdomens actually appear to be favoured among males eclosing early, resulting in minimal fitness for males eclosing early and having small abdomens. Perhaps the most striking aspect of the regression solution is the suggestion that the relationship between DOE and prepreproductive survival is contingent upon abdomen size, being virtually absent among males with large abdomens.

Figure 3.

 Contour plots of estimated correlational selection on components of body size and Julian date of eclosion to adult (DOE) through prereproductive survival in Aquarius remigis at Mont St-Hilaire, Quebec. Correlational selection for females (left panel) is estimated using the empirically derived equation [prereproductive survival  = 1 + 0.22(DOE) + 0.12(THORAX × DOE)]. Correlational selection for males in 1994–95 (right panel) is estimated from the empirically derived equation [prereproductive survival  =  0.92 + 0.41(DOE) – 0.21(ABDOMEN × DOE)]. The traits have been standardized. Darker shades of grey indicate higher survival.

Male reproductive lifespan and mating frequency

We found no evidence of any relationship between body size and reproductive lifespan (all models with P > 0.10). In the analysis of mating frequency, there was a significant interaction between standardized total length and population (F = 5.53, d.f. = 1, 585, P = 0.019), and thus the two populations were analysed separately. On West Creek there was no significant relationship between body size and mating frequency (all P > 0.16). On South Creek, the models with standardized total length were not quite significant, but the linear model suggested a trend favouring shorter males [fitness = 1 – 0.092(total), F = 3.52, d.f. = 1, 396, P = 0.061]. The components of body size (genital, abdomen, thorax and mesofemoral lengths) significantly predicted mating frequency (F = 1.88, d.f. = 14, 383, P = 0.027). Reduction by stepwise regression revealed that sexual selection favoured males with shorter thoraxes and longer external genitalia [fitness = 1 – 0.18(thorax) + 0.15(genital)]. The model with genital and pregenital lengths is a better predictor of fitness (F = 3.76, d.f. = 5, 392, P = 0.0024): shorter males with longer external genitalia were favoured [fitness = 1 – 0.18(pregenital) + 0.15(genital)]. Thus, significant sexual selection on South Creek favoured smaller males with larger external genitalia.

Female net adult fitness

Because daily fecundity varied with time in the second generation, our estimates of daily fecundity could not be meaningfully multiplied by reproductive lifespan as an estimate of lifetime fecundity. Therefore, net adult fitness in females was only estimated for the first generation. Both linear selection gradients for total length are positive in that generation, though one is very small and neither is significant (Table 3). Even after combining the data for the two populations we found no significant selection on total length or any component of body size (all P > 0.32).

Female prereproductive survival

The linear model with standardized total length and DOE was significant (F = 6.34, d.f. = 2, 1312, P = 0.0018), but stepwise regression reduced the model to [fitness = 1 + 0.20(DOE)], suggesting that total length was unimportant. To confirm this, we regressed relative survival on standardized total length by itself: this model was not significant (F = 1.03, d.f. = 1, 1313, P = 0.31), while the addition of standardized DOE significantly improved it (F = 11.63, d.f. = 1, 1312, P < 0.001), and therefore there was no significant selection on total length.

However, the linear model containing components of body size was significant (F = 3.82, d.f. = 5, 1309, P =  0.0019), reducing to [fitness = 1 + 0.21(DOE) + 0.14(genital)] after stepwise regression. Therefore, females with longer external genitalia were significantly more likely to survive. The full model was also significant (F = 1.67, d.f.  = 20, 1297, P = 0.032), and reduced to [fitness =  1 + 0.22(DOE) + 0.14(genital) + 0.12(thorax ×DOE)]. The interaction between standardized thorax length and DOE indicates that the effect of thorax length depends upon DOE ( Fig. 3, left panel): thorax length is positively related to prereproductive survival for females that eclose late, but this relationship is negative among early eclosing females. In spite of this interaction, the parametric solution suggests that fitness is maximized for late-eclosing females with long thoraces.

Female reproductive lifespan and fecundity

As in males, no significant selection through reproductive lifespan was found (all P > 0.39). Fecundity was also independent of total length or any component of body size (all P > 0.23), and therefore female body size appears to have been selectively neutral during the reproductive season.


Using fitness measured through the entire adult lifetime, we found that selection on total length differed between males and females in a manner consistent with the SSD: mean male total length is smaller than that of females, and significant selection favoured smaller male total length. In contrast, though selection was not significant on females, the selection coefficients for female total length were positive (Table 3). This suggests that we are not merely missing selection for smaller females due to lack of statistical power. These results are consistent with optimum male size being smaller than optimum female size, a conclusion that Preziosi & Fairbairn (in press) arrived at for one of our populations through the two generations immediately preceding this study.

Surprisingly, the difference between the sexes in net adult fitness functions for total length did not result primarily from differences in the reproductive season. The only significant selection on female body size was found in the prereproductive survival episode, including selection favouring longer external genitalia. The same selection on female external genitalia was significant in one of two generations studied by Preziosi & Fairbairn (in press); however, the external genitalia of females make up only about 5% of the total length ( Fairbairn, 1992), and therefore this selection did not significantly influence total length in either study. The only episode with significant selection on male total length was prereproductive survival: smaller males were favoured in the second generation, and the trend was the same in the first generation, though not significant. Although a nonsignificant trend favouring smaller male total length through mating frequency on South Creek was found, the significant selection through prereproductive survival in one generation, coupled with the high mortality before the reproductive season (83%), suggests that the selection through net adult fitness favouring smaller males was due primarily to selection acting in the prereproductive episode.

Although prereproductive survival has generally been found to be independent of total length in A. remigis ( Blanckenhorn, 1994; Preziosi & Fairbairn, in press), some samples have shown significant selection for larger female total length ( Blanckenhorn, 1994), or nonlinear selection on male total length ( Preziosi & Fairbairn, in press). There are few life history differences between the sexes before the reproductive season that might be responsible for differences in the fitness functions. Males and females occupy the same habitat ( Rubenstein, 1984; Fairbairn & Brassard, 1988; Krupa & Sih, 1993) and are similar in development time ( Fairbairn, 1990), DOE (this study), patterns of movement ( Fairbairn, 1985b; Fairbairn & Brassard, 1988), feeding behaviour ( Blanckenhorn, 1991b) and adult prereproductive survival (this study). Galbraith & Fernando (1977) found that females emerged from diapause earlier than males in a small stream in southern Ontario, but no significant difference between the sexes was found in this study. Female bodies contain more lipids, both absolutely and as a proportion of body weight ( Lee et al., 1975 ), suggesting physiological differences between the sexes. However, how this might contribute to the apparent difference in selection on size in males and females during the prereproductive season is unclear.

The only significant quadratic selection in this study was correlational selection found in the prereproductive episode: for lower (abdomen length × DOE) in the second generation of males and higher (thorax length × DOE) in females. Correlational selection on body size in A. remigis has generally been found to be weak and nonsignificant ( Fairbairn & Preziosi, 1994, 1996), and for this reason correlational gradients are sometimes not reported (e.g. Preziosi & Fairbairn, in press). However, DOE has not previously been included in these analyses, and therefore the generality of the patterns found in this study is impossible to assess. Nevertheless, the difference between the sexes reported here is consistent with the hypothesis that males and females experienced different selective pressures through prereproductive survival.

We expected our estimate of mating success, which is sensitive to encounter rate, to yield an estimate of sexual selection that favoured smaller males, or at least favoured larger males less than in previous studies, and that is what we found on South Creek. The target of selection for larger males in previous studies, genital length, was also found to be under sexual selection for greater size in this population. At the same time, selection strongly favoured smaller pregenital body length, resulting in weak net selection for reduced total length. Such a result is consistent with the female-biased SSD in total length and pregenital length, as well as the male-biased SSD in genital length. The tendency for smaller males to have more opportunities to mate ( Krupa & Sih, 1993; Blanckenhorn et al., 1995 ) could account for the advantage that smaller males appeared to have in acquiring mates on South Creek in this study. If small males are better at finding mates, small size is likely to be favoured when the density of the population is low and food is limiting ( Ghiselin, 1974; Blanckenhorn et al., 1995 ). This sensitivity to population density and food level means that the balance between selection for longer genitals and for smaller overall size in A. remigis is expected to vary over space and time. This may account for some of the previously reported variation in direction and intensity of sexual selection in this species ( Fairbairn, 1988; Kaitala & Dingle, 1993; Fairbairn & Preziosi, 1994, 1996; Sih & Krupa, 1995; Preziosi & Fairbairn, 1996).

Our results demonstrate spatial and temporal heterogeneity in fitness functions, for mating frequency and prereproductive survival. In males, patterns of sexual selection differed significantly between the two populations, and selection through prereproductive survival differed between generations. Our results also differ from those of Preziosi & Fairbairn (in press), who found net adult stabilizing selection on total length in both sexes on South Creek through one of two generations, attributed to opposing selection pressures through different episodes of selection. Preziosi & Fairbairn (in press) report significant net adult stabilizing selection only on female total length in their other generation, while the pattern of opposing selection pressures producing net adult stabilizing selection was lacking in males. We found no evidence of opposing selection pressures across episodes, or of stabilizing selection.

Although the differences between our study and that of Preziosi & Fairbairn (in press) suggest temporal heterogeneity of fitness functions on South Creek, these apparent differences could arise from either methodological differences between the studies or actual shifts in the body size distributions (rather than in the fitness functions) among generations. The methodological differences between the studies may explain the stronger selection for smaller pregenital body size detected in the present study, as our assay of sexual selection was designed to include selection through encounter rate. Differences in methodology might also account for some differences between the studies in estimates of the other components of reproductive fitness: we measured both daily fecundity and reproductive lifespan with less precision (i.e. fecundity over 2 rather than 3 days, and lifespan at weekly rather than twice-weekly intervals). This reduction in precision would be expected to increase the error variance of our estimates, and thus reduce the probability of detecting statistically significant selection. It may, for example, account for the lack of significant fecundity selection favouring longer abdomens on South Creek in our study. However, differences in methodology cannot explain our finding of significant directional selection through prereproductive survival, which Preziosi & Fairbairn (in press) did not find, because this component of fitness was estimated in exactly the same way in the two studies. Thus, methodological differences are insufficient to explain all the observed temporal variance in fitness functions.

The second alternative is that the underlying fitness functions remained relatively constant over the four generations, but the distributions of body sizes changed between generations. For example, if the mean total length was near the optimum in one generation, and much larger the next, selection could appear to be stabilizing in the first generation, but directional (favouring smaller size) in the second. If this was responsible for the differences between our results and those of Preziosi & Fairbairn (in press), we would expect the mean total lengths in our study to be higher in males and lower in females than in Preziosi & Fairbairn (in press). This is not the case (Table 4). Therefore, the most plausible explanation for the differences between studies remains temporal variance in fitness functions, particularly for prereproductive survival. Temporal and spatial variance in fitness functions has been reported for other taxa ( Grant & Grant, 1989; Arnqvist, 1992; Endler & Houde, 1995; Blanckenhorn et al., 1999 ), and is doubtless typical for selection acting in local populations.

Table 4.   Comparison of patterns of net adult selection with means for total length. All data are from Aquarius remigis on South Creek at Mont St.-Hilaire, Canada. Thumbnail image of

In conclusion, although spatial and temporal heterogeneity in fitness functions was observed, we found net adult selection generally favouring greater SSD, caused mainly by differences between the sexes in the fitness functions for prereproductive survival. However, the results of our studies of multivariate selection in this species indicate that selection acts antagonistically on different body size components, and therefore estimates of selection on total length can give misleading results. Patterns of selection on the main components of total length (genital, abdomen and thorax length) are consistent with the dimorphisms in those components, while selection on total length is often weak and sometimes inconsistent with the SSD (Table 1). Thus, fitness in A. remigis may be relatively insensitive to overall body size (total length). Our results suggest, instead, that variation in the pattern and intensity of selection on total length observed in this and previous studies primarily reflects interactions between phenotypic distributions of components of body size and locally variable fitness functions.


We wish to thank J. Blicker, K. Brennan-Alpert, G. Gentile, A. Rigler, A. Savard, C. Schluter, E. Abouheif, J. Brennan, S. Gallant, B. MacDonald, R. Preziosi, J. Reeve and D. Yadlowski for assistance with various technical aspects of this study. Gratitude is also due J. Grant, R. Preziosi, and J. Reeve for providing useful comments on the manuscript. We wish especially to acknowledge R. Preziosi and J. Reeve for many helpful discussions. This research was supported by a Natural Sciences and Engineering Research Council grant to D.J.F.