Age‐dependent changes in infidelity in Seychelles warblers

Extra‐pair paternity (EPP) is often linked to male age in socially monogamous vertebrates; that is, older males are more likely to gain EPP and less likely to be cuckolded. However, whether this occurs because males improve at gaining paternity as they grow older, or because “higher quality” males that live longer are preferred by females, has rarely been tested, despite being central to our understanding of the evolutionary drivers of female infidelity. Moreover, how extra‐pair reproduction changes with age within females has received even less attention. Using 18 years of longitudinal data from an individually marked population of Seychelles warblers (Acrocephalus sechellensis), we found considerable within‐individual changes in extra‐pair reproduction in both sexes: an early‐life increase and a late‐life decline. Furthermore, males were cuckolded less as they aged. Our results indicate that in this species age‐related patterns of extra‐pair reproduction are determined by within‐individual changes with age, rather than differences among individuals in longevity. These results challenge the hypothesis—based on longevity reflecting intrinsic quality—that the association between male age and EPP is due to females seeking high‐quality paternal genes for offspring. Importantly, EPP accounted for up to half of male reproductive success, emphasizing the male fitness benefits of this reproductive strategy. Finally, the occurrence of post‐peak declines in extra‐pair reproduction provides explicit evidence of senescence in infidelity in both males and females.

tive appearance/disappearance (i.e., the age of entry into/exit from the reproductive population, respectively) of individuals with consistently different ability to gain EPP (van de Pol & Verhulst, 2006).
Indeed, very few studies have attempted to disentangle within-from between-individual effects on EPP (Hsu et al., 2017;Schroeder et al., 2016). Clearly, more longitudinal studies are needed if we are to understand the factors that shape male age-dependent variation in EPP and, therefore, better understand the evolution of infidelity.
The relationship between age and extra-pair reproduction in females remains markedly understudied and unclear. Many reasons have been suggested as to why females may seek extra-pair copulations, including the acquisition of direct benefits (e.g., fertility insurance; Sheldon, 1994) or indirect genetic benefits (e.g., high-quality or compatible genes in offspring; Brown, 1997;Hamilton & Zuk, 1982;Zeh & Zeh, 1996). Older females may have fewer extra-pair offspring because they are more capable of obtaining a better quality social male (Wagner, Schug, & Morton, 1996), and thus do not need to seek extra-pair copulations. Alternatively, they may be more experienced at avoiding or resisting unwanted copulation attempts (Morton & Derrickson, 1990). On the other hand, older females may have more extra-pair offspring because they are better at avoiding mate-guarding, and at obtaining copulations with other males-for "good genes" or other reasons (Bouwman & Komdeur, 2005). Additionally, older females may be more likely to produce extra-pair offspring because they are better at overcoming constraints imposed by male retaliation to perceived paternity loss (the "constrained female" hypothesis; Dixon, Ross, O'Malley, & Burke, 1994;Gowaty, 1996). The few studies that have investigated the relationship between female age and the production of extra-pair offspring have provided contrasting results, showing a positive relationship (Bouwman & Komdeur, 2005;Dietrich, Schmoll, Winkel, Epplen, & Lubjuhn, 2004;Kempenaers, Congdon, Boag, & Robertson, 1999), a negative relationship (Moreno et al., 2015;Ramos et al., 2014;Stutchbury et al., 1997) or no relationship (Lubjuhn, Gerken, Brün, & Schmoll, 2007;Wagner et al., 1996). However, none of these studies distinguished withinand between-individual age effects.
Senescence-the progressive deterioration in performance in late life (Medawar, 1952;Williams, 1957)-is an interesting and important within-individual process related to age. There have been numerous studies assessing the fitness consequences of senescence, most of which have focused on declines in survival and reproduction with age (reviewed by Nussey, Froy, Lemaitre, Gaillard, & Austad, 2013).
However, to our knowledge only one study has addressed, albeit not explicitly, senescence in EPP (Hsu et al., 2017), and has focused only on males.
A within-individual senescent decline in EPP success in late life is compatible with both the good genes and the competitive ability hypotheses. Specifically, in a good genes scenario, the oldest males are the most attractive because they are of highest intrinsic quality (as evidenced by highest longevity) and thus are preferred by females; however, if senescence causes lower fertilization ability (e.g., because of lower sperm competitiveness; Dean et al., 2010), very old males may gain less extra-pair (and total) paternity than younger (less attractive) males. In this case, annual EPP success would be impacted by both a between-individual age effect, represented by a positive association between annual EPP and longevity (because males that live longer are preferred by females), and by a within-individual age effect, resulting in a decline in EPP in all males in late life (because at very old ages males are in lower physical condition). According to the competitive ability hypothesis, all males initially increase their reproductive success as they age (due to increasing experience and/or body condition) but, if senescence occurs, this initial increase in reproduction with age is expected to turn into a decline in late life. In this case, annual EPP would be predicted only by a within-individual age effect, resulting in an increase in EPP at young ages followed by a decline at old ages (while in the absence of senescence EPP should asymptote when added experience, or condition, does not lead to further improvements in the ability to gain EPP). Given that the acquisition of EPP may change with age and/or experience, and also show senescence, it is likely that the contribution of EPP to total reproductive success will vary considerably with age. Numerous studies have investigated how EPP alters male reproductive success (e.g., Albrecht et al., 2007;Lebigre, Arcese, Sardell, Keller, & Reid, 2012), but only a few have done so in relation to age (e.g., Girndt, Chng, Burke, & Schroeder, 2018;Hsu et al., 2017). To our knowledge, only one of these studies disentangled within-and between-individual age effects (Hsu et al., 2017), although this study did not explicitly test for post-peak senescence in EPP.
Here, we investigate extra-pair offspring production in relation to male and female age in the Seychelles warbler (Acrocephalus sechellensis). This long-lived passerine has a mean life expectancy of 5.5 years after fledgling (Komdeur, 1991) and a maximum observed lifespan of 19 years (Hammers & Brouwer, 2017). Seychelles warblers display facultative cooperative breeding (Komdeur, 1992;Richardson, Burke, & Komdeur, 2007): dominant pairs occupy a territory on their own or (in ~30% of territories) are joined by subordinates of either sex . Clutches typically consist of one egg, but ~13% of nests contain one or two additional eggs, often laid by subordinate females (Richardson, Burke, & Komdeur, 2002;Richardson, Jury, Blaakmeer, Komdeur, & Burke, 2001). Individuals are socially monogamous, but ~44% of young are sired by males other than the social male (Hadfield, Richardson, & Burke, 2006;Richardson et al., 2001). Over 97% of all paternity is gained by dominant males (Hadfield et al., 2006;Raj Pant, Komdeur, Burke, Dugdale, & Richardson, 2019;Richardson et al., 2001) either in their own territory (within-group paternity: WGP) or with females from another territory (extra-group paternity: EGP). Thus, in this species EGP is virtually the equivalent of EPP.
In the Seychelles warbler, females that live in larger groups are more likely to produce extra-group offspring (EGO, i.e., offspring sired by extra-group males; Raj Pant et al., 2019). In subordinate mothers, but not in dominant mothers (which account for over 85% of offspring), the likelihood of producing EGO is positively linked to the genetic relatedness to the dominant male in the group (Raj Pant et al., 2019). However, this potential inbreeding avoidance mechanism in subordinate females does not prevent the occurrence of inbreeding in the population (Richardson, Komdeur, & Burke, 2004). In Seychelles warblers, there is evidence that dominant males actively seek EGP during extra-territorial forays (Komdeur, Kraaijeveld-Smit, Kraaijeveld, & Edelaar, 1999). Past research has also shown that females initiate successful copulations, are able to resist mating attempts (Komdeur et al., 1999) and are more likely to gain extra-group fertilizations from more MHC (major histocompatibility complex)-diverse males when paired with males of low MHC diversity (Richardson, Komdeur, Burke, & von Schantz, 2005), which suggests that female choice plays an active role in EGP. However, the relative role of female choice (pre-or post-copulatory) and male-male competition (including sperm competition) in determining patterns of extra-group fertilizations remains unknown in this system.
We use an 18-year longitudinal data set from the Seychelles warbler to determine the patterns of extra-group reproduction in relation to age in males (dominant) and females (dominant or subordinate).
Our isolated study population on Cousin Island provides an excellent system for such investigations: inter-island migration is virtually absent (Komdeur, Burke, Dugdale, & Richardson, 2016;Komdeur, Piersma, Kraaijeveld, Kraaijeveld-Smit, & Richardson, 2004), extrinsic mortality is low (Hammers et al., 2015) and >96% of individuals have been DNA-sampled and individually colour-ringed since 1997 (Brouwer et al., 2010). Accurate parentage assignment and precise estimates of survival and individual reproductive output (throughout life) are therefore available. We assess how patterns of extra-group reproduction are affected by within-individual changes with age and between-individual selective appearance and disappearance effects. We also test for declines in extra-group reproduction in late life (senescence). Finally, we quantify the relative contribution of EGP and WGP success to annual reproductive success in males.
By undertaking the analyses outlined above, we provide evidence to distinguish between different hypotheses as to why females engage in extra-pair mating and improve our understanding of the factors driving the evolution of infidelity.

| Study system
The Seychelles warbler is an insectivorous passerine endemic to the Seychelles archipelago. The population on Cousin Island (29 ha, 04°20′S, 55°40′E) has been monitored as part of a long-term study, which started in 1981 and intensified in 1997 Komdeur, 1992;Richardson, Komdeur, & Burke, 2003). Since then, virtually all breeding attempts have been followed each year during the main breeding season (June-September). As many birds as possible were captured every year, either using mist-nets or as nestlings. Newly caught individuals were assigned a unique combination of three colour rings and a British Trust for Ornithology metal ring.
Inter-island dispersal is virtually absent (<0.1%) in the Seychelles warbler  and individual resighting probability per season on Cousin Island is very high (~92-98%, Brouwer et al., 2010); therefore individuals not seen over two consecutive seasons can safely be assumed to be dead (Hammers, Richardson, Burke, & Komdeur, 2013).
During each breeding season, group membership, social status and territory boundaries were assigned for all birds using observations of foraging and singing locations, non-aggressive social interactions, and aggressive territorial interactions (e.g., Bebbington et al., 2017). Within groups, dominant pairs were identified via pair and courtship behaviours. Subordinate birds, which are often offspring that have delayed dispersal (Kingma, Bebbington, Hammers, Richardson, & Komdeur, 2016), are classified as "helpers" or "nonhelpers" based on their participation in incubation (females only) and in feeding offspring (both males and females; Hammers et al., 2019;Komdeur, 1994).
Reproduction is seasonally constrained by invertebrate prey availability and offspring are fed for up to three months after fledging (Komdeur, 1991). We refer to the dominant male in a group as the "social male" of any reproductively mature females in his group (dominant and subordinate), as males can fertilize both dominant and subordinate females in their territory.

| Data selection
We gathered previously generated parentage data for 934 Seychelles warblers that were assigned parentage with high confidence (≥80%) and that hatched on Cousin Island during main breeding seasons in the period 1997-2014 (Edwards et al., 2018;Hadfield et al., 2006;Richardson et al., 2001;Sparks et al.,2020). We used these data to assess the age-dependent production of EGO by females and the age-dependent risk of cuckoldry for their social male partner (the dominant male in the group). We first tested if the likelihood that an offspring was sired by a male outside the breeding group ("EGP likelihood") was related to the age of the mother (dominant or subordinate) and/or the age of the dominant male. The ages of dominant males and females are only weakly correlated in the Seychelles warbler (r = 0.16; Hammers et al., 2019). Given that Seychelles warbler females do not lay eggs in nests outside their own territory (Richardson et al., 2002), EGP likelihood will capture female infidelity.
We compiled 1,364 annual records of all dominant males alive between 1997 and 2014 that were genetically assigned at least one offspring across the whole data period (including entries of males siring no young in single years, n = 237 males). For each male, we determined the annual number of EGO (i.e., EGP success) and within-group offspring (i.e., WGP success). We then estimated the annual proportion of EGO sired by each dominant male (535 annual records from 233 males, excluding cases in which a male had an annual reproductive success of zero). In our paternity measures, we included only offspring that survived for at least three months to remove any potential bias on annual reproductive estimates resulting from differing catching efforts across years (which cause offspring to be caught at different ages in different years). Using these data, we assessed the relationship between male age and WGP, EGP, and the annual proportion of EGO sired by each male (i.e., the contribution of EGP to annual reproductive success).

| Statistical analyses
We quantified within-individual effects of age on the production of EGO (i.e., longitudinal changes throughout an individual's lifetime).
To separate out between-individual (population-level cross-sectional) effects of age (i.e., due to selective appearance and disappearance), we employed the method developed by van de Pol and Verhulst (2006). By including age of first reproduction and age of last reproduction (or longevity) in addition to age in the same mixed model, this approach allows us to quantify the within-individual effect of age (expressed by the age term) while controlling for selective appearance and disappearance (expressed by age of first and last reproduction/longevity, respectively). Here, we modelled selective appearance using the age of first dominance for males, to account for when they could potentially start breeding (virtually all paternity is obtained by dominant males in the Seychelles warbler; Raj Pant et al., 2019). Because females can reproduce before gaining dominance, we used the age at which females were first assigned an offspring as subordinates or the age of first dominance-whichever came first (subsequently termed "age of first dominance" for simplicity). Age at death (longevity) was used to model selective disappearance for both males and females. Individuals of unknown longevity (i.e., birds translocated to other islands or those that had not died by the end of 2018) were excluded from the analyses. The chronological age of individuals was always included as a fixed effect alongside age of first dominance and longevity so that it represents the withinindividual effect of age on EGP. Chronological age, age of first dominance and longevity were all measured in years (integers).
Reproductive performance can change shortly before death, independently of age (Bowers et al., 2012;Coulson & Fairweather, 2001). Therefore, to avoid confounding any age-related effects with an age-independent terminal effect, we included a binary variable in models indicating whether a bird died before the subsequent breeding season.
We performed statistical analyses in R (3.6.3) with generalized linear mixed models (GLMMs) fitted using the lme4 (1.1-20) package (Bates, Mächler, Bolker, & Walker, 2015). We built separate GLMMs to analyse the following variables (summarized in Table 1): (1) offspring EGP likelihood-i.e., the likelihood that the offspring was sired by a male outside the group (yes/no binary variable), in relation to the age of the mother (n = 852 offspring) or the age of the dominant male in the group (n = 848 offspring); (2) annual paternity obtained by each male (n = 1,364 male/years) split into (2a) EGP success (number of EGO sired), (2b) WGP success (number of within-group offspring sired) and (2c) total reproductive success (i.e., extra-group + within-group young sired); and (3) annual proportion of EGO (i.e., the number of EGO over the number of total offspring per male per year) sired by each dominant male that was assigned at least one offspring in a given year (n = 535). Models analysing EGP likelihood and the annual proportion of EGO sired by each male were constructed with a binomial error structure and logit link function, while models of paternity success (EGP/ WGP/annual reproductive success) were built with a Poisson error structure and log link function. Furthermore, we built a GLMM to perform a population-level comparison between the age of the social father (i.e., the cuckolded male) and the age of the genetic father (i.e., the extra-group sire) of each EGO (n = 395). The GLMM was built with a Poisson error structure and log link function. We checked for collinearity between fixed effects using the variance inflation factor (VIF) and found none (VIF ≤ 3). We standardized (mean-centred and scaled to one standard deviation) continuous predictors and used the "BOBYQA" nonlinear optimization (Powell, 2009) to aid convergence of models. The statistical significance of model terms was assessed using parametric bootstrap p-values (nsim = 3,000) from full models containing all biologically meaningful predictors of interest.
We assessed female and social male age effects on offspring EGP likelihood in separate models to avoid any potential bias caused by the non-independence of female and social male age over time (within our data set, 51% of females that reproduced in multiple years and 53% of social males that raised offspring in multiple years had more than one social partner). Both female and social male models contained four random effects (random intercepts): year, territory, mother's identity and social male's identity (pair identity explained zero variance and was not included as an additional random effect). Female and social male models also included two fixed effects in common: the age difference between the mother and her social male and the offspring's natal group size (offspring EGP likelihood is higher in larger groups; Raj Pant et al., 2019).
The model analysing female age effects on EGP likelihood included as additional fixed effects the mother's age (linear and quadratic), age of first dominance, longevity and terminal effect.
To check for any potential bias caused by inbreeding avoidance effects occurring in subordinate females (Raj Pant et al., 2019), we re-ran this model on offspring produced only by dominant females (n = 759) and compared results with those from the model run on the full data set (i.e., offspring produced by dominant and subordinate females, n = 852).
The model analysing social male age effects on EGP likelihood contained the corresponding additional fixed effects of social male instead of female age traits (i.e., a social male's age, age of first dominance, longevity and terminal effect). Furthermore, we built a model to test for any differences between the age of cuckolded males and the age of the extra-group sires that cuckolded them (population-level analysis). This model featured male age as the response variable, male status (i.e., extra-pair or cuckolded male) as a fixed effect and three random effects (random intercepts: social father, genetic father and mother identity). Models analysing annual paternity success (EGP/WGP/reproductive success) per male and the annual proportion of EGO sired by each male included five fixed predictors-male age (linear and quadratic), age of first dominance, longevity and a terminal effect-and three random effects (random intercepts)-year, territory and male identity. Because annual EGP and WGP may affect one another, when analysing EGP and WGP, we also included as fixed effects either WGP or EGP respectively, and the interaction between WGP/ EGP and male age.
A negative quadratic relationship between reproductive components and age does not necessarily indicate that a late-life decline in these components exists but may just represent that an increase early in life levels off at later ages (Bouwhuis, Sheldon, Verhulst, & Charmantier, 2009). To determine whether EGP likelihood and paternity success exhibit true late-life declines consistent with senescence, we tested for linear age effects after the peak age for each of these components. We estimated peak ages from the linear and TA B L E 1 Summary of the response variables addressed in our analyses of age-related changes in the reproduction of the Seychelles warbler, highlighting the age-effects we detected quadratic coefficients of age, as (−β linear )/(2 × β quadratic ), from models we built with non-standardized data. We compiled subsets of individuals with ages ≥ the peak age for offspring EGP likelihood (female age effects: n = 319; social male age effects: n = 346), paternity success (within-group: n = 598; extra-group and total: n = 360), or the proportion of EGO sired (n = 97) per male. We ran models regressing EGP likelihood, paternity success (extra-group/within-group/total) or the proportion of EGO sired over the linear age (post-peak) of individuals and other predictors included in previous models, except the quadratic age term. For simplicity, when analysing EGP/WGP in the post-peak subsets we also excluded WGP/EGP and the interaction between WGP/EGP and male age, which were all nonsignificant predictors in full data set analyses (see Table 4).

| Offspring EGP likelihood and female age
The proportion of offspring sired by an extra-group male was 42% in the population. There was a negative quadratic effect of maternal age on offspring EGP likelihood, which increased from a predicted ~29% for mothers in their first year to ~46% when the mother was 5.6 years old, after which it decreased to ~10% for the oldest mothers ( Figure 1 and Table 2). Furthermore, the older a female was relative to the dominant male in her group, the higher the likelihood was that she produced an EGO (Table 2). Regarding senescent effects, the EGP likelihood of offspring produced by females ≥6 years old (i.e., past the peak age of EGP likelihood) declined with female age (β ± SE = −0.41 ± 0.20, p = .039; Table S1). EGP likelihood was not affected by the mother's age of first dominance, longevity or a terminal effect (Table 2). EGP likelihood was positively related to group size ( Table 2). All results remained quantitatively similar when repeating the analysis of EGP likelihood using only offspring produced by dominant mothers (n = 759; Table S2). This indicates that any inbreeding avoidance effect occurring via extra-group reproduction by subordinate females does not bias our results on age-dependent production of EGO by females.

| Offspring EGP likelihood and social male age
When analysing offspring EGP likelihood-i.e., the probability that a male was cuckolded (on an offspring by offspring case)-in relation to male age, we found a positive quadratic effect of male age. The likelihood of being cuckolded decreased within individuals, from a predicted ~44% in young males to ~34% in males of 6.2 years of age ( Figure 2 and Table 3). Despite the positive quadratic effect of age there was no post peak senescent effect-i.e., that males were increasingly more likely to be cuckolded when they were ≥6 years old (β ± SE = 0.05 ± 0.19, p = .808; Table S3). Males that lost WGP were on average 1 year younger than the extra-group

F I G U R E 1
The likelihood of offspring extra-group paternity (EGP) in relation to maternal age in the Seychelles warbler (n = 852 offspring). Means of raw data (points) and their standard error (bars) are shown with associated sample sizes. The black line represents the prediction from the GLMM (Table 2)  β Male status (extra-group sire) = 0.23 ± 0.03, p < .001; Figure S1). The positive relationship between the probability of WGP loss and the female-social male age difference did not reach statistical significance (p = .067). The probability of being cuckolded was not associated with male age of first dominance or a terminal effect, and only showed a non-significant tendency to decrease with male longevity (p = .063). The probability of WGP loss was positively associated with group size (Table 3).

| Annual paternity success and male age
When analysing annual paternity success (extra-group, within-group and total) in relation to male age, we found an early-life increase and a late-life decline within individuals (Table 4). Specifically, there was a negative quadratic effect of male age on EGP success; the predicted number of extra-group offspring sired per annum increased from ~0.06 in males in their first year to peak at ~0.24 at 7.8 years and decreased thereafter to ~0.03 in the oldest males ( Figure 3 and Table 4). There was also a negative quadratic effect of male age on annual WGP gained; the predicted number of within-group offspring sired increased from ~0.23 per annum in males in their first year to ~0.32 at 6.1 years and declined to ~0.09 in the oldest males ( Figure 3 and Table 4).
As a result of the age-related changes in EGP and WGP outlined, total predicted annual reproductive success increased with male age from ~0.29 offspring in first-year males up to ~0.58 at 7.7 years, before declining to ~0.14 in the oldest males ( Figure 3 and Table 4). The post-peak reduction in male reproduction in late life was confirmed by the significant negative linear relationships between age and annual EGP in males ≥ 8 years (β ± SE = −0.32 ± 0.16, p = .046), WGP in males ≥ 6 years (β ± SE = −0.27 ± 0.10, p = .013) and total paternity success in males ≥ 8 years (β ± SE = −0.25 ± 0.11, p = .021; Table   S4). Male annual extra-group, within-group and total reproductive success were not affected by male longevity, age of first dominance or a terminal effect (Table 4). We found no evidence of a trade-off between EGP and WGP: WGP and its interaction with male age were not related to EGP success, and EGP and its interaction with male age did not predict WGP success (Table 4). When analysing the proportion of a male's annual reproductive output obtained outside his own group, this increased with age, from a predicted ~19% in firstyear males to a peak of ~49% at 8.6 years ( Figure 4 and Table 5).
Despite finding a significant negative quadratic effect of age, there was no significant senescent decline in the proportion of reproductive success resulting from EGP with age in males ≥9 years old (β ± SE = −0.23 ± 0.25, p = .343; Table S5). This suggests that the proportion of EGO sired remained relatively stable after peaking, probably as the decline in the amount of extra-and within-group offspring sired were similar in late life. The annual proportion of EGO sired was not influenced by male longevity, age of first dominance or a terminal effect (Table 5). and-as a result of EGP and WGP patterns-in total reproductive success. Moreover, the likelihood of being cuckolded decreased within males from early to midlife. Extra-group reproduction accounted for ~50% of annual reproduction for males at their reproductive peak. No age-dependent differences among individuals, due to selective appearance or disappearance, were detected in relation to extra-group reproduction in either males or females, or in relation to within-group and total reproductive success in males. We detected F I G U R E 2 The likelihood of withingroup paternity loss in relation to male age in the Seychelles warbler (n = 848 offspring). Means of raw data (points) and their standard error (bars) are shown with associated sample sizes. The black line represents the prediction from the GLMM (Table 3) (Table 4)

| Age-dependent female extra-group reproduction
The likelihood of producing EGO changed with age within females, increasing until females were 5.6 years old and declining thereafter (Figure 1), but there were no selective appearance or disappearance effects (between-female age effects). Our findings are consistent with some cross-sectional studies that have found a positive association between female age and infidelity (Bouwman & Komdeur, 2005;Dietrich et al., 2004;Kempenaers et al., 1999), while other cross-sectional studies have shown a negative relationship (Moreno et al., 2015;Ramos et al., 2014;Stutchbury et al., 1997) or no relationship with age (Cordero, Wetton, & Parkin, 1999;Li & Brown, 2000;Lubjuhn et al., 2007;Veiga & Boto, 2000;Wagner et al., 1996). To our knowledge, no other studies have separated within-from between-individual age effects on the production of EGO by females.
The age-related increase in female extra-group reproduction we observed may be due to increases in experience and/or body condition of females with age (until they approach 6 years).
In female Seychelles warblers, breeding and helping experience, which accrue with age, enhance the number of offspring raised to independence (Komdeur, 1996). Moreover, female reproductive success increases until they reach 6.5 years of age (Hammers, Richardson, Burke, & Komdeur, 2012), suggesting that a female's physical condition (and experience) improves until this point. It is possible that females at this peak of reproduction are more attractive to males seeking EGP (which may perceive them as successful reproducers) and that they are targeted for extra-group fertilizations. In that case, the detected within-female change in reproduction may be mostly, or even entirely, male-driven. Another possibility is that, as they grow older, females improve their ability to avoid mate-guarding and engage in extra-group copulations, thanks to experience or improved body condition (Bouwman & Komdeur, 2005). Additionally, greater experience and/or condition may allow older females to cope better with any reduction in paternal investment that may occur when males perceive a loss of paternity, thus allowing females to gain more extra-group fertilizations (the constrained female hypothesis; Gowaty, 1996).
However, indirect evidence suggests that female extra-group reproduction is not constrained by male retaliation in the Seychelles warblers. In territories with cooperative breeding, helpers provide load-lightening to the dominant pair (van Boheemen et al., 2019;Hammers et al., 2019) and this might liberate dominant females from the costs imposed by male retaliation (Mulder, Dunn, TA B L E 4 Parameter estimated from GLMMs of annual (A) extra-group paternity (EGP) success, (B) within-group paternity (WGP) success and (C) total reproductive success (RS) in relation to male age in the Seychelles warbler (n = 1,364). Variance (σ 2 ) and number of observations (n) are shown for each random effect.

Fixed effects (A) EGP success (B) WGP success (C) Total RS
Significant (p < .05) terms are shown in bold.
Abbreviations: AFD, age of first dominance; X, interaction between two terms.
Cockburn, Lazenby-Cohen, & Howell, 1994). Contrary to the expectation based on this logic, the presence of helpers is not asso- When analysing female age effects, we also found that the older a female was compared to the dominant male in her group, the higher the likelihood was that she would produce an EGO. This is in accordance with other studies in which the production of extra-pair offspring was based on the combination of the female's age and that of her social male (Bouwman & Komdeur, 2005;Dietrich et al., 2004;Ramos et al., 2014;Rätti, Lundberg, Tegelström, & Alatalo, 2001)but see Lubjuhn et al. (2007) and Moreno et al. (2015). This result further suggests that females may be targeted more by extra-group males-and/or more easily avoid mate-guarding-when socially

F I G U R E 4
The proportional contribution of extra-group paternity (EGP) to the annual reproductive success of dominant male Seychelles warblers (siring ≥ 1 offspring), in relation to age (n = 535 annual observations from 233 males). The means of raw data (points) and their standard error (bars) are shown with associated sample sizes. The black line represents the prediction from the GLMM (Table 5)   paired with a male that is considerably younger than the female (i.e., a young male that is not skilled at mate-guarding and/or defending his territory from intruders).

| Female benefits of infidelity
One key hypothesis suggests that females may seek extra-group fertilizations to obtain good paternal genes for their offspring (Hamilton & Zuk, 1982) and age is expected to reflect individual quality via viability (Trivers, 1972). Consequently, the fact that many (cross-sectional) studies have shown that older males gain more paternity through extra-pair reproduction than younger males (Ackay & Roughgarden, 2007;Hsu et al., 2015) has often been put forward as support for the good genes hypothesis (Forstmeier et al., 2014). In the Seychelles warbler, we found that male paternity gain (and loss) varied with age within individuals, and that age-related changes were not explained by selective appearance or disappearance effects. Similar results were found in the two other studies that have separated within-from betweenindividual age effects on EPP success and within-pair paternity loss (Hsu et al., 2017;Schroeder et al., 2016). This lack of any  (Brouwer et al., 2010;Richardson et al., 2005). However, any female (pre-/post-copulatory) preference for more MHC-diverse extra-pair males would not cause a between-individual effect of male age on EPP in the Seychelles warbler, because the survival benefit of higher MHC diversity is only observed in juveniles. In adult males, there is no differential survival linked to MHC diversity (older adult males are not more MHCdiverse than younger adult males).
Further work is now required to understand the mechanisms through which males improve their ability to gain EGP with age, and whether this also provides any benefits to females. Females may also engage in extra-pair mating to gain other types of benefits (Forstmeier et al., 2014), such as fertilization assurance (Sheldon, 1994) in case they are socially paired with truly infertile males (Hasson & Stone, 2009).
In the Seychelles warbler, individual males both gain EGP and lose WGP. This indicates that males that become cuckolded are not infertile but does not rule out that extra-group copulations could have evolved to guard against any rare cases of infertility (although totally infertile males have never been identified in the Seychelles warbler).
Another reason why females may seek extra-pair fertilizations is to acquire indirect genetic non-additive benefits (e.g., compatible genes in offspring; Brown, 1997;Zeh & Zeh, 1996). However, unlike "good genes" benefits, other benefits are not normally expected to be signalled by male viability.
Alternatively, it is possible that infidelity may not provide any benefits for females and instead may have evolved as a by-product of positive selection on genetically correlated traits in males (between-sex correlation) or in females themselves (Arnqvist & Kirkpatrick, 2005;Forstmeier, Martin, Bolund, Schielzeth, & Kempenaers, 2011;Forstmeier et al., 2014;Halliday & Arnold, 1987).
This idea, which has received little attention so far, may constitute a promising avenue in unveiling the evolution of infidelity in socially monogamous species, but assessing this hypothesis is beyond the scope of the current study.

| Male age-dependent paternity gain and loss
Both extra-group and within-group paternity success increased within individual male Seychelles warblers in early life and declined in late life. Moreover, the likelihood of being cuckolded decreased within males at young ages and remained stable from midlife onward.
These within-male changes in reproduction and cuckoldry, coupled with the lack of between-male differences due to selective appearance and disappearance, do not provide evidence for the good genes hypothesis (Hamilton & Zuk, 1982) but support the competitive ability hypothesis . This hypothesis argues that the improvement in male paternity success with age is due to increasing experience (Hsu et al., 2015;Westneat & Stewart, 2003) or body condition    Cleasby & Nakagawa, 2012) and with a longitudinal study on house sparrows which identified age-dependent increases in EPP and within-pair paternity success (Hsu et al., 2017). However, our results are particularly insightful because they clearly show that these age-related changes in extra-and within-group paternity occur within individuals and not as a result of preferred males living longer.
There was no evidence of a trade-off between WGP and EGP gain in Seychelles warblers and both WGP and EGP success increased in early life and declined in late life ( Figure 3); this indicates that when males obtain more WGP success, they do not do so at the expense of EGP gains, and vice versa. The combined result of EGP and WGP is that annual reproductive success changes with male age, increasing until 7.7 years and declining thereafter (Figure 3).
Such within-individual variation in reproductive success (an increase in early life followed by a decline in late life) is common in vertebrates (Nussey et al., 2013). In the Seychelles warbler, annual EGP success displayed a particularly steep increase at young ages, thus strongly intensifying the spike in annual male reproductive success at 7.7 years of age ( Figure 3). It is widely recognized that senescence is an important age-related process occurring in the wild. Numerous studies have assessed senescence in multiple traits (Hayward et al., 2015), including survival (e.g., Cameron & Siniff, 2004) and reproductive output (e.g., Dugdale, Pope, Newman, MacDonald, & Burke, 2011), across a number of species. In the Seychelles warbler, senescence in female reproductive success has been detected in the past (Hammers et al., 2012). To our knowledge, however, no studies have assessed senescence in extra-pair reproduction in females, and only one study has addressed (and found) senescence in extra-pair reproduction in males (Hsu et al., 2017), although this study did not explicitly test for senescent post-peak declines in EPP.
Here, we analysed senescent post-peak declines in extra-group reproduction and found these to occur in both male and female Seychelles warblers. In males, we also assessed and found evidence for senescence in WGP and total paternity success. Our results highlight the importance of the role that senescence plays in the alternative pathways to reproductive success in this and possibly other species.

| CON CLUS IONS
The lack of between-male age effects on extra-group reproduction emerging from our study undermines the often cited suggestion that male age-dependent patterns of EPP success support the good genes hypothesis for the evolution of female infidelity. Our results provide support for the idea that infidelity may be important to females for other reasons, such as the acquisition of compatible genes in offspring, or that infidelity evolved because of genetic constraints (i.e., genetic correlation between infidelity and traits under positive selection). Our analyses also provide, to our knowledge, the first explicit evidence that senescence in extra-group reproduction occurs not only in males, but also in females. Finally, our work shows that EGP explains a large proportion of the annual reproductive success of males, and that age-specific changes in EGP amplify age-dependent patterns of reproduction. Further work is now needed to understand how this affects male variance of reproductive success and therefore selection for infidelity.

ACK N OWLED G EM ENTS
We are grateful to the Republic of Seychelles Department of Environment and the Seychelles Bureau of Standards for permission to undertake this research and permits to export samples. We are also very grateful to Nature Seychelles for the opportunity to conduct fieldwork on Cousin Island and their support with logistics. We thank all the many enthusiastic fieldworkers who have contributed to the long-term data collection in the Seychelles warbler project, Owen Howison for maintenance of the Seychelles warbler database and Marco van der Velde for microsatellite genotyping. We are also grateful to five anonymous reviewers for providing construc-