Henrik Jensen, Department of Biology, Realfagbygget, Norwegian University of Science and Technology, N-7491 Trondheim, Norway. Tel: + 47 73596949; Fax: + 47 73591309; E-mail: email@example.com
1In this study we relate variation in lifetime reproductive success (LRS) of male and female house sparrows Passer domesticus to morphological characteristics.
2Our analyses demonstrated no sex-specific difference in the distribution of LRS. The variance in LRS was influenced mainly by variation in individual annual reproductive success, and to a lesser extent by variation in individual lifespan.
3Phenotypic traits explained a significant proportion of the variation in LRS in males, but not in females. The effect of male morphology on LRS operated mainly through an effect on the number of recruiting daughters.
4The size of the patch of black feathers on the chest of males (badge size) and male bill length were both positively associated with LRS. Lifespan and bill length were positively related and reproductive success increased with badge size. In females, number of recruiting daughters was positively related to bill length, body mass and body condition index due to the positive effect of these traits on annual production of daughters.
5These results indicate that identifying factors causing the large individual variation in LRS, which is likely to be closely related to fitness, will be important to understand microevolutionary processes in this metapopulation, and hence their demographic feedbacks.
Lifetime reproductive success (LRS) of an individual is defined as the number of recruits to the following generation that the individual produces over its entire lifespan (Clutton-Brock 1988; Newton 1989a). A general pattern is that the distribution of LRS is often very skewed (see reviews in Clutton-Brock 1988; Newton 1989a). This means that most breeding individuals do not succeed in producing a recruit. As a consequence, only a small proportion of the breeding population contribute to future generations. Because LRS in natural populations is generally assumed to be a relatively good estimate of fitness (Grafen 1988; Newton 1989b; but see Benton & Grant 2000), identifying phenotypic characteristics of those successful individuals will provide important insight into the evolutionary processes in the population.
Basically, LRS consists of two components (Barrowclough & Rockwell 1993). One is the average number of recruits an individual produces in each of its years as an adult. The second is the number of reproductive seasons which is, of course, related closely to the adult life expectancy. Thus, a long adult lifespan and high annual reproductive success can both contribute to a high LRS. In general, life expectancy is an important determinant of LRS in birds (Gustafsson 1986; Newton 1989a; Grant & Grant 2000; Merilä & Sheldon 2000; Krüger & Lindström 2001) as well as in other taxa (Clutton-Brock 1988; Wauters & Dhondt 1995; Bérubé, Festa-Bianchet & Jorgenson 1999; Kruuk et al. 2000). However, the relative contribution of individual variation in adult survival rate to differences in LRS seems to differ among species. For instance, in splendid fairy-wren (Malurus splendens) lifespan accounted for 44 and 66% of the variation in male and female LRS, respectively (Rowley & Russel 1989). On the other hand, lifespan accounted for only about 5% of the variation in male and female LRS in red-billed gulls (Larus novaehollandiae scopulinus) (Mills 1989). This suggests that we must understand how a character affects survival as well as reproductive success to understand fully its effects on LRS.
Several studies have indicated large intra- and interspecific sexual differences in the distribution of LRS. Such variation may be related to, for example, different degrees of polygamy (Orians & Beletsky 1989; Weatherhead & Boag 1997). However, if the level of polygamy is low, and survival equal in males and females, the variance in LRS of both sexes is expected to be similar (Newton 1989b) as found in, for example, pied flycatchers (F. hypoleuca) (Sternberg 1989).
Trivers & Willard (1973) suggested that when males have higher variance in LRS than females, it would pay for high-quality individuals to produce a high-quality son in preference to producing a high-quality daughter, because a high-quality son gives higher fitness returns. In contrast, poor-quality individuals should produce more daughters. Such differential investment in sons and daughters may occur both before and after birth (e.g. Maynard Smith 1980; Hewison & Gaillard 1999). Accordingly, there are indications that males of a number of animal species need a higher amount of resources to survive or grow (Clutton-Brock, Albon & Guinness 1985; Griffiths 1992). As an alternative to the theory of Trivers & Willard (1973), high-quality individuals are expected to produce more surviving sons because their average level of provisioning is higher (Clutton-Brock 1991), and not as a consequence of differential investment in sons and daughters per se among high- and poor-quality individuals. Both theories predict that individual quality should be related positively to number of sons they produce, but the second theory also predicts a positive relationship with number of daughters.
It has generally been quite difficult to measure LRS in natural populations. First, long-term studies are often required to follow a sufficient number of individuals throughout their lives (Newton 1989a). Secondly, having recorded the lifespan of a sufficient number of individuals, it is often difficult to measure the number of recruits an individual contributes to the population, both because it may be difficult to determine parentage of recruits (Newton 1989a) and because young individuals may emigrate and recruit into populations outside the study area (Clarke, Sæther & Røskaft 1997; Lambrechts et al. 1999). In our study metapopulation of house sparrows (Passer domesticus L.) off the coast of northern Norway these problems are negligible, because a large proportion of all individuals on the study islands have been individually marked (Ringsby et al. 1999) and genotyped. Thus, we could identify the genetic parents of recruits through DNA analyses (Ringsby et al. 1999; Jensen et al. 2003). Moreover, the main study islands were surrounded by an archipelago of 13 other islands where a considerable proportion of the house sparrows had also been individually banded. Consequently, both birds that recruited on their natal island and birds that emigrated from one of the main study islands before recruitment had a high probability of being recorded (see also Altwegg, Ringsby & Sæther 2000).
The purpose of the present study was to examine the following questions.
1How is the individual variation in LRS of the two sexes related to differences in annual reproductive success and life expectancy?
2Does variation in LRS relate to morphological characteristics of males and females?
3Do morphological characteristics of males and females affect lifetime production of sons and daughters differently?
This study was carried out from 1993 to 2001 in an island metapopulation of house sparrows off the coast of northern Norway (66°N 13°E; see map in Ringsby et al. 1999). Five islands constitute our main study area. In addition, we observed and captured birds on 13 neighbouring islands in the archipelago covering an area of more than 1600 km2 every summer and also searched for emigrants from our study area several 10 km along the coastline on the mainland (see also Altwegg et al. 2000). Further description of the study area is given in Ringsby et al. (1999, 2002) and Sæther et al. (1999).
More than 70% of all birds on our five main study islands were marked individually with a numbered metal ring and coloured plastic rings from 1993 to 2001. In addition, more than 40% of the birds on the 13 surrounding islands in the archipelago were marked individually in most years (i.e. in 75% of all island-years). Birds were marked either as chicks in nests just prior to fledging, as fledged juveniles, or as adults. During the summer field season (from beginning of May to middle of August), all birds outside nests could be classified as either juvenile or adult. Some of the birds caught during the autumn field-season (from the end of September to the beginning of November) were, however, difficult to classify. This applied to 26% of the birds (103 of 391) caught for the first time during the autumn. However, because of the high proportion of marked adult birds in the populations, these individuals were assumed to be juvenile.
Each time an adult bird was caught we used slide callipers to measure tarsus length, bill depth and bill length to the nearest 0·1 mm, and a ruler to measure wing length to the nearest mm (Method 3, Svensson 1992). Body mass was measured to the nearest 0·1 g with a Pesola spring balance. Slide callipers were also used to measure length and width (to the nearest mm) of the black throat badge of adult males. The size of this black badge (i.e. badge size) was defined as the area covered by black feathers and feathers with black bases and grey tips on the throat and chest when the bird was held with its bill pointing at a right angle to its body. Badge size was then calculated according to Møller (1987)
Badge size = 166·67 + (0·45 × x × z),(1)
where x is badge length and z is badge width.
Only measurements taken during the adult life of each individual were used. Moreover, to avoid any seasonal variation in morphological traits (e.g. Summers-Smith 1988; Cramp & Perrins 1994), only measurements taken from January to August were used for wing length, whereas only measurements taken during the summer (i.e. from May to August) were used for body mass, bill depth, bill length and badge size. To standardize measurements taken by each fieldworker, each new fieldworker was first taught how to measure the different traits. After the initial training period, the general relationship between measurements of T. H. Ringsby and the new fieldworker were estimated in a linear regression model that included at least 20 birds measured independently by the two. All measurements taken by each fieldworker were then corrected by the respective equations from these linear regressions. The repeatability of male badge size was 0·31 [± 0·049 (1 SE)]. For the other traits, the repeatability was relatively similar across sexes and ranged from 0·53 (body mass) to 0·95 (tarsus length) in males, and from 0·29 (body mass) to 0·92 (tarsus length) in females (see Jensen et al. 2003). Body condition index was calculated as the unstandardized residual form a regression of individual body mass upon tarsus length, in a model that also included sex as a fixed factor and the interaction between sex and tarsus length (model: F = 52·579, d.f. = 3, P < 0·001; sex: P = 0·99; tarsus length: P < 0·001; sex × tarsus length: P > 0·7; n = 685). In our study population data suggest that wing length of both sexes, bill length of females and badge size of males increase with age. Thus, we standardized these measurements in our analyses by running generalized linear models (GLM; SPSS Inc. 1997) that included individual number as fixed factor, age and age2 as covariates, and the morphological trait as response variable. There were significant effects of age on wing length of both males (model: F = 5·060, d.f. = 188, P < 0·001; age: P < 0·001; age2: P < 0·001, n = 381) and females (model: F = 6·986, d.f. = 172, P < 0·001; age: P < 0·001; age2: P = 0·015, n = 391), on bill length in females (model: F = 5·590, d.f. = 164, P < 0·001; age: P < 0·001; age2: P = 0·005, n = 365) and on badge size in males (model: F = 2·000, d.f. = 181, P < 0·001; age: P = 0·002; age2: P = 0·080, n = 358). Each individual was given a value for the trait by summing up its intercept, the intercept of the total model and the parameter estimates of age and age2, thus standardizing its trait size to size at age one. For each of the other traits we used the average of all measurements on the individual taken as an adult.
estimation of reproductive success
The parentage of the offspring was determined by genetic analyses of DNA extracted from a small blood sample taken from each bird at first handling. Eight highly polymorphic microsatellite loci were used in the genotyping procedure (see Jensen et al. 2003 for further details on the genotyping procedures). The parenthood analysis software cervus 2·0 (Marshall et al. 1998) was used to determine the mother and father of birds hatched on the islands during the breeding seasons from 1994 to 1999. An adult bird captured or observed on an island in a given year was classified as a potential parent on the island that year. Birds not observed in a given year, but that had hatched or been captured a previous year, and were captured or observed in a later year, were also classified as potential parents. In addition to this information, cervus used information on the estimated proportion of adult birds on an island in a given year that was marked, as well as information on allele frequencies at the eight microsatellite loci. For each fledgling or juvenile on the island, the program then calculated a LOD-score (log-likelihood ratio) for every potential parent. These scores were compared with a criterion generated in the simulation module of cervus, resulting in an assigned maternity or paternity that was correct in 95% of cases (Marshall et al. 1998). Each island was analysed separately, and maternity was determined before we ran analyses to determine paternity (see Jensen et al. 2003 for further description of the parenthood-assignment procedures).
In our analyses we examined only correlates with reproductive success of adult birds from the four cohorts (i.e. that hatched in the years) 1993 (n = 16 males; n = 21 females), 1994 (n = 35 males; n = 25 females), 1995 (n = 15 males; n = 12 females) and 1996 (n = 18 males; n = 23 females). A total of 10 males and seven females from the four cohorts were still alive in 2000. We included these birds in our analyses because we were able to account for the fact that these birds could also have produced recruits after the breeding season in 1999 (see below). The lifespan of a bird was defined as the number of years from hatching to the last year it was either observed or recaptured, as this year was assumed to be the last it was alive. Moreover, the lifespan of the 17 birds that still were alive in 2000 was defined as the number of years from hatching to 1999, but their status was defined as ‘censored’ and thus accounted for in the models where we examined factors explaining variation in lifespan (see below). In reality, the lifespan of these birds was at least 4 years, but could have been more than 8 years for some of them, because one of the males and one of the females hatched in the 1993 cohort were still alive in 2001. Moreover, because of the high recapture rate demonstrated previously in this metapopulation (Ringsby et al. 1999), there was only about a 17% chance that a bird alive in a given year was not observed or recaptured during that field season. Thus, we feel confident that the potential bias introduced by not taking into account the recapture probability in our estimates is small.
A fledgling or juvenile in a given year was defined as a recruit and sexed if it was recaptured or observed on one of the main study islands, on a neighbouring island in the archipelago or on the mainland adjoining the archipelago during a field season subsequent to the year it hatched. Because of the high recapture rate in this study metapopulation (Ringsby et al. 1999) we are confident that birds that hatched in a given year, but that were not recaptured or observed during the subsequent two or more study years, were classified correctly as non-recruits. We were able to identify the mother of 76% and the father of 72% of the 461 recruits that hatched on the five main study islands from 1994 to 1999. The parenthood analyses suggested that among the recruits for whom we could not determine the parents, a marked potential mother and father may have been the parent of only 11 recruits each. This was because two or more potential parents were genetically compatible with the recruit. The remaining 21 and 26% of the recruits thus most probably had a mother or father that was not genotyped, respectively.
The LRSg of an individual can be partitioned into two components, namely lifespan (L) and average production of recruits each year (i) the individual is alive ( ) (the subscript g indicates either T: total number, M: male recruits or F: female recruits). To examine the effects of morphological traits on LRSg, we first explored the effects of these factors on lifespan (L). Secondly, we investigated how the morphological traits affected . When there is no covariance between L and , one can combine the separate effects of each morphological trait on L and to obtain an estimate of the total effect of this trait on LRSg.
If one assumes that the survival probability of individuals is constant after adult life is reached (see e.g. Ringsby et al. 1999; Loison et al. 2002; Nisbet & Cam 2002), lifespan can be expected to have an exponential distribution (Larsen & Marx 1986). This assumption can be handled by accelerated failure time models (proc lifereg; SAS Institute 1989) with an exponentially distributed response variable (i.e. lifespan). In this model, the log of the expected lifespan is modelled as a linear function of the parameters of interest,
where α represents a vector of unknown coefficients and x is a vector of covariates (i.e. explanatory variables). The procedure estimates the parameters using maximum likelihood and allows for right censoring of the data, i.e. inclusion of individuals that are alive at the end of the study period (after the 1999 breeding season in this study). This procedure allowed us to examine how morphological traits affected lifespan, and to test whether lifespan varied significantly between islands. The analyses were initiated with a general model that included island and the main effects of all the measured morphological traits. For the general model and all the models nested within it we computed a modification of Akaike's information criteria suitable for small sample sizes (AICc) to determine the final and most parsimonious model (Burnham & Anderson 1998).
We modelled the effects of morphological traits on and tested whether varied significantly between cohorts and islands as follows. Our objective was to estimate some model for the expected average annual reproductive success E( ), for instance
where β is a vector of unknown coefficients, x is a vector of covariate values (i.e. explanatory variables). However, the variance of (the mean of several stochastic variables) is likely to vary and thus violate assumptions of standard regression models (Larsen & Marx 1986). The variance of LRSg, conditional on L, in contrast, is approximately equal to L× Var( ), and thus approximately proportional to the expectation of LRSg. This suggests that it is more reasonable to work with LRSg as the response variable in a quasi-Poisson regression (proc genmod; SAS Institute 1996). If we do this and estimate a model for E(LRSg|L) on the form
where ln(L) is included as an offset term, then the estimate of the parameter vector β in eqn 3 is equal to the estimate of β′ from eqn 4 because
A model for the total effect of morphological traits x on LRSg is then
again provided that Cov( , L) = 0. This kind of count data, consisting of numbers of chicks surviving to the age of 1 year, can possibly be non-independent. Such interrelationships will tend to inflate the variance of LRSg. In addition, reproductive success in individual years need not follow a Poisson distribution. Such overdispersion was accounted for in the model by standard methods (McCullagh & Nelder 1989; SAS Institute 1996).
To select the most parsimonious model that described variation in we started with a general model including cohort, island and all morphological traits. Furthermore, we used a modification of Akaike's information criteria that accounted for overdispersion and was suitable for small sample sizes (QAICc) to discriminate between all the models nested within the general model (Burnham & Anderson 1998).
The parameter estimates from the analyses modelling factors affecting lifespan and factors affecting , can be written for each model as a linear predictor µj = β1x1 + β2x2 + … + βmxm, where βk denotes the parameter estimate of variable xk, and µj denotes the linear predictor for model j (proc lifereg; proc genmod; SAS Institute 1989, 1996). In both analyses the expectation of the response variable is modelled as . The expectation of LRSg is the expectation of L multiplied by the expectation of , provided that and L are independent conditional on the values of explanatory variables. This assumption of independence was tested by a standard correlation test applied to the standardized residuals from each regression (i.e. between most parsimonious models explaining L and , and , respectively). There was no significant correlation between those residuals in either males (for all three analyses: r < 0·15, P > 0·18, n = 84) or females (for all three analyses: r < 0·15, P > 0·18, n = 81). The combined effects of each morphological trait can then finally be obtained by summing the separate parameter estimates from the two models (i.e. the numerator in eqn 7), whereas the standard error of this estimate was estimated as the denominator in eqn 7. We calculated Z-scores to test whether the combined estimates were significantly different from zero
where xj was the parameter estimate of one of the morphological traits in the most parsimonious model explaining variation in lifespan and xk was the parameter estimate of the same trait from the most parsimonious model explaining variation in . Furthermore, σj and σk were the respective standard errors. Thus, a morphological trait that was not included in the most parsimonious model was given a parameter estimate and an estimate of its standard error equal to zero. To test for significance the resulting Z-scores were tested against a large sample standard normal distribution.
Because of the strong positive correlation between body mass and body condition index (see Table 1), all analyses were run with either body mass or body condition index in the models. These analyses gave similar results. Thus, for the other morphological traits we present only results from models with body mass included. All statistical tests were two-tailed, and parameter estimates are given ± 1 standard error.
Table 1. Correlations between morphological traits in male (below diagonal, n = 84) and female (above diagonal, n = 81) house sparrows. See Methods for description of how traits were standardized and for how structural body size and body condtion index were estimated
Large individual variation was found in the LRSg of both males and females (Fig. 1). The maximum number of total recruits by a male was six, whereas the corresponding number for a female was eight recruits. The median LRST was one in both sexes. Although females had higher maximum number of recruiting offspring during their lifetime (Fig. 1) the variance in LRST was large, but similar in the two sexes (coefficients of variation: 1·24 in males and 1·23 in females). The variation in LRSM and LRSF was also large in both sexes (Fig. 1, coefficients of variation: LRSM: 1·35 and 1·45; LRSF: 1·66 and 1·49, in males and females, respectively). Considering only male recruits, the maximum number of recruits was four and five for male and female parents, respectively. Furthermore, the most successful male was able to recruit five females, one less than the most successful female. The median number of sons and daughters was zero in both sexes.
components of lrsg
LRSg was positively related to lifespan in both sexes (Figs 2 and 3). Moreover, the relationship between LRSg did not deviate significantly from proportionality (see Methods). To examine the relative amount of variation in LRSg explained by variation in lifespan and average annual reproductive success, we ran separate linear regressions of LRSg on each of its components and recorded the amount of variance explained (r2). Regressions indicated that variation in average annual reproductive success explained most of the variation in LRSg, and that it generally explained twice as much as variation in lifespan (Table 2).
Table 2. Approximate proportion of variance in lifetime reproductive success (LRS) accounted for by lifespan and annual reproductive success in male and female house sparrows. LRST denotes total lifetime reproduction, LRSM and LRSF denotes lifetime production of sons and daughters, respectively
lrst in relation to morphology
LRST of males was positively related to both bill length and badge size (Table 3, Fig. 4). The effects of these traits on LRST were associated with a positive effect of bill length on lifespan (β = 0·64 ± 0·27, χ2 = 5·74, d.f. = 1,77, P = 0·017, Fig. 4a), and a positive effect of badge size on (β = 0·0058 ± 0·0028, χ2 = 4·18, d.f. = 1,82, P = 0·041, Fig. 4b). Furthermore, there was no significant temporal variation in LRST, but a tendency for LRST of males to differ among islands (Table 3). The tendency for LRST to vary among islands was due to the variation in male lifespan among islands (χ2 = 9·12, d.f. = 4,77, P = 0·058; see also Ringsby et al. 1999). The most parsimonious model explaining variation in male lifespan included a non-significant negative association between tarsus length and lifespan (β = −0·32 ± 0·17, χ2 = 3·32, d.f. = 1,77, P = 0·069). Thus, there was a tendency for LRST to be influenced by tarsus length as well (Table 3).
Table 3. Spatial variation and effect of morphological traits on lifetime production of recruits in house sparrows. LRST denotes total lifetime reproduction, LRSM and LRSF denote lifetime production of recruiting sons and daughters, respectively. The estimates are from a combination of the most parsimonious models explaining variation in lifespan and yearly production of recruits (total, males and females) within each sex. The general models included island, cohort (in models explaining variation in annual production of recruits), tarsus length, wing length, bill length, bill depth, body mass (or body condition index), and badge size (for males). Model selection procedures based on AIC were used to select the most parsimonious models (see Methods for further description of procedures)
To quantify the effect of large badge size on production of recruits we compared the lifetime production of recruits of the 20% of males with largest and smallest badges, respectively (n = 17 in each group). The results from this comparison demonstrated that the large-badge males on average fathered 2·29 recruits (median = 2), whereas the small-badge males on average fathered 1·06 recruits (median = 0). Thus, males with a large badge fathered significantly more recruits than males with a small badge (Mann–Whitney U-test: P = 0·041).
Our analyses demonstrated that no morphological traits in females were significantly associated with either lifespan or . Consequently, none of the morphological traits affected LRST of females. Moreover, there was no significant spatial or temporal variation in LRST of females.
To investigate whether the effects of morphology that we demonstrated in males were significantly different from the lack of effects in females we tested for significant interactions between sex and the morphological traits. The most parsimonious model for effects of morphology on lifespan in males was used as the basis for the analysis. In addition, we included the main effect of sex and the interactions between sex and tarsus length and bill length. The analysis showed that the sex by trait interactions were non-significant (sex × tarsus length: P = 0·26; sex × bill length: P = 0·24). Both interactions were also non-significant when the other interaction was excluded from the model (both interactions: P > 0·3).
lrsm and lrsf in relation to morphology in males and females
The only morphological trait that significantly affected number of male recruits (i.e. LRSM) in males was bill length (Table 3), which affected LRSM through its positive effect on male lifespan (see above). Similarly, production of daughters (LRSF) was also positively related to a positive association of bill length with male lifespan. Furthermore, increased with badge size (β = 0·0085 ± 0·0036, χ2 = 5·74, d.f. = 1,82, P = 0·020). As expected from this, LRSF was also positively related to badge size (Table 3). The relationship between badge size and was, however, not significant (model with badge size only: β = 0·0032 ± 0·0032, χ2 = 0·96, d.f. = 1,82, P = 0·33). Thus, badge size was excluded from the final model explaining variation in and LRSM (see Table 3). Moreover, there was a tendency for both LRSM and LRSF to differ among islands, related to a tendency for an inter-island variation in male lifespan (see above). Finally, both LRSM and LRSF tended to decrease with male tarsus length, due to a negative but non-significant relationship between tarsus length and male lifespan (see above).
LRSM of females did not vary either spatially or temporally, neither was it affected by female morphology (Table 3). On the other hand, LRSF differed among islands (through an effect on : χ2 = 13·16, d.f. = 4,74, P = 0·011), and was significantly positively related to both female bill length, body mass and body condition index (Table 3). This variation in LRSF was related to a positive effect of each trait on (bill length: β = 0·67 ± 0·29, χ2 = 5·27, d.f. = 1,74, P = 0·021; body mass: β = 0·14 ± 0·07, χ2 = 4·34, d.f. = 1,74, P = 0·037; body condition index: β = 0·15 ± 0·07, χ2 = 4·14, d.f. = 1,74, P = 0·042).
The relationship between LRSF and bill length, body mass and body condition index were not significant in males. However, non-significant sex by trait interactions in a model based on the most parsimonious model for females (see Table 3) but that also included the main effect of sex, and the interactions between sex and bill length and body mass (or body condition index), showed that the effect of these traits were not significantly different in females compared to males (sex × bill length: P > 0·9; sex × body mass: P = 0·10; sex × body condition index: P = 0·16). The sex by trait interactions were also non-significant when they were included in the model separately (sex × bill length: P > 0·7; sex × body mass: P = 0·09; sex × body condition index: P = 0·15).
This study demonstrates that lifetime reproductive success (LRS) in male and female house sparrows is affected mainly by variation in individual annual reproductive success, and to a lesser extent variation in individual lifespan (Table 2). A significant proportion of the variance in the LRS of house sparrows (Fig. 1) was explained by variation in phenotypic traits in males (Table 3, Fig. 4). Furthermore, we found that the effect of male morphology on LRS operated mainly through an effect on the number of recruited daughters (Table 3). In contrast, overall LRS of females was independent of their phenotypic characteristics, although the number of recruiting daughters was affected by female morphology (Table 3). The effect of morphology on lifespan and annual reproductive success appeared to differ in males and females, but these differences were not significant.
The importance of variation in annual reproductive success for variation in LRS is in accordance with studies on other passerines (van Balen, van Noordwijk & Visser 1987; McCleery & Perrins 1988, 1989; Gustafsson 1989; Merilä & Sheldon 2000). On the other hand, in studies where estimates of LRS are based on fledgling production instead of the production of recruits, adult lifespan often emerges as the most important component of variation in LRS (Bryant 1989; Dhondt 1989; Grant & Grant 2000). This is generally because those estimates do not account for the large variation in reproductive success that is due to variation in the probability of survival from fledging to recruitment (see Newton 1989a). Previous studies of our house sparrow metapopulation have documented large temporal variation in the survival rate of juveniles, closely related to variation in climate (Ringsby et al. 1999, 2002). Thus, it seems likely that variation in juvenile survival rate is one important reason why variation in annual reproductive success, and not lifespan of adults, is the major component of LRS in house sparrows in northern Norway. Similar effects of juvenile survival on variation in annual reproductive success, and consequently on LRS have also been documented in, for example, great tits (Parus major) (van Balen et al. 1987; McCleery & Perrins 1988, 1989) and in red deer (Clutton-Brock et al. 1988). Estimates of LRS based on the number of fledglings and not on recruits, may therefore not reflect the true relative importance of lifespan and annual reproductive success for the contribution of an individual to future generations (i.e. its fitness) (Grafen 1988).
Very few studies exploring factors related to LRS include dispersing recruits in the estimates of individual LRS (Lambrechts et al. 1999). This may result in considerable biases in estimates of LRS because some dispersal is the rule rather than the exception in most animals (Stacey, Johnson & Taper 1997). Because theories predict that either high- or low-quality individuals should disperse (Clobert et al. 2001), not including dispersing recruits may preclude detection of any effects of, for instance, morphological characteristics of parents on the LRS. In contrast, in our study we included recruits that dispersed in the estimates of LRS because the main study islands were surrounded by an archipelago of 13 islands covering an area of more than 1600 km2 where house sparrows were also captured and observed. In addition, we searched for emigrants from our study area several 10 km along the coastline on the mainland (see Methods). This suggests that sexual differences in natal dispersal, as is common in many bird species (Clarke et al. 1997), cannot account for the differences in the effects of morphology on the production of male and female recruits (Table 3).
The influence of male morphological characteristics on variation in LRS (Table 3, Fig. 4) suggests that the phenotypic characteristics of the male are important determinants of the LRS, even in such a variable environment as is found at the coast of northern Norway (Ringsby et al. 2002). This corresponds with results from studies on other bird species, that have documented significant associations between body size and either annual reproductive success or lifespan in males and/or females (Bryant 1988, 1989; Johnson & Johnston 1989; Grant & Grant 2000; Krüger & Lindström 2001). In some species a major part of the variation in juvenile survival and growth, and consequently also adult morphology, may have a stochastic component that is so considerable that the link between, for example, adult body size and parental quality is weak. This may be the case in, for instance, collared flycatchers on Gotland (Gustafsson 1989; Merilä & Sheldon 2000) and great tits in England (McCleery & Perrins 1989). Nevertheless, as suggested by the results in this and some other studies (e.g. Kruuk et al. 1999; Festa-Bianchet, Jorgenson & Réale 2000; Kruuk et al. 2002), morphological characteristics of parents may reflect individual quality despite any effects of a stochastic environment on morphology.
A few other studies have focused on the relationship between characteristics of sexually selected traits and LRS (e.g. Sheldon & Ellegren 1999; Kruuk et al. 2002). Furthermore, recent reviews of the effect of sexually selected traits on survival of offspring (Møller & Alatalo 1999) and male viability (Jennions, Møller & Petrie 2001) have indicated that sexually selected traits in general are associated positively with these factors, and should thus also be significantly related to LRS. In the house sparrow, Møller (1988, 1989, 1990) demonstrated that there was a positive relationship between the size of the sexually selected black badge of males and their annual mating success. Thus, our results support Møller's findings but are in disagreement with most other studies on house sparrows, which have found either no (Veiga 1993; Kimball 1996; Cordero, Wetton & Parkin 1999; Whitekiller et al. 2000; Václav & Hoi 2002) or a negative effect of badge size on annual reproductive success (Griffith, Owens & Burke 1999a). All these studies have, however, examined the effect of male badge size only on reproductive success within a single breeding episode, and some have not taken extra-pair offspring into account (but see Cordero et al. 1999; Griffith et al. 1999a; Whitekiller et al. 2000). One possible reason for our contrasting result is the fact that our estimates of LRS are of genetic offspring (both intra- and extra-pair) averaged over all breeding episodes of a male. Consequently, we could account for the contribution of any extra-pair young to male LRS. In addition, if badge size of males on average has a small positive effect on number of recruits each breeding season, perhaps through increased frequency of extra-pair offspring or survival of offspring, such effects will be enhanced when summing the success from all breeding episodes during the entire male lifespan. This positive effect of badge size on LRS (Table 3, Fig. 4b) suggests that badge size is an indicator of male quality, and consequently that sexual selection (Andersson 1994) may operate in the house sparrow. The positive relationship between badge size and LRS suggests that the male badge either indicates indirect (i.e. genetic) benefits that promote offspring survival or direct benefits such as, for instance, enhanced fertility, high level of parental care or a good territory or nest site (Andersson 1994; Møller 1994). We cannot, however, distinguish between these two alternatives on the basis of our results. Regardless of the ultimate mechanism, our results agree with the general belief that the expression of sexually selected traits has an effect on male fitness (e.g. LRS), as documented recently in collared flycatchers (Gustafsson et al. 1995; Pärt & Qvarnström 1997; Sheldon et al. 1997), red deer (Kruuk et al. 2002) and lions (West & Packer 2002).
The positive effect of male badge size on LRS was associated with a significant positive effect on the number of recruiting daughters, whereas it did not significantly influence the number of sons (Table 3). In females, none of the morphological characters affected the number of recruiting sons (Table 3). However, bill length, body mass and body condition of females had a positive effect on number of daughters (Table 3). In accordance with studies on other passerines (Clarke et al. 1997), it has been documented previously that dispersal among female juvenile house sparrows in our study metapopulation is almost twice as high as for males (Altwegg et al. 2000). The positive association between morphological characters of adults and number of female recruits may thus result in increased production of dispersers by these birds as well.
A number of studies on both birds and mammals have documented differential production of sons and daughters (see reviews in Sheldon 1998; Hewison & Gaillard 1999). In many cases this has been attributed to mechanisms by which parents adjust the sex ratio according to the differential fitness returns implemented in production of sons and daughters (Trivers & Willard 1973), sometimes resulting in increased production of sons by good quality parents (Ellegren, Gustafsson & Sheldon 1996; Nager et al. 1999; Sheldon et al. 1999). However, such mechanisms often depend on the assumption that variance in male reproductive success is higher than for females (Trivers & Willard 1973; Frank 1990; but see also Gowaty 1993), a condition that was not fulfilled in this study (see Fig. 1). High quality males may, however, also be predicted to invest more in daughters because such investment may be selected for by female choice (e.g. Seger & Trivers 1986). Accordingly, although we currently have no knowledge of how much different parents invest in male and female offspring in house sparrows, our results seem to imply that the level of investment in daughters is related to the morphological characteristics of adults (see Table 3). However, one can speculate whether the reason for this apparent positive relationship between investment in daughters and adult morphology may be a result of equal investment in male and female offspring, but that males with a large badge and females with long bills, high body mass and good body condition are of ‘better quality’ and thus perhaps produce more fledglings with higher average survival. Furthermore, if male offspring on average have more variable survival because they need more resources during growth and development (see Cordero et al. 2000; Westneat et al. 2002), and are consequently more sensitive to environmental conditions during the juvenile period (see e.g. Clutton-Brock et al. 1985; Sheldon et al. 1998; Lindström 1999; Badyaev 2002), the observed relationship between parental quality and production of daughters may arise without the need to invoke differential allocation by parents (see also Griffiths 1992).
Our results demonstrate that the morphology of parents affect their LRS (Table 3), a measure linked closely to fitness (e.g. Grafen 1988; Newton 1989b). Such effects of morphology will have evolutionary consequences if the traits are heritable (i.e. that they have additive genetic variance that selection can act upon) (Falconer & Mackay 1996; Lynch & Walsh 1998). In a previous study of the same metapopulation we demonstrated that bill length had a heritability of around 0·55 in both sexes, and tarsus length a heritability of 0·31 in males (Jensen et al. 2003). Moreover, we estimated that the heritability of female body mass and body condition index was 0·12 and 0·18, respectively. A cross-fostering experiment by Griffith, Owens & Burke (1999b) indicated that male badge size was not heritable, but affected mainly by environmental factors (see also Veiga & Puerta 1996; Griffith 2000). Møller (1989), however, estimated that the heritability of badge size was 0·6, but his estimate was based on a very small sample of individuals raised by their natural parents. Data from our study population indicate that the heritability of badge size is low (approximately 0·3), but significant (H. Jensen, unpublished results). Consequently, the morphological characteristics of adults that are related to production of offspring (i.e. body mass and condition in females, bill length in both sexes, and tarsus length and badge size in males) may be transmitted to their offspring. Accordingly, studies of other taxa have demonstrated that traits related to LRS may be heritable. This was demonstrated, for example, for the size of the white forehead patch in collared flycatchers (Merilä & Sheldon 2000), bill length in snow petrels (Pagodroma nivea) (Barbraud 2000) and antler mass in red deer (Kruuk et al. 2002). If selection on a trait due to its relationship with LRS is not counteracted by selection on correlated traits, the traits will evolve as a result of their effect on LRS (Falconer & Mackay 1996). In our house sparrow population, selection on traits due to their association with LRS is not counteracted by negative phenotypic correlations (Table 1). However, we have demonstrated previously the existence of a strong negative genetic correlation between bill length and body condition index in females, and a strong positive genetic correlation between bill length and tarsus length in males (Jensen et al. 2003). The negative genetic correlation between bill length and body condition index in females indicates that the positive association between these traits and production of female recruits may not result in evolutionary change in any of the traits. Moreover, because male tarsus length and bill length are constrained by a positive genetic correlation, the same will be true for these two traits in males because they appear to be selected for in opposite directions (see Table 3). Similarly, badge size in males may also be constrained by genetic correlations with other traits related to LRS (H. Jensen, unpublished results).
In this study we have documented large individual variation in LRS and identified important characteristics of individuals that are most successful in reproduction over their entire adult life. These characteristics are heritable and some may evolve as response to their effect on LRS, whereas the evolution of others may be constrained by genetic correlations with other characteristics that affect LRS. Moreover, these individual characteristics generally have a positive influence on the production of female recruits, which is the dispersing sex in house sparrows and most other passerines. Our study thus suggests important consequences of individual variation in morphology not only on the distribution of the contribution that each individual has to the gene pool in the next generation, but that this variation also may affect the evolutionary processes within local populations and probably the dynamics of the whole metapopulation.
We are grateful to R. Altwegg, T. Berge, T. Jensen, T. Kolaas, S. Krogstad, N. M. Pedersen, K. Solbakken, E. J. Solberg, H. Staff, B. Staven, K. Sørensen and I. R. K. Stewart for assistance in the field, and to C. Pélabon and two anonymous referees for helpful comments on the manuscript. We would also like to send our greetings to the inhabitants in our study area at Helgeland, whose hospitality made this study possible. This study was supported by grants from the Norwegian Research Council, the Norwegian Directorate for Nature Management and the EU (project METABIRD).