SEARCH

SEARCH BY CITATION

Keywords:

  • courtship feeding;
  • Empididae;
  • male choice;
  • ornamentation;
  • Rhamphomyia;
  • sex-role reversal;
  • stabilizing selection;
  • trade-offs

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

Ornamental traits function by improving attractiveness and are generally presumed to experience directional selection for mating success. However, given the greater investment of females in offspring than males, female-specific ornaments can in theory signal fecundity yet be constrained by fecundity costs. Theoretical work predicts that such constraints can lead to stabilizing selection via male choice for intermediately ornamented females. Female dance flies Rhamphomyia longicauda (Diptera: Empididae) display two female-specific ornaments in mating swarms – inflatable abdominal sacs and pinnate tibial scales. We investigated the intensity and form of sexual selection on female traits including ornaments and found no evidence for directional sexual selection. Instead, we found marginally nonsignificant quadratic selection for all three measures of ornament expression. Canonical analysis confirmed that the strongest vectors of nonlinear selection were associated with ornamental traits, although the significance of the quadratic coefficients associated with these vectors depended on the statistical approach. Direct Mitchell-Olds and Shaw tests for the location of the maximum fitted fitness value for both raw morphological traits and canonical axes revealed only one marginally nonsignificant result for the multivariate axis loading most heavily on pinnate leg scales. Together, these results provide the first tentative support for stabilizing selection on female-specific ornaments.


Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

Elaborate ornaments, such as showy morphological or behavioural traits, can arise via sexual selection if they increase the reproductive success of the bearer during contest competition for mates or make the bearer more attractive to mates (Darwin, 1871). As producing eggs is costly and males are usually readily available, the number of gametes a female can produce usually plays a larger role in determining her reproductive success than access to mates (Trivers, 1972). However, sexual selection on females can arise if males are in short supply (i.e. if the operational sex ratio, is biased towards males), which may arise when mating confers certain kinds of costs to males (Parker, 1983; Gwynne, 1991; Owens & Thompson, 1994; Kokko & Monaghan, 2001). Costs to males can come with the effort in producing or obtaining nuptial gifts (Simmons, 1990) or performing long courtship or copulatory behaviours (Saeki et al., 2005), as well as increased male mortality from these activities or from other sources (Jiggins et al., 2000). In such cases, males may engage in mate choice for the highest quality females. Females may exhibit variation in quality through variation in fecundity (Kvarnemo & Simmons, 1999), stage of egg development (Funk & Tallamy, 2000) or access to resources (Heinsohn, 2008). Males of various taxa have demonstrated mating preferences for an indicator of fecundity or for body size, which usually correlates with fecundity (review in Bonduriansky, 2001).

Sexual selection on females is reasonably common (Clutton-Brock, 2007), but examples of sex-specific female ornamentation are relatively scarce (Amundsen, 2000; Funk & Tallamy, 2000; Amundsen & Forsgren, 2001). Amundsen (2000) outlined three hypotheses to explain female ornamentation in general. In the first hypothesis, sexual selection is not operating directly on females, but female ornamentation is an artefact of selection for male ornament genes expressed in females. However, this hypothesis does not hold in species where the ornament is only expressed in females. In such cases, ornament expression is probably the result of direct selection on females for ornaments that function either to attract mates (e.g. female-specific bright-yellow bellies in two-spotted gobies Gobiusculus flavescens: Amundsen & Forsgren, 2001) or in sexual or social competition with other females. An example of the latter is the female-specific bright blue and red coloration in a parrot, Eclectus roratus, which is important in female–female competition for scarce nesting hollows (Heinsohn et al., 2005; Heinsohn, 2008).

If ornaments evolved in the context of mate attraction, one might predict linear selection for large ornament expression as is often reported for male sexually selected traits. However, investment in the expression of female ornaments could come at a cost to gamete production and therefore her reproductive success (Fitzpatrick et al., 1995). This potential fecundity cost of ornaments could undermine their signalling value to males; males should maximize reproductive success by preferring females with smaller ornaments if these females are more fecund (Fitzpatrick et al., 1995). However, if males cannot directly assess female fecundity and must rely on an indirect fecundity indicator, male mate choice for female ornaments might be adaptive in spite of any fecundity costs (Chenoweth et al., 2006). In such systems, females with a low level of ornament expression may suffer from a lack of mating success, whereas females with a high expression would suffer low fecundity (from both direct selection against such overinvestment and presumably being avoided by choosy males). Chenoweth et al. (2006) therefore predict stabilizing selection on ornament expression in species with showy female ornaments under some conditions, a notion that has been supported in Drosophila serrata (Chenoweth et al., 2007; Rundle & Chenoweth, 2010). In contrast, male preferences for colour in many females appears to be directional (see e.g. Griggio et al., 2009; reviewed in Amundsen, 2000).

Empidine flies (Diptera: Empididae) include species exhibiting a diversity of elaborate female ornamentation – including modified wings, abdomens and legs (Cumming, 1994; Svensson, 1997). Rhamphomyia longicauda is one of the most ornamented species in this group with two female-specific ornaments that are displayed in mating swarms at dusk and dawn. First, females inflate eversible pleural sacs by swallowing air just prior to swarming. The result is an inflated abdomen three to four times wider than that of an uninflated abdomen (Funk & Tallamy, 2000). Females also have pinnate leg scales, which render their legs much thicker in appearance than males. While displaying in courtship swarms, females position their legs alongside their inflated abdomens, greatly exaggerating the size of their body distal to the thorax (Newkirk, 1970). Males enter the swarms from below and appear to assess the females’ silhouettes against the twilight above, showing a preference for larger silhouettes (Funk & Tallamy, 2000). Although males may differentiate between the two ornaments, it is possible that, in such low-light conditions, these ornaments may be perceived as a single trait and have a combined effect of exaggerating the size of the female silhouette.

Several factors may have led to reversed mating roles observed in this species, that is, male mate choice and female–female competition. Female R. longicauda vary greatly in the maturity of their eggs, a potential indicator of female quality; even late in the season, all stages of egg development are present within swarming females (Funk & Tallamy, 2000; J. Wheeler, unpublished data). Male R. longicauda and other empidines also present females with a nuptial gift of insect prey prior to copulating. Females in some species appear to have lost the ability to hunt and must rely upon gift nutrients to develop their eggs to maturity (Downes, 1970). The restricted (swarming) time interval when nuptial gifts are available, the variance in female mate quality (egg maturity) and the presumed male effort in obtaining nuptial gifts may thus all contribute to the apparent reversal of the mating roles of this species and thus sexual selection on females (Funk & Tallamy, 2000).

Multivariate selection analyses provide standardized measures of the partial correlation between different aspects of the phenotype and fitness and therefore allow comparative analyses of how morphological diversity covaries with estimates of selection in wild populations (Arnold & Wade, 1984a,b; Brodie et al., 1995). Such measures are a valuable complementary approach to manipulative experiments such as those conducted by Funk & Tallamy (2000) on R. longicauda, because they allow comparisons of selection on traits differing in scale both within and among species. Bussière et al. (2008) conducted selection analyses comparing wild-caught mated and unmated R. longicauda of both sexes and found significant linear selection for larger wings and smaller tibiae in females. However, they did not include ornament measures in their selection analyses. Although larger wings may be a part of female ornamentation in this species (wings appear to be under sexual selection in another dance fly, Empis borealis: Svensson & Petersson, 1987, 1988), it is imperative to perform analyses incorporating the abdominal and leg ornaments to fully evaluate sexual selection on females in R. longicauda.

This study aims to assess the form of multivariate sexual selection for mating success among females in a natural population of R. longicauda. We predict one of two patterns will emerge: either positive directional sexual selection for ornaments, as is found for most male sexually selected traits and for female tibial scale ornaments in the congeneric dance fly Rhamphomyia tarsata (LeBas et al., 2003), or stabilizing selection on ornament expression, with individuals that show intermediate levels of expression achieving the highest probability of mating (Chenoweth et al., 2006).

Methods

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

Biology of R. longicauda

Rhamphomyia longicauda gather in mating swarms from early June to early July. Swarms form under gaps in the tree canopy at dusk and dawn only, usually next to rivers and streams. Although the sex ratio of adults in the population is unity (Gwynne & Bussière, 2002), swarms are comprised mainly of females (on average, the swarm is composed of 88% females; Gwynne et al., 2007), presumably because females are usually eager to forage on nuptial gifts, whereas only a fraction of males can provide one at any given time. Nevertheless, males still demonstrate mating preferences regardless of the adult sex ratio (Wheeler, 2008). Once a male has ascended through the swarm and copulation occurs, the nuptial gift is transferred between the mating pair and they leave the main swarm. Unlike many other empidines (e.g. Svensson & Petersson, 1987; Preston-Mafham, 1999; LeBas et al., 2003, 2004; Daugeron & Grooterat, 2005), copulation occurs ‘on the wing’ so that inflated mating females remain in flight while consuming the nuptial gift.

Sample collection

Samples were collected from 11th of June to 3rd of July in 2007. The collection site is located on the banks of the Credit River, near Glen Williams (Halton Co., Ontario, Canada: 43°41′117′′N, 79°55′34′′W). The site has been used in a number of previous studies on this system (Gwynne & Bussière, 2002; Gwynne et al., 2007; Bussière et al., 2008). At each swarming event, mating pairs were caught in a 15 × 12 cm dipping net and immediately transferred to a plastic test tube. The pair was then doused with 100% ethanol super-cooled in dry ice and immediately placed in a cooler filled with dry ice. A single female from the swarm was simultaneously caught and treated in the same manner. The mating female was recorded as a ‘paired’ female (N = 115), and the singleton female netted from the swarm as an ‘unpaired’ female (N = 149). As some samples were damaged during collection or processing, there are unequal numbers of paired and unpaired females.

This sampling approach produced a conservative estimate of the strength of sexual selection acting on females for two reasons. First, because many more females are seeking mates than are mating at any given time, this approach can substantially underestimate selection gradients relative to the real values obtained from analyses of representative numbers in each mating class (Blanckenhorn et al., 1999). Unfortunately, we do not have exact numbers of nonmating vs. mating females in individual swarms, and therefore, we were unable to correct for this bias in our selection analyses. Furthermore, most of our ‘unpaired’ females had previously mated, such that we cannot accurately describe them as individuals that failed to secure a mating. In the absence of molecular information on mating frequency (e.g. from genotyping sperm stores, Bretman & Tregenza, 2005), we argue that females observed in the act of mating are likely to mate (and to have mated) more frequently than unpaired rivals. Thus, even though our data for mating status are binomial, it should nevertheless reflect mating frequency, which is probably a strong correlate of fitness because of the nutritious food gifts received from males at each copulation. Moreover, the comparison of paired and unpaired individuals in a cross-sectional sample has been shown to be a reliable method for assessing sexual selection intensities in wild populations of other sex-role reversed insects (e.g. in Mormon crickets, Anabrus simplex; Robson & Gwynne, 2010).

All samples were transferred to a −10 °C freezer upon return from the field. A few days later, samples were placed in alcoholic Bouin’s solution for a minimum of 24 h, to permanently harden the pleural regions of the abdomen in the inflated position (Funk & Tallamy, 2000). Samples were then preserved in 70% ethanol until measurements could be taken.

Morphological measurements

Morphological traits were measured using a dissecting microscope fitted with a digital camera (LeicaDFC290) connected to an iMac and using Image J (Rasband, 2008), a digital imaging program. We took the following morphological measurements: wing length, tibia length, abdomen area, thorax length (all four measured as described in Bussière et al., 2008), as well as pinnate scale length (from base to tip of the longest scale) and pinnate scale area (the area of scales on the inferior surface of the tibia). We measured both right and left structures for paired traits and took an average, except when one of the body parts was injured in which case the undamaged side was the sole measurement considered. Virgin females may either behave differently than mated females, or males may prefer virgin females, and this could be a confounding factor in our analysis of phenotypic correlates of mating success. To investigate whether females had mated previously, the spermatheca was dissected and the presence or absence of sperm recorded. Only 6% of the females did not have sperm in their spermatheca, and one-third of these females were in the ‘mated’ category.

Statistical analyses

Both area measurements (scale and abdomen area) were square-root-transformed to ensure that all traits were measured in the same units (mm). All continuous traits were normally distributed as revealed by the Shapiro–Wilk test of normality (> 0.1).

To investigate the relationship between morphology and mating success, we conducted selection analyses using multiple regression techniques (Lande & Arnold, 1983) with standardized trait values and relative fitness estimates. We also report mean-standardized estimates of our parameter estimates to allow more effective assessment of the strength of selection in our data (Hereford et al., 2004).

High positive correlations between phenotypic traits can sometimes interfere with statistical detection of selection in multivariate analyses. This is because the partial effect of each of several collinear traits can be small even if selection on the phenotypic axis represented by the collinear assemblage of traits is much larger. High correlations between traits are expected whenever groups of characters are functionally related or depend in similar ways on resource acquisition. To determine the level of collinearity in our system, we examined variance inflation factors (VIF) as suggested by Zuur et al. (2009) using the vif function from the car package (Fox & Weisberg, 2010) of R software (R Development Core Team, 2010). If any VIF in our linear model exceeded 10, we subsequently sequentially removed the variable with the highest VIF, reran the analysis and repeated the process until all VIFs were below 10. The nonlinear model was built using the same models that provided a satisfactory linear model for consistency; VIF is not informative in models with quadratic and cross-product terms because those terms are necessarily collinear with main effects.

The linear selection gradients were extracted from regressions featuring only linear terms, whereas the quadratic and correlational coefficients come from regressions including all terms. All coefficients (linear, correlational and quadratic gradients) were obtained from multiple regressions, but we employed logistic regression to test for significance in the regression of mating success on morphology. The coefficients of quadratic terms were doubled prior to inclusion in the gamma matrix of nonlinear selection gradients reported in the results (Stinchcombe et al., 2008).

To further examine nonlinear selection on the multivariate phenotype [which is estimated conservatively by the standard regression approach (Box & Draper, 1987; Blows & Brooks, 2003; Phillips & Arnold, 1989)], we used PopTools for Microsoft Excel (version 3.0.3, Hood, 2008) to perform a canonical analysis of the matrices of nonlinear selection. These analyses use matrix algebra to find the major axes of nonlinear selection as summarized by a diagonalized matrix, M; because the off-diagonal terms in this matrix are zero, we have more power to detect nonlinear selection using regression. Each canonical rotation yielded five major axes (because five traits remained in the analysis after one was excluded owing to collinearity: see below), the strength of selection along which is indicated by the eigenvalues, with the loadings of the original traits on each axis are provided by the eigenvectors (Blows & Brooks, 2003). We calculated the m-score for each individual on each major axis based on the trait loadings, and these m-scores and their squares were used as the independent variables in additional linear regression models, with the appropriate fitness surrogate as the dependent variable. A logistic regression model including all m-axes, quadratic m-terms and cross-products was employed to test the significance of the coefficients of squared terms obtained from the model for mating success (Bisgaard & Ankenman, 1996).

Although the approach described above is ideal for testing the significance of quadratic terms, and therefore of nonlinear selection, it does not explicitly test for stabilizing selection. We therefore also applied the Mitchell-Olds and Shaw test (Mitchell-Olds & Shaw, 1987) to our raw scores of ornament phenotype and the axes of multivariate phenotypic space along which there was statistical support for nonlinear selection. The Mitchell-Olds and Shaw method explicitly tests the null hypothesis that the predicted ‘hump’ for a nonlinear relationship is at either the maximum or the minimum value for the predictor variable. We used the MOStest function provided in the ‘vegan’ package (Oksanen et al., 2009) of R software (R Development Core Team, 2010) to execute these tests and to visualize nonlinear relationships between phenotypic traits and mating success.

Results

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

Selection analysis: mating success

There was evidence of strong collinearity in the linear model including all six morphological variables. After excluding wing length, however, all VIFs were below 10. Consequently, our linear and nonlinear models exclude wing length in the analyses below.

Two alternative models of ornament evolution predicted different forms of selection (linear and stabilizing, respectively) on the two female ornaments: scale size and inflated abdomen area. We found no evidence of significant directional selection on ornaments or any of the other female morphological traits (Table 1, left-hand side). Instead, the primary form of selection we observed was nonlinear. All three quadratic terms for ornamental traits were negative as expected for characters under stabilizing selection, although in each case the coefficient was marginally nonsignificant (see Table 1). There was also a significant, positive correlational selection gradient (another form of nonlinear selection as it acts to change the variance rather than the mean trait value) for the cross-product of scale area and scale length, and a significant negative correlational selection gradient for the product of tibia length and scale length. For completeness, we also report mean-standardized selection gradients in Table 2 (see Hereford et al., 2004). Whereas the variance-standardized selection gradients reported in Table 1 are useful for assessing the strength of selection among individuals in the population, mean-standardized gradients offer a measure of the strength of selection relative to selection on fitness itself (βμ for relative fitness = 1).

Table 1.   The vector of variance-standardized linear selection (βσ) and the matrix of variance-standardized quadratic (on the diagonal) and correlational (off-diagonal) selection gradients (γσ) for mating success. Significance tests come from logistic regressions.
Mating success N = 264βσγσ
Tibia lengthThorax lengthScale length√Scale area√Ab area
  1. P < 0.10; *P < 0.05.

Tibia length−0.02590.0776    
Thorax length0.14410.4313−0.6532   
Scale length0.0821−0.7669*0.1074−0.9610†  
√Scale area−0.01070.14110.06250.9594*−0.4270† 
√Ab area−0.0543−0.09300.0880−0.16260.2987−0.3079†
Table 2.   The vector of mean-standardized linear selection (βμ) and the matrix of mean-standardized quadratic (on the diagonal) and correlational (off-diagonal) selection gradients (γμ) for mating success.
Mating success N = 264βμγμ
Tibia lengthThorax lengthScale length√Scale area√Ab area
Tibia length−0.40250.3067    
Thorax length2.10001.8661−0.7734   
Scale length0.9100−0.80850.0620−0.0300  
√Scale area−0.13820.36780.08850.3334−0.4127 
√Ab area−0.4510−0.80790.4186−0.18850.8556−1.4702

To assist in the interpretation of the form of multivariate selection, we conducted a canonical rotation of the γ matrix of nonlinear variance-standardized selection gradients that resulted in two major axes with positive eigenvalues and three with negative eigenvalues (Table 3). We derived m-scores using the eigenvectors of our canonical analysis as loadings and regressed the linear, quadratic and cross-products of m-scores on mating success. This analysis revealed significantly negative quadratic selection along m5 (see Table 3) only. Reynolds et al. (2010) showed that the double-regression approach to analysing the canonical axes can suffer unacceptably high type 1 error rates and proposed an alternative analytic approach that uses permutation tests to evaluate the significance each eigenvector. When we conducted these tests, axis m5 was no longer significant, but axis m3 (which was nonsignificantly different from zero in the logistic regression) was marginally nonsignificant (see Table 3). Although the main purpose of the canonical analysis is not to interpret the nature of traits under univariate selection, but rather to help visualize the total selection on the multivariate phenotype, it is nevertheless possible to use the loadings of the original traits on the m-axes to ask questions about which traits are principally aligned with strong nonlinear selection. The two axes for which we have some evidence of quadratic selection, m5 and m3, both loaded heavily on ornamental traits and relatively weakly on the other morphological traits in our analysis. Axis m5 loaded strongly on both scale area and scale length, whereas axis m3 loaded most heavily on the other ornamental trait, abdomen area.

Table 3.    The M matrix of eigenvectors from the canonical analysis of γσ for mating success. The quadratic (λi) gradients of selection along each eigenvector are given in a separate column. We provide two sets of significance values because they differ in outcome: the first set comes from logistic regressions including all linear, quadratic and cross-product terms, and the second from permutation tests (see Reynolds et al., 2010).
 MSelectionP (logistic)P (perm. tests)
Tibia lengthThorax lengthScale length√Scale area√Ab areaλ
m10.69820.1462−0.6377−0.2888−0.30831.55380.23870.7280
m20.48300.46710.36770.55820.31870.02780.63900.7280
m3−0.2163−0.1203−0.2857−0.05170.9244−0.53280.23510.0590
m4−0.36440.8528−0.0344−0.3724−0.0057−1.84190.28080.3570
m5−0.31560.1364−0.61260.6808−0.2074−4.15950.03550.1990

Although positive and negative quadratic selection gradients are often interpreted as evidence for disruptive and stabilizing selection, respectively, formal tests of disruptive and stabilizing selection involve rejecting the null hypothesis that the minimum or maximum fitness value is found at one extreme of the scale of phenotypic variation. We applied Mitchell-Olds and Shaw tests to each of our raw phenotypic traits and to the m-axes obtained following canonical analysis. The null hypothesis states that the location of the fitness peak for each function is located at either the maximum or minimum value for the x-variables; samples could not be rejected for any of these axes (see Table 4). However, the few marginally nonsignificant outcomes were associated with m3 and m5, the axes of m-space that loaded most heavily on ornamental traits and for which we had the strongest (although not consistently significant) evidence of quadratic selection. In Fig. 1, we plot the data, fitted quadratic model and approximate 95% confidence regions (see Oksanen et al., 2001) for both of these canonical axes.

Table 4.   Results of Mitchell-Olds and Shaw tests for the location of a quadratic extreme along axes of original phenotypic space and after canonical rotation. Significant P-values indicate that one can reject the null that the hump of the function is located at the minimum, maximum or both extremes of the x-variable.
Phenotypic axisz (Min)P (Min)z (Max)P (Max)P (combined)
  1. P < 0.10; *P < 0.05.

Original trait space
 Tibia length−2.9160.9542.7970.3970.972
 Thorax length−2.4600.7332.8040.5260.873
 Scale length−2.3950.1583.2820.6080.670
 √Scale area−4.1690.8902.7650.5050.946
 √Abdomen area2.2820.3052.4450.4260.601
m-space
 m11.5940.4991.5940.6440.822
 m2−5.2130.7245.9010.6030.890
 m3−2.0020.1332.2290.049*0.175
 m4−1.0070.2741.0010.5330.661
 m5−1.0730.044*1.0720.016*0.060†
image

Figure 1.  The relationship between mating success and two multivariate canonical axes, m3 (in the left-hand panel) and m5 (right). The solid line plots the fitted polynomial, and the dashed lines represent approximate 95% confidence regions. See text for detailed interpretation.

Download figure to PowerPoint

Discussion

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

The adaptive significance of elaborate ornamentation in females remains puzzling because of the plausible direct fecundity costs of overinvestment in ornaments, both for females bearing the traits and for the males who mate with them. Recent theoretical work (Chenoweth et al., 2006) predicted stabilizing selection on female ornaments when they exert a fecundity cost yet enhance fitness by signalling fecundity. Consistent with this prediction, our study found marginally nonsignificant negative quadratic selection for three female-specific ornaments in R. longicauda that are displayed in mating swarms and function in the context of mate attraction (Funk & Tallamy, 2000). Furthermore, canonical analysis to determine the multivariate axes of nonlinear selection revealed compelling (albeit not consistently distinguishable from zero) negative quadratic selection along the multivariate dimensions loading heavily on ornamental traits; females with intermediate levels of ornament expression achieve the greatest mating success. Explicit tests of the intermediate location of a fitness peak or trough were (marginally nonsignificantly) associated with multivariate axes that loaded on either scale length and area (m5) or abdomen area (m3).

Our results therefore provide tentative support for the hypothesis (Fitzpatrick et al., 1995; Chenoweth et al., 2006) that intermediate levels of expression of female-specific traits can be adaptive and the first evidence for stabilizing sexual selection on a female-specific ornaments in a wild population. Laboratory studies on D. serrata show that females experience stabilizing sexual selection on cuticular hydrocarbons (Chenoweth & Blows, 2005; Rundle & Chenoweth, 2010). Although this species offers one of the few examples of stabilizing selection on a female ornament, this ornament is not female-specific; the cuticular hydrocarbons of male D. serrata are also subject to sexual selection, albeit linear selection. The pinnate leg scales of female R. tarsata are under significant sexual selection (LeBas et al., 2003). However, this female-specific ornament is subject to escalating, quadratic selection: females with relatively large scales have higher mating success. The differing patterns of selection on scale ornaments in the two Rhamphomyia species may lie in the aerial mating of R. longicauda in low-light conditions. Rhamphomyia tarsata mates on vegetation during daylight and lack the inflated abdomen ornament, which may allow males of this species to assess fecundity directly.

The female ornaments of R. longicauda may play an important role in signalling fecundity. Males cannot directly assess females’ fecundity when they enter the mating swarm. Our results complement Funk & Tallamy’s (2000) experiment that showed male R. longicauda assess the female silhouette when choosing a mate. Thus, as Chenoweth et al. (2006) predict, the presumed costs to fecundity of investment in ornaments may be mediated by the benefits of attracting males in R. longicauda. Many female dance flies require the nutrition from nuptial gifts to develop their eggs (Downes, 1970). As noted earlier, the majority of sampled females in our study had sperm in their spermathecae, indicating that most females achieve at least one mating. However, females may require the nutrition gained from multiple matings to develop their eggs. We have observed marked females returning to swarms in subsequent evenings even after mating (unpublished data), suggesting that mating could be an important part of a female’s foraging activity. If this is true, investment in ornaments that help secure matings may be critical to reproduction.

Our data do not support the hypothesis of directional selection for larger scales and abdomens and thus appear to contradict findings that male R. longicauda approach female models of large silhouettes more frequently than small ones (Funk & Tallamy, 2000; Wheeler, 2008). There are at least two candidate explanations for this apparent discrepancy. First, within the natural size range of fly models used in the silhouette experiments, choice of intermediate-sized females may not have been detected because males were presented only with large and small models. Second, males may be responding to some aspect of the shape made by the two traits in forming the female silhouette (recall that swarming females hold their scaled legs alongside their inflated abdomen) or their sizes relative to other unmeasured aspects of female morphology. We selected traits that might capture much of the variation in relevant morphology, but it is possible that we have inadvertently neglected crucial dimensions of the silhouette phenotype that were not correlated with any of the measured traits or multivariate axes we studied.

We found support for selection on all three measures of female ornamentation, with the quadratic raw scores and the canonical analysis. The evidence for selection on abdominal sacs appears to be weaker than that on pinnate scales (in terms of both the magnitude of the selection gradient in Table 1 and the weaker, nonsignificant axis (m3) that loaded heavily on abdominal sacs in Table 3). However, the permutation tests suggest that this numerically weaker effect may be more statistically robust (albeit still nonsignificant). If selection on abdomens is really weaker in magnitude, this is unexpected given the positive relationship between abdomen size and egg number and size (Funk & Tallamy, 2000; Wheeler, 2008). One potential explanation is that benefits to males of mating fecund females with larger inflated abdomens may be offset by the potential costs of lower paternity, thereby weakening selection for inflated abdomen size. As mentioned above, females in some empidine dance fly species appear to rely on nuptial prey nutrition to develop eggs (Downes, 1970) and so must mate in order to achieve this. Females with small, inflated abdomens and fewer mature eggs may obtain fewer matings and meals than females with larger abdomens, leading to greater sperm competition in the latter.

Selection for female abdomen size may also be constrained by male load-lifting capacity. Although the male R. longicauda is dorsal to the female during copulation, it is only the male that keeps the pair in flight during the initial pairing and copulatory positioning (Marden, 1989; J. Wheeler, personal observation). Male empids have a maximum mass that they can carry for a given flight muscle mass (Marden, 1989). Thus, fecund females with large inflated abdomens may have a more limited pool of males able to carry them than females with small abdomens, and actual mating success for large females could therefore be smaller than predicted by attractiveness of silhouettes alone. This hypothesis predicts size-assortative mating. Bussière et al. (2008) did not find any significant correlation between male and female phenotypes of coupled R. longicauda; however, they did not include abdominal ornaments in these analyses. The potential effect of female ornamentation on male load-lifting capacity is an intriguing area for further study.

Like many other assessments of sexual selection in insects including dance flies (Sadowski et al., 1999; LeBas et al., 2003, 2004; Bussière et al., 2008), our study focussed on mating success as a primary determinant of fitness. In males, this convention is justifiable because variation in mate number usually covaries closely with fitness. In many female dance flies including R. longicauda, this argument is also plausible because each mating event is associated with the transfer of direct benefits from males to females. Nevertheless, it is possible that our cross-sectional approach does not accord well with total sexual selection on ornamentation. Furthermore, we note that our measure of selection on female ornaments is incomplete; we assessed the influence of these traits on mate acquisition, but the current work cannot evaluate their effect on the quality of nuptial gifts (e.g. if ornamented females attract males with high-quality food gifts) or test how the ornaments covary with lifetime reproductive success (e.g. the size and number of clutches produced and the survival probability of females). As a consequence, we cannot combine our fitness measures to estimate total female fitness without making arbitrary and unjustified assumptions about the covariance between lifetime fitness and mating success, egg size, and egg number in a cross-sectional sample of flies. Direct tests of how nuptial feeding affects fecundity are not yet available for dance flies (unlike for other insects featuring nuptial gifts, see Gwynne, 1981; Robson & Gwynne, 2010), but our approach is a conservative and useful first step in identifying how mate competition may shape phenotypes in dance flies.

Future research that refines our approach will be complicated because there are several plausible causal links between ornament and fecundity that would affect observed correlations between these traits. A trade-off between ornament expression and fecundity (Fitzpatrick et al., 1995; Chenoweth et al., 2006) could give rise to a negative correlation only if these traits arise from a common resource pool. However, ornaments could also cause increases in fecundity by promoting mating success and access to male-provided nuptial gifts. Our measures of abdominal area could also be affected by fecundity in turn if mature eggs exaggerate the apparent abdominal size over and above what is achieved via inflation. More comprehensive measures of female fitness and a careful consideration of all the possible causal relationships across traits will be needed in order to evaluate these possibilities and more reliably estimate the effect of morphology on fitness across different episodes of life history (Hunt et al., 2009).

Acknowledgments

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

We thank Vicki Simkovic for her assistance in the field. Göran Arnqvist, Andre Gilburn, Matthew Hall, John Hunt, Ken Kraaijeveld, Helen Rodd, Locke Rowe, Rob Baker and an anonymous reviewer provided insightful advice and comments that greatly improved the quality of this manuscript. This work was supported by an Ontario Graduate Scholarship to JW, and the research was funded by an NSERC Discovery Grant to DTG. LFB was supported by the University of Stirling.

References

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References