Christine W. Miller, Department of Entomology and Nematology, University of Florida, PO Box 110620, Steinmetz Hall Natural Area Drive, Gainesville, FL 32611-0620, USA. Tel.: +1 352 273 3917; fax: +1 352 392 0190; e-mail: firstname.lastname@example.org
Sexual selection arises from social interactions, and if social environments vary so too should sexual selection. For example, male–male competition often occurs either in the presence or in the absence of females, and such changes in the social environment could affect the form and strength of sexual selection. Here we examine how the presence of a female influences selection arising from male–male competition in a leaf-footed cactus bug, Narnia femorata, which has a resource defence mating system. Males compete for territories on cacti because females lay eggs on the cactus plants. Females are not always present when this competition first occurs; however, the presence or absence of the female matters. We found that both the form and strength of selection on male traits, those traits that influenced success in intrasexual competition, depended on the social context. When a female was not present, male size and the area of the sexually dimorphic hind legs was only marginally important to winning a contest. However, males with larger overall size and leg area were more likely to win in the presence of a female. There was also positive quadratic selection on these traits when a female was present with both the largest and the smallest males winning. The implication is unexpected alternative strategies when females are present. Our results support the notion that sexual selection should be studied under all relevant social contexts.
In this study, we examined male–male competition under different social conditions. The species we studied, the leaf-footed cactus bug Narnia femorata (Fig. 1), exhibits resource defence polygyny in the wild, with males establishing territories on Opuntia and Cylindropuntia sp. cacti with fruit and attracting females. Male N. femorata compete for territories on this cactus, especially around fruits, and females use cacti pads and (preferentially) fruit for oviposition. Male–male competition can occur before a female arrives and so can be independent of any female effect. Thus, we are measuring one component of sexual selection; it is likely that total sexual selection will involve a combination of first male–male competition and then female choice (Hunt et al., 2009). Competition for a cactus can be described as competition over a mating resource, as territories are held on these cacti and reproduction requires cacti for egg laying and subsequent juvenile development. However, males that lose this initial competition do not necessarily disperse, or competition may not necessarily occur. Male–male competition may also (or instead) take place once a female arrives. Competition under these conditions likely differs, reflecting direct competition for mates when females are present as well as potential breeding sites. Sexual selection arising from male–male competition therefore must incorporate both situations.
We measured the effect of variable social environment in N. femorata by measuring selection gradients (Lande & Arnold, 1983) and visualizing selection (Schluter, 1988) resulting from male–male competition under different social conditions. Estimating selection gradients and visualizing the selection surface is a standard method to examine the selection acting on traits that influence ‘winning’ a competition, one potential component of fitness (e.g. Moore, 1990; Brodie et al., 1995; Brodie & Janzen, 1996; Hunt et al., 2009). Our goal was to determine whether selection arising from male–male competition differs under the two common social conditions that are experienced by N. femorata males. Thus, our approach is analogous to a partitioned selection approach (Arnold & Wade, 1984; Wade & Kalisz, 1989; Hunt et al., 2009). We predicted that selection on male traits that are related to success in male–male competition would be similar in form (directional) under both resource and mating competitions. However, we also predicted stronger (steeper) selection gradients in the presence of a female because of the ‘added value’ of winning in that context; that is, because the probability of immediately mating was greater.
Materials and methods
Narnia femorata is a member of the leaf-footed bugs (family Coreidae) and is between 10 and 18 mm in length. This genus of insects is native to the south-western United States, Mexico and parts of Central America, where it lives on Opuntia and Cylindropuntia sp. cacti, and was only recently introduced to Florida (Baranowski & Slater, 1986). In our study area in north-central Florida, USA, it lives on O. humifusa. Males establish territories and defend areas on cacti by physical competition. The form of this competition is not ritualized. Competitions can end by one male retreating after just antennal contact, after physical contact or after fully escalated competition. During escalated competition, males face rear to rear and grab each other using their hindmost legs (Fig. 1), as is common throughout the Coreidae (Mitchell, 1980; Miyatake, 1997; Eberhard, 1998; Miller & Emlen, 2010a,b).
We collected N. femorata from the wild at the Ordway-Swisher Biological Station, University of Florida (29°41′N, 82°W), and brought them to the laboratory to form populations. The majority (73%) of the males were first-generation laboratory insects, whereas the remainder were second generation. Adults were paired so as to maximize genetic diversity in the population. The O. humifusa for the study was collected from both the Ordway-Swisher Biological Station and also the Camp Blanding Joint Training Centre (29.95°N, 81.98°W).
Both male and female N. femorata complete five nymphal instars before eclosion. Until 4th instar, nymphs were kept in communal ‘deli’ containers with soil, cladode and ripe fruit in heated incubators at 26 (±1)°C and on a light cycle of 14 h : 10 h light/dark. At 4th instar, we separated nymphs into their own containers, in which there was soil, cladode and ripe fruit. We then moved them to a greenhouse on a 15 h : 9 h light/dark cycle with mean temperature of 26 °C.
Experiment 1: male competition in the absence of a female (MM)
The arenas in which we measured selection were containers containing soil, a cactus pad (cladode) and a fruit. Where possible, we used fruit naturally attached to the pad; when this was not possible, we artificially attached fruit to the top of the cladode with toothpicks. We randomly selected two males, marked one with water-soluble paint and placed both into an arena together. We allowed them to acclimatize and establish dominance for 2 h prior to observations. Males were unrelated and matched as closely as possible in age post-eclosion (median = 0 days, mean = 1.86 days) and always within 7 days in age. All males were over 14 days of age to ensure sexual maturity, with a median age of 22 days post eclosion (maximum 45 days). Observations commenced after the males had been in the arena for the 2-h acclimatization period and lasted for 2 h in a room maintained at 24 °C. A single observer conducted the 85 behavioural trials. Not all pairings resulted in competitions (see Results). We chilled bugs after observations, placed them into labelled tubes to be frozen and later measured morphological traits.
Experiment 2: male competition in the presence of a female (MMF)
The experimental set-up was the same as Experiment 1 except that after the males had been together for 2 hours, a marked, sexually mature female was added and then observations were begun. Females were selected randomly from the available population and were unrelated to either male. Female age varied more widely than in selected males, with female median age 35 days post eclosion (minimum = 14 days; maximum = 97 days). A single observer again conducted 85 behavioural trials, and again, not all pairings resulted in competitions. Different males were used for experiment 1 and experiment 2. We froze bugs after observations and later measured morphological traits.
Outcome of interactions as a component of fitness
We used two measures to score fighting success. We scored ‘absolute male competitive success’, with the bug that withdrew most per interaction scored the loser (fitness = 0) and the bug that withdrew the least per interaction scored the winner (fitness = 1). We also used a continuous measure of fitness, ‘relative male competitive success’, calculated as [1 − (proportion of withdraws)]. In the end, both measures provided identical qualitative results, including patterns and statistical significance, so we only present the results from absolute male competitive success. A zero-one (win-lose) measure is a commonly used fitness surrogate for male–male competition, so adopting this measure facilitates comparisons (Hunt et al., 2009). Fitness was transformed to relative fitness for selection analyses (Brodie et al., 1995), although for zero-one scoring this does not matter (Brodie & Janzen, 1996).
This measure of fitness we use is indirect; that is, it assumes that males compete for territories that later influence mating success. Although we did not then measure male mating success or female mate choice of males that competed, this is a parsimonious assumption giving the resource defence polyandry system of mating for these bugs. Male–male competition in the absence of a female may commonly represent the first component of sexual selection, followed by sexual selection in the presence of a female. Our measures of selection therefore reflect a partition of sexual selection (e.g. Moore, 1990).
We dissected frozen bugs into their component parts (legs and body) to facilitate measurement of characteristics. Each component was then photographed using a Canon EOS 50D digital camera attached to a dissecting microscope along with a reference measurement. We used ImageJ 1.42q (http://rsb.info.nih.gov/ij/, 1997–2009) to measure straight line and area measurements from the photographs. We measured the pronotum width, head length, body length (thorax + abdomen), front femur length and front tibia length using a single linear measure. We also used ImageJ to trace the outline of the rear (elaborated) leg segments and obtain measures of the rear femur area and rear tibia area (Fig. 1). For leg characters where there were right and left measurements, we measured both and used the mean in our analyses.
We determined the form and strength of selection by using parametric regressions to estimate linear (β) and nonlinear (γ) coefficients (Lande & Arnold, 1983; Arnold & Wade, 1984). Linear coefficients were determined using a multiple linear regression of all the single terms and nonlinear coefficients using a multiple linear regression of single terms, squared terms and interaction terms (Brodie et al., 1995). Regression coefficients for nonlinear coefficients were doubled to give the correct γ value (Stinchcombe et al., 2008). All morphological variables were transformed to mean = 0, standard deviation = 1 before analysis. Although we chose morphological measurements that can be independently derived during development, extreme correlation of variables is problematic for selection analyses (Lande & Arnold, 1983; Mitchell-Olds & Shaw, 1987) and so we used Pearson’s product–moment correlations to examine the extent of correlation between variables. Highly correlated variables were converted to a single composite measure with principal component analysis. Because fitness was measured as two categories, we used logistic regression to determine the significance of the selection coefficients (Janzen & Stern, 1998).
Using multiple regression to estimate the form of selection has limitations, as it fits a specific model (Schluter, 1988; Brodie et al., 1995). Therefore, we also used a nonparametric method to determine whether more complex selection, especially nonlinear such as disruptive or stabilizing, was present. Nonparametric estimations of the fitness function (f) were made using cubic splines (Schluter, 1988; Brodie et al., 1995), with λ set to 1, using jmp 8.0; SAS Institute Inc. Cary, NC, USA.
We found no statistically significant differences between males used in the MM or MMF experiments in any of the morphological measurements (Table 1) (pronotum width: t402 = 1.468, P = 0.143; head length: t402 = 1.855, P = 0.064; body length: t402 = 1.313, P = 0.190; front tibia length: t402 = 1.641, P = 0.102; front femur length: t402 = 1.629, P = 0.104; hind tibia area (HTA): t402 = 1.375, P = 0.170; hind femur area (HFA): t402 = 1.612, P = 0.110). We also found no difference in any of the variances between males in the two treatments (Table 1) (Levene’s test; pronotum width: F1,402 = 0.077, P = 0.781; head length: F1,402 = 0.844, P = 0.359; body length: F1,402 = 0.348, P = 0.555; front tibia length: F1,402 = 0.210, P = 0.647; front femur length: F1,402 = 0.572, P = 0.450; HTA: F1,402 = 0.004, P = 0.984; HFA: F1,402 = 0.005, P = 0.985). Competition between males was no more likely in MM (68.2% of pairs expressed dominant/subordinate behaviour) than in MMF (57.6%; χ21 = 1.614, P = 0.204).
Table 1. Summary statistics of untransformed phenotypic characters. All means (SD in parentheses) are given in mm except for areas, which are in mm2.
Front tibia length
Front femur length
Hind tibia area
Hind femur area
We found strong correlations between all measured variables (Table 2). Given the strong correlations, we used PCA to reduce our measures to a single composite measure of size (Table 3). PC1 had strong positive loadings on all traits, explained 87% of the variance and provides a robust measure of general body size. PC2 had a single strong loading on head length, with all other traits having negative but weaker loadings, but this component had an eigenvalue less than one and explained only 9% of the variance. No other PC explained more than 2% of the variance. We therefore only used PC1 in further analyses.
Table 2. Pearson’s product–moment correlation coefficients for all measured traits.
PW, pronotum width; HL, head length; BL, body length; FTL, front tibia length; FFL, front femur length; HTA, hind tibia area; HFA, hind femur area. All pairwise correlations are highly statistically significant. Measurements from N = 404 individuals.
Table 3. Principal component analysis of linear morphological measurements, explaining 96.2% of the variance.
Front femur length
Front tibia length
% Variance explained
Experiment 1: male competition in the absence of a female
When competitive outcomes were scored as absolute competitive success, we found no statistically significant linear or nonlinear selection coefficients (Table 4). We did find that positive linear selection approached statistical significance for overall size (0.071) and nearly reached conventional significance for tibia area (P = 0.059). Thus, if any selection arises from male–male competition when no female is present, it is weak and for larger size. This interpretation is supported by the visualizations of selection provided by the univariate cubic splines (Fig. 2). The selection surface for all traits, although often rugged, was primarily directional.
Table 4. Linear (β) and nonlinear (γ) selection coefficients (Lande & Arnold, 1983) on size (PC1) and elaborated traits during male–male competition when there was no female present. Fitness was measured as lose/win (0, 1). Significance was determined by logistic regression (Janzen & Stern, 1998).
Hind tibia area
Hind femur area
PC1 × hind tibia area
PC1 × hind femur area
Tibia area × femur area
Experiment 2: male competition in the presence of a female
In contrast to the results found when no female was present, when a female was present we found statistically significant positive linear and nonlinear selection on overall size (Table 5). We also found significant positive linear selection on HTA and significant disruptive nonlinear selection on HFA (Table 5). These suggest a combination of positive linear selection and divergent selection (positive selection of trait variance), a pattern that was supported by the visualizations of selection using splines (Fig. 3). For all traits, there were multiple peaks on the selection surface.
Table 5. Linear (β) and nonlinear (γ) selection coefficients (Lande & Arnold, 1983) on size (PC1) and elaborated traits during male–male competition with a female present. Fitness was measured as lose/win (0, 1). Significance determined by logistic regression (Janzen & Stern, 1998).
Hind tibia area
Hind femur area
PC1 × hind tibia area
PC1 × hind femur area
Tibia area × femur area
Comparison of selection in two environments
Determining the extent of the differences in selection in the two environments is complicated. We can test for differences in the selection gradients, but this is a very conservative test as it reduces selection to just the linear or nonlinear components described by regression coefficients. In our experiments, the linear selection gradients on the HTA were statistically significantly different (two-tailed t-test, d.f. = 400, P = 0.035), as was linear selection on HFA in the two environments (d.f. = 400, P = 0.029). Linear selection gradients on PC1 were not statistically significantly different (d.f. = 400, P = 0.321). None of the nonlinear selection gradients were statistically significantly different from each other based on a t-test (all d.f. = 400; PC1, P = 0.174; HTA, P = 0.879; HFA, P = 0.874; PC1 × HTA, P = 0.878; PC1 × HFA, P = 0.287; HTA × HFA, P = 0.566). However, visual inspection of the selection surfaces supports the less conservative interpretation based on which gradients were statistically significantly different from zero, and which were not. Directional selection is stronger in the presence of a female, and there also tends to be more disruptive selection when a female is present.
The shape and strength of selection reflected the social context. Qualitatively, we found that linear selection on overall size (PC1) was weaker when no female was present than when a female was in the environment (Tables 4 and 5). In addition, when a female was present there were significant nonlinear components, suggesting disruptive selection, that were not seen when no female was present. We also examined selection on the rear leg area, the part of the males that is most elaborated and where we expected the strongest sexual selection arising from male–male competition. An even stronger pattern was seen for selection on the elaborated traits: both qualitatively and quantitatively, there was stronger linear selection in the presence of a female. We also found more evidence for a nonlinear component indicating disruptive selection on rear leg area. Although the elaborated traits were highly correlated with our overall size measure, there was no indication of correlational selection integrating size and elaboration, or integrating the two elaborated leg areas. These findings are more nuanced than our original predictions.
We did not predict selection would change form, only that the strength might differ. Both the cactus and the female are mating resources for a male; therefore, we predicted that the two resources together should represent a higher value resource. Females are expected to incite male–male competition when possible to ensure they mate with more dominant males (Cox & Le Boeuf, 1977; Pizzari, 2001); however, we would expect incitation of competition to result in stronger linear gradients in the presence of a female. There is no a priori reason to predict that selection should be nonlinear. We suggest the difference between our observations and initial predictions reflects a failure to consider that directly achieving access to females is essential to reproduction, while successfully defending a host plant in the absence of a female is one additional step removed. The evidence for quadratic selection in the presence of a female suggests that males in this situation may have alternative strategies for reproduction. Our findings might also help explain why females are bigger than males. It is very likely that there is fecundity selection for larger size on females, as this is true for most insects (Reeve & Fairbairn, 1999), but selection on males is not simply for larger size. For males, we found that bigger is not always best.
The difference in selection surfaces under different social conditions suggests variation in mating tactics with body size in N. femorata. There are many examples of alternative-mating strategies associated with sexual selection (Ra’anan & Sagi, 1985; Shuster & Wade, 1991; Akagawa & Okiyama, 1993; Sinervo & Lively, 1996; Calsbeek et al., 2002; Hankison & Ptacek, 2007); some of which vary with body size (Ra’anan & Sagi, 1985; Shuster & Wade, 1991). Male–male competition is costly to males (Andersson & Iwasa, 1996), and so males might be expected to change their tactics depending on the necessity and importance of the resource over which they compete. Fights in N. femorata are not without costs. During our observations, competition resulted in the severing of body parts on several occasions. Given that the smallest males have very little chance of establishing and maintaining a territory for very long, it may be to their benefit to risk injury only in the presence of a female, where there is likely to be an immediate gain of fitness after the fight. Larger males could expect to maintain a territory in the future and so have a significantly higher future reproductive success than small males, therefore reducing the benefit of potentially costly competition immediately. This would result in increased aggressiveness of small males in the presence of a female, which may explain their observed increase in competitive success. It is still of greater benefit to be large rather than small in the presence of a female (Fig. 3).
These findings add further support to the increasing body of evidence suggesting that the environment in which selection occurs has a large bearing on the form of selection itself. Social environments can be much more dynamic than other forms of environmental variation that affect sexual selection. Most existing studies of temporal variation in sexual selection have shown change over breeding seasons (Jann et al., 2000; Punzalan et al., 2008, 2010) or between years (Ferguson & Fairbairn, 2000; Preziosi & Fairbairn, 2000; Gosden & Svensson, 2008), but our study system could vary in social environment and therefore selection on male–male competition in a matter of hours. Social information can provide powerful cues to adaptive behaviour (Dall et al., 2005). Where studied, rapidly varying social environments appear to be important influences on sexual selection arising from male behaviour. For example, the social composition within a group influences male pheromonal communication and sexual selection in Drosophila melanogaster (Kent et al., 2008; Krupp et al., 2008). ‘Audience effects’ influence male Poecilia mexicana courtship (Plath et al., 2008a,b). This suggests that social contexts should be measured and varied whenever sexual selection is studied. For field studies, estimates of the proportion of competitions with or without a female present can add an important layer of information, suggesting how selection may lead to further evolution of male traits. Incorporating these factors into sexual selection can be accomplished by multiple partitioning of selection (Arnold & Wade, 1984; Wade & Kalisz, 1989). This then can be combined with selection arising from mate choice. Ultimately, to determine total sexual selection (Hunt et al., 2009), we will also have to examine how females influence mating success subsequent to the outcome of male–male competition. It is clear that a complete picture of sexual selection can only be gained when sexual selection is measured under all socially and ecologically relevant conditions.
Finally, we found that we could not predict a target of contemporary selection. Enlarged hind legs are important in male–male competition throughout the Coreidae (Miyatke, 1997; Eberhard, 1998; Miller & Emlen, 2010a), so selection on size of the elaborated hind leg areas could be expected in N. femorata. We found strong correlations between all measurements of hind leg dimensions and other morphological variables (Table 2), so there is no signal of these being isolated as separate targets of selection. This does not preclude a role for leg elaboration in determining the outcome of male–male competition at some point in the past; it is just that current associations between body size and leg size are complete.
These results suggest many follow-up experiments and manipulations that could provide a more refined view of what is occurring with sexual selection on males. Here we present one experiment that we ran as a quick follow-up to the results we found and present here. Males in the study described above were randomly paired, lessening our power to detect the independent role of leg size in contest outcome. Therefore, we later size matched males (difference of 5% or less) based on pronotum size (a reliable indicator of overall size for many insects, including N. femorata, that is easily measured) but not femur size. Males with larger legs won 31 contests but males with smaller legs won 27, regardless of the presence or absence of a female. However, the larger the difference in weapon size, the more likely the male with the larger femur won the contest (Fig. 4). Thus, although there may be little or no contemporary selection on weapon size, it appears as if the femurs can and therefore perhaps did, in the past, influence the outcome of male–male competition. Although this is an attractive and parsimonious explanation, the experiment we present here is just a ‘quick and dirty’ summary. Manipulative experiments under all social conditions are required to disentangle the contributions of size and elaboration to male success.
We thank the members of the Miller Lab, especially Allison Bechard and Stephanie Gillespie, for help throughout the investigation. Stephanie ran the follow-up experiment. Tom Pizzari and an anonymous referee provided very helpful comments that refined our arguments and presentation. NSF grant IOS-0926855 (to CWM), along with grants from the European Social Fund and NERC (to AJM), supported this research.