Older males attract more females but get fewer matings in a wild field cricket

The age of potential mates has been proposed to be an important target for mate choice by females. Alternative hypotheses predict preferences in either direction. Females might be expected to prefer older males because such males have demonstrated their capacity to survive. Alternatively, they might prefer younger males that have not accumulated deleterious mutations. Preferences in both directions have been observed in laboratory experiments, suggesting that this is an issue that needs to be understood within its ecological context. We measured individual behaviour and reproductive success in a natural population of the ﬁ eld cricket Gryllus campestris over 10 years. We found that in this annual insect, a male's age relative to his peers was poorly correlated with his life span. This suggests that there is limited potential for selection to favour female choice for older males because a strategy of choosing older males would not signi ﬁ cantly increase a female's likelihood of mating with a long-lived male. Older males were more successful at pairing up with females at a burrow, but once paired they were less likely to mate with them. By genotyping the next generation of adults we con ﬁ rmed that observations of both pairing up with a female and matings were associated with successful offspring production. However, there was no relationship between how old a male was at mating and how many adult offspring he had. This lack of evidence for

Females are expected to exercise choice among potential mates if that choice leads to higher quality offspring. One male trait on which females may base these decisions is the age of a potential mate. However, there are contrasting (but not exclusive) predictions regarding potential advantages or disadvantages for females choosing older or younger males as mates. Germ-line mutations accumulated during a male's life have the potential to lead to declines in fertility and offspring viability (Johnson & Gemmell, 2012). Such declines should select for avoidance of mating with older males in order to maximize fertilization and development of females' eggs (Aitken & De Iuliis, 2007;Beck & Promislow, 2007;Preston, Saint Jalme, Hingrat, Lacroix, & Sorci, 2015). In contrast, females may benefit from choosing older mates if survival to old age is indicative of genetic quality, which will be inherited by their offspring (Brooks & Kemp, 2001;Kokko & Lindstr€ om, 1996). These two opposing selection pressures will act simultaneously, with the expectation being that female preferences will reflect the balance between costs and benefits of mating with older or younger males.
Organisms tend to exhibit physiological deterioration with age, a process known as senescence (Rodríguez-Muñoz, Boonekamp, Liu, Skicko, Haugland Pedersen, Fisher et al., 2019;Rose, 1991). This deterioration is generally thought to be the result of trade-offs between somatic maintenance and investment in reproduction (Kirkwood & Holliday, 1979;Partridge & Barton, 1993;Rodríguez-Muñoz, Boonekamp, Liu, Skicko, Fisher, Hopwood et al., 2019;Williams, 1957). In this context there is a clear prediction that individuals that are better adapted to their environment or that carry other forms of 'good genes' should be able to sustain increased costs of both somatic and reproductive investment. As a result, these individuals are expected to have a greater capacity to survive to an old age. Consequently, females who preferentially mate with older males should increase the probability that their eggs will be fertilized with sperm carrying genes that confer high fitness on their offspring (Andersson, 1994;Trivers, 1972). However, the predictions made by models examining the value of old age as an indicator of high quality depend on the assumptions upon which they are built. Hansen and Price (1995) dismissed the potential role of age as an indicator of high genetic quality. They invoked potential declines in fertility and increases in mutation load with age (as discussed above) and argued that survival to old age is not an indicator of genetic quality. In contrast, a simulation model by Kokko and Lindstr€ om (1996) supported the possibility of age acting as a good-genes indicator. This latter conclusion seems more likely to be correct, as discussed by Kokko (1998), who demonstrated that Hansen and Price's (1995) model did not allow for differences in quality between males. Nevertheless, Kokko and Lindstr€ om's (1996) finding relied on large differences in genetic quality between males expressed in their life span. In such a case, females can benefit from a mating rule that makes them increasingly likely to mate with a male according to his age. However, in assessing the likely importance of such a mechanism, note that although older males are known to be long-lived, young males may also go on to be long-lived. Hence the strength of the correlation between the age of a male relative to his peers and how long he lives may not be as strong as it intuitively appears to be. This is an issue that becomes particularly important in populations where the assumption that there is a stable age distribution is not met, for instance in annual organisms with a single breeding season.
Across taxa, populations vary from those in which males continually enter the adult population over the course of 2 years or more, as in most vertebrates, to those with discrete generations in which males become adult during a brief annual time window ( Sniegula, Gołą b, & Johansson, 2016). This variation in seasonal life history determines how accurately age reflects life span. At one extreme, if all males are born simultaneously, they will all have the same age, so at any point in time when a female is choosing a mate, the age of the males available to her will have no relationship to their longevity. Annual species, such as most temperate insects where adults emerge during a short time window, are not quite at this extreme, but nevertheless, the relative age of the males available as mates will provide a more limited amount of information about their longevity than in continuously reproducing populations. Females could delay mating in order to sample only old males, but they risk dying and deteriorating in the meantime and they would have to delay producing offspring. Furthermore, delaying mating would select for males that emerge later, which might reduce the power of delaying mating as a strategy to select for longer-lived adult males.
Species where there is high variance in the date of birth of males and in their longevity include many birds (Akçay & Roughgarden, 2007), for example great bustards, Otis tarda (Alonso, Magaña, Palacín, & Martín, 2010), numerous large mammals, for example Soay sheep, Ovis aries (Clutton-Brock & Pemberton, 2004), and many reptiles, including Lacerta monticola (L opez, Arag on, & Martín, 2003). In these species, there may be strong relationships between age and longevity (Reid et al., 2010). The demographic structure of these species contrasts sharply with that of annual species with discrete generations, meaning that evidence of female choice for older males from the former should not be generalized to indicate its likely prevalence in the latter.
Empirical analyses on the effect of male age on female mate choice provide conflicting results. In the laboratory, there are studies showing female preference for older (Avent, Price, & Wedell, 2008;Somashekar & Krishna, 2011), intermediate (Jones, Balmford, & Quinnell, 2000;Jones & Elgar, 2004;Liu, Xu, He, Kuang, & Xue, 2011) and younger males (Prokop, Stuglik, _ Zabi nska, & Radwan, 2007), and even no preference for any particular age of male (Gasparini, Marino, Boschetto, & Pilastro, 2010;Martin, Leugger, Zeltner, & Hosken, 2003). Studies in natural or seminatural populations are scarcer, and report preferences for older males (L opez et al., 2003;Simmons & Zuk, 1992). Studies on the effect of age on fertilization and embryo or offspring viability are more consistent in their findings. Negative effects of male age on fitness are found both in the laboratory (Jones & Elgar, 2004;Jones et al., 2000;Liu et al., 2011;Preston et al., 2015;Priest, Mackowiak, & Promislow, 2002) and in natural or seminatural environments (Richard, Lecomte, De Fraipont, & Clobert, 2005;Schroeder, Nakagawa, Rees, Mannarelli, & Burke, 2015), and include insect species where the range of possible ages within a cohort is just a few weeks or even days (Jones & Elgar, 2004). A positive effect has been reported by Avent et al. (2008), although they argued that their results were best explained by direct benefits to females due to older males transferring greater quantities of sperm and accessory gland proteins.
Crickets have been proposed as a potential model system for the study of interactions between sexual selection and senescence (Archer & Hunt, 2015). Males sing to attract females and females can discriminate between males based on their calls (Archer & Hunt, 2015;Fitzsimmons & Bertram, 2011;Jacot, Scheuber, & Brinkhof, 2007;Verburgt, Ferreira, & Ferguson, 2011). However, opposing results in relation to female preferences for male age have been found even within the same species: in the cricket Gryllus bimaculatus, males' songs declined in energetic quality as they aged and females preferred younger males in a laboratory setting (Verburgt et al., 2011), whereas in a natural population, paired males were older on average than unpaired singing males (Simmons & Zuk, 1992). However, female preferences in relation to male age, and their associated fitness consequences, have never been studied directly in crickets or any other wild insect. We monitored a wild population of the field cricket Gryllus campestris in a meadow in northern Spain for 12 annual generations, recording the individual date of adult emergence and death, monitoring behaviours including matings, and building a pedigree across generations. Here we use these data to analyse (1) how reliable male age is as a signal of male longevity, (2) whether female crickets select older or younger males as mates, and (3) whether the sire's age influences female reproductive success.

METHODS
Our WildCrickets project (see WildCrickets.org) has monitored a population of wild G. campestris in a meadow in northern Spain for 12 consecutive years, 2006e2017 (Rodríguez-Muñoz, Boonekamp, Fisher, Hopwood, & Tregenza, 2019;Rodríguez-Muñoz, Boonekamp, Liu, Skicko, Haugland Pedersen, et al., 2019;Rodríguez-Muñoz, Bretman, Slate, Walling, & Tregenza, 2010). G. campestris has a single annual generation; nymphs of both sexes dig burrows in the autumn and overwinter in them, emerging to resume foraging and growth in late spring. The first adults appear in our meadow in late April. A few days after becoming adult, males begin to call to attract females, and both sexes begin moving around the meadow seeking matings. This movement results in frequent switching of burrows, and adults spend 0.45 ± 1.05 days (mean ± SD) between movements. Burrows provide a refuge from predation and bad weather, but are too narrow to allow crickets to mate (Rodríguez-Muñoz, Bretman, & Tregenza, 2011). When two members of the same sex meet at a burrow, usually one of them immediately leaves, or there is a fight, followed by the loser leaving. When members of the opposite sex meet, fights are very rare; normally either one of them leaves, or the pair temporarily cohabit at the burrow, often mating repeatedly during this period (Rodríguez-Muñoz et al., 2011). After an average ± SD of 0.64 ± 1.44 days, one of the pair moves away from the burrow; we do not know which individual is responsible for initiating these fissions since a male departure might be provoked by some behaviour by females, but pairs that mate more frequently tend to stay together for longer (Rodríguez-Muñoz et al., 2011). Males do not appear to attempt to prevent females from leaving, suggesting that females remaining at a burrow with a male are choosing to do so. Both sexes frequently have multiple mating partners throughout their lives (Rodríguez-Muñoz et al., 2010), and females store sperm (Tyler, Harrison, et al., 2013;Tyler, Rodriguez-Munoz, & Tregenza, 2013).
The WildCrickets meadow is managed in a similar way every year, with the grass being mowed in mid-March and again in July or August. Between August and March, the grass is kept short with additional mowing. Weekly searches for burrows are made from February until the end of the breeding season sometime in July, when the last adult cricket dies. Each burrow is flagged with a unique number that will identify it for the whole breeding season. By mid to late April, usually before the adults start to emerge, we install up to 133 infrared day/night cameras. The cameras use motion-activated digital video recording software (i-Catcher, icodesystems.co.uk) to continuously record the activity around each burrow entrance onto servers housed in a building adjacent to the meadow.
A few days after each individual emerges as an adult, we trap it in its burrow and glue a PVC tag onto its pronotum. The tag has a unique one-or two-character code allowing us to identify the individual on video. We also collect a drop of haemolymph (sampled by piercing the membrane at the hind leg joint) and a small piece of the tip of one of the hind legs. These samples are later used to provide individual DNA profiles (see below). Marked and sampled crickets are weighed, photographed and released back into the same burrow from which they were collected.
Occupied burrows sometimes outnumber cameras, and adult crickets regularly move around the meadow occupying different burrows. To track crickets at burrows lacking a camera we carry out direct daytime observations every 1e2 days. We record the ID of any adult present or whether a nymph is in residence. This allows us to record accurate adult emergence dates even if burrows are not directly monitored by video at that particular time (nymphs and recently emerged adults rarely move between burrows so the presence of an adult where there was a nymph the day before indicates an emergence). Video watching is still to be concluded for 2 of the 12 years, and pedigree is unavailable for another year, so that the data included in this study are for either 9 or 10 years depending on the specific analysis.

Relationship Between Longevity and Relative Male Age
As discussed above, between-male variation in adult age at any particular point in time arises only as a result of variation in the date of reaching adulthood and is then modulated by differences in adult life span. For convenience we refer to adult age (which we measure from the final moult) and adult life span as age and life span. Variation in emergence date and life span will create a correlation between age and life span within the cohort of individuals alive at any point in time. The strength of this correlation will increase with variance in both life span and emergence date. In our meadow, the adult emergence period across years lasted 21.9 ± 6.5 days (mean ± SD, N ¼ 10 years). The interquartile range of emergence date within years was 4.7 ± 1.6 days (mean ± SD, N ¼ 10 years). Average male life span including all years was 27.5 ± 18.5 days (mean ± SD, N ¼ 431, range 0e76). These distributions show that male age will vary substantially at any given point in the season (Fig. A1).
To determine the extent to which females could use a male's age to predict how long he will live, we tested the relationship between the mean relative male age of each male and his life span. This approach assumes that females choose a mate from those that are available at any given time. To calculate the relative age of a male on any particular day, we subtracted from his age the mean age of the rest of the males also alive on that day. A male's mean relative age across his life span was then the mean of these daily values across his life span. We analysed the relationship between mean relative male age and life span by first running an overall linear mixed model using lme4 (Bates, M€ achler, Bolker, & Walker, 2015), and then an independent linear regression test per year. For the mixed model, individual life span was the response variable, mean relative age a fixed effect and year a random effect. We estimated R 2 following the approach of Nakagawa and Schielzeth (2013). For the independent linear regression tests for every year, we again included individual life span as the response variable and mean relative age as the predictor. The strength of this relationship will determine how effective a female strategy of mating with the relatively oldest males is as a method of selecting the longest-lived males.

Female Preference for Males in Relation to Their Age
Females might use three alternative methods to choose mates based on age differences. If females can identify the age of a male in isolation from other males, they could employ a mating rule that makes them more, or less, likely to mate with a male that they encounter according to his absolute age (as envisaged in Kokko & Lindstr€ om, 1996). However, in an annual species where the average age of males increases throughout the season, this type of rule would mean that early in the season females would be unwilling to mate with any male, and later on, all males would be extremely attractive. This is clearly not the situation in our crickets where females begin mating as soon as they are sexually mature. A more appropriate mating rule would be that females are more, or less, likely to mate with a male according to his age relative either to her own age or to the age of other males that are also available to the female at that time. The former possibility would require females to know their own age, whereas the latter would only require females to be able to estimate a male's age relative to other males for which information is available at that time. Females will only directly encounter a small proportion of the male population. However, adult males spend a lot of their time producing calling songs audible at distances several multiples of the length of our meadow. The songs of other species of Orthoptera change as males age (Hartley & Stephen, 1989) and female bush crickets have been shown to discriminate between the songs of males that differ in age (Ritchie, Couzin, & Snedden, 1995). This suggests that females may have the opportunity to sample the male population remotely as well as through direct encounters.
We analysed female choice for each period of uninterrupted male occupancy at a burrow (i.e. if the male left the burrow and returned more than 5 min later, the two periods were considered separately). For each of these observation periods, we calculated relative male age as the difference between the age of the target male (averaged over the period of occupancy) and the median age of all adult males alive within that observation period. Only males whose age was known to within 2 days were used in the analysis.
We examined the relationship between relative male age and female mate choice based on two bivariate response variables: Paired and Mated. Paired was scored as 1 when a female was observed sharing a burrow with a male at any time within an observation period and as 0 when the male was on his own over the whole period. Mated is a subset of the pairing variable, representing whether a mating occurred during a period of pairing (scored as 1) or not (scored as 0). In our study population, sexual activity starts a few days after adult emergence. To remove this latency period, our analyses include only crickets older than 5 days (with age measured from the day of adult emergence). Over the entire period of our study, sexually active males and females were monitored in total for about 150 000 h and 178 000 h, respectively. The portion of time that individuals were unpaired was very similar in both sexes: 73% for males and 72% for females. This provides ample latitude for females to express a choice of pairing with available males or remaining unpaired.
We analysed female preferences by running mixed models for Paired and Mated using lme4 (Bates et al., 2015). For the analyses of Paired, whenever more than one female was seen with a male over the same observation period, we included an extra record per female for that observation period. Crickets often visit the same burrows many times; this increases the likelihood that two individuals meet repeatedly at the same burrow just because that burrow is within their shared home range. To avoid the potential for pseudoreplication created by this spatial effect, we only included the first observation period of each pair at any particular burrow in our analysis. For the analysis of Mated, we included all periods when a pair were seen together, regardless of whether it was their first encounter at any given burrow or not. Similar to the Paired analysis, when a male mated with more than one female over the same observation period, we included an extra record per female in the data set for that observation period. We included relative male age and the duration of the observation period as fixed effects, and male identity, burrow identity and year as random effects.

Genetic Analyses and Parentage Assignment
We performed genetic profiling with microsatellite loci to conduct parentage analysis. Previous genotyping in the Wild-Crickets project used 14 loci, and the details of DNA extraction and genotyping are described elsewhere (Bretman, Rodriguez-Munoz, Walling, Slate, & Tregenza, 2011;Rodríguez-Muñoz et al., 2010). These loci were originally developed in a related species, G. bimaculatus, and a number of them had null alleles segregating at high frequency in G. campestris (Gbim21) or alleles that were sex linked (Gbim59, Gbim71), making them unsuitable for parentage analysis. Therefore, we retained only 11 of the loci in this study (Gbim4, Gbim15, Gbim26, Gbim29, Gbim33, Gbim49, Gbim52, Gbim53, Gbim57, Gbim66 and Gbim72). Furthermore, for cohorts 2010 onwards loci Gbim26, Gbim29, Gbim33, Gbim53, Gbim66 and Gbim57 were replaced by a set of 10 new loci, identified from RNAseq data generated from G. campestris individuals from the WildCrickets meadow. These loci, termed L2590, L5292, L6980, L8550, L9077, L9737, L26448, L26654, L29153 and L30428, are all autosomal and are not segregating for high-frequency (>0.1) null alleles. Details of microsatellite discovery and genotyping are provided in the Appendix. Genotyping was performed on an ABI3730 capillary sequencer, using standard protocols (Ball et al., 2010) and scoring was performed using GeneMapper v3.7 software.
Parentage analysis was performed using genotype data combined with spatial and mating data in a Bayesian framework using the MasterBayes package (Hadfield, Richardson, & Burke, 2006;Koch, Hadfield, Sefc, & Sturmbauer, 2008). This program allows the simultaneous use of both genetic and phenotypic data to infer the pedigree, improving the statistical power of pedigree reconstruction (Hadfield et al., 2006). We used three sources of phenotypic information from our video and direct observations, all relating to the likelihood that a male and a female mated to produce offspring observed in the following year: (1) the Euclidean distances between the average location of a male and a female during a breeding season; (2) whether or not a male and a female were ever observed together ('paired'); and (3) whether or not a male and a female were ever observed to mate.
When using genetic data to infer parentage it is important to consider genotyping error rate (Hadfield et al., 2006;Marshall, Slate, Kruuk, & Pemberton, 2003;Wang, 2004). The majority of pedigree inference programs allow an estimate of error rate to be included in the model. However, MasterBayes allows simultaneous estimation of the genotype of an individual, the pedigree and the relationship between phenotypic data and the pedigree. This allows the genotyping error rate to be estimated rather than assumed and should increase statistical power to reconstruct the pedigree (Hadfield et al., 2006;Koch et al., 2008).
This comes at the cost of a considerable increase in the computational complexity of the models to be estimated (Koch et al., 2008). To accommodate this, we estimated the pedigree on a year by year basis rather than as a single run (single year runs took ca. 30 days running as the only process on a desktop with a 2.0 GHz intel Xeon processor). This is appropriate here as field crickets are annual and thus generations are not overlapping. Finally, the size of the unsampled population can have a considerable influence on the power of pedigree reconstruction and this parameter can also be estimated simultaneously with the pedigree in MasterBayes (Koch et al., 2008). Thus, in assigning parentage we simultaneously considered genetic and phenotypic data and estimated the genotyping error rate and the unsampled population size.
The exact details of the models run varied from year to year (Table A1), but all models were run for long enough and with a large enough interval between samples to minimize autocorrelation in parameter estimates, assessed from plots of model outputs and ensuring autocorrelation between successive samples was low (<0.1). Phenotypic information was considered for all years apart from 2015 as these data have not yet been collated. Genotypes and genotyping error rates were estimated for all models apart from the models estimating parentage for 2013 and 2014. For these years, the models estimating genotypes showed poor mixing (high autocorrelation between successive saved samples of estimates of the phenotypic effects in the model; Table A2). This possibly results from assignment rates in these 2 years being very high; the estimated unsampled population size of male parents was 0.116 (95% confidence interval, CI ¼ 0.0014e1.61) in 2013 and 0.356 (95% CI ¼ 0.0028e2.50) in 2014. As a result, any update to the genotype of an individual during model estimation resulted in a worse fit to the data, constraining the parameter space available for the model to sample. We therefore ran the models for 2013 and 2014 assuming fixed genotypic error rates per locus. A thousand draws of the pedigree were saved for each year and the modal paternity and maternity assignment for each individual were used to assign parentage. Estimates of genotyping error rates for each microsatellite marker are given in Tables A3 and A4.

Effect of Male Age on Offspring Production and Fitness
Female crickets mate and lay eggs throughout their adult lives. They usually mate with more than one male, with separate periods of pairing and mating with a particular partner. To assess the effect of male age on reproductive success, we tested whether a pair's likelihood of producing adult offspring was related to the relative age of the sire over the period when the pair were observed to mate. This approach of considering each temporary pairing that a male engages in during his life is necessary in order to track the offspring that he may sire in relation to his age. For pairs that mated repeatedly over a period of several days (often including several paired observation periods), we calculated mean relative sire age over the period between the first and the last observed matings. We assessed the possibility that age has a quadratic rather than a linear relationship to adult offspring production by introducing a quadratic term in the analysis. Also, because field cricket females are usually polyandrous, sperm competition can influence fertilization success. Evidence from the closely related G. bimaculatus indicates that there is no sperm precedence in relation to mating order in field crickets (Bretman, Newcombe, & Tregenza, 2009;Bretman, Wedell, & Tregenza, 2004). Therefore, we accounted for this effect by including a term that weighted the number of matings of a particular pair relative to the total number of matings observed for the female of that pair. Each pair provided a bivariate variable indicating whether they contributed any offspring to the following generation (1) or not (0). Because females that did not have any offspring are uninformative in relation to the relative success of the males with whom they mate, we only included pairs where the female produced at least one offspring. We ran this analysis by building a full model including the proportion of matings for that pair (PropMatings) and the mean sire age (both as a linear and a quadratic term) as fixed effects, and sire identity and year as random effects. We then compared the full model with three simpler models that excluded, in this order, the quadratic sire age term, the proportion of matings and the year. We based the comparison on the difference in the Akaike information criterion (AIC) values among models, with differences smaller than seven indicating a similar model fit (Burnham, Anderson, & Huyvaert, 2011).
To explore potential long-term effects of male age on offspring fitness, we repeated this analysis but using the production (or not) of grand-offspring as the response variable. We restricted our analysis to only those pairs that produced adult offspring and used the model with the best fit in the previous analysis.

Ethical Note
The crickets used in this study are removed from the meadow for a period of a maximum of a few hours during which time we take a small haemolymph sample and attach a plastic tag by gluing it to the pronotum. Observations of individuals immediately after these procedures indicate that they exhibit normal behaviours within a few minutes of being released, and as far as we know, crickets have never died as a result of any of the procedures used in this study. Our tagged crickets live out their natural lives in the meadow.

Relationship Between Longevity and Relative Male Age
We found a positive relationship between relative male age and longevity (t ¼ 5.88, P < 0.001, N ¼ 431), but the effect was small, with a marginal R 2 of 0.064 (Fig. 1). The independent least squares regressions within years were nonsignificant in most cases, with R 2 values ranging from 0 to 0.262, with a mean of 0.091 (Table 1).

Female Preference for Older Males
Relatively older males attracted more females, but once paired they were less likely to mate. Both relative male age and the duration of the observation period had a positive relationship with the probability of sharing a burrow with a female (Fig. 2a, Table 2). In contrast, although the length of time a pair stayed together was positively associated with their probability of mating, there was a significant negative relationship between relative male age and the probability of mating at least once (Fig. 2b, Table 3). Among the random effects affecting the probability of being paired, burrow identity explained over seven times more variance than male identity, while year had a very small effect (Table 2). However, the effect of burrow identity on the probability of mating once a pair met decreased to about half of the effect of male identity (Table 3).

Parentage Assignment
The phenotypic data included in the pedigree reconstruction models were generally informative. In the final model, whether a male and female were observed together or not positively influenced their probability of producing offspring together across all years (Table A5). If a male and female were also observed to mate this additionally increased their probability of producing offspring together, although this increase was less than having been seen together in the first place (Table A5). The mean Euclidean distance between a male and a female's burrow also predicted the likelihood of these two individuals producing offspring together, with closer The bias in favour of males of younger relative ages (more than half the points are relative ages <0) reflects the fact that younger males are more numerous in the population because older males have to be young males earlier in their lives, but young males do not all go on to be old males. The age of a male, relative to the other males alive at any point in time, is only a very weak predictor of his longevity. neighbours being more likely to have offspring in all years apart from 2007. Using the modal parentage assignment for each individual, maternity was assigned to a sampled individual in the population for 1267 of 1465 individuals (86% of the population) and paternity to a sampled individual for 1380 individuals (94% of the population). The median confidence of maternity assignments to known individuals was 98.7% (interquartile range, IQR ¼ 30.4; Fig. A2) and in paternity assignments was 98.9% (IQR ¼ 32.1; Fig. A2). A number of individuals were assigned an unsampled parent as the most likely parent (Table A5, Fig. A2), with the size of the unsampled population varying across years, but generally being low.

Effect of Male Age on Offspring Production and Fitness
The best fitting model included the proportion of matings a male got from his partner, but not the quadratic term of relative male age (Table A6). Having a higher proportion of matings with a female increased the probability of producing offspring but being relatively older did not (Table 4). Offspring fitness (the probability that a sire had grand-offspring) was also unrelated to relative male age (Table 5). For offspring fitness, year explained much more variance than male identity because of large differences in population size between years (Table 5).

DISCUSSION
Exactly how females choose among potential mates is a challenging question in its own right. The females in our meadow physically encounter a maximum of a few males a day, but may hear the calls of every singing male in the population. We therefore characterized females as choosing their mates from among all the males alive at any particular time. When we examined the cohorts of males alive each day, we found an overall positive relationship between the longevity of each male and his age relative to the other males in his cohort. However, this relationship was very weak; in 6 of 10 years there was no statistically significant relationship at all (Table 1), and the average proportion of the variance in male longevity predicted by relative age was 0.091. This number represents an absolute upper limit on the power of female choice to select for longevity since it assumes that females can accurately determine male age and can choose the oldest male from all the males alive on any day. This assumption will very rarely be met in any system. In our meadow, females can gather information about males through their calls, but they can only actually mate with  Results of a mixed model using the lme4 R package with a binomial error distribution. The model includes relative male age and the duration of the monitored period as fixed effects, and year, burrow and male identities as random effects. Coefficients with significant P values are highlighted in bold.

Table 3
Analysis of the relationship between relative male age (as compared to the mean age of the other males in the population) and the probability of mating at least once within each period when a male was monitored while sharing a burrow with a female Results of a mixed model using the lme4 R package with a binomial error distribution. The model includes relative male age and the duration of the observation period as fixed effects, and year, burrow and male identities as random effects. Coefficients with significant P values are highlighted in bold. males that they encounter. Our observations reveal that only a small proportion of individuals ever actually encounter one another (Fisher, Rodríguez-Muñoz, & Tregenza, 2016a;2016b), placing severe constraints on the capacity of females to select long-lived males based on their relative age. The extent to which female crickets are able to detect male age is unknown; Verburgt et al. (2011) reviewed studies of the effect of age on calling song in crickets, finding evidence of changes in at least one calling song parameter with age in four of 14 studies. These included Jacot et al.'s (2007) study of G. campestris, which identified a difference in carrier frequency of just under 5% between 8-day-old and 30-day-old males, with 19 of 32 males showing a decrease in carrier frequency and eight of 32 males showing an increase. Our population of crickets would appear to be a fairly typical annual insect in relation to the parameters we investigated. Most annual insects have a breeding period lasting for a few weeks or months and females may encounter hundreds of males in their lifetimes, but would very rarely have the opportunity to compare the ages of a large number of males. Given the multiplication of all these factors, it is difficult to avoid the conclusion that female mating preferences are unlikely to exert significant selection in relation to male age in wild insects.
We found that on any given day, older males in the population were more likely to be sharing their burrow with a female than were younger males. Since it appears that male age cannot be used by females to estimate longevity, it seems likely that this effect relates to some other characteristic of males which correlates with their relative age. There is an obvious temptation to speculate that this may be a general reflection of the male's quality. However, such a conclusion is tempered by our finding of a negative relationship between a male's age and the probability that he actually mates with the female with whom the burrow is shared (Fig. 2b, Table 3), which is more consistent with a reproductive advantage for younger males. Our Bayesian parentage assignment approach found that, when burrow sharing is taken into account, adding whether or not a pair actually mated to the hierarchical model only marginally improved the fit. However, this probably reflects the fact that pairs that share a burrow generally mate, and that when the number of burrows was larger than the number of cameras, we could collect data only on whether a pair were together at a burrow and not whether they had mated.
Burrow identity was a strikingly better predictor of whether a male was paired or not than was his own identity. Burrows were monitored at a similar rate irrespective of their spatial location (and matings are very clearly identifiable) so this finding is not an artefact of differences in our capacity to observe burrows. The large effects of burrow on pairing success suggest that burrow characteristics could be a more important determinant of reproductive success than some aspects of the phenotype including a male's age. This corresponds with Niemel€ a and Dingemanse's (2017) observation that burrow identity has a large effect on male behaviour. We can only speculate about the burrow features that females might be attracted to, with candidates including spatial location, aspect, physical dimensions and the appearance of the burrow entrance.
We found no relationship between how old a male was when he mated with a particular female and the number of offspring or grand-offspring he contributed to subsequent generations. These analyses were less powerful than others in our study: only 143 males contributed data to the analysis of number of direct offspring and only 89 to the analysis counting grand-offspring. Nevertheless, the fact that there was no hint whatsoever of an effect in either direction suggests that the age of a male when a female mates with him does not have a strong effect on his success in fertilization or the subsequent survival and reproduction of his offspring.
Overall, our observation of relationships between male age and measures of pairing and mating do not provide convincing evidence of female preference for older males in our system. This is presumably why no evidence for increased reproductive success of older males in terms of offspring was apparent. Our study demonstrates that although, in theory, females could gain information about a male's life span from his relative age, in natural insect populations, particularly of annual species, it seems unlikely that a sufficiently reliable signal of longevity is available to females. It is possible that the preferences of females for older males identified in a small proportion of laboratory studies of crickets (Verburgt et al., 2011) reflect the slightly greater correlation between age and longevity expected in multivoltine species. However, this increase (over the very weak correlation we observed) is not expected to be large, suggesting that nonadaptive explanations for female preferences for older males deserve consideration.

Data Availability
All data are available from the authors upon request, including a fasta file of all the contigs used in microsatellite discovery.

Acknowledgments
We thank L. Rodríguez and M. C. Muñoz for unconditional support, providing access to facilities including the Wild-Crickets study meadow. The following people contributed to video processing and data recording: Liu Xingping, Thor Veen, Carlos Rodríguez del Valle, Alan Rees, Hannah Hudson, Sophie Haugland Pedersen, Geetanjali Mishra, Jasmine Jenkin, Lauren Table 4 Analysis of the relationship between relative male age (as compared to the mean age of the other males in the population) and the probability of having at least one offspring Results of a mixed model using the lme4 R package with a binomial error distribution. The model includes relative male age and the proportion of matings of the target male (from among the total number of matings of his partner) as fixed effects, and year and male identity as random effects. Coefficients with significant P values are highlighted in bold.

Table 5
Analysis of the relationship between relative male age (as compared to the mean age of the other males in the population) and the probability of having at least one grand-offspring Results of a mixed model using the lme4 R package with a binomial error distribution. The model includes relative male age as a fixed effect, and year and male identity as random effects.
Microsatellite repeats were detected using a modified version of Sputnik (http://wheat.pw.usda.gov/ITMI/EST-SSR/ LaRota). Sputnik parameter settings were e u 2 e v 5 e s 20 e p e L 20, meaning that motifs between 2 and 5 base pairs were reported, provided they were at least 20 bp long and had a score of at least 20. Sequences with a repeat purity of at least 90% were retained. This process yielded a total of 219 unique microsatellite loci (72 dinucleotides, 93 trinucleotides, 35 tetranucleotides and 19 pentanucleotides).
PCR primers were designed for 27 loci, using Primer 3 (Rozen & Skaletsky, 2003). PCR products were tested in a panel of 24 crickets. Loci that failed to produce PCR products, that appeared to be monomorphic in the test panel or were hard to score reliably were excluded from downstream analyses. A set of 10 loci were retained and scored in all cohorts. The microsatellites in the new panel of loci (Table A7) were labelled Lxxxx where xxxx represents the contig that the microsatellite was mined from. A fasta file of all the contigs is available on request.
Each PCR reaction contained 1 ml of air-dried DNA at a concentration of approximately 10 ng/ml, 1 ml of primer mix (fluorescently labelled forward and reverse) at 0.2 mM and 1 ml of QIAGEN multiplex PCR mix (QIAGEN Inc, Manchester, U.K.). The PCR program had an initial denaturation step at 95 C for 15 min, followed by 45 cycles of 30 s at 94 C, 90 s at 57 C and 60 s at 72 C, followed by a final extension step at 60 C for 30 min. PCR products were diluted 1:4000 with doubledeionized water, before resolving them on an ABI 3730 48well capillary DNA Analyser (Applied Biosystems, Foster City, CA, U.S.A.). Allele scoring was performed using GeneMapper v3.7 (Applied Biosystems, CA, U.S.A.). The distribution of microsatellite allele lengths was examined using the R package Msatallele (Alberto, 2009), to ensure that binning of estimated fragment lengths (given to two decimal places) to different alleles (scored as integers) was consistent. All individuals that possessed alleles that differed by 1 bp were genotyped at least twice, along with additional control individuals of known genotype, to check for allele sizing errors. Observed and expected heterozygosities were Data are presented from runs assigning parentage to individuals from 2013 and 2014 either without (left) or with (right) genotyping error rates being estimated. The autocorrelation between successive samples when estimating genotyping error rate is considerably higher than when not estimating error rate, suggesting models estimating error rate are not mixing well. All analyses in the paper are based on models that do not estimate genotyping error rates for these two 2 years only. calculated and null allele frequencies estimated using CERVUS v3.0.3 (Kalinowski, Taper, & Marshall, 2007), and deviations from HardyeWeinberg Equilibrium (HWE) and linkage equilibrium were assessed with GENEPOP v4.0.10 (Rousset, 2008;Raymond & Rousset, 1995; Table A8).

Table A5
Parameter estimates for the pedigree reconstruction models  As described in the Methods, this was done for years 2007e2009, 2011e2012 and 2015. The markers used in 2007e2009 differ from those used in the other years. Estimates are posterior modes and their 95% confidence intervals (95% CI). Genotyping error rates are estimated as two parameters: E1, the probability of heterozygote dropout, and E2, the probability of alleles being mis-scored. E1 is an estimate of the probability of heterozygote dropout, and E2 is an estimate of the probability of alleles being mis-scored.

Table A6
Model comparison for the analysis of the relationship between relative male age (as compared with the mean male age in the population) and offspring production The full model includes the proportion of matings by the male from among those recorded for the female whose offspring production is considered (PropMating), the mean relative male age over the period when he was seen mating with that female both as a linear (SireAge) and a quadratic (SireAge 2 ) term as fixed effects, and male identity (ID) and year (Year) as random effects. The