Age-related change in breeding performance in early life is associated with an increase in competence in the migratory barn swallow Hirundo rustica

Authors


Javier Balbontín, Departamento de Biología Animal, Universidad de Extremadura, E-06071 Badajoz, Spain. E-mail: jbalare@unex.es

Summary

  • 1We investigated age-related changes in two reproductive traits (laying date and annual fecundity) in barn swallows Hirundo rustica L. using a mixed model approach to di-stinguish among between- and within-individual changes in breeding performance with age.
  • 2We tested predictions of age-related improvements of competence (i.e. constraint hypothesis) and age-related progressive disappearance of poor-quality breeders (i.e. selection hypothesis) to explain age-related increase in breeding performance in early life.
  • 3Reproductive success increased in early life, reaching a plateau at middle age (e.g. at 3 years of age) and decreasing at older age (> 4 years). Age-related changes in breeding success were due mainly to an effect of female age.
  • 4Age of both female and male affected timing of reproduction. Final linear mixed effect models (LME) for laying date included main and quadratic terms for female and male age, suggesting a deterioration in reproductive performance at older age for both males and females.
  • 5We found evidence supporting the constraints hypothesis that increases in competence within individuals, with ageing being the most probable cause of the observed increase in breeding performance with age in early life. Two mechanisms were implicated: (1) advance in male arrival date with age provided middle-aged males with better access to mates. Yearling males arrived later to the breeding grounds and therefore had limited access to high-quality mates. (2) Breeding pairs maintaining bonds for 2 consecutive years (experienced pairs) had higher fecundity than newly formed inexperienced breeding pairs.
  • 6There was no support for the selection hypothesis because breeding performance was not correlated with life span.
  • 7We found a within-individual deterioration in breeding and migratory performance (arrival date) in the oldest age-classes consistent with senescence in these reproductive and migratory traits.

Introduction

Empirical studies of free-living vertebrates suggest strongly that reproductive performance increases with age, reaching a maximum output at middle age, followed by a clear deterioration at advanced age due to senescence (Clutton-Brock 1988; Forslund & Pärt 1995). This association between age and fecundity has been observed in fish (Hodder 1963), amphibians (Salthe 1969), reptiles (Tinkle & Ballinger 1972), mammals (Clutton-Brock 1988) and birds (Newton 1989; Sæther 1990; Forslund & Pärt 1995). Three major groups of non-exclusive hypotheses based on progressive appearance or disappearance of individual phenotypes among age-classes (delayed breeding and selection hypotheses; Perrins & Moss 1974; Curio 1983; Newton 1989), age-related improvements of competence (constraint hypotheses; Curio 1983; Nol & Smith 1987; Pärt 2001) and optimization of reproductive effort (restraint hypotheses; Williams 1966; Pianka & Parker 1975; Clutton-Brock 1988) have been proposed to explain the increase in reproductive output with age in early life. It is important to realize that changes in phenotypic traits could result from within-individual patterns (e.g. ageing) and from between-individual changes in the quality of individuals entering or leaving the population (e.g. selection). The constraint and restraint hypotheses focus on the phenotypic changes that take place within individuals, and the selection or delayed-breeding hypotheses focus on phenotypic changes among individuals.

The selection hypotheses assume that there is always a difference in quality among individuals in a population. The progressive disappearance of individuals with poor reproductive performance occurs if such individuals have lower survival probabilities than individuals with superior reproductive success, and therefore a positive correlation between reproductive performance and survival or longevity is expected at the population level. Although natural selection is involved in this process, humans can cause mortality on cohorts that reverse the pattern due to natural selection (Balbontín, Penteriani & Ferrer 2005). Most studies have reported associations between age and reproductive traits based on cross-sectional comparisons. Because individuals within a population differ in their ability to acquire and utilize resources and in their optimal allocation of such resources, between-individual variation in quality can mask within-individual ageing patterns in cross-sectional studies (Nol & Smith 1987; Forslund & Pärt 1995; Nussey et al. 2006; van der Pol & Verhulst 2006). In particular, problems of estimating age-effects for reproductive traits within individuals arise when there is a difference in quality among individuals causing a difference in longevity. Using longitudinal data on reproductive traits of individuals over their lifetime, the selection hypothesis could be tested readily. Recently, van der Pol & Verhulst (2006) suggested that by using linear mixed models appearance or disappearance effects could be tested easily by adding age at first reproduction (appearance effects) or longevity (i.e. age at last reproduction, disappearance effects) as a covariate in a model in which age is kept as a fixed effect for investigating their effects on reproductive traits. Correlations between reproductive traits within individuals are controlled by adding individual identity as a random effect in these models.

The constraints hypotheses suggest that with increasing age, individuals improve in skills that improve reproductive performance. Many different ways of increasing competence have been proposed. For instance, a prolonged period of learning is needed to perform all reproduction tasks correctly to raise healthy young successfully. Therefore, previous breeding experience has been suggested to influence breeding performance positively. Increased experience while ageing could also be due to learning to cope with predators or competitors, familiarity with the local environment or within-individual improvement in feeding behaviour, all of which could cause an increase in reproductive output with age (Burger 1988; Marchetti & Price 1989; Desrochers 1992). Other extrinsic mechanisms may prevent young individuals from reproducing as successfully as older individuals. For instance, improvement in breeding success with age could be due to limited access to high-quality mates or territories for young individuals. Therefore, limited access to resources by young individuals due to constraints on migratory performance or dominance status might cause poor reproductive performance when young.

The restraint hypotheses suggest that individuals enhance reproductive performance as they grow old because of a decrease in residual reproductive value (Williams 1966; Gadgil & Bossert 1970; Curio 1983; Charlesworth 1994). Life history theory predicts a trade-off between reproduction and survival or between current and future reproduction (Stearns 1992). Assuming such trade-offs, individuals may invest less in reproduction when young because they trade investment of resources between current reproduction and survival to the next breeding season.

Most studies of wild vertebrates failed to describe the effect of age on fecundity in the last part of the life of an individual. Studies rarely tested whether declines in reproductive performance in later life are due to changes in within- or between-individual performance. Theories of senescence deal with within-individual changes in reproductive performance with age. However, when using cross-sectional data, changes in performance between individuals might mask patterns of senescence in reproductive output. For example, if individuals that increase reproductive effort in early life pay a cost in terms of reduced longevity or accelerated reproductive deterioration in late life, then the observed pattern of reproductive senescence would be due to the disappearance of high-quality breeders rather than a phenotypic change within a given individual. Therefore, separation of within- and between-individual variation in age-related breeding performance constitutes a major challenge for the study of senescence at the population level in the wild (Forslund & Pärt 1995; Nussey et al. 2006; van der Pol & Verhulst 2006).

Our main aim was to test hypotheses explaining age-related increase in two reproductive traits (e.g. lay date and annual fecundity) early in life, using the barn swallow (Hirundo rustica L.) as a model species. The barn swallow is a small (c. 20 g), socially monogamous, semicolonial passerine feeding on insect prey. It is a long-distance migratory bird covering up to > 10 000 km between breeding and wintering sites. Barn swallows breed on farms, gaining permanent access through open doors and windows. Females lay between one and three clutches containing one to six eggs per breeding season. Breeding dispersal is very low, because less than 0·1% of > 10 000 banded adults (studied during 30 years in populations in Spain, Italy and Denmark, including the population studied here) moved between breeding colonies from one year to the next, and all but one among 450 local recruits was captured in the first year of life (Møller, de Lope & Saino 2005). Therefore, age at first reproduction shows little or no variation, with most individuals starting reproduction at the age of 1 year. On average, only 35% of breeding individuals survive to the next breeding season (Møller 1994a; Saino et al. 1999), with very few reaching very old age (i.e. 1·5% of first-time breeding birds reached the age of 5 years). None the less, longevity or age at last reproduction shows considerable variation (range: 1–8 years) compared with age at first reproduction, and therefore only disappearance effects are likely to mask within-individual changes in age-related fecundity in this short-lived species.

hypothesis testing (age-related increase in breeding performance before senescence)

Selection: progressive disappearance hypothesis

Breeding performance increases with age because poor-quality breeders die young. We predicted annual breeding success to be correlated positively with longevity (i.e. age at last reproduction) and negatively with lay date. This association between age at last reproduction and a reproductive trait (i.e. annual fecundity or lay date) must hold when measuring breeding performance controlling for age. This hypothesis assumes a difference in quality between individuals with different life spans. Long-lived individuals should be better breeders than individuals with short life spans.

Constraint: breeding experience hypothesis

More experienced breeders or breeding pairs reproduce better than less-experienced breeders or breeding pairs. Breeding experience should increase final reproductive output. Because there is no variation in age at first breeding, all similarly aged individuals should have similar breeding experience, and therefore we used pair-bond duration as an index of breeding experience. We predicted breeding pairs maintaining bonds to reproduce better than newly formed breeding pairs. Because improvement in competence occurs within individuals, we controlled for between-individual changes by incorporating age at last breeding into the analyses when testing this hypothesis.

Constraint: limited access to high-quality mates or territories hypothesis

Improvement in breeding success with age is due to limited access to high-quality mates or preferred nest sites at younger ages. We predicted that migratory performance (measured as the date that male and female arrived at the breeding grounds) should decrease with age, at least in early life. Early-arriving individuals should have better access to mates and preferred nest site than late-arriving ones. We assumed that early arrival provides males and females with a greater advantage in terms of mating success, laying date and annual reproductive output than late-arriving individuals, as demonstrated previously in several studies on this species (Møller 1994a,b; Saino et al. 1997; Kose & Møller 1999; Kose et al. 1999; Møller, de Lope & Saino 2004). Because change in improvement in competence occurs within individuals, we controlled for between-individual changes by incorporating age at last breeding into analyses when testing this hypothesis.

Finally, we tested whether deterioration in reproductive performance (laying date and annual fecundity) and migratory performance (arrival date) in late life was due to trade-offs between reproduction and survival by introducing curvilinear relationships between age at last reproduction and breeding performance, as documented previously in wild vertebrates (Reid et al. 2003).

Methods

field procedures

We studied barn swallows at Badajoz, south-west Spain (38°50′ N, 6°59′ W). The study site was mainly open farmland with pasture, cereals and fruit plantations (de Lope 1983). Early during the breeding season we captured adults at dawn by using mist nets across windows and doors of the breeding sites. We studied barn swallows in four different colonies: Potosi (POT), Almendral (ALM), Tres Arroyos (TA), and Virgen de Guadalupe (VG) within our study area (mean ± SD distance between colonies: 10781 m ± 8505, range: 1706–20 338 m). We made weekly captures of birds during every breeding season until 98–100% of breeding individuals were captured. Each bird was identified with a numbered metal ring and a combination of coloured PVC rings, so we could recognize each adult visually. We measured right and left outermost tail feathers with a ruler to the nearest 0·5 mm. Tail length was determined as the mean value of left and right characters. Individuals with broken tails were excluded from the analyses, because the rounded tip of the outermost tail feathers reveals clearly whether it is broken. Body mass was recorded with a Pesola spring balance to the nearest 0·5 g. All measures were taken by the same observer (F. de L.), which eliminates any noise in the data due to interobserver variability. All birds were provided with an individual combination of colour markings on their belly feathers using stamp ink. Individuals were sexed from the presence (females) or absence (males) of a brood patch and from observation of breeding behaviour during the courtship and incubation period.

We tracked all reproductive events and identified visually each individual of pairs using binoculars from inside a hide, to avoid disturbance during reproduction. We checked all nests in our colonies every 2 days during the breeding season with the aim of recorded laying date, clutch size and brood size up to the third brood. Once females started laying we conducted daily observation sessions inside a hide, observing each occupied nest, with the aim of identifying individuals based on colour rings and ink marks. We were able to identify each individual in the pair and their nest for the majority of the occupied nest sites. Because these populations have been subject to several experimental studies (tail-length manipulation), we eliminated from our data set any individuals involved in these experiments (89 manipulated males and 32 manipulated females). However, manipulated birds differed from non-manipulated birds for several traits, resulting in a non-random subset of individuals being involved in experiments. Females involved in experiments arrived later [mean (SE): –0·02 (0·05)], laid later [standardized lay date: –0·05 (0·05)] and had shorter tails [84·6 mm (0·30)] than non-manipulated females [mean (SE), standardized arrival date: –0·48 (0·08), standardized laying date: –0·48 (0·06), tail length: 86·9 mm (1·10 mm)], and manipulated males arrived later [mean (SE): –0·02 (0·05)] and were older [mean (SE): 2·00 (0·06) years] than non-manipulated males [standardized arrival date, mean (SE): –0·37 (0·06), age: 1·67 (0·09) years] (t-tests employed to check for differences on tail length, standardized arrival date, standardized lay date, standardized annual breeding success and age). Because tail manipulation was performed as birds arrived at our study areas, starting a little later during the breeding season, it could result in the sample of manipulated birds being skewed towards later arrival date, which could explain the observed differences in arrival date, lay date, male age and tail length between manipulated and non-manipulated individuals. However, the sample of non-manipulated birds (i.e. those used in the present study) encompassed the entire range of the population for arrival date, tail length, lay date and age. Therefore, excluding manipulated birds would have reduced the number of later-arriving and breeding individuals and the number of females with shorter tails from our final samples. In total, we identified individual males and females for pairs for 322 breeding events during 1994–2006. Two different measures of breeding success were employed. Annual fecundity was measured as the total number of chicks raised to fledging age every year (in up to three different annual broods), and laying date was estimated as the date of the first egg of the first brood, relative to day 1 (1 March). Migratory performance (arrival date) was measured as the first day an individual was captured in our consecutive capture–recapture sessions, relative to day 1 (1 February). The precision of this estimate has been checked previously by observations from the beginning of the breeding season and calculating a second measure of arrival date as the first day an adult bird was identified from its colour rings and belly ink marks. The correlation between these two estimates was very high (Pearson's correlation coefficient: range: 0·98–0·99, calculated in 7 different years for both sexes separately; Møller et al. 2004).

Several studies have demonstrated that barn swallows show high site fidelity and rarely leave the colony. Therefore, disappearance of colour-ringed breeders from the colony indicated mortality rather than dispersal. Once a 1-year-old individual returned from its first migration, it selected a breeding area and remained loyal to it for the rest of its life. Therefore, we could assign the age of individuals with accuracy in our study colonies, assuming unringed birds to be yearlings at first capture (Møller 1992; Møller et al. 2005).

statistical analyses

We used linear models to investigate the effect of age on breeding success and migratory performance. The key feature of longitudinal data is that individuals are measured repeatedly through time. Linear mixed effect models (LME) are particularly useful when there is temporal pseudo-replication (repeated measurement). We used the LME procedure of S-Plus 2000 (Mathsoft 1999) to investigate the effect of age of the female and its mate on breeding performance, timing of breeding (lay date) and migratory performance (arrival date). We controlled for interannual variation on breeding and migratory performance by standardizing the first egg-lay date, annual fecundity and arrival date by subtracting the annual population mean from each observed value and dividing by the annual population standard deviation (Zar 1999). We used a normal error distribution with an identity link function to model standardized arrival, first egg-lay date or annual fecundity as response variables. Full and reduced models were fitted by using a maximum likelihood (ML) method and final parameters for final minimal adequate model were estimated using the restricted maximum likelihood (REML) method (McCullough & Nelder 1989; Crawley 2002). There was missing information for some individuals, which resulted in slightly varying sample sizes in different analyses.

Because of high adult mortality (> 65%, Møller 1994a; Saino et al. 1999) the duration of pair-bonds was always short, and therefore we could not specify pair identity as a random factor. Accordingly, when modelling reproductive traits, we introduced bird identity (female or female's mate identity) as a random effect, incorporating identity of the female in a first attempt and thereafter repeating the analysis using identity of the female's mate, with the aim of checking if conclusions for fixed effects were similar using these two different approaches. Explanatory variables were female's and mate's age, female's and mate's age at last reproduction (ALR) (defined as the age a breeding individual was last recorded in the breeding colony), colony (factor with four levels, POT, ALM, TA and VG) and pair-bond duration (i.e. the number of breeding seasons a pair had bred together). We also fitted full models with female's and mate's body mass (g) and female's and mate's tail length (mm) when modelling standardized laying date. Also, female's and mate's arrival and laying date was included to model standardized annual fecundity, because they have been shown previously to affect these reproductive traits (review in Møller 1994a). When the response was standardized arrival date (for either males or females), explanatory variables introduced into the models were age and ALR of either sex. All these variables were treated as fixed-effects in all models.

The selective disappearance hypothesis was investigated following van der Pol & Verhulst (2006). ALR was left in all models independent of its effect, in such a way that we could test specifically for within-individual age effects in the presence of a selective disappearance effect (the estimated slope of ALR). Investigation of the dispersion plots suggested incorporation of female's and mate's age as second-order polynomials in a maximal model. Curvilinear relationships between ALR and breeding performance were also tested, because a significant effect could suggest trade-offs between reproduction and survivals (Reid et al. 2003; Nussey et al. 2006). Therefore, we first included second-order main fixed effects and possible two-way interactions in a maximal model, reducing it by eliminating non-significant terms from the fixed structures (Crawley 2002). The statistical significance of each covariate and all possible two-way interactions among fixed effects were tested in turn, using a backward stepwise procedure to select the most parsimonious model. Models with different fixed structures were compared using F-tests, Akaike's information criterion (AIC) and L-ratio tests (Akaike 1973; Pinheiro & Bates 2000; Crawley 2002). The final model was considered to have been reached when all variables (except ALR) had a significant effect at P < 0·05.

Results

female and male age effects on timing of breeding

The age of the two members of a breeding pair was related to the start of breeding. LME showed that the most parsimonious model included both main and quadratic terms for the female and its mate's age, while accounting for the known source of variability due to among-individual variation (Table 1). This model was significant (L-ratio = 95·91, P < 0·0001). Middle-aged females (3 years) started incubation earlier than young (1–2 years) and older females (4 years or older). The age of the female mate's was related to the start of reproduction in a similar way. Females that mated with middle-aged (3 years) males laid eggs earlier than those mated with either young (1–2 years) or older mates (4 years or older), independently of their own age (Fig. 1). The interaction between female and male age was not statistically significant. There was no effect of disappearance of low-quality individuals (estimated βs for main and quadratic terms on female's ALR: P > 0·2, Table 2). Female tail length was related to start of reproduction, with long-tailed females starting to reproduce earlier than the average females. The duration of the pair-bond was not related significantly to the onset of reproduction (P > 0·2). Neither male tail length nor male or female body mass were related significantly to the onset of reproduction. There was considerable between-individual variation in first-egg lay date, with female identity accounting for 43% of the total variance (calculated in a model without fixed effects).

Table 1.  Linear mixed models of standardized first egg-lay date as the response variable. Full models built with age and age at last reproduction (ALR) of female, pair-bond duration, colony, age and ALR of female's mate, female and male body mass and tail length as explanatory variables. Only significant terms are shown, except for female ALR, which was retained in the model. Sample size is 322 breeding events for known-age breeding pairs
 Random effects
SD95% CI
Female identity0·5660·451–0·710
Residual0·5810·493–0·685
 Fixed-effects
EstimateSEd.f.FP
Intercept (β0)3·4610·8141, 234 0·000·96
Female age (β1w)–0·6630·1581, 6743·04<0·0001***
Female age22w)0·0870·0261, 6729·34<0·0001***
Female ALR (β3s)–0·0680·1611, 67 0·000·95
Female ALR24s)0·0110·0251, 67 0·070·77
Mate's age (β5)–0·7710·1501, 6717·99<0·0001***
Mate's age26)0·1130·0271, 6716·270·0001***
Female tail length (β7)–0·0190·0091, 67 4·340·04
Figure 1.

(a) Effect of female age on laying date; (b) effects of mate's age on laying date having accounted for female age effect (i.e. mate's age effect on residuals from the regression of female age on laying date).

Table 2.  Final linear mixed model obtained for standardized annual fecundity. Full model fitted with female age, female age at last reproduction (ALR), pair-bond duration, colony, age and ALR of female's mate, male and female body mass and tail length as explanatory variables. Only significant terms are shown, except for female ALR, which was retained in the model. Sample size is 312 breeding events for known-age breeding pairs
 Random effects
SD95% CI
Female identity0·2370·036–1·536
Residual0·9310·805–1·078
 Fixed-effects
EstimateSEd.f.FP
Intercept (β0)–0·2800·8471, 2030·200·65
Female age (β1w) 0·4430·2211, 663·580·06
Female age22w)–0·0760·0381, 667·350·008**
Female ALR (β3s)–0·0450·1831, 660·400·52
Female ALR24s) 0·0120·0271, 660·470·49
Pair-bond (β5) 0·5320·2391, 664·880·03*
Male body mass (β6) 0·0970·0431, 665·140·02*

female and male age effects on annual fecundity

Only age of the female in the breeding pair was related significantly to annual fecundity. The most parsimonious model retained main and quadratic terms of female age, while accounting for the known source of variability due to among-individual variation (Table 2). The final model was significant (L-ratio = 21·57, P = 0·0014). Average annual fecundity for middle-aged females (3 years) was 6·74 ± 2·95 fledglings/year, which was larger than average annual fecundity of young females (1–2 years, mean ± SD = 5·44 ± 2·66 fledglings/year) or older females (4 years or more, mean ± SD = 5·88 ± 2·99 fledglings/year) (Fig. 2). Neither the effect of male age nor the interaction between female and male age was retained in the model. There was no significant effect of disappearance of low-quality individuals (estimated βs for main and quadratic terms on the female's ALR: P > 0·4, Table 2). The duration of the pair-bond was related significantly to final reproductive output. Thus, the average annual fecundity of pairs that bred together twice was significantly larger (mean ± SD = 7·32 ± 3·03 fledglings/year, n = 19) than that of pairs that remained together for just one breeding season (mean ± SD = 5·41 ± 2·74 fledglings/year, n = 219, Fig. 3). Male body mass was also retained in the final LME. Male body mass was related positively to the number of chicks fledged. Neither female (L-ratio = 0·81, P = 0·36) nor male arrival date (L-ratio = 2·86, P = 0·09) or lay date (L-ratio = 2·15, P = 0·14) were retained in the minimal adequate model. However, if we excluded from the full model the main and quadratic effects of female age, the effect of laying date and arrival date become highly significant (estimate ± SD: laying date = –0·184 ± 0·06, L-ratio = 8·29, P = 0·004; female arrival date = –0·167 ± 0·059, L-ratio = 7·77, P = 0·005; male arrival date = –0·182 ± 0·05, L-ratio = 9·83, P = 0·001). Between-individual variation in annual fecundity (calculated for female identity) accounted for 16% of the total variance (calculated in a model without fixed effects).

Figure 2.

Effect of female age on standardized annual fecundity. Breeding females aged 1, 2, 3, 4, 5, 6 and 7 years old. Sample size is 313 breeding events.

Figure 3.

Box plots for standardized annual fecundity for experienced (i.e. breeding together for two consecutive years, n = 19 breeding attempts) and inexperienced breeding pairs (i.e. breeding together for just 1 year, n = 219 breeding attempts).

male and female age effects on migratory performance

Linear mixed models showed that age was related to the date males returned to their breeding grounds. Variation in male arrival date was best explained by a model including male age with significant main and quadratic terms, while accounting for the known source of variation due to among-male identity (Table 3). This model was significant (L-ratio = 73·76, P < 0·0001). Middle-aged males (3 years) arrived earlier than either young (1–2 years) or older males (4 years or older) (Table 3, Fig. 4a). There was no effect of disappearance of low-quality individuals in early life (main effect of ALR, P > 0·2). The quadratic terms of ALR were also not significant (P = 0·21, Table 3). There was considerable between-individual variation in male arrival date, with male identity accounting for 33% of total variance (calculated in a model without fixed effects).

Table 3.  Linear mixed models of standardized male and female arrival date as the response variable, and age and age at last reproduction (ALR) as explanatory variables. Only significant terms are shown except for female ALR. Sample size is 319 and 320 arriving males and females, respectively
 Standardized male arrival date Random effectsStandardized female arrival date
SD95% CISD95% CI
Male/female identity Residual0·4050·223–0·7360·4910·306–0·790
0·8020·688–0·9350·8070·682–0·954
 Fixed-effects 
EstimateSEFPEstimateSEFP
Intercept (β0) 1·7410·245 0·10 0·74 0·7500·222 0·060·80
Age (β1w)–1·0700·20036·40<0·0001–0·7540·19313·880·0004
Age22w) 0·1580·03532·77<0·0001 0·1130·03311·320·001
ALR (β3s)–0·2900·198 1·04 0·30 0·1280·183 1·100·29
ALR2 (β4s) 0·0380·030 1·55 0·21–0·0300·028 1·140·28
Figure 4.

(a) Effect of male age on standardized male arrival date; (b) effect of female age on standardized female arrival date. Curvilinear relationships (i.e. model prediction) are shown. Sample size was 320 females and 319 males.

Age was related in a quadratic fashion to the day females returned to their breeding grounds. The final LME model included both linear and quadratic terms for female age (L-ratio = 74·65, P < 0·001). Female age was related to their arrival date at the breeding grounds in a similar way to males, with an optimum age of arrival at 3 years. Younger (1–2 years) and older females (4 years or more) arrived later at the breeding grounds (Table 3, Fig. 4b). There was no significant effect of disappearance of low-quality individuals on the date of female arrival [estimate (βs) for main and quadratic terms P > 0·2; Table 3]. There was considerable between-individual variation in female arrival date, with female identity accounting for 40% of total variance (calculated in a model without fixed effects).

Discussion

The main findings of this study were (1) that annual fecundity increased with age in early life, reaching a maximum at mid-life and decreasing in late life; (2) the association between breeding performance and age was due mainly to a change in within-individual ageing rather than to a change in between-individual pattern or selection; (3) the association between age and annual fecundity was due mainly to a female age effect; (4) both female and male age affected the timing of reproduction; (5) two reproductive traits (i.e. annual fecundity and laying date) and one migratory performance trait (i.e. arrival date) deteriorated in late life; (6) between-individual variation in individual phenotypic quality did not affect change in migratory performance with age; and we found that (7) some evidence supporting ‘constraints’ hypothesis were found because breeding performance increases with breeding experience and migratory performance improve while ageing. Each of these main findings are discussed briefly.

Numerous empirical studies have established that reproductive performance in birds generally improves with age (Clutton-Brock 1988; Forslund & Pärt 1995; Sanz & Moreno 2000; Green 2001; Pyle, Sydeman & Hester 2001; Laaksonen, Korpimäki & Hakkarainen 2002). However, the reasons for such increases are still poorly understood. Experimental and observational studies in birds and mammals have found evidence consistent with the increase in competence and selection hypotheses for explaining age-related fecundity in early life (De Steven 1978; Desrochers 1992; Komdeur 1996; Balbontín, Penteriani & Ferrer 2003; Ferrer & Bisson 2003; Penteriani, Balbontín & Ferrer 2003). Reid et al. (2003), in a review of the bird literature, reported evidence supporting the differential mortality hypothesis (i.e. selection) in seven of 24 (29%) studies. Most empirical studies were based on cross-sectional comparisons in which, in many cases, these two groups of hypotheses were confounded. Increases in competence have been found to be an important factor in explaining the increase in reproductive success with age. For instance, some studies have found experienced breeders to perform better than inexperienced breeders while controlling for age, providing support for the breeding experience hypothesis (Ainley, LeResche & Sladen 1983; Pyle et al. 1991; Forslund & Larsson 1992; Pärt 1995; Dittmann & Becker 2003). In contrast, other studies did not find this effect of experience on breeding success (Perdeck & Cavé 1992; Newton, Marquiss & Moss 1981; Boekelheide & Ainley 1989; Raleigh & Rendell 2001).

Here, we found evidence suggesting that increases in competence within individuals most probably caused the observed increase in breeding performance with age, at least until reaching middle age. In contrast, we found no support for the selection hypothesis because there was no correlation between life span and breeding performance. These conclusions were not confounded by between-individual differences in quality, because we controlled for such effects in the analysis concerning within-individual changes in breeding success or laying date with age. Furthermore, we identified two different mechanisms associated with an increase in competence. First, male and female barn swallows advance their arrival date at the breeding grounds as they age until reaching middle age. The fitness benefits of early arrival have been documented widely in birds (Forstmeier 2002; Dittmann & Becker 2003), including this species (Møller et al. 2004). Early-arriving individuals have a higher probability of mating, start to reproduce earlier and have higher fecundity than the average individual (Møller 1994a,b). Arrival date has been shown to be condition-dependent, with males in prime condition arriving early at the breeding grounds. Therefore, ageing males and females may increase the skills needed to perform correctly the tasks related to make the long-distance migratory journey safely between winter quarters in Africa and the breeding grounds in Europe. This increase in competence in migratory performance with age would finally result in both an advance in laying date and an increase in breeding success because of the positive relationship between arrival date, laying date and breeding success (Møller 1994a,b). However, it was difficult to disentangle the effects of age from those of arrival and laying dates on annual fecundity in this correlative study. Although the minimal adequate model retained main and quadratic terms of female age, the main effect of female age became non-significant (P = 0·10) when including female and male arrival date and laying date as three new explanatory variables. This was because age, arrival dates and laying dates are correlated with each other, which precludes us from finding an association between these variables and annual fecundity on the predicted direction at the same time.

Secondly, we also reported an increase in competence due to an increase in breeding experience, as demonstrated by the larger annual fecundity of breeding pairs that reproduced together for 2 consecutive years, compared with that of breeding pairs that reproduced only in 1 year and hence were less experienced. Conversely, Saino et al. (2002), studying a population of barn swallows in Italy, did not find a reproductive advantage in terms of clutch size, hatching date, fledging success or offspring phenotype between pairs that re-mated for 2 consecutive years compared with pairs that divorced from one year to the next. These results and those that we have obtained here are not comparable, because in the study by Saino et al. (2002), divorced pairs were considered to be those in which both members survived but did not re-mate, while in the present study less experienced pairs did not re-mate mainly because one of the pair members died between one breeding season and the next.

A previous study of age-dependent changes in reproductive traits and senescence in barn swallows has shown that reproductive success increased in early life, reaching a plateau in middle age and decreasing in old age (Møller & de Lope 1999). However, that study did not explore the effect of female and male age on breeding success separately and therefore could not establish the independent effect of female and male age on fecundity. The present study confirmed age-dependent relationships with fecundity and showed that it was due specifically to an age effect of females. Our study was based on a large sample size, and therefore the power of the statistical tests was large. Studies considering the age of both sexes suggested that performance can vary more closely with either sex depending on species-specific reproductive roles of males and females. Examples for a larger male effect were found in passerines (Nol & Smith 1987; McCleery et al. 1996; Green 2001), auklets (Pyle, Sydeman & Hester 2001) and birds of prey (Espie et al. 2000) and a larger female effect in water-birds (Forslund & Larsson 1992) and bee-eaters and passerines (Lessells & Krebs 1989; Desrochers & Magrath 1993; Smith 1993; Komdeur 1996). In barn swallows, parents feed the young until a few days after fledging, but females contribute more to feeding than their mates (Saino et al. 2002), providing a possible explanation as to why female age effects were more important than male age effects in age-dependent fecundity, at least during the early part of life.

A novel finding of the present study was that female and male age was related to timing of breeding in barn swallows. As far as we know, no association between female and male age and laying date has been shown previously in this or other species. Interestingly, the age of males was related to timing of breeding in a similar way to females. This means that females could benefit by mating with middle-aged males and males by mating with middle-aged females, because this will advance the onset of reproduction of the pair. Therefore, it would be adaptive for each individual to mate with middle-aged mates because this is the optimal age for which lying date was earliest. Future research could focus on individual mating preferences with respect to age, with the aim of trying to disentangle age effects from mating preferences for phenotype traits (for example, tail length). Experimental studies could manipulate the tail length of males and females of different age-classes. For example, tails of individuals of different age could be shortened or elongated to an average of middle-aged individuals, allowing to test for mating preference with respect to age while controlling for tail length. Mating preferences for tail length could be tested by manipulating tail length within different age-classes, thereby controlling for age effects.

Future research might also focus upon testing the optimization of reproductive effort (restraints hypothesis), which was not evaluated in the present study. This hypothesis predicts that the relative amount of resources allocated to reproduction should increase with age. If survival probability decreases only late in life, as expected in the barn swallow, reproductive effort should increase at old age. Observational studies could evaluate investment in reproduction of individuals belonging to different age-classes, with special focus on old age classes. Experimental studies, for example, could manipulate brood size for breeding individuals of different ages. Old individuals should invest more in reproduction than young individuals, specifically when caring for an enlarged brood.

Our long-term study allowed us to quantify within-individual deterioration in two reproductive traits (i.e. annual fecundity and laying date) and one migratory trait (i.e. arrival date) late in life, while controlling for between-individual variation. This finding provides evidence of the existence of senescence in fecundity and arrival date, as shown previously (Møller & de Lope 1999), and in laying date. Linear mixed models revealed considerable variation among individuals in laying date, annual fecundity and arrival date. Individual differences in laying date, annual fecundity and female or male arrival date were not associated with age at last reproduction because neither linear nor quadratic terms were significant in LME. Therefore, the quality of female breeders with respect to timing of breeding or breeding success or the quality of male and female with respect to arrival date was represented equally among the overall range of age-classes.

In conclusion, the observed increase in breeding performance with age in early life was related to a within-individual increase in breeding experience and migratory performance as individuals aged, providing support for the ‘constraints hypothesis’. Conversely, we did not find evidence for the disappearance of poor-quality individuals among age-classes and, hence, the ‘selection hypothesis’ was not supported. Fecundity was affected specifically by age of the breeding female, probably because females invest more in the late stage of reproduction, feeding offspring more extensively than their mates. The age of the two members of a breeding pair affected the onset of reproduction in a similar manner, which could have consequences for the mating preferences of individuals with respect to age. We confirmed the existence of senescence in fecundity and arrival date and found that laying date also deteriorated in late life.

Acknowledgements

We are grateful to all the people that help to obtain data in the field: F. Mateos, C. Navarro, P. Ninni, J. Cuervo, A. Barbosa and S. Merino. The study was supported by the Spanish Ministry of Education and Science (CGL 2006–2913).

Ancillary