Who wears the pants in a mute swan pair? Deciphering the effects of male and female age and identity on breeding success


Correspondence author. E-mail: jauld@wcupa.edu


  1. Traditionally, many breeding traits (e.g. the timing and size of clutches) were considered to be female-only traits in that males played little-to-no role in their expression. Although the contribution of males to such breeding traits, as well as other aspects of reproduction, is increasingly recognized, few studies have demonstrated the effects of male age and life history on breeding traits and, importantly, whether these effects are underlined by additive-genetic variation.
  2. Here, we take advantage of a long-term data set on mute swans (Cygnus olor) to demonstrate that the ages of both the male and female parents play significant roles in the timing and size of clutches, although recruitment success did not show similar effects. Individual males varied significantly in their influence on the timing of egg laying.
  3. We decomposed this variation using an ‘animal model’; competing models that were the source of this variation as additive-genetic or permanent-environmental variation was not statistically distinguishable.
  4. Our results add to the growing evidence that reproductive performance should be considered as a product of the identity and condition of both parents.


The idea that males can influence their mates' reproductive performance is now pervasive in most taxa, and has been particularly discussed and tested in the context of maternal investment. Stemming from the fundamental role of anisogamy, sexual selection theory predicts that maternal investment in reproduction depends on the mate's characteristics and especially on male attractiveness (Trivers 1972), such that females mated with more attractive males display higher investment and, in particular, larger broods or clutches (differential allocation hypothesis; Burley 1986). In support of this hypothesis, experiments in vertebrates (e.g. de Lope & Møller 1993) and invertebrates (e.g. Fox, McLennan & Mousseau 1995) show female reproductive investment increasing with male size or attractiveness. Although most of these studies cannot decipher between phenotypic and genetic male effects, there is some evidence that males can vary genetically in their indirect influence on female reproductive investment (e.g. Pischedda et al. 2011).

In the particular case of birds, reproductive parameters such as laying date and clutch size are classically considered female-specific traits as they are not expressed by males. A small number of studies have tested for among-male persistent variation in their mates' laying date or clutch size, as well as for indirect effects of male age or other life-history traits (e.g. life span) on these female reproductive parameters. Overall, the vast majority of studies conducted on short-lived passerines have concluded that there is no influence of male identity on a pair's laying date or clutch size (Sheldon, Kruuk & Merilä 2003; McCleery et al. 2004; Gienapp, Postma & Visser 2006; Browne et al. 2007; Caro et al. 2009), although these traits can be related to male life history, such as age at first reproduction (Auld & Charmantier 2011). However the picture could be very different for long-lived socially monogamous birds where reproduction is more classically correlated with the age or breeding experience of both partners as, for example, in the sparrowhawk Accipiter nisus (Newton & Marquiss 1984), the oystercatcher Haematopus ostralegus (van de Pol & Pettifor 2006; van de Pol et al. 2006) or the mute swan Cygnus olor (Charmantier et al. 2006a,b). Such differences between short- and long-lived species may have several underlying causes, but one mechanism may be the investment by males in territory establishment and guarding. In long-lived species there may be more invested in maintaining a high-quality territory across breeding seasons. If females benefit in terms of security or food resources from pairing with a certain male, especially if these interactions take place prior to nest establishment, the male may have strong effects on the female's reproductive investment and output.

Deciphering male and female phenotypic and genetic contributions to laying date and clutch size is a key element to understanding the evolution of these life-history traits, yet it requires long-term data sets that combine phenotypic observations and pedigree data to conduct quantitative-genetic analyses. To date, only two studies have demonstrated indirect genetic effects of males on avian timing of breeding. First, a long-term study of common gulls Larus canus showed that direct female and indirect male genetic effects explained 14·5% and 4·8% of the variance in laying date, respectively (Brommer & Rattiste 2008). The quantitative-genetic framework in this study also suggested a sexual conflict arising from a negative genetic correlation between direct (female) and indirect (male) genetic effects. Second, heritability of laying date was found insignificant in females, but significant in males in a long-term study of red-billed gulls Larus novaehollandiae scopulinus (Teplitsky et al. 2010). In this latter case, the authors interpret this strong indirect genetic effect as a probable result of additive-genetic variance in the feeding ability of males during courtship. Regarding breeding performance as measured by the number of offspring fledged or recruited, these are no longer considered sex-limited traits, and male influence is more pervasive in the avian literature, especially male age (e.g. Green 2001; Hatch & Westneat 2007). However, investigations on male age effects in late life, that is male senescence in reproductive performance, are still very scarce (e.g. studies by Velando, Drummond & Torres 2006; Brommer, Wilson & Gustafsson 2007; Reed et al. 2008). The fact that reproductive senescence has by and large been mostly investigated in females (e.g. see review in Nussey et al. 2008) could well be a major limit in our knowledge of the evolution of senescence in vertebrates.

Here, we build on previous investigations using long-term data in a mute swan Cygnus olor colony to investigate how male phenotype and genotype are related to female reproductive performance, from the timing of laying to the number of offspring recruited. Mute swans in Abbotsbury, Dorset, UK, have been individually monitored since the late 1970s, and the data collected in this colony has already proven highly useful in uncovering age-specific patterns of reproductive traits across individual lifetime, and especially senescence patterns in females. Female mute swans show delays in laying date and declining clutch size after 12 years of age (McCleery et al. 2008) suggesting a senescent decline in performance as selection favours early and large clutches (Charmantier et al. 2006a). As mentioned above, variation among males in their mate's laying dates was significant (Charmantier et al. 2006a), yet the origin of this variation was unexplored. Although swans often pair for several years, new pairings represent 30–55% of breeding attempts (Perrins & McCleery 1997), hence in the full data set, 29% of females (320/681) had more than one partner, and mean number of male partners was 1·4 (range: one to six). Male mute swans secure the nest territory in early spring after agonistic interactions with their competitors; they contribute to parental care until the cygnets gain their independence, which may be in the autumn of their year of birth or later during the winter (Birkhead & Perrins 1986). This involvement of males suggests that their genetic and phenotypic characteristics could play an important and hitherto ignored role in breeding performance, a question we explored by analysing our longitudinal data with linear mixed models and quantitative-genetic animal models.

Materials and methods

Study system details

We analysed a data set collected over 30 years (1979–2008) from a semi-natural population of colonially nesting mute swans (Cygnus olor) that inhabit the Fleet at Abbotsbury, Dorset, UK; earliest records of this population date to the thirteenth century (Perrins & Ogilvie 1981; Perrins, McCleery & Ogilvie 1994; McCleery et al. 2008). The majority of both males (78%) and females (91%) were born into the colony; other individuals are immigrants. All birds are individually ringed, so their identity and age are known. Age of immigrants is based on coloration of plumage and beak at first capture, which can discriminate between birds of 1, 2 and 3 years or more. From 1979 to 2008, an average of 104 pairs have bred each year and for each breeding attempt the identity of both parents, the laying date of the first egg (1 March = 1) and the clutch size have been recorded. Chicks are ringed before fledging and since 1989 they are web-tagged at hatching. Subsequently, recruitment is assessed as the number of individuals from any given clutch that attempt to breed within the population. Emigration from the population is rare and it has never been recorded after a bird has commenced breeding (McCleery et al. 2008). We restricted our analysis of the number of recruits to clutches from 1989 to 2004 because records prior to web-tagging from 1979 to 1988 were less standardized and parentage less certain, and breeding attempts commencing in 2005–2008 were likely to contain incomplete information on successful recruitment (most swans start breeding at 3 or 5 years of age, but some start later). The age at last reproduction for each individual male and female has been calculated based on annual records of breeding; if a bird is not observed to breed for two consecutive years, it is assumed to be dead (see also Charmantier et al. 2006a; McCleery et al. 2008); it is extremely rare for an individual to miss >2 consecutive years of reproduction and re-commence breeding (Table 4 in Perrins & McCleery 1997).

We analysed laying date, clutch size and the number of recruits for each brood as a function of the characteristics of both members of a breeding pair (i.e. male and female). Explanatory variables belong to three categories: an environmental variable (i.e. year), permanent characteristics of the father and mother (i.e. individual identity, age at last reproduction and resident/immigrant status) as well as the ages of the male and female when each clutch was laid. Any given male or female is associated with a single brood in a given year, yet normally each male and female (or the same pair) has produced several broods over a span of years. Previous studies have examined the influence of female age on breeding success in this mute swan population (e.g. Charmantier et al. 2006a,b,c; McCleery et al. 2008); here we were primarily interested in examining the role of paternal age and identity, and how paternal age interacts with maternal age. We included age at last reproduction to control for selective disappearance that can bias estimates of age effects (Cam et al. 2002; van de Pol & Verhulst 2006; McCleery et al. 2008); age at first reproduction was not included in our analyses because of multicollinearity among explanatory variables.

Statistical analyses

It is widely appreciated that mute swans are socially monogamous, and thus there is a potential problem of collinearity between paternal and maternal age as explanatory variables (i.e. assortative mating by age). For example, it is known that even if an old female loses her mate (e.g. if he dies), she will tend to choose an old male (i.e. a male of roughly equivalent age) to replace him (Perrins & McCleery 1997). In the present data set, a simple linear regression of male age on female age explains almost three-fourth of the variation in male age (r2 = 0·74; P < 0·001). Collinearity between explanatory variables leads to fragility of the statistical model, often observed through large variation around the estimated effects. To address this problem, we have analysed the data set in two ways. First, we analysed and reported the analyses of male and female age treating these two variables as independent. The variance inflation factor for these variables was 3·8 (i.e. 1/(1 − r2), where r2 is from a linear regression of one on the other), which is below the common cut-off value of five above which multicollinearity is considered too high (i.e. r2 = 0·8; Davis et al. 1986). Along with this, we did not observe any problems of model convergence or inconsistent results during the stepwise regression model selection, so this approach seems justifiable. That is to say that when collinearity between explanatory variables is ‘too high’ regression coefficients and their standard errors typically change erratically between similar models; we never observed this. However, to be sure that the effects of male age are not statistical artefacts, we employed a second, highly conservative means of controlling for the collinearity of male and female age. That is, we also analysed (as reported in the Supporting Information) the residuals from the regression of male age on female age. These male-age residuals were used in place of male age in an analysis that included female age and the other variables described above. Thus, this analysis reveals any effects of male age that are not explained by female age. Results of models using male-age residuals rather than male age are reported in supplementary tables, but are not thoroughly discussed in the main text as overall the main conclusions remain unchanged (i.e. the role of male and female age and identity remains qualitatively the same). Although we recognize that this sort of analysis is not a perfect way to resolve the issue of collinearity, we feel it is justifiable and akin to analyses using residuals to distinguish, e.g. body size and fitness (e.g. Krebs & Singleton 1993; Jakob, Marshall & Uetz 1996). Lastly, we were unable to simultaneously explore the role of experience as a couple (i.e. the number of subsequent years that a given male and female bred together) because it was collinear with age; breeding experience may play a role, but our focus here is on senescence (i.e. direct effects of age).

In the first step of our analyses, we constructed linear mixed models to test the linear, quadratic and interactive effects of male and female age on laying date, clutch size and the number of recruits. Age at last reproduction (ALR) for both the male and female was included along with all pairwise interactions with male age (linear and quadratic). Quadratic effects of age are used to test for curvature in the relationship of a trait and age, e.g. improvement to a certain age followed by senescent decline. Resident/immigrant status of both parents was also included as a fixed effect. Nestlings from some clutches were placed inside a pen for protection against aggressive adult males; as this may affect the number of surviving recruits, we included Pen as a factor in our analyses of the number of recruits. Random effects included male identity, female identity and year; this was done for male/female identity to account for repeated measures across individuals and for year because we do not expect a directional effect of annual variation. We used stepwise regressions to select the best models: Non-significant fixed effects were sequentially deleted starting with those of highest order.

Linear mixed models were fit in SAS (version 9·2); laying-date and clutch-size models were fit using PROC Mixed (a restricted maximum likelihood estimation procedure), whereas models for the number of recruits were fit using PROC Glimmix (a pseudo-likelihood estimation procedure) with Poisson error structure. As it is widely recognized that these traits are highly correlated, we controlled the effects of age on other (previously expressed) traits by including these traits as covariates. That is, we analysed clutch size including laying date as a covariate and we analysed the number of recruits including both laying date and clutch size as covariates (see also Bouwhuis et al. 2009; Auld & Charmantier 2011). All models were re-fit without these covariates; results are reported in the online supplement.

As quadratic age effects over whole life data are insufficient to conclude with confidence on senescence, we explored the late-life effects of male and female age by repeating all analyses excluding all records where male or female age were <11 years. In these ‘step-two’ analyses, the saturated models included all of the same terms as described above for the analyses of laying date, clutch size and the number of recruits, but we excluded all quadratic-age terms: Once birds reach 11 years of age, the initial ‘improvement’ phase of the life cycle is over and senescent dynamics are expected (McCleery et al. 2008). It is worth noting that in this ‘step-two’ data set, the collinearity between male and female age is much lower than in the entire data set (r2 = 0·32).

Significant phenotypic variation among individual males (or females) can be decomposed into genetic and environmental components using a quantitative genetic ‘animal model’ (reviewed by Kruuk 2004; Postma & Charmantier 2007). Such analyses use pedigree information to determine the amount of phenotypic variation that can be attributed to specific causes (e.g. additive-genetic variation). Whenever we found significant variation among individual males in the step-one analysis, we decomposed the male (indirect) and the female (direct) effects into additive-genetic and permanent-environmental components in a ‘step-three’ analysis, using the lifelong data, and a 1989–2008 social pedigree. The pedigree spanned over four generations with 2734 individuals, of which 1931 were of known parents, with 300 fathers and 270 mothers forming 339 distinct pairs with at least one recruited offspring. In the absence of web-tagging on the cygnets right after hatching, parentage assigned before 1989 was less reliable hence disregarded although individuals born in these years were integrated in the pedigree, with unknown parents. Decomposition of variance was done using a Restricted Maximum Likelihood (REML) animal model as described in Brommer & Rattiste (2008), with the full model including random effects for year, female and male additive-genetic effects (Vaf and Vam, respectively) and female and male permanent-environment effects (Vpef and Vpem, respectively). Fixed effects included were derived from the final step-one model. This animal model was solved using ASReml (VSN International), which estimates the variance for each random effect, with associated standard errors. Significance of random effects was assessed by comparing hierarchical mixed models of increased complexity, using likelihood-ratio tests.


Male age and female age were positively related (Fig. 1) indicating positive assortative mating by age. Thus, we analysed the effects of male and female age on laying date, clutch size and number of recruits in two separate ways to reveal patterns of improvement and senescence that can be linked to each parent independently. The results from models that include both male age and female age are reported in Tables 1 and 2; similar tables are provided in the online supplement (Tables S1 and S2, Supporting Information) to report the results of models that include the residuals from a regression of male age on female age. Despite a few disparities, most of our main results are similar under both methods of analysis, and reveal that male identity and age can play a role in the expression of these traits. The independent effects of male age (i.e. over and above the collinearity with female age) do not appear to be very important late in life.

Table 1. Final linear mixed models testing for age effects across all ages (i.e. Step-One Analysis) for laying date, clutch size and number of recruits are reported by the estimated effects of each term (Est.) with associated standard errors and P-values. Non-significant fixed effects (ns) were deleted from the final model. Linear male and female age terms were always retained; Age² represents quadratic terms. ALR is the age at last reproduction. Resident Status was coded 0 (resident) or 1 (immigrant), thus showing the relative effect of being an immigrant. Pen was included as a factor only for the number of recruits (see text). P-values for random effects are from a Wald's Z-test
 Laying dateClutch sizeNumber of recruits
Est. SE P Est. SE P Est. SE P
Fixed effectsIntercept 62·04 2·25 <0·001 6·96 0·31 <0·001 −0·650·580·263
♂ Age−0·500·730·493 0·27 0·09 0·002 0·0010·030·986
♂ Age² 0·09 0·03 0·008 −0·02 0·01 <0·001 ns
♀ Age −1·61 0·59 0·007 0·19 0·08 0·017 0·040·030·194
♀ Age² 0·12 0·03 <0·001 −0·01 0·004 0·004 ns
♂ Age * ♀ Agensnsns
♂ ALR −0·27 0·12 0·024 nsns
♀ ALR0·240·190·2010·010·010·498 0·05 0·02 0·004
♂ Age * ♂ ALRnsnsns
♂ Age² * ♂ ALRnsnsns
♂ Age * ♀ ALR −0·12 0·03 <0·001 nsns
♂ Age² * ♀ ALRns 0·001 0·0003 0·011 ns
♂ Resident statusnsns −0·57 0·24 0·020
♀ Resident statusnsnsns
Pen 0·77 0·11 <0·001
Laying date −0·08 0·003 <0·001 −0·04 0·01 <0·001
Clutch size 0·11 0·04 0·015
Random effects♂ ID 22·24 4·05 <0·001 0·050·060·2150·100·080·12
♀ ID 22·50 4·28 <0·001 0·33 0·07 <0·001 0·120·090·08
Year 28·42 8·18 <0·001 0·07 0·03 0·007 0·33 0·15 0·015
Residual 54·12 2·39 <0·001 1·53 0·06 <0·001 0·89 0·04 <0·001
Sample size410450347
Table 2. Final linear mixed models testing for age effects for breeding attempts where both male and female ages are ≥11 years (i.e. Step-Two Analysis) for laying date, clutch size and number of recruits. All abbreviations and procedures are otherwise identical to Table 1 (i.e. Step-One Analysis)
 Laying dateClutch sizeNumber of recruits
Est. SE P Est. SE P Est. SE P
Fixed effectsIntercept 38·06 7·09 0·001 10·58 0·58 <0·001 −4·131·320·20
♂ Age−0·270·640·668−0·010·050·8460·090·080·313
♀ Age 2·19 0·62 <0·001 −0·020·040·554−0·080·080·345
♂ Age * ♀ Agensnsns
♂ ALRnsnsns
♀ ALR −1·11 0·47 0·019 ns 0·17 0·06 0·005
♂ Age * ♂ ALRnsnsns
♂ Age * ♀ ALRnsnsns
♂ Resident statusnsnsns
♀ Resident statusnsnsns
Pen 1·08 0·26 <0·001
Laying date −0·09 0·01 <0·001 ns
Clutch size 0·27 0·08 <0·001
Random effects♂ ID 53·86 18·70 0·002 0·120·300·3500·010·270·489
♀ ID13·1517·010·2200·410·340·1130·120·270·333
Year9·955·800·043 0·10 0·06 0·05 0·32 0·23 0·083
Residual 56·29 7·65 <0·001 1·28 0·12 <0·001 0·84 0·12 <0·001
Sample size10115287
Figure 1.

Male age plotted as a function of female age within breeding pairs. The diameter of the symbol is proportional to sample size (minimum = 1; maximum = 299).

Laying date

Laying date decreased (i.e. improved) and subsequently increased (i.e. senesced) as a function of both male and female age (Fig. 2a). We detected a quadratic effect of male age independent of the linear and quadratic effects of female age (Table 1), but no interaction between male age and female age (N.B. this interaction is significant in the analysis of male-age residuals; Table S1, Supporting Information). However, we did observe a negative interaction between female age at last reproduction and male age indicating that the effects of male age were stronger when females had long reproductive lives. We verified that this male age by female ALR interaction was not driven by a few old birds by repeating the analysis after excluding all birds whose age was ≥17 years; this interaction was still significant (P = 0·016). Male age at last reproduction also had an effect on laying date, with long-lived males having earlier laying dates. We observed significant effects of male identity and female identity, indicating that the identity of both parents explains some variation in the timing of egg laying. All these effects on laying date were consistent regardless of whether we analysed male age or male-age residuals (Table 1 and Table S1, Supporting Information).

Figure 2.

Linear mixed model predictions for laying date (a) and clutch size (b) plotted as a function of male age nested within female age (ages in years). Therefore, male age effects (for a given female age) are viewed by the small curves whereas female age effects (for a given male age) are viewed across the entire figure. Predictions are based on average values for the age at last reproduction.

The step-two analysis also showed significant differences among individual males for their effects on laying date late in life (i.e. > 11 years old; Table 2), but no differences among individual females. This is identical to the analyses presented in Table S2, Supporting Information. Despite the quadratic effect of male age on laying date (Table 1), there was not a significant linear effect of male age late in life.

In the step-three analysis using a REML animal model on the 1573 observations of laying dates (corresponding to 480 pairs), there were two best models based on likelihood-ratio tests (models 5 and 7 in Table 3). Both of these models included significant annual variation as well as variation for both males and females that is attributed to permanent-environmental effects. The difference between these two indistinguishable models lies in whether the among-male variance can be attributed to additive-genetic variance or whether it is completely attributable to permanent-environmental effects. Neither of these models included an additive-genetic effect for females. In both models, the permanent-environment female (direct) effect explained 17·7% of the total phenotypic variance estimated by the model; the among-male variance explained 17·5% of the total phenotypic variance. The model attributing some of this among-male variance to additive-genetic effects ascribed 11·6% of the phenotypic variance to additive-genetic variance. Hence, although we can conclude on an absence of evidence for additive-genetic effects in females, the evidence for such indirect effects from males remains inconclusive.

Table 3. Selection of mixed models decomposing the variance displayed in 1573 mute swan laying dates, by 401 males and 352 females over 30 years. The estimated variance components (and associated standard errors) were provided by ASReml. Fixed effects are identical to the final laying date model in Table 1 and are not detailed here. Random effects included residual variance (Vr), variance attributed to annual fluctuations (Vyear), to permanent-environmental differences between females and males (Vpef and Vpem) and to additive-genetic effects in females and males (Vaf and Vam). Significance of χ² distributed likelihood-ratio tests to compare models of increasing complexity is indicated as follows: ***P < 0·001, **P < 0·01, *P < 0·05. In all tests, the difference in degree of freedom between two models was 1. In bold are the two best models
Model V r V year Vpe f Va f Vpe m Vam Deviance of model Test Log-likelihood test statistic
  1. a

    The female additive-genetic component was negative if unconstrained and the standard error not estimated.

1113·9 (4·1)9056·48  
293·5 (3·4)22·8 (6·8)8815·861 vs. 2 240·62***
362·9 (2·6)28·9 (8·4)32·0 (4·0)8571·582 vs. 3 244·28***
461·0 (2·5)26·8 (7·8)33·1 (3·9)8561·522 vs. 4 254·34***
5 54·1 (2·3) 28·4 (8·2) 22·5 (4·5) 22·2 (4·3) 8502·02 4 vs. 5 59·5***
654·1 (2·3)28·4 (8·2)22·5 (4·5)0a22·2 (4·3)8502·025 vs. 6 1
7 54·1 (2·3) 28·5 (8·2) 22·7 (4·5) 7·5 (10·5) 14·8 (10·6) 8499·9 5 vs. 7 2·12

Clutch size

Clutch size increased (i.e. improved) and subsequently decreased (i.e. senesced) as a function of both male and female age over lifetime (Table 1 and Fig. 2b). The linear effect of male age on clutch size remained significant in the analysis of male-age residuals, but the quadratic effect did not (Table S1, Supporting Information). However, no age effect remained significant in the step-two late-life analysis (Table 2). Clutch size was negatively correlated with laying date, and thus we report the results for clutch size that control for this correlation with laying date (Table 1 and Table S1, Supporting Information). We repeated the analyses excluding laying date as a covariate and obtained the same results (cf. Tables S3–S6, Supporting Information). There was no significant variation among individual males for their effect on clutch size; significant variation among individual females and annual fluctuations were detected in the lifelong analysis only (Table 1).

Number of recruits

Variation in the number of recruits was largely explained through correlations with laying date and clutch size as well as female reproductive longevity (Tables 12). However, this fitness estimate did not display persistent differences between males, with (Tables 1 and 2) or without (Tables S3 and S4, Supporting Information; see also Tables S5 and S6, Supporting Information) including laying date and clutch size as covariates. Long-lived females had increased numbers of recruits, which suggests an effect of selective disappearance with lower performing breeders dying earlier. Male immigrants also suffered reduced numbers of recruits compared with male residents.


Our study provides evidence that in a relatively long-lived bird species, the Mute swan, males can influence their partners' reproductive traits in various substantial ways. First, we found that despite the high correlation between breeding partners' ages, a female's laying date and clutch size varied across male age with an improvement in male early life (earlier laying, larger clutches) followed by a decline in male late life. Second, we found that male identity had an important indirect effect on a female's laying date, and that this indirect effect was not only potentially underlined by additive-genetic variance among males but also by persistent permanent-environmental effects associated with each male. Finally, we show that these strong male effects on laying date do not translate into male phenotypic or genetic influences on our best fitness estimate, the number of offspring recruited, suggesting either a possible compensatory mechanism along the cycle by the breeding pair or that other stochastic factors may affect eventual recruitment into the population. It is important to bear in mind that extra-pair paternity is expected to be low to moderate in this species (0–17% of extra-pair cygnets were found in closely related species; see Birkhead & Møller 1996; Kraaijeveld et al. 2004). The existence of extra-pair paternity could result in slight underestimation of the additive-genetic variance in the quantitative genetic analysis. If the rate of extra-pair paternity is related to parental age or length of the partnership (e.g. Kempenaers, Verheyen & Dhondt 1997; Richardson & Burke 1999), this could also alter the age-specific pattern of genetic reproductive success. Hence, our analyses are focused on the effects of the ‘social father’ rather than on age-specific patterns of genetic reproductive success because of the lack of data on genetic paternity.

Several constraints due to the nature and size of our data set limited our power to decipher male from female effects on reproductive performance, and especially age effects. Mute swans in the Abbotsbury colony display strong assortative mating by age (Fig. 1), and this correlation is not only due to young birds pairing up together when they reach breeding age but also because older birds seeking a replacement for a lost mate also pair with a bird of similar age to themselves (Perrins & McCleery 1997). Hence, although 29% of females were paired with more than one male, fewer were paired with males with dissimilar ages from theirs. To circumvent this issue, we analysed the effects of male and female age on laying date, clutch size and number of recruits in two separate ways to reveal patterns of improvement and senescence that can be linked to each parent independently. Most of our main results are similar under both manners of analysis, which reveals that male identity and age can play a role in the expression of these traits, especially timing of breeding. We also acknowledge that caution should be used when interpreting the results from the late-life (step-two) analysis; the reduction in sample size along with generally small effects of age potentially explain why quadratic-age effects in the initial analysis are not retained as significant linear effects in the step-two analysis (e.g. age effects on clutch size). However, the lack of significant effects of male age late in life may indicate that the primary effects of males on laying date and clutch size are restricted to early in ontogeny. Keeping the limitations of reduced sample size in mind, we will discuss below the main robust results from our study.

Variation among males or across male life history was strongest for the timing of egg laying. Our study shows that both female and male ages matter to determine the timing and size of a clutch: similar to what was previously described in female mute swans (McCleery et al. 2008), middle-aged males have partners with earlier and larger broods than average whereas females breeding with old males have later and smaller clutches (Fig. 2). These results echo previously demonstrated effects of selective disappearance whereby long-lived females had earlier laying dates compared with relatively short-lived females (McCleery et al. 2008). In addition to these indirect effects of male age and life history on their female's timing of reproduction, we also found substantial variation among males in the timing of breeding, with 17·5% of the variance in laying date explained by a male (indirect) influence vs. 17·7% by a female (direct) influence. These variances may have some additive-genetic basis in males, yet there is no such basis in females, which is in concordance with previous findings restricted to females (Charmantier et al. 2006b). This latter study showed that laying date displayed significant heritability for females in late life only (12 years and older), a stage at which many individuals will have died and selection is much weaker. The indirect additive-genetic variance in males expressed in their partners' timing of reproduction implies that this trait has some evolutionary potential in this species (Roff 1997), yet it will act mainly through indirect selection on males rather than females. Nevertheless, this suggests that permanent differences among males of environmental or genetic origin influence a female's timing of reproduction. Such indirect effects of males can be mediated by different processes. Variation among males suggests that a permanent male characteristic, either heritable genetic effects or otherwise (e.g. his condition, his ornamentation or the quality of his territory if these are repeatable), influence the female in her timing of breeding. Hence, differential timing between females could be a consequence of either the environmental phenology in the territory secured by the male, differential allocation to reproduction (Sheldon 2000) where a female changes her timing of reproduction according to the male's age or attractiveness (de Lope & Møller 1993) or song quality (Gil et al. 2004), or some combination of these factors. Most of the above-cited male characteristics (male territory quality, attractiveness, song quality) can be influenced by differences in the genetic or environmental backgrounds.

Although incubation in mute swans is undertaken by the female only, males have an active role early in the breeding season as they take part in sometimes violent agonistic interactions to secure and defend a territory (Birkhead & Perrins 1986). Long-term selection analyses in this colony show strong directional selection for earlier laying (and larger clutches; Charmantier et al. 2006a), even though mute swans in Abbotsbury are fed over the breeding season. This suggests that the 50–150 nests established in close proximity over the one hectare or so colony, probably vary in territory quality, and that earlier established pairs will benefit from the highest quality territories. Our analyses would thus imply that males with highest quality territories are middle aged and long lived but also that some males can repeatedly secure the best territories.

Similar effects of male age, life history or identity could have been expected later in the breeding cycle as mute swans display prolonged parental care by protecting the cygnets from conspecifics until fledgling in autumn (Rees et al. 1996). However, the number of cygnets recruited in the population from a given breeding attempt was not influenced by male or female age or identity. This is in contrast to previous results hinting at declining breeding success in very old birds in the same colony (Perrins, McCleery & Ogilvie 1994). This discrepancy could be explained by the higher complexity of our mixed models accounting for several fixed and random effects. In particular, males of different origin did have differential reproductive success as birds born in Abbotsbury had higher recruitment, and long-lived females also had higher annual recruitment (Table 1). Another potential explanation of the lack of male influence on recruitment is that perhaps these effects may be masked or compensated for by the combination of penning and feeding the cygnets. Future studies exploring the potential for male effects on fitness traits including recruitment in natural systems will be useful in determining whether these somewhat artificial conditions may obfuscate male effects. However, the influence of male age and life history was restricted to traits expressed early in the breeding cycle in a similar analysis of a short-lived, wild passerine (Cyanistes caeruleus ogliastrae; Auld & Charmantier 2011).

Males' roles in reproduction are increasingly recognized, even in species where it was originally thought that the male contribution was restricted to providing genes. A recent comparative meta-analysis shows that in species with bi-parental care, female birds classically increase their clutch size when mated to preferred males (Horvathova, Nakagawa & Uller 2012) suggesting that indirect male effects are in fact mediated via direct differential investment by females. However, experimental approaches are only starting to unveil the mechanisms that underlie differential allocation in females. For example in the Houbara bustard Chlamydotis undulate undulate male courtship behaviour modulates maternal allocation (Loyau & Lacroix 2010). Future work is needed to unveil the mechanisms underlying these indirect phenotypic and genetic effects. Another potential factor that may underlie some of the age effects observed in this study is the role that time together/breeding experience of the couple plays in, e.g. the improvement phase. This is a factor that will be ideally addressed in a species that is not as socially monogamous as the Mute swan, thereby allowing a discrimination of male age, female age and their joint experience as a couple.


We thank Mrs. Charlotte Townshend for allowing this study on swans, as well as the staff and volunteers who have collected data over many years. We thank O. Gimenez for statistical advice as well as D. Nussey, C. Teplitsky and two anonymous reviewers for comments on the manuscript. J.R.A. received Post-Doc funding from the French C.N.R.S. (award to A.C. and P. David) as well as the U.S. National Evolutionary Synthesis Center (NESCent), NSF #EF-0905606. A.C. was funded by the Agence Nationale de la Recherche (grant ANR-08-JCJC-0041-01).