Components of lifetime mating success and body size in males of a scrambling damselfly

ANIMAL BEHAVIOUR, 2000, 59, 339–348 doi: 10.1006/anbe.1999.1309, available online at http://www.idealibrary.com on

Components of lifetime mating success and body size in males of a scrambling damselfly R. STOKS

Department of Biology, University of Antwerp (RUCA/UIA), Belgium and Department of Biological Sciences, Dartmouth College, Hanover (Received 14 June 1999; initial acceptance 28 July 1999; final acceptance 26 October 1999; MS. number: 6254)

Sexual selection is hypothesized to favour small body size in males of scrambling species, that is, those in which males obtain matings by actively searching for females. I tested this hypothesis in a natural population of the scrambling emerald damselfly, Lestes sponsa. Mating efficiency (matings/visit to the pond) was the most important factor explaining variation in male lifetime mating success (LMS; 71%). This suggests a large potential for sexual selection. Path analysis of male LMS suggested a quality factor that positively affected both mating efficiency and life span. In contrast with the small-male mating advantage hypothesis, part of this potential for sexual selection was realized as stabilizing selection on male body size, indicating that there may also be a lower limit to body size for mating efficiency. This also illustrates that the constancy of body size may be explained by sexual selection alone. Survival explained about 20% of the variation in LMS and random processes were potentially important for determining LMS. My results show the problems of using mating efficiency as a measure for the intensity of sexual selection and the need to distinguish between potential and realized selection pressures, especially when comparing the importance of natural and sexual selection. I discuss mechanisms that may have caused the intermediate-male mating advantage in this scrambling species. 

lifetime data to disentangle these selection pressures (e.g. Arnold & Wade 1984b). Such studies are, however, very time consuming and have been done on relatively few species, especially vertebrates (e.g. 20 out of 25 species in Clutton-Brock 1988). Furthermore, there is a bias towards studies dealing with animals showing mate choice or male contests. As a result, there is a need for studies dealing with phenotypic correlates of LMS in scrambling species (Andersson 1994; Andersson & Iwasa 1996). Scrambling can be defined as a mating system where males obtain matings by actively searching for females (Thornhill & Alcock 1983). The joint study of the variance partitioning of LMS and potential phenotypic correlates may shed light on the ongoing controversy of whether sexual selection and natural selection can operate in the same direction, favouring the same traits (Willson 1990). Darwin (1859, 1871) himself wrote that natural selection sometimes opposes sexual selection and sometimes reinforces it. Body size is an important fitness correlate in animals (e.g. Clutton-Brock 1988). Constancy of body size over time and across populations, despite the presence of additive genetic variance, as well as observations of low fitness for extreme phenotypes, suggests that stabilizing selection

Variation in lifetime mating success (LMS) can be generated by several nonexclusive processes: sexual selection, natural selection and/or random processes (Koenig & Albano 1986; Hubbell & Johnson 1987). Sexual selection is considered here as arising from competition for matings, and natural selection covers all other forms of selection in the wild (Andersson 1994). To reveal the separate effects of sexual selection and natural selection, many empirical studies have tried to partition the variation in mating success into, respectively, variation arising from mating efficiency on the one hand and from life span on the other (see references in Clutton-Brock 1988). However, the contributions of both mating efficiency and life span to varation in LMS give only the potential importance of the respective selection pressures. Indeed, unless phenotypic correlates between any of these components and a phenotypic character are found, no selection can be proven, and the existing variation may be explained completely by random processes (e.g. Fincke 1986; Bradbury & Andersson 1987; Downhower et al. 1987; Clutton-Brock 1988). Ideally, one should use Correspondence: R. Stoks, Department of Biological Sciences, Dartmouth College, Hanover 03755, New Hampshire, U.S.A. (email: [email protected]). 0003–3472/00/020339+10 $35.00/0

2000 The Association for the Study of Animal Behaviour

339



2000 The Association for the Study of Animal Behaviour

340

ANIMAL BEHAVIOUR, 59, 2

on size is common in nature (e.g. Endler 1986). Stabilizing selection has been shown to result from opposing sexual selection pressures (e.g. Moore 1990), from opposing natural selection pressures (e.g. Schluter & Smith 1986), or from a combination of natural and sexual selection pressures working in opposition (e.g. Neems et al. 1998). Andersson (1994) hypothesized that in scrambling species sexual selection would select for reduced male body size (see Discussion). Given a constancy in body size, this would necessitate a counteracting natural selection pressure. The majority of studies partition LMS using the method of Arnold & Wade (1984a, b). They suggest that for each stage during the life span one should calculate the opportunity for selection. The contributions of each stage and the covariances between them can then be expressed as percentages of the total opportunity for selection. Brown (1988) gave some statistical critique and described an alternative approach. He used the exact relationship between the variance in LMS and the variances and covariances of its components (for worked-out examples of both methods see Clutton-Brock 1988). Nevertheless, there are still drawbacks to both methods. They both require that the components of LMS are multiplicative, which may cause spurious covariations between them (e.g. Fincke 1986). Such covariances between factors (which may be statistical artefacts) tend to inflate or deflate the amount of variance explained by each of these factors and may even result in factors explaining a negative portion or a portion larger than 100% of the variation in LMS (e.g. Fincke 1986; Cordero 1995; D. Brown, personal communication). Furthermore, both methods only assume direct effects and ignore any causality in the relationships between components. A powerful tool to circumvent these problems is path analysis which was originally proposed by Wright (1920) to test explicit hypotheses about causal relationships. Path analysis is in essence a sequence of multiple regressions and correlations that are structured by an a priori hypothesis (Li 1975; Bollen 1993; Wootton 1994; Hayduk 1996; Schumacker & Lomax 1996; Loehlin 1998). The method is designed to explain covariances in causal terms, often through the use of unmeasured factors (Crespi & Bookstein 1989). Because it recognizes the importance of unmeasured factors in biological explanation (Crespi & Bookstein 1989; Kingsolver & Schemske 1991), it enables us to test for the existence of such factors. Moreover, it has the advantage that there is no longer a need for the variables included to be multiplicative. For the path analysis, I formulated two competing models, generated a priori, by postulating causal links between the predictor variables and the dependent variable LMS (see Fig. 3 in the Results). As a basic model I started with the three components of LMS suggested by Koenig & Albano (1986): adult survival (represented by life span), proportion of time devoted to reproduction (represented by number of visits to the pond) and mating efficiency. Because damselflies are not present at the breeding site every day of their reproductive life span, I used matings/visit to the pond as a measure of their mating efficiency as proposed by Fincke (1986, 1988).

Because matings typically occur at the pond, life span is thought to act only indirectly on LMS through number of visits to the pond. The other component affecting LMS is mating efficiency. To check for an unmeasured quality index that influences both life span and mating efficiency I included a correlation between them. This is equal to a path model that explicitly contains the quality factor with the correlation between life span and mating efficiency being replaced with direct effects of the quality factor on life span and mating efficiency (see Li 1975, pp. 146–148). As in the convention for path diagrams, a singleheaded arrow indicates a causal link. Where two factors covary through the influence of unknown or unmeasured background factors they are connected by a doubleheaded curved arrow. Arrows not originating at a variable represent residual (unexplained) variance (Li 1975; Loehlin 1998). Each causal link in the model is quantified by a path coefficient, which is equal to the standardized partial regression coefficient of the dependent variable on the predictor variable. Path coefficients for noncausal links are equal to the correlation between the two variables. I present here data on the variance partitioning of LMS in males of the scrambling emerald damselfly, Lestes sponsa (Hansemann, 1823), and examine the nature and direction of the selection pressures acting on body size. Among scrambling species, damselflies are ideal organisms to study in this context because their size is fixed at emergence, making it possible to distinguish between size and age effects within cohorts. Specifically, I focus on the following questions. (1) What is the potential role of sexual and natural selection in shaping LMS and what is the role of chance? (2) Do sexual and natural selection on body size reinforce or oppose each other? This allows us to test Andersson’s (1994) hypothesis that sexual selection should favour small body size in males of scrambling species. METHODS

Study Species and Study Site Lestes sponsa is widespread in large parts of Europe where it is the most abundant species of lestid, occurring mainly at ponds and fens (Askew 1988). It is a sexually dimorphic, medium-sized damselfly. After emergence, the damselflies leave the pond for ca. 20 days and return when mature (Stoks 1999). They spend the night away from the pond and visit it only for reproduction. Oviposition occurs mainly in Juncus effusus (Jo ¨ dicke 1997). Lestes sponsa is a typical scrambling damselfly where males and females meet at the oviposition site in European populations (Corbet 1999). Watanabe & Matsunami (1990) reported in a Japanese population of L. sponsa a lek-like mating system with the sexes meeting in the wood away from the oviposition sites. In contrast to populations in Europe (Stoks et al. 1997), males at the oviposition site do not harass tandem pairs and do not try to take over the female (Watanabe & Matsunami 1990). This striking geographical difference in mating system

STOKS: MALE MATING SUCCESS AND SIZE IN A SCRAMBLER

may be linked with an associated difference in habitat use. While in Europe populations typically occur in open ponds in meadows and heathland, and never in woods (Askew 1988; personal observation), in Japan they occur only in woods (M. Watanabe, personal communication). I studied a population of L. sponsa at a small pond (perimeter 124 m) near the University of Antwerp, northern Belgium, in 1994 which is part of a park landscape dominated by bushes and pastures. To the best of my knowledge the population at the study pond is highly isolated. Very few mature males wandered to another nearby pond. In a range of 2 km there is only one other pond inhabited by L. sponsa (ca. 50 m away). During five intensive searches at this pond (23, 31 July and 4, 11 and 14 August) only six individuals (out of 572) from the population were retrieved here (two males on 31 July and four males on 4 August). None of these was seen again at the study population, despite the high recapture rates of marked individuals in this study population (Stoks 1999, see also Jo ¨ dicke 1997). This is in agreement with the fact that lestids are very site faithful (Utzeri et al. 1984), like other sexually mature damselflies in general (e.g. Parr 1973; Banks & Thompson 1985a). I therefore assume that disappearance of an individual means its death rather than dispersal (see also Anholt 1991; Cordero 1995).

Monitoring of Lifetime Mating Success The first mature individuals were observed at the pond on 5 July. On 15 and 16 July I marked all those present at the pond and its surroundings. All males were individually marked with a number written on both hindwings with a permanent marker (Staedtler Pancolor permanent liner), which made it possible to recognize individuals without capturing them. I started monitoring mating success on 17 July. During the entire monitoring period I continued marking all unmarked males that entered the population. Since individuals did not visit the pond daily, to obtain a cohort of individuals whose reproductive life span was included in the sampling period I selected only those males that were first seen at the pond more than two median interval lengths between visits to the pond after 14 July (median 2.2 days, N=243 males). This means that all males seen before 19 July were excluded from the analysis of LMS (see also Fincke 1982; Cordero 1995). I monitored the population daily from 17 July to 11 August. The population size was constant until 1 August with ca. 260 males and 130 females (Stoks 1999). Afterwards the population decreased steadily, leaving only 14 individuals on 11 August. The following 2 days were rainy, and thereafter no individuals were seen at the study pond. This means that the flight season was about 5 weeks at the study pond, which is rather short for this species (Zettelmeyer 1986). Both population sex ratio (ca. 67% males) and daily operational sex ratios (>87.5% males) were male biased in my study population (Stoks et al. 1997; Stoks 1999) which is the general pattern observed in other scrambling damselflies (e.g. Fincke 1982, 1986; Anholt 1991; Cordero 1995).

I measured body size (abdomen length) with dial callipers (precision 0.05 mm). Repeatability, determined by measuring 29 males twice on different visits, was quite high (r=0.947, P<0.001). Each day at least two persons continuously walked slowly around the pond during the active period of the day (0900–1800 hours) recording the identification code, sex, time of day and reproductive state (mating versus nonmating) of every individual. Total reproductive life span was taken as the length from first to last visit, augmented with one half median interval length (0.6 days) (see also Fincke 1988). Throughout the text, life span refers to reproductive life span. Contact mate guarding in this species is very effective (Stoks et al. 1997). Therefore, individuals in tandem or copula were scored as having acquired a mate (as in Anholt 1991). I am confident that almost every mating activity was seen because (1) I focused on tandem pairs (most of them were noticed several times during their visit), (2) oviposition was restricted to discrete bushes of J. effusus (N=9 bushes), Phalaris arundinacea (N=4) and Alisma plantago-aquatica (N=2), and (3) an average pair spent more than 1 h in tandem. Pairs in which at least one individual was unmarked were captured and the unmarked damselflies were given a number. Because this disrupted the tandem link and the release of solitary females may inflate the observed number of matings, I excluded subsequent matings resulting from this operation. I measured selection on mating success, although this should ideally be done on reproductive success. By doing so, I ignored variation in fertilizations/mating. Several of the sources of variation in this component of total reproductive success are also potentially attributable to sexual selection (e.g. mate guarding, male-induced fertility, cryptic female choice; Fincke 1988). However, Fincke (1988) showed in another scrambling damselfly that variation in fertilizations/mating was numerically much less important than the components I did consider.

Statistical Analyses To test for effects of age on the probability (0 or 1) of mating during a visit I performed a mixed regression with individuals as a random effect and with a binomial error structure. I fitted the model using the MIXED procedure of SAS 6.12 (Littell et al. 1996). Correct degrees of freedom were obtained with the Satterthwaite option. To test whether mating was random, I compared observed mating frequencies with expected frequencies generated by a Poisson distribution with a mean calculated from the observations. I tested for goodness-of-fit between frequencies with a G test and pooled classes with expected frequencies smaller than 5 (Sokal & Rohlf 1995). I corrected G values following the Williams’s procedure; these corrected values are assumed to be chi-square distributed with the degrees of freedom equal to the number of classes minus two (because the mean was generated from the data; Sokal & Rohlf 1995). To test the goodness-of-fit of the model to the data and to compare the descriptive power of the constructed

341

ANIMAL BEHAVIOUR, 59, 2

nested path diagrams, I used structural equation modelling in the path analysis approach (Mitchell 1992). I have chosen to present a chi-square value and a rho value for an assessment of fit of a given model. When the chisquare value is nonsignificant, the null hypothesis of a fit cannot be rejected, and it is said that the corresponding model fits the data. On the other hand, when the chisquare test is significant, the null hypothesis of a fit between model and data is rejected. Rho values greater than 0.90 are considered indicative of a good practical fit (see Mitchell 1992). I tested a nested, restricted model against the basic model by comparing the chi-square values. The reduction in chi-square between the basic model and the nested one, where a single constraint is relaxed, is also distributed as chi-square with degrees of freedom equal to the difference in degrees of freedom between the two models. This gives a direct test of the significance of the constraint not shared by the two models (Mitchell 1992, 1994). For the same reason the goodness-of-fit test of the nested model has one degree of freedom more than the basic model. The effects of a predictor variable on LMS may be both direct and indirect. A degree of determination is always a round trip for that variable via another variable(s) that determines it (Li 1975). For the degree of determination of LMS in the basic model there are four round trips from LMS to LMS via visits or mating efficiency. These four trips are LMS–visits–LMS, LMS–mating efficiency–LMS, and then the complete round trip which can be done in both directions (hence twice; see Li 1975, page 116). The direct partial coefficients of determination of a predictor variable were calculated as the square of the respective path coefficient. Indirect partial coefficients of determination were calculated as twice the product of the coefficients along the round trip linking the predictor variable with LMS (see Li 1975; Weis et al. 1989). Life span, number of visits and LMS were log(x+1), transformed and mating efficiency was arcsine transformed to make the residuals of the path models normal. After the given transformations there was no curvature in the plots of the residuals from the multiple regression analysis for each independent variable included in the regressions, indicating that all responses were linear (Neter et al. 1996). Since path analysis allows only for linear effects and the relationships between size and some components of LMS were nonlinear, size could not be entered in the path analysis (Kingsolver & Schemske 1991). Therefore, I tested for a size effect on LMS and its components by performing separate multiple regressions with size and squared size as independent variables. The latter checks for stabilizing selection. This, however, holds only whenever the relationship between size and LMS (or one of its components) shows a clear mode within the range of observed phenotypes (Mitchell-Olds & Shaw 1987; Schluter 1988). To eliminate the possibility that other optima may exist beyond the range of the data, I also performed constrained regression analyses where I forced the optimum to be outside the range of the data (see Mitchell-Olds & Shaw 1987). All means are presentedSE.

100 50 Survival (%)

342

20 10 5

5

10

15

20

25

Age (days) Figure 1. Survivorship curves of mature male L. sponsa (N=326). The Y axis has a logarithmic scale.

RESULTS In total I caught 403 males of which 322 (80%) were reseen. This is the highest percentage reported among Lestidae (Jo ¨ dicke 1997). I selected 326 males for studying LMS (see Methods); of these 260 had been measured.

Visiting Pattern to the Pond and Life Span The mean number of visits that a male made to the pond was 5.420.23 (N=326). The majority of males (74.5%) returned after their first visit. These I called residents. The proportion of males that mated during their first visit to the pond did not differ between those visiting the pond only once (16.9%, N=83) and residents (16.0%, N=243; Fisher’s exact test: df=1, P>0.1). More males (83) did not return to the pond after the first visit than after their second visit (41;
21 =14.23, P<0.001), suggesting a marking effect. There was hardly any effect of age on interval lengths between successive visits of males to the pond (Spearman rank correlation: rS =
0.068, N=1446, P=0.01). The average reproductive life span was 10.000.41 days for all males and 13.070.38 days for resident males. The survivorship curve is a type III curve, indicating that daily survival rate is dependent on age with older individuals having a higher probability of dying (Fig. 1).

Lifetime Mating Success and Mating Efficiency I recorded 271 matings by the selected cohort of 326 males. Only six males mated twice on the same day. A majority of the males (51.8%) did not mate during the study period, while few were able to mate more than once (Fig. 2). Mean LMS was 0.840.06 for all males and 1.070.08 when including only residents (medians 0 and 1, respectively). Mating success deviated significantly from a Poisson distribution (G test: Gadj,2 =18.38, P<0.001; Fig. 2). When I included only residents I obtained the same pattern (G test: Gadj,3 =22.08, P<0.001). Much of the deviation from randomness in male mating success is a result of

STOKS: MALE MATING SUCCESS AND SIZE IN A SCRAMBLER

visit did not increase with age (mixed regression: r=
0.0200.011, t300.04 =
1.78, P=0.076).

Percentage of individuals

50 40

Components of Variance in Lifetime Mating Success

30 20 10 0

0

1 2 Lifetime mating success

>2

Figure 2. Observed ( ) and predicted ( ) lifetime mating success of male L. sponsa (N=326).

larger-than-expected numbers of nonmating males and of males that mated more than twice. I compared the distributions of LMS within groups with a similar number of visits to the pond. To obtain enough degrees of freedom I pooled males within six visiting classes (Table 1). In all five visiting classes with enough degrees of freedom, observed male mating frequencies did not deviate from frequencies expected under random mating (Table 1). Mating efficiency was 0.1500.013 matings/visit for all males and 0.1440.010 when excluding transients. Mating efficiency was significantly higher for males that visited the pond only once than for residents (0.1690.041 versus 0.1440.010 matings/visit; Mann– Whitney test: Z=4.7, N1 =83, N2 =243, P<0.001). The probability of remating of males first captured in copula (66.7%, 26/39) did not differ from that of those first captured while not in copula (51.0%, 104/204; Fisher’s exact test: df=1, P=0.081). The life spans of males captured in copula and those captured while not in copula also did not differ (14.640.42 versus 12.77 0.87 days; Mann–Whitney test: Z=1.77, N1 =39, N2 =204, P=0.076). Moreover, the probability of mating during a

To circumvent the marking effect in males and to overcome spurious negative correlations between life span and mating efficiency I continue the analysis only for residents (see also Fincke 1986). As a basic model for the path analysis of the sources of variation in LMS I started with the one shown in Fig. 3. My basic model adequately describes the data of male LMS (goodness-of-fit test:
22 =1.73, P=0.42). I also tested a reduced model, without the correlation between life span and mating efficiency. This made the fit of the model significantly poorer and even resulted in a nonsignificant path model (goodness-of-fit test nested model:
23 =11.8, P=0.008; comparison basic model versus nested model:
21 =10.07, P=0.0015). The biological relevance of the selected basic path model is supported by an R2 of 0.875 (this is (1
unexplained variance)=1
0.3542; see Fig. 3) and a rho value of 0.998. Based on the selected path diagram, mating efficiency has the greatest direct influence on male LMS, with a path coefficient of 0.786, whereas the coefficient of visits is 0.401 (Fig. 3). The compound path coefficient of the indirect effect of life span is 0.303 (0.7570.401). Table 2 shows the partial coefficients of determination. The number of visits determines 16.1% of the variance of LMS directly, mating efficiency determines 61.8% directly, and life span and mating efficiency jointly determine 9.6% (two times the round trip: 2(0.7570.401 0.7860.202)). The latter is the variation in LMS that can be traced to an unmeasured quality factor. When direct and indirect effects are summed mating efficiency explains 71.4%, and life span 18.8% (two times the round trip+(0.7570.401)2) of the variation in LMS. Note that the direct effects of visits and mating efficiency together

Table 1. Comparison of observed mating frequencies with expected frequencies, generated by a Poisson distribution with the same mean, for males with a similar number of visits to the pond Number of matings Visits

N

Mean

0

1

2

3

>3

Gadj

df

P

1–2

118

0.15

1

0.76

39

0.69

1.11

1

0.29

7–8

50

1.12

2.07

2

0.36

9–10

38

1.58

0.25

2

0.88

>10

40

2.18

0 (0.0) 0 (0.2) 1 (0.1) 2 (1.4) 5 (2.9) 8 (7.1)

0.1

5–6

0 (0.1) 0 (0.9) 0 (1.1) 1 (3.8) 3 (5.1) 6 (7.8)

—

0.63

0 (1.1) 6 (4.3) 3 (4.7) 12 (10.2) 9 (9.8) 13 (10.7)

0

41

18 (15.2) 14 (13.8) 16 (13.5) 21 (18.3) 12 (12.4) 7 (9.9)

—

3–4

100 (101.6) 21 (21.8) 19 (19.6) 14 (16.3) 9 (7.8) 6 (4.5)

1.13

3

0.77

N: Sample size; mean: average lifetime mating success; expected frequencies are given within parentheses. Cells with expected frequencies less than five were pooled in the analysis.

343

ANIMAL BEHAVIOUR, 59, 2

0.654 ± 0.032** 0.757 ± 0.027** Life span Visits 0.40 1± 0.0 29 **

1.4

0.354 ± 0.021**

Mating efficiency

.0

6 .78

±0

87

0.8

30

35 3

0.6 0.4

0

0.2

Figure 3. Selected path diagram showing relationships between components of lifetime mating success (LMS) for male L. sponsa. Causal links are shown by straight arrows and noncausal links by curved arrows. Vertical arrows indicate the path between the variable and unknown or unmeasured causes; the corresponding coefficient is the square root of the residual variance. Numbers accompanying the arrows are path coefficients given with 1 SE. Significance of path coefficients is indicated: *P<0.01; **P<0.001. Table 2. Partial coefficients of determination of variance in male lifetime mating success (LMS)

2

0.0 0.25 Mating efficiency

0.202 ± 0.062*

23

12

1.0 LMS

LMS **

91

1.2

Males

0.20 0.15 0.10 0.05 0.00

Direct

Indirect

Total†

Visits Mating efficiency Life span

0.161 0.618 —

0.096* 0.096 0.188

0.257 0.714 0.188

*It may seem counterintuitive that there is an indirect effect of visits on LMS through life span and mating efficiency, as the connecting path in this direction would pass through a unidirectional link in the opposite direction. This indirect effect can best be understood by partitioning the correlation between visits and LMS into its direct and indirect components. The correlation between visits and LMS is 0.52 and can be divided into the direct effect (0.401) and an indirect effect (0.757×0.202×0.786=0.12) working through life span and mating efficiency (see Fig. 3). †The totals in this column do not sum to unity. The three nonoverlapping components in the path diagram are the direct effects of visits (0.161) and mating efficiency (0.618) and the indirect joint effect they have in common (0.096). These effects sum to 0.875, the R2 of the path model. The indirect effect (0.096) has been added to both the direct effect of visits and mating efficiency, while the indirect effect of life span should not be considered separately because it is included completely in the effects of visits and mating efficiency.

6 5 Visits

Source

4 3 2 1 0 12

Life span (days)

344

11 10 9 8 7 6

2

with their joint contribution add up to the R value of the model (0.161+0.618+0.096=0.875, see Methods). Including males with only one visit to the pond gives qualitatively similar results.

Body Size and Components of Lifetime Mating Success There was no relationship between size and date of marking (r258 =0.01, N=260 males, P=0.86). Larger males visited the pond more often (r258 =0.14, N=260 males, P=0.023). There was no monotonic increasing or decreasing relationship beween size and life span, LMS or mating efficiency (r258 =0.09, N=260 males, P=0.14, r258 =0.09, P=0.17 and r258 =0.04, P=0.52). Mating was

27

28

29

30

31

32

33

Abdomen length (mm) Figure 4. Relationships between body size and components of lifetime mating success (LMS) in male L. sponsa. Error bars represent standard errors and numbers above error bars are sample sizes.

not assortative by size (r211 =0.04, N=213 matings where male and female size is known, P=0.47). Males that mated were less variable in size than unmated males (variances 1.37 and 0.77, respectively; Levene’s test: F1,258 =10.67, P=0.0016). On a finer scale, intermediate males had a higher LMS than males with extreme sizes (partial regression coefficient for the squared component of size: =
6.002.36, P=0.012; Fig. 4) indicating stabilizing selection on body size with

STOKS: MALE MATING SUCCESS AND SIZE IN A SCRAMBLER

respect to LMS. There was also stabilizing selection on body size with respect to mating efficiency (=
5.03 2.38, P=0.036) and number of visits to the pond (=
5.012.36, P=0.034; Fig. 4). Note that the latter shows that the largest males made fewer visits to the pond, despite the significant positive correlation between size and number of visits (see above). Although males of intermediate size also seemed to live longer, this effect was not significant (=
2.512.39, P=0.29; Fig. 4). In all regressions showing significant quadratic coefficients, both with the optimum set above and below the range of the data, the unconstrained regressions with the intermediate optimum provided the better fit (all P<0.05). Despite the significant stabilizing effects on body size relative to several components of LMS, and LMS itself, these effects were very weak (coefficients of determination, LMS: 3.6%; mating efficiency: 1.9%; number of visits: 4.1%). DISCUSSION Partitioning the variance in male LMS showed that mating efficiency was the most important contributing factor (71%), suggesting a large potential for sexual selection, and a low potential for natural selection. Path analysis suggested the presence of a quality factor positively affecting both life span and mating efficiency. Unexpectedly, the results do not corroborate Andersson’s (1994) hypothesis that small males should have a higher mating efficiency and show that stabilizing selection on male body size was completely due to sexual selection. I discuss below the relative importance of the different selective forces and random processes in shaping LMS and the finding of an intermediate-male mating advantage in this scrambler.

Determinants of Variation in Lifetime Mating Success Just over half the males (52%) never mated. This is in accordance with the male-biased operational sex ratio in my study population (Stoks et al. 1997; Stoks 1999) which is probably the major force causing the deviation of the male distribution of LMS from Poisson (see Sutherland 1987). A surplus of males at the pond resulted in a low mating efficiency, with many males not mating during a visit. As a result, male LMS was primarily (71%) determined by mating efficiency which creates the potential for sexual selection to have an important role. However, this is only a potential (upper limit) and might not have been realized. Even when a mating distribution does deviate from Poisson, there may be a substantial random (Poisson) component inflating the variance in LMS. In Poisson processes the variance should equal the mean. Mackenzie et al. (1995a, b) suggested excluding the Poisson component from the variance in LMS by simply subtracting the mean from the variance; this subtracts that part of the variation in LMS that can be explained by random processes. This would mean that of the total variance in male LMS (1.30), about 64% (mean 0.84) can

be explained by Poisson processes and even all of the variance within groups of males with a similar number of visits to the pond. This indicates that random encounters with females may be enough to explain variation in the mating efficiency of males (see also Banks & Thompson 1985b; Sutherland 1985; Fincke 1986; Cordero 1995) unless a phenotypic correlate of mating efficiency can be determined. Path analysis of the components of male LMS suggests the presence of an unmeasured quality factor positively affecting both life span and mating efficiency and explaining ca. 10% of the variation in LMS. This suggests that sexual selection and natural selection reinforce each other. Cordero (1995) also reported a positive relationship between reproductive life span and mating efficiency in one of his two populations of Ischnura graellsii, but he attributed this to a learning effect. This alternative explanation does not hold in my study population, however, since there was no relationship between age and mating efficiency. The absence of a learning effect may be due to the low probabilities of a male encountering a female and/or counteracting ageing effects (see also Fincke 1982). Given the phenotypic correlate of male LMS, not all of the variation in LMS was due to random processes. Furthermore, the ca. 10% explained by the suggested quality factor gives a lower limit for the importance of sexual selection because it considers only that phenotypic part that positively influences both life span and mating efficiency. Other unmeasured characteristics may exist that influence only mating efficiency. However, I could trace only a small portion of the variation in male mating efficiency to phenotypic differences between males. Therefore, only a small portion of the potential for sexual selection might have been realized. The hypothesis that a large part of the variation in mating efficiency and LMS is due to random processes cannot therefore be rejected. Life span explained a rather small part of the variation in LMS (ca. 20%), which is much smaller than the part explained by mating efficiency (71%). This is often interpreted as showing that sexual selection is more important than natural selection in shaping LMS (see e.g. Fincke 1988; Fincke et al. 1997). On the other hand, the fact that the deviation of the distribution of LMS from a random distribution disappeared within groups of males with a similar number of visits to the pond (hence a similar life span) suggests that most of the super-Poissonian variance in LMS was due to variation in life span. This would mean that the realized sexual selection pressure is lower than the realized natural selection pressure. Unfortunately, given the small portion of the variation in life span that can be explained, my results can prove only that a small fraction of the potential for natural selection is realized. The unmeasured quality factor positively influenced life span and there was a nonsignificant trend for larger males to live longer. Several theoretical (e.g. Sutherland 1985, 1987; Koenig & Albano 1986; Downhower et al. 1987; Hubbell & Johnson 1987; Mackenzie et al. 1995a, b) and empirical studies (e.g. Banks & Thompson 1985b; Fincke 1986; Michiels & Dhondt 1991; Grether 1996) have argued

345

346

ANIMAL BEHAVIOUR, 59, 2

against using the variation in LMS as a measure and evidence for sexual selection because it includes both natural selection and random processes. Koenig & Albano (1986) argued that variation in mating efficiency would better represent the potential for sexual selection, but that it might underestimate it in some circumstances. Clearly my results show that it may also overestimate it (see also Fincke 1986). Blindly using the variation in mating efficiency instead of LMS as a measure of sexual selection may therefore only shift the problem because both natural selection (if mating efficiency is age dependent, e.g. through learning; see Grether 1996) and random processes may still be included. Furthermore, my results argue against using relative contributions of mating efficiency and life span for making inferences over the relative roles of sexual and natural selection in shaping the variation in LMS, as done by several authors, because the pattern of the realized pressures may even reverse.

Male Size and Components of Lifetime Mating Success I found stabilizing selection on size relative to LMS and mating efficiency. This could have been an artefact of emigration or immigration of males. However, even in the unlikely event that severe emigration of sexually mature males occurred (see Methods), this probably did not seriously affect this finding. Indeed, in field enclosures, excluding any emigration, I also found the same relationships with size in this species (Stoks 1999). Furthermore, only natal dispersal is thought to be important in damselflies (Corbet 1999) and Anholt (1990) showed that larger, freshly emerged males disperse more. As a result I might have missed the largest males in my population. Including these males would probably make the drop in mating efficiency in the larger male classes even more significant. Therefore, I believe the observed relationship is real. In his review of studies showing a relationship between mating success and phenotypic characteristics, Andersson (1994) reported only four (out of 232) studies showing such a link in scrambling species. Given that in a scrambling species rapid location of the mate is crucial, Andersson (1994) hypothesized that characters linked with searching and mobility would be important and that sexual selection would favour a small body size. Several studies support this hypothesis (e.g. Banks & Thompson 1985b; McLachlan & Allen 1987; Neems et al. 1990; Blanckenhorn et al. 1995; Dunn et al. 1999). However, my results do not completely confirm this prediction and suggest that there may be a lower limit to size for mating efficiency, resulting in sexual selection for an intermediate body size. As far as I know, my study is the first to show stabilizing sexual selection on body size in a scrambling species. The constancy of body size can thus be explained by sexual selection alone, without the need for a counteracting natural selection pressure. Indeed, no counteracting natural selection pressures on size could be detected. Stabilizing total selection on size or mass relative to male LMS has been shown explicitly in two other

scrambling damselflies: Coenagrion puella (Banks & Thompson 1985b) and Enallagma hageni (Fincke 1982, 1988); in addition, the figures for the relationship between size and male LMS for the 1986 study population of Enallagma boreale (Anholt 1991), for Ischnura elegans (Cordero et al. 1997) and both study populations of I. graellsii (Cordero 1995) show an intermediate optimum. Both of these explicit studies, however, argued that the stabilizing effect on size relative to LMS was a combination of natural and sexual selection. Banks & Thompson (1985b) found that larger males had a longer life span, but a lower daily mating rate, indicating directional natural selection working in opposition to directional sexual selection. Fincke (1988) suggested that small size was more advantageous for survivorship than for mating efficiency. She showed that selection was stabilizing only on LMS and visits and not on life span and mating efficiency. In contrast with the findings of these studies, I have shown that sexual selection alone is enough to create a stabilizing selection pressure relative to LMS on size (see Moore 1990 for a territorial dragonfly). Small size may affect mating efficiency in several nonexclusive ways. First, smaller males may be more manoeuvrable, and therefore better at detecting and ‘scrambling’ females (Banks & Thompson 1985b; McLachlan & Allen 1987; Neems et al. 1990; Anholt 1991). Second, small males may have lower flight costs per unit time than large males and therefore may allocate more time to active searching for females instead of foraging (Ghiselin 1974; Reiss 1989): the so-called ‘Ghiselin–Reiss small-male hypothesis’ (Blanckenhorn et al. 1995; Blanckenhorn & Viele 1999). This may be associated with an increased willingness to mate in small males (e.g. Dunn et al. 1999). Third, an ignored factor in this context may be sizelinked thermoregulatory capabilities. Several flying insects have limited ability to avoid overheating (for L. sponsa see Watanabe & Taguchi 1993) and several studies have shown that in insects a substantial increase in body temperature above ambient can occur with a relatively small increase in body size (e.g. Carroll & Quiring 1993). During the very hot study period of my study, smaller males may therefore have spent more time actively searching for females before they became overheated compared with larger males (Willmer 1991). The latter two mechanisms can be seen as increasing the duration and/or the searching intensity per unit of time during a single visit. On the other hand, being too small may prevent successful capture of a female and tandem formation (e.g. Anholt 1991; Dunn et al. 1999). Furthermore, smaller males will cool down more rapidly (Cossins & Bowler 1987) which may be important in ectotherms during periods with strong temperature fluctuations, for example when it is cloudy. Since several of these mechanisms may operate simultaneously and potentially oppose each other, the appearance of a small-male mating advantage will depend on their relative importance. This may depend on factors such as weather and food availability but also on population parameters such as sex ratio. For example, Carroll & Salamon (1995) showed that with increasing sex ratios mate-searching

STOKS: MALE MATING SUCCESS AND SIZE IN A SCRAMBLER

capabilities became increasingly important for male mating efficiency in scrambling bugs. Clearly there is a need for more experimental testing of the exact mechanisms linking body size with lifetime mating success in scrambling species and how these depend on other factors. This knowledge may reveal in which circumstances a small-male mating advantage in males of scrambling species is to be expected. Acknowledgments I am grateful to the following for their help and stamina in the field: M. De Block, L. De Bruyn, S. De Vocht, A. Stoks, J. Van Acken, B. Van de Vijver, B. Van Hooydonck and G. Van Reyn. I thank T. Benton, A. Cordero, L. De Bruyn, E. Matthysen, N. Michiels, H. Van Gossum and an anonymous referee for comments on the manuscript, D. Brown, A. Mackenzie, Mark McPeek and Jean Richardson for stimulating discussions, and D. Brown and M. Watanabe for personal communications. I was financially supported by an aspirantship of the Fund for Scientific Research-Flanders (FWO-Flanders). References Andersson, M. 1994. Sexual Selection. Princeton, New Jersey: Princeton University Press. Andersson, M. & Iwasa, Y. 1996. Sexual selection. Trends in Ecology and Evolution, 11, 53–58. Anholt, B. R. 1990. Size-biased dispersal prior to breeding in a damselfly. Oecologia, 83, 385–387. Anholt, B. R. 1991. Measuring selection on a population of damselflies with a manipulated phenotype. Evolution, 45, 1091–1106. Arnold, S. J. & Wade, M. J. 1984a. On the measurement of natural and sexual selection: theory. Evolution, 38, 709–719. Arnold, S. J. & Wade, M. J. 1984b. On the measurement of natural and sexual selection: applications. Evolution, 38, 720–734. Askew, R. R. 1988. The Dragonflies of Europe. Essex: Harley Books. Banks, M. J. & Thompson, D. J. 1985a. Emergence, longevity and breeding area fidelity in Coenagrion puella (L.) (Zygoptera: Coenagrionidae). Odonatologica, 14, 279–286. Banks, M. J. & Thompson, D. J. 1985b. Lifetime mating success in the damselfly Coenagrion puella. Animal Behaviour, 33, 1175– 1183. Blanckenhorn, W. U. & Viele, S. N. T. 1999. Foraging in yellow dung flies: testing for a small-male time budget advantage. Ecological Entomology, 24, 1–6. Blanckenhorn, W. U., Preziosi, R. F. & Fairbairn, D. J. 1995. Time and energy constraints and the evolution of sexual size dimorphism: to eat or to mate? Evolutionary Ecology, 9, 369–381. Bollen, K. A. 1993. Testing Structural Equation Models. Newbury Park: Sage. Bradbury, J. W. & Andersson, M. B. 1987. Sexual Selection: Testing the Alternatives. New York: J. Wiley. Brown, D. 1988. Components of lifetime reproductive success. In: Reproductive Success (Ed. by T. H. Clutton-Brock), pp. 439–453. Chicago: University of Chicago Press. Carroll, A. L. & Quiring, D. T. 1993. Interactions between size and temperature influence fecundity and longevity of a tortricid moth, Zeiraphera canadensis. Oecologia, 93, 233–241. Carroll, S. P. & Salamon, M. H. 1995. Variation in sexual selection on male body size within and between populations of the soapberry bug. Animal Behaviour, 50, 1463–1474.

Clutton-Brock, T. H. 1988. Reproductive Success. Chicago: University of Chicago Press. Corbet, P. S. 1999. Dragonflies, Behaviour and Ecology of Odonata. Colchester: Harley Books. Cordero, A. 1995. Correlates of male mating success in two natural populations of the damselfly Ischnura graellsii (Odonata: Coenagrionidae). Ecological Entomology, 20, 213–222. Cordero, A., Carbone, S. S. & Utzeri, C. 1997. Male mating success in a natural population of Ischnura elegans (Vanderlinden) (Odonata: Coenagrionidae). Odonatologica, 26, 459–465. Cossins, A. R. & Bowler, K. 1987. Temperature Biology of Animals. London: Chapman & Hall. Crespi, B. J. & Bookstein, F. L. 1989. A path-analytical model for the measurement of selection on morphology. Evolution, 43, 18–28. Darwin, C. 1859. On the Origin of Species by Means of Natural Selection. London: Murray. Darwin, C. 1871. The Descent of Man, and Selection in Relation to Sex. London: Murray. Downhower, J. F., Blumer, L. S. & Brown, L. 1987. Opportunity for selection: an appropriate measure for evaluating variation in the potential for selection? Evolution, 41, 1395–1400. Dunn, D. W., Crean, C. S., Wilson, C. L. & Gilburn, A. S. 1999. Male choice, willingness to mate and body size in seaweed flies (Diptera: Coelopidae). Animal Behaviour, 57, 847–853. Endler, J. A. 1986. Natural Selection in the Wild. Princeton, New Jersey: Princeton University Press. Fincke, O. M. 1982. Lifetime mating success in a natural population of the damselfly, Enallagma hageni (Walsh) (Odonata: Coenagrionidae). Behavioral Ecology and Sociobiology, 10, 293–302. Fincke, O. M. 1986. Lifetime reproductive success and the opportunity for selection in a non-territorial damselfly (Odonata: Coenagrionidae). Evolution, 40, 791–803. Fincke, O. M. 1988. Sources of variation in lifetime reproductive success in a non-territorial damselfly (Odonata: Coenagrionidae). In: Reproductive Success (Ed. by T. H. Clutton-Brock), pp. 24–43. Chicago: University of Chicago Press. Fincke, O. M., Waage, J. K. & Koenig, W. D. 1997. Natural and sexual selection components of odonate mating patterns. In: The Evolution of Mating Systems in Insects and Arachnids (Ed. by J. C. Choe & B. J. Crespi), pp. 58–74. Cambridge: Cambridge University Press. Ghiselin, M. T. 1974. The Economy of Nature and the Evolution of Sex. Berkeley: University of California Press. Grether, G. F. 1996. Sexual selection and survival selection on wing coloration and body size in the rubyspot damselfly Hetaerina americana. Evolution, 50, 1939–1948. Hayduk, L. A. 1996. LISREL Issues, Debates, and Strategies. Baltimore: Johns Hopkins University Press. Hubbell, S. P. & Johnson, L. K. 1987. Environmental variance in lifetime mating success, mate choice, and sexual selection. American Naturalist, 130, 91–112. Jo ¨ dicke, R. 1997. Die Binsenjungfern und Winterlibellen Europas. Magdeburg: Westarp. Kingsolver, J. G. & Schemske, D. W. 1991. Path analysis of selection. Trends in Ecology and Evolution, 6, 276–280. Koenig, W. D. & Albano, S. S. 1986. On the measurement of sexual selection. American Naturalist, 127, 403–409. Li, C. C. 1975. Path Analysis: a Primer. California: Boxwood Press. Littell, R. C., Milliken, G. A., Stroup, W. W. & Wolfinger, R. D. 1996. SAS System for Mixed Models. Cary: SAS Institute. Loehlin, J. C. 1998. Latent Variable Models: an Introduction to Factor, Path and Structural Analysis. 3rd edn. Mahwah, New Jersey: L. Erlbaum. Mackenzie, A., Reynolds, J. D. & Sutherland, W. J. 1995a. Measuring sexual selection. Nature, 376, 471.

347

348

ANIMAL BEHAVIOUR, 59, 2

Mackenzie, A., Reynolds, J. D., Brown, V. J. & Sutherland, W. J. 1995b. Variation in male mating success on leks. AmericanNaturalist, 145, 633–652. McLachlan, A. J. & Allen, D. F. 1987. Male mating success in diptera: advantages of small size. Oikos, 48, 11–14. Michiels, N. K. & Dhondt, A. A. 1991. Sources of variation in male mating success and female oviposition rate in a non-territorial dragonfly. Behavioral Ecology and Sociobiology, 29, 17–25. Mitchell, R. J. 1992. Testing evolutionary and ecological hypothesis using path analysis and structural equation modelling. Functional Ecology, 6, 123–129. Mitchell, R. J. 1994. Path analysis: pollination. In: Design and Analysis of Ecological Experiments (Ed. by S. M. Scheiner & J. Gurevitch), pp. 211–229. London: Chapman & Hall. Mitchell-Olds, T. & Shaw, R. G. 1987. Regression analysis of natural selection: statistical inference and biological interpretation. Evolution, 41, 1149–1161. Moore, A. J. 1990. The evolution of sexual dimorphism by sexual selection: the separate effects of intrasexual selection and intersexual selection. Evolution, 44, 315–331. Neems, R. M., McLachlan, A. J. & Chambers, R. 1990. Body size and lifetime mating success of male midges (Diptera, Chironomidae). Animal Behaviour, 40, 648–652. Neems, R. M., Lazarus, J. & McLachlan, A. J. 1998. Lifetime reproductive success in a swarming midge: trade-offs and stabilizing selection for male body size. Behavioral Ecology, 9, 279–286. Neter, J., Kutner, M. H., Nachtsheim, C. J. & Wasserman, W. 1996. Applied Linear Statistical Models. Chicago: Irwin. Parr, M. J. 1973. Ecological studies of Ischnura elegans (Vander Linden) (Zygoptera: Coenagrionidae). II. Survivorship, local movements, and dispersal. Odonatologica, 2, 159–174. Reiss, M. J. 1989. The Allometry of Growth and Reproduction. Cambridge: Cambridge University Press. Schluter, D. 1988. Estimating the form of natural selection on a quantitative trait. Evolution, 42, 849–861. Schluter, D. & Smith, J. N. M. 1986. Natural selection on beak and body size in the song sparrow. Evolution, 40, 221–231. Schumacker, R. E. & Lomax, R. G. 1996. A Beginner’s Guide to Structural Equation Modelling. Mahwah, New Jersey: L. Erlbaum. Sokal, R. R. & Rohlf, F. J. 1995. Biometry. 3rd edn. New York: W. H. Freeman. Stoks, R. 1999. Natural and sexual selection in the damselfly Lestes sponsa. Ph.D thesis, University of Antwerp.

Stoks, R., Matthysen, E. & De Bruyn, L. 1997. The adaptiveness of intense contact mate guarding by males of the emerald damselfly Lestes sponsa (Odonata: Lestidae): the male’s perspective. Journal of Insect Behavior, 10, 289–298. Sutherland, W. J. 1985. Chance can produce a sex difference in variance in mating success and explain Bateman’s data. Animal Behaviour, 33, 1349–1352. Sutherland, W. J. 1987. Random and deterministic components of variance in mating success. In: Sexual Selection: Testing the Alternatives (Ed. by J. W. Bradbury & M. B. Andersson), pp. 209–219. New York: J. Wiley. Thornhill, R. & Alcock, J. 1983. The Evolution of Insect Mating Systems. Cambridge: Harvard University Press. Utzeri, C., Carchini, G., Falchetti, E. & Belfiore, C. 1984. Philopatry, homing and dispersal in Lestes barbarus (Fabricius) (Zygoptera: Lestidae). Odonatologica, 13, 573–584. Watanabe, M. & Matsunami, E. 1990. A lek-like system in Lestes sponsa (Hansemann), with special reference to the diurnal changes in flight activity and mate-finding tactics (Zygoptera: Lestidae). Odonatologica, 19, 47–59. Watanabe, M. & Taguchi, M. 1993. Thoracic temperatures of Lestes sponsa (Hansemann) perching in sunflecks in deciduous forests of the cool temperate zone of Japan (Zygoptera: Lestidae). Odonatologica, 22, 179–186. Weis, A. E., Wolfe, C. H. & Gorman, W. L. 1989. Genotypic variation and integration in histological features of the goldenrod ball gall. American Journal of Botany, 76, 1541–1550. Willmer, P. 1991. Thermal biology and mate acquisition in ectotherms. Trends in Ecology and Evolution, 6, 396–399. Willson, M. F. 1990. Sexual selection in plants and animals. Trends in Ecology and Evolution, 5, 210–214. Wootton, J. T. 1994. Predicting direct and indirect effects: an integrated approach using experiments and path analysis. Ecology, 75, 151–165. Wright, S. 1920. The relative importance of heredity and environment in determining the piebald pattern of guinea pigs. Proceedings of the National Academy of Sciences U.S.A., 6, 320–332. Zettelmeyer, W. 1986. Populationso¨ kologische Untersuchungen an der Kleinlibelle Lestes sponsa HANS. in einem Moorgebiet der Egge, Nordrhein-Westfalen. Telma, 16, 113–130.