Introduction

How fitness changes with age is an intriguing question for a number of reasons. Much of this research is focussed on actuarial and functional senescence and the evolutionary processes that may drive these phenomena, such as mutation accumulation or antagonistic pleiotropy (Medawar 1952; Williams 1957). Studies of changes in reproductive performance with age have attracted special attention because of additional evolutionary explanations that relate to sexual selection (Bonduriansky et al. 2008). Age-based indicator mechanisms are postulated to explain female preferences for older males (who have by virtue of their old age proven their ability to survive in the current environment, and therefore represent males of high genetic quality; Brooks and Kemp 2001). Although there are several theoretical and empirical studies supporting these mechanisms (e.g. Kokko 1998; Pike et al. 2007; Zuk 1987), the relative importance of age-based mate choice for indirect benefits remains uncertain, in part because it is often difficult to rule out confounding factors associated with age (such as paternal ability) that might drive female preferences. There are also empirical and theoretical examples of situations in which females prefer younger or intermediately aged males (Ritchie et al. 1995; Jones et al. 2000; Beck and Powell 2000). Moreover, system specific complexities may cause age-related mate choice for reasons that are rather unrelated to good-genes sexual selection in the strict sense, such as inbreeding avoidance (De Luca and Cocroft 2008; Keller et al. 2008).

In addition to the above-mentioned complexities, evaluating how one particular aspect of a male’s phenotype, such as age, influences his mating success is difficult because phenotypic characters are usually intercorrelated, and sexual selection rarely occurs in a single episode. There are many studies of how male characters covary with mating success, but relatively few of these focus on how male traits affect events occurring before courtship (e.g., long-distance attraction; Brooks et al. 2005) and after copulation (e.g., post-copulatory sexual selection; Bussière et al. 2006). Selection in these episodes need not concur in sign or magnitude with selection on courtship for mating success (Danielsson 2001), and consequently it is often hard to evaluate whether selection for pair formation is representative of sexual selection as whole Hunt et al. 2009.

Bumblebee males have features that make them less vulnerable to many of these problems. First, most morphological characters involve sclerotized structures which do not change with bumblebee age, and can therefore be assessed relatively independently of it. In most species, female bumblebees are not known to remate in the field (Estoup et al. 1995; Schmid-Hempel and Schmid-Hempel 2000; Paxton et al. 2001; Payne et al. 2003) and so post-copulatory sexual selection and any potential interaction between sperm performance and ageing (Radwan 2003; Reinhardt 2007) are unlikely to play a substantial role in selecting on male traits. Furthermore, since male bumblebees acquire most of their developmental resources thanks to the efforts of their nursing sisters, their early developmental progress probably depends less on their own morphology than is the case for many other species. Notwithstanding some understudied aspects of male patrolling and pheromone communication (Baer 2003; Goulson 2010), these facts together suggest that measures of selection based on mating success are much more likely to represent the total intensity of sexual selection in male bumblebees than for most other insect species in which male morphology is more constrained by selection for resource acquisition and success in sperm competition.

At a colony level, there are a number of aspects of the reproductive strategies adopted by bumblebees that are not yet fully understood. Firstly, they generally seem to have a highly male-biased sex-ratio which cannot readily be explained, particularly since kin-selection theory predicts that, in monogamous species such as most bumblebees, workers should favour a 3:1 investment in females relative to males (while the queen should favour equal investment in each) (reviewed in Bourke 1997; Goulson 2010). Secondly, observations of the timing of production of males versus new queens in bumblebee nests suggest that males are produced considerably earlier than one might expect, unless females show a strong preference for old males. Protandry appears to be the norm in bumblebees (reviewed by Bourke 1997) and many other insects, and is thought to optimise the number of virgin females that a male will encounter during his life. However, based on studies of B. terrestris, it seems that bumblebee colonies vary greatly in the timing of their reproductive phase (Duchateau and Velthuis 1988): some colonies switch from rearing workers to rearing reproductives while the colony is still small, approximately 10 days after emergence of the first workers. In contrast other colonies switch later, an average of 24 days after emergence of the first workers. Early-switching colonies produce mainly males, and late-switching colonies produce mainly new queens. Because males take ~26 days to develop from egg to adult, while queens take ~30 days (Duchateau and Velthuis 1988), on average, the first males produced by early-switching colonies become adult 18 days before the first new queens emerge. Males are fully fertile and reproductively active after just 5 days (Greeff and Schmid-Hempel 2008), and leave the nest. There is, therefore, no clear explanation as to why males are produced so early, since they then have to survive for many days before they are likely to get a chance to mate, a period during which they could succumb to predation or disease.

One possibility is that older males have a mating advantage that selects for early male development. Alternatively, it has been suggested that colonies switch early and specialize in male production in response to stress or low resource availability. Beekman and Van Stratum (1998) argue that this strategy is adaptive if male size is less closely linked to reproductive success than is female size. This hypothesis seems plausible since small queens suffer high mortality in hibernation (Beekman et al. 1998). However, the relationship between male size and reproductive success has not been investigated.

Here we study the relationships between male morphology, male age and mating success under a range of experimental conditions. We test how age and mass affect the speed with which a male bumblebee mates, the duration of mating and his propensity to mate a second time. We also perform experiments to determine the influence on mating success of the phenotype in a multivariate context. This experiment allowed randomly selected males to compete for single females, and assessed the relative importance of several different aspects of the phenotype once correlated characters were controlled for statistically.

Methods

Study Insects and Rearing Conditions

A total of 32 B. terrestris colonies were imported from Koppert UK limited, and the colonies were allowed to produce sexuals (new queens and males) in a 24-hour dark climate room at 28 ± 1°C, 50 ± 5% RH. We used red light for observations and animal husbandry. The bee diet consisted of ad libitum pollen and sugar solution. The newly emerged sexuals were removed daily and kept in nestmate groups in mesh screen cages (30 × 30 × 30 cm) with a typical number of 50 males or 20 queens in each cage. We stored male and queen cages in separate rooms at 25 ± 1°C and 50 ± 5% RH with ad libitum pollen and sugar solutions until individuals were used in experiments. Thirty-one colonies produced males and 29 produced queens, and a total of 2,164 queens and 3,571 males were harvested.

Mating Conditions

Mating experiments were carried out in a climate room at 25 ± 1°C and 50 ± 5% RH. All matings were staged in transparent plastic boxes (10 × 8 × 8 cm) which had mesh screens for ventilation. Observations were made under white light (Philips MCFE65-80 W/29, Poland) positioned 0.5 m above the mating cages. Queens and males present in a mating cage were from different colonies.

Effect of Male age on Mating Behaviour

This experiment was done with four different age groups of males viz. 6, 9, 12, and 15 days old post-eclosion, with 7 day old queens. In each mating cage a virgin queen and a virgin male were allowed to interact freely. If the couple had not started mating within 1 h, the trial was aborted and the pair was discarded. For trials in which mating occurred, we recorded the mating latency (time elapsed between the introduction of male and female into the mating cage until initiation of copulation) and the duration of mating (time elapsed between copulation initiation to termination by each pair). As soon as they completed mating, the mated queen was taken out and replaced with a virgin queen, and the new pair were observed for the next hour. The remating latency (time elapsed between the introduction of the second queen into the mating cage until initiation of copulation with her), and the duration of the second copulation were also recorded. If any male did not remate within 1 h then the pair was excluded from analyses of remating behaviour. For each age group, we conducted trials until 20 males had mated once. All of the mated males were used for remating. The number of males which mated a second time within 1 h were n = 17, 17, 15 and 15, for the age groups 6, 9, 12 and 15 days-old, respectively.

Effect of Male Mass on Mating Behaviour

To observe the effect of male body mass on mating behaviour we conducted a separate experiment using 7 day-old queens and 7 day-old males, which are at the peak of receptivity for mating (Duchateau and Marien 1995). We measured male body mass using a digital top loader balance (Fisher Scientific SG-202, LA NORME NMB003, DU Canada) to an accuracy of 0.01 g. Mass was measured immediately before mating trials took place. We placed a virgin female together with a virgin male and recorded mating latency, mating duration, remating latency and remating duration as described above. Data were collected from a total of 105 mated males of which 84 mated twice.

Effect of Sex Ratio on Mating Behaviour

To test how the number of competing males for each queen in the mating cage affects mating behaviour, we observed mating in female: male sex ratios of 1:5, 1:10, 1:15 and 1:20. All males and queens were 7 day old virgins, and 16–17 replicates were carried out for each sex ratio. For each queen to male ratio, the animals were allowed to interact in the mating cage freely for 1 h, and the above methods were employed for estimating mating latency and duration.

Male Phenotype and Mating Success Under Competition

Because our previous experiments revealed intriguing effects of age and mass on mating behaviour, we sought to determine the relative impact of phenotypic traits (including age) on male mating success in a competitive context. We conducted a series of mate choice trials using virgin males of varying age and morphology. Males were between 7 and 24 days old and the queens were all 7 days old. Each male was used only once, and was randomly paired with a rival (except that no siblings were used in the same trial). A total of 406 males and 203 queens were used. We observed mating by placing one virgin queen and two randomly selected males that were weighed immediately prior to mating trials in a mating cage (10 × 8 × 8 cm); one male was given a spot (Nice day, Euro Office Ltd. London) on the thorax so that they could be distinguished. Males and queens were allowed 1 h to mate; if no mating occurred within 1 h they were discarded. After the completion of mating, the bees were killed by freezing, and several morphological characters were subsequently measured. We weighed all bees on an analytical precision balance (Denver Pinnacle Series 0.04 Digital, BCBP 100, Germany). We removed wings and all legs from the bees, and photographed them using a binocular microscope connected to a digital camera (Moticam 2000, 2.0 M Pixel USB2.0, China). We measured the length of wings and all tibiae (see Supplementary Figs. 1a and b) using Motic Images Plus software (Version 2.0, Novex, Holland).

Statistical Analysis

An analysis of variance (one way ANOVA) with post hoc Duncan’s Multiple Range Test (DMRT) was applied to compare the mating behaviour of different age groups of males. Comparisons were made between mating and remating latency, and duration between first and second mating for each age group using Student’s t-tests (independent samples). The effect of queen to male ratio on mating latency and duration was analysed by one way ANOVA with post hoc Duncan’s Multiple Range Test. We used a Pearson’s correlation to examine the relationship between body mass and mating latency and duration. All the above analyses were performed using SPSS (PASW Statistics 17).

For the analysis of sexual selection on male phenotypes, we used standard linear and nonlinear approaches for estimating selection gradients (Arnold and Wade 1984a, 1984b). The parameter estimates we report are derived from standard regressions, but the p-values associated with these parameters are derived from logistic regressions because the response variable is binomial. We doubled quadratic coefficients when reporting quadratic selection gradients as recommended by Stinchcombe et al. (2008). Because our design featured two males in every trial, the outcome for each male was not independent of the outcome for his partner. We therefore used bootstrap resampling methods as suggested by Mitchell-Olds and Shaw (1987) to test for bias in our estimated gradients based on the full sample of males. In each iteration of this procedure, we sampled one male from each trial, selected at random, and executed the regression analysis to generate an unbiased estimate of the true parameters. We then repeated the resampling procedure 10,000 times and observed the distribution of parameter estimates to generate 95% confidence intervals for the true parameters. These analyses were all performed using R version 2.10.1 (R Development Core Team 2009). We visualized selection on individual phenotypic traits using nonparametric cubic splines created in the vegan package (Oksanen et al. 2009) of R software.

Because the standard multiple regression approach can systematically underestimate the strength of nonlinear selection on phenotypes, we also used a canonical rotation of the γ matrix of nonlinear selection gradients (Phillips and Arnold 1989; Blows and Brooks 2003) to isolate the axes of multivariate phenotypic space along which nonlinear selection was strongest. This allowed us to assess with more confidence whether age was important in a multivariate nonlinear context. We visualized the two statistically significant axes of this nonlinear selection (which were also the axes along which nonlinear selection was strongest) using a nonparametric thin-plate spline. We modelled the spline using the fields package (Fields Development Team 2006) of R software, adopting the smoothing parameter (lambda) that minimized the generalised cross-validation (GCV) score.

Results

Effect of Male Age on Mating Behaviour

In all the age groups of males, some started mating within 1 min of their introduction into the mating cage, and the number of matings increased with time (Fig. 1). The mean mating latency varied from 11.1 ± 12.1 min for 6 day old males to 22.9 ± 17.1 min for 15 day old males, with age having a significant positive effect on mating latency (mean ± SD, ANOVA: F3, 76 = 2.73, p < 0.05). The mean remating latency also varied significantly with age, with older males remating after a longer interval (Fig. 1, F3, 60 = 2.77, p < 0.05). There was no significant difference in mating latency for first and second matings (t35 = 0.27, p = 0.90; t35 = 0.29, p = 0.31; t33 = 0.33, p = 0.28; t33 = 0.25, p = 0.56 for age group 6, 9, 12 and 15 days, respectively). Mating duration also differed between age groups, being longer in older males in both their first mating (F3, 76 = 2.68, p < 0.05) and in their second (ANOVA: F3, 60 = 2.75, p < 0.05) (Fig. 2), and there was no significant differences between first and second mating durations (t35 = 0.89, p = 0.13; t35 = 0.67, p = 0.53; t33 = 1.42, p = 0.66; t33 = 0.67, p = 0.72 for age group 6, 9, 12 and 15 days, respectively).

Fig. 1
figure 1

Effect of male age on the mating and remating latency (mean ± se) of B. terrestris. Mating and remating latency both increased significantly with age (F3, 76 = 2.73, p < 0.05 and F3, 60 = 2.77, p < 0.05, respectively)

Full size image
Fig. 2
figure 2

Effect of male age on the mating and remating duration (mean ± se) of B. terrestris. Mating and remating duration both increased significantly with age (F3, 76 = 2.68, p < 0.05 andF3, 60 = 2.75, p < 0.05, respectively)

Full size image

Effect of Male Mass on Mating Behaviour

Heavier males tended to have lower mating latencies (r =−0.231, p < 0.05, Fig. 3a). We found significant negative correlations between male body mass and remating latency (r =−0.251, p < 0.05, Fig. 3b), mating duration (r =−0.402, p < 0.01, Fig. 3c) and remating duration (r =−0.346, p < 0.01, Fig. 3d).

Fig. 3
figure 3

Relationship between male body mass and a mating latency, b remating latency, c Mating duration and d Remating duration of B. terrestris. All four relationships are significant (r =−0.231, p < 0.05; r =−0.251, p < 0.05; r =−0.402, p < 0.01; r =−0.346, p < 0.01)

Full size image

Effect of Sex Ratio on Mating Behaviour

The sex ratio in the mating cages significantly affected mating latency; it was significantly longer at high male:queen ratios compared to low ratios (ANOVA: F3, 62 = 3.20, p < 0.05) but mating duration was not significantly different (ANOVA: F3, 62 = 1.52, p = 0.21) (Fig. 4).

Fig. 4
figure 4

Effect of queen to male ratio on mating latency and duration of B. terrestris (mean ± se). Latency increased when more males were present (F3,62 = 3.20, p < 0.05) but mating duration did not change significantly (F3,62 = 1.52, p = 0.21)

Full size image

Sexual Selection on Age and Morphology

Table 1 lists the linear and nonlinear selection gradients estimated via these mating trials. The bootstrapped 95% confidence regions around these gradients (unbiased by the nonindependence of males in the same mating trial, see Methods) are indicated below the selection gradients in Table 1 of the online supplementary material.

Table 1 The vector of standardized linear selection gradients (β) plus the bootstrapped confidence intervals for them, and the matrix of quadratic and correlational selection gradients (γ) on male bumblebee age and morphology. The confidence intervals around the quadratic and correlational selection gradients are provided in Table 1 of the online supplementary material. Randomization probabilities: ° P < 0.1; * P < 0.05; ** P < 0.01
Full size table

Our multivariate analysis revealed significant linear selection for larger fore- and hind-tibiae, but no significant selection on mass or age once other aspects of the phenotype were accounted for (Table 1). There was little evidence of quadratic or correlational selection on morphology, with only two of the 21 nonlinear terms in the regression approaching significance (both P < 0.1; Table 1). Neither of these terms featured age; instead they represented the cross products of mass and hind tibia length, and wing length and mid tibia length.

To further explore nonlinear selection in our data set, we conducted a canonical rotation of the γ matrix of nonlinear selection gradients to isolate the main multivariate axes of nonlinear selection (Phillips and Arnold 1989). This procedure revealed two axes of multivariate phenotypic space that were significantly related to fitness, one of which was primarily disruptive, and the other primarily stabilizing (the details of this canoncial analysis are provided in Table 2 of the online supplementary material). Age did not load strongly on either of the two axes representing most of the nonlinear selection.

As a final confirmatory step, we visualized selection on leg lengths and male age using nonparametric splines that were not constrained to fit straight lines. The linear selection on male leg lengths was clearly evident in these plots, and present both in the raw data (morphological traits plotted against mating success, Fig. 5) and in partial plots illustrating selection once other traits were taken into account (not shown). By contrast, there was no hint of any directional or nonlinear selection on age in spite of the findings described above related to mating latency (Fig. 5). A thin-plate spline visualization of the nonlinear selection suggested by our canonical analysis revealed that this nonlinear selection was mainly due to minor curvature in the fitness surface, rather than classically stabilizing or disruptive selection (see Fig. 2 of the online supplementary material).

Fig. 5
figure 5

Cubic splines illustrating positive linear sexual selection on fore tibia and hind tibia lengths in male B. terrestris, but no apparent sexual selection on age. These splines are on raw data rather than residuals, but the patterns for residuals are the same (data not shown)

Full size image

Discussion

In early-switching nests (which produce most male bumblebees) males emerge ~18 days before queens emerge from late-switching nests, and one might therefore predict that older males have a mating advantage. However we found that when a single male and queen were enclosed together, older bumblebee males were slower to initiate mating and copulated for longer, and that this pattern was repeated if males were given a chance to mate with a second queen. In separate experiments, male body mass was a significant predictor of the speed of initiation of mating and of copulation duration, with large males performing more quickly. Taken in isolation, these results would suggest that bumblebee nests should be under selective pressure to produce large males, and to produce them only a short time interval before the first virgin queens emerge (optimally about 5 days, the time it takes for males to mature and leave the nest, Greeff and Schmid-Hempel 2008). Our results therefore suggest that males from early-switching nests may be at a significant disadvantage in competition for mates with the smaller number of younger males produced from late-switching colonies.

Based on these data on mating latency, when two males were enclosed with a single female in a competitive situation, we predicted that younger, larger males would be most successful in obtaining mates. However, this was not the case; age did not predict mating success. The only factors which did influence mating success were the fore and hind tibia lengths (each contributing significant explanatory power to the model), with males having longer legs being more likely to mate. When two males were confined with a single virgin female, antagonistic interactions were frequent with the males often wrestling for position on top of the female and occassionally displacing each other. During courtship, male bumblebees climb on to the back of the female and grasp her thorax with their forelegs while stimulating her abdomen with their hind legs (MRA, pers. obs.). Males are considerably smaller than females, and relatively long legs may facilitate both gripping the female and stimulating her simultaneously. Of course it is also possible that long fore and hind legs are not themselves mechanistically responsible for the mating advantage we observed, and that instead some unmeasured trait or traits that covary strongly with the lengths of these legs mediates the outcome of sexual competition in bumblebees. It has been previously suggested that larger males may have larger genitalia that act as stronger forceps for attaching to the queen’s sting apparatus (Williams 1985; Duvoisin et al. 1999). In the dung fly Scatophaga stercoraria, Simmons and Parker (1992) reported that larger males are able to pull the aculeus of the queen more forcefully and hence mated more quickly. However, we did not investigate the dimensions of the male genitalia, and how they covary with leg lengths is unknown.

The magnitude of directional selection we report is slightly less than the mean of values commonly reported for selection in nature (Kingsolver et al. 2001), and the fact that selection is rather weak may explain the persistence of variation in male performance traits even if these traits feature substantial genetic variation. Alternatively, it is plausible that balancing selection within or across the sexes acts to maintain variation in leg lengths among male bees. For example, leg length may negatively influence aerodynamics, which may be important in male patrolling behaviour (reviewed in Goulson 2010), or it may be that longer-legged males may be more susceptible to becoming caught in spider’s webs. Given that they may have to survive for several weeks in the field before they encounter a mate, any factors increasing mortality could readily offset a mating advantage. At present we have no information on the heritability of leg length in bumblebees, so there is a clear need for further research in this area.

With regard to the reproductive strategies adopted by bumblebee nests, our results do not explain why early-switching colonies produce males long before queens are likely to be available to be mated, but they do illustrate that older males (up to 24 days) may suffer no disadvantage when competing with young males. They also somewhat call into question the argument that male size is less closely linked to reproductive success than is female size (Beekman and Van Stratum 1998). The evidence to date suggests that size is likely to be positively correlated with reproductive success in both sexes, but in which it has the strongest relationship remains unknown. In the context of the more general question concerning evolutionary explanations for reproductive ageing, our data suggest caution in interpreting associations between male age and reproductive success based on no choice trials or without appropriate consideration of possibly confounding correlated phenotypic traits. More studies adopting a multivariate perspective are needed, ideally featuring realistic conditions that simulate natural levels of mate competition.

The copulation durations observed in our study are similar to those reported in previous studies (e.g. Roseler 1973; Amin et al. 2009), with a mean duration of copulation varying from ~28 to 36 min for the youngest and oldest bees, respectively. Duvoisin et al. (1999) observed the copulation behaviour of B. terrestris using 10 day old queens and 15 day old males and reported that copulations lasted on average 37 min but that most sperm were transferred into the queen genital tract within the first 2 min. The remaining time in copulation may be required for transfer of the ‘mating plug’, a sticky substance that completely fills the queen bursa copulatrix after mating and prevents remating (Baer et al. 2000, 2001). Males transfer the mating plug to the female within 10–30 min of the onset of copulation (Duvoisin et al. 1999). The differences of mating duration among the age groups in our study may be due to differences in the speed of sperm and mating plug migration, or to differences in the amount of material transferred. It seems intuitive that older males who take longer to initiate mating may transfer sperm more slowly, but in the beetle Propylea dissecta older males mate for longer but transfer more sperm than younger males (Pervez et al. 2004). In agreement with our finding of a significantly longer copulation duration when mating for a second time, Roseler (1973) reported that B. terrestris males may mate up to eight times, and show an increase in copulation duration across subsequent matings. Similar patterns have been observed in other insects (e.g. Hughes and Chang 2000).

Our studies of mating latency in relation to sex ratio are of some practical relevance to those wishing to induce mating in bumblebees, for they suggest that intermediate ratios of 1:5–1:10 females to males produce the most rapid onset of copulation. The results are also interesting from a behavioural viewpoint; when only a single male is present, mating latency can be prolonged often because the male is inactive. The presence of other males seems to stimulate activity (MRA, pers. obs.). However, when large numbers of males are present, latency increases. In this situation several males are usually attempting to mate with the female simultaneously, and it seems likely that the increase in mating latency is due to interference between males.

There have now been many studies of the mating behaviour of B. terrestris, and of colony-level reproductive strategies. Some aspects remain poorly understood, notably the male-biased sex ratio, the factors which determine whether nests adopt an early or late-switching strategy, and why early-switching nests produce males long before there ought to be mates for them to copulate with. However, our understanding may be hampered by the heavy bias towards lab studies which may not accurately reflect the situation in the field; the reproductive strategies adopted by natural, wild nests have rarely been studied, and mating behaviour is exceedingly difficult to study under natural conditions, providing substantial challenges for future researchers.