Molecular and pedigree measures of relatedness provide similar estimates of inbreeding depression in a bottlenecked population

Authors


Correspondence: Sheena M. Townsend, Department of Zoology, University of Otago, PO Box 56, Dunedin 9054, New Zealand.

Tel.: +64 3 479 7986; fax: +64 3 479 7584; e-mail: townsend.sheena@gmail.com

Abstract

Individual-based estimates of the degree of inbreeding or parental relatedness from pedigrees provide a critical starting point for studies of inbreeding depression, but in practice wild pedigrees are difficult to obtain. Because inbreeding increases the proportion of genomewide loci that are identical by descent, inbreeding variation within populations has the potential to generate observable correlations between heterozygosity measured using molecular markers and a variety of fitness related traits. Termed heterozygosity-fitness correlations (HFCs), these correlations have been observed in a wide variety of taxa. The difficulty of obtaining wild pedigree data, however, means that empirical investigations of how pedigree inbreeding influences HFCs are rare. Here, we assess evidence for inbreeding depression in three life-history traits (hatching and fledging success and juvenile survival) in an isolated population of Stewart Island robins using both pedigree- and molecular-derived measures of relatedness. We found results from the two measures were highly correlated and supported evidence for significant but weak inbreeding depression. However, standardized effect sizes for inbreeding depression based on the pedigree-based kin coefficients (k) were greater and had smaller standard errors than those based on molecular genetic measures of relatedness (RI), particularly for hatching and fledging success. Nevertheless, the results presented here support the use of molecular-based measures of relatedness in bottlenecked populations when information regarding inbreeding depression is desired but pedigree data on relatedness are unavailable.

Introduction

Breeding between related individuals is frequently observed to result in declines in traits influencing fitness. This phenomenon, termed inbreeding depression, has long been recognized in a wide variety of taxa (Darwin, 1876; Wright, 1977; Charlesworth, 1987). Although the effects are highly variable, it is now well established that inbreeding depression, acting across a variety of traits, has the potential to affect the growth and persistence of wild populations (Crnokrak & Roff, 1999; Keller & Waller, 2002; Frankham, 2005). Inbreeding depression is a central concern in conservation genetics, in part, because restricted population sizes may result in an increase in inbreeding over time even under random mating (Keller & Waller, 2002). Negative effects of inbreeding may be further compounded by the other genetic complications of small population size such as increased genetic drift leading to an overall loss of diversity (Lande, 1995; Lynch et al., 1995; Frankham et al., 2002).

Despite this, more than a decade on from a highly cited review of studies of inbreeding depression in wild populations (Keller & Waller, 2002) and the flood of studies that followed, the study of inbreeding depression can still be difficult to conduct in practice, when wild populations are involved. The increased expression of rare deleterious recessive alleles among inbred individuals is thought to be the most important genetic contributor to inbreeding depression (Wright, 1921, 1977; Charlesworth & Charlesworth, 1999). This is due to increased genomewide homozygosity expected as a result of the proportion of loci that are identical by descent (IBD) in offspring produced by consanguineous matings. Since the inheritance of at least a proportion of the genetic load responsible for inbreeding depression may be largely dependent on stochastic events and population or lineage specific mutations, the effects of inbreeding depression are expected to be, to some extent, population, trait, environment, and also importantly, family specific (Kristensen & Sørensen, 2005). Attempts to generalize the results of studies of inbreeding in captive or laboratory reared populations may, therefore, underestimate the potential effects of inbreeding experienced under wild conditions (Crnokrak & Roff, 1999). Captive and laboratory based studies will also fail to provide information on the frequency of inbreeding expected to occur naturally.

Individual-based estimates of the degree of inbreeding or parental relatedness from pedigrees provide a critical starting point for studies of inbreeding depression (Pemberton, 2008). Molecular genetic data are now more readily available than ever before, improving our ability to reconstruct wild pedigrees using genotypes when accurate parental records are unavailable (Pemberton, 2008; Jones et al., 2010). However, accurate and reliable wild pedigrees are still difficult to obtain and typically involve a long-term commitment to intensive data collection.

As inbreeding is expected to reduce heterozygosity by increasing the number of loci with alleles that are IBD, within populations inbred individuals are expected, on average, to have lower genomewide levels of heterozygosity than less inbred individuals (Wright, 1922). It is therefore intuitively and practically appealing to use relatively easy to obtain molecular marker data as an indirect measure of an individual's level of inbreeding. This exploits the theoretical expectation that when inbreeding depression is present, individuals with higher pedigree inbreeding coefficients will be more homozygous and experience depression of the trait of interest (Falconer & MacKay, 1996; Charlesworth & Charlesworth, 1999). In this way, inbreeding depression would be expected to generate observable correlations between heterozygosity and traits. These are commonly referred to as heterozygosity-fitness correlations (HFCs) when traits are associated with fitness (Houle, 1989; Hansson & Westerberg, 2002; Balloux et al., 2004; Slate et al., 2004).

Although both HFCs and inbreeding depression have received well-deserved attention, wild pedigrees are still rarely available due to the difficulty of obtaining data (Pemberton, 2004, 2008; Grueber & Jamieson, 2008). There is, therefore, relatively little empirical evidence available to address the ability of HFCs to reflect inbreeding under a variety of demographic scenarios. When wild pedigrees are available, they present a rare opportunity to assess the ability of both pedigree and molecular methods to detect inbreeding depression for the same population and traits.

Here, we first assess evidence for inbreeding depression using pedigree-determined measures of relatedness. Then, we examine HFCs in the same individuals using molecular data based on a panel of 35 microsatellite loci. We focus on measures of breeding success in three life-history stages in a remnant population of Stewart Island robins in New Zealand. A reintroduced island population of this species has been studied and monitored intensively since it was established on Ulva Island off Stewart Island, in 2000 (Jamieson, 2011). Previous studies, which included pedigree and fitness data up until 2006 when juvenile recruitment rates were high and the population was still growing, documented relatively weak levels of inbreeding depression (Laws et al., 2010; Laws & Jamieson, 2011). Here, we add four additional breeding seasons, but focus on the period between 2005 and 2010 during which juvenile recruitment rates declined as the island habitat approached its carrying capacity for territorial robins (I. Jamieson, unpublished data).

We generate standardized effect size estimates for both the effect of microsatellite heterozygosity and pedigree-determined inbreeding f on hatching success, fledging success and juvenile survival using an information-theoretical approach. Our findings support the conclusion that HFCs in this population reflect the weak inbreeding depression revealed by examining the pedigree.

Materials and methods

The Stewart Island robin is a subspecies of the South Island robin (Petroica australis; Miller & Lambert, 2006), a small, New Zealand endemic forest passerine (Higgins & Peter, 2002). Although South Island robins are listed by the IUCN as a species of least concern (IUCN, 2011), the Stewart Island subspecies is listed as ‘Nationally Endangered’ by the New Zealand Department of Conservation (Hitchmough et al., 2002; Miskelly et al., 2008). Declining numbers of robins on the mainland of Stewart Island prompted the relocation of individuals to Ulva Island (46°55′ S, 168°08′ E; 259 ha), a pest-free sanctuary located in the Paterson Inlet of Stewart Island, New Zealand (Greer, 2000; Towns & Broome, 2003; Laws & Jamieson, 2011). The Ulva Island robin population is isolated from the mainland and was established with five male and seven female genetic founders starting in 2000 (Jamieson, 2011).

Since its establishment, the Ulva Island robin population has been closely monitored during the breeding season from August to February. Individuals begin breeding in their first year and may lay up to four clutches of one to three (typically two) eggs per season at low population densities, although the number of clutches typically reduces to one or two per year at higher densities. Although the lifespan of the species has not been established, adult mortality is low (approximately 12%; Masuda & Jamieson, 2013). Once chicks have hatched, both the male and female participate in the care of chicks and fledglings (Higgins & Peter, 2002).

All offspring produced in the population are colour ringed while under parental care (either in the nest or during the fledgling period). During ringing, a blood sample was collected via brachial venipuncture for the purposes of obtaining DNA. Records of paternity and maternity (see below) have been maintained since the population was founded in 2000. We calculated the pedigree inbreeding coefficient (f) for all individuals that survived to reach ringing age in the Ulva robin population using the functions provided in the R package pedigree (Coster, 2011). We also assigned kinship coefficients (k) to all pairs breeding in the population using the same method, where k is equal to f for offspring produced by a pair.

We included in the analysis all robins in the Ulva population that were actively breeding (i.e. laid at least one clutch between 2005 and 2010) and for which we had both pedigree data and microsatellite-derived genotypes (see details below). Extra-pair paternity, determined on the basis of microsatellite genotyping is low (approximately 0–2%) in this species and population (Taylor et al., 2008; S. M. Townsend and I. G. Jamieson, unpublished data). Therefore, robins on Ulva Island were considered socially and genetically monogamous and birds observed at the nest were assumed to be the genetic parents.

We considered factors affecting the nesting success of breeding pairs over three separate life-history stage transitions: hatching, fledging and juvenile survival. We defined hatching success as survival of at least one egg to the nestling stage as confirmed by the observation of parental feeding in the nest, fledging success as survival of at least one nestling to fledging as confirmed by the observation of parents with fledglings in the natal territory, and juvenile survival as the survival of at least one fledgling to the beginning of the following breeding season (see also below). Survival was confirmed by a population census at the start of each breeding season and observation during the season (Laws & Jamieson, 2011). Following the 2010 breeding season, rats reinvaded Ulva Island and juvenile survival was significantly lower than in previous years (Masuda & Jamieson, 2013). Therefore, survival rates for the 2010 cohort have been excluded from this analysis.

For each of the life-history stages examined, survival was determined at the brood level and scored as a binary variable (success = 1, failure = 0). Stewart Island robins lay clutches of between one and three eggs. As nests are frequently inaccessible to humans (between 25% and 50% are inaccessible in any one season; S. Townsend personal observation) it was not always possible to determine the number of eggs or chicks in a nest. The sample size of available nests with known binomial numerators and denominators (i.e. total hatched and total number of eggs for example) would have been unacceptably low and biased towards accessible nests. To avoid this bias, a nest was considered successful to the next life-history stage (e.g. egg to nestling) if at least one individual was confirmed to survive, and unsuccessful only if no chicks hatched, fledged or survived to the first breeding season. Occasionally, due to the timing of nest checks, it was not possible to determine the stage at which nest failure occurred. Such nests were also excluded from the relevant life-history stage transition (details below).

Estimating lethal equivalents using pedigree-determined measures of inbreeding

For each of the life-history stages under investigation we constructed a global generalized linear mixed-model (GLMM) containing six linear predictor variables and two interaction terms. Because nest success is likely to vary across breeding seasons, year was included as a predictor variable. The variable ‘year’ was centred by subtracting 2007 from the observed nesting season to improve interpretability of model coefficients (Schielzeth, 2010). To allow comparison of model coefficients for the remaining predictors, especially in the case of interactions, the remaining five continuous predictors were converted to scores with a mean of zero and standard deviation of 0.5 (Gelman, 2008). We included the pedigree-determined inbreeding coefficient of the brood (fbrood), which is equivalent to the kin coefficient (k) of the pair, along with the pedigree-determined inbreeding coefficients of the father (fpat) and the mother (fmat) as measures of inbreeding.

The inclusion of fpat and fmat in the model allows for interpretation of the lethal equivalents (LE) due to fbrood controlled for the inbreeding level of the father and mother. Previous studies of inbreeding depression in the Ulva Island population have suggested that age, in particular the age of the mother, may interact with the level of inbreeding (Laws et al., 2010). As a result we also included male and female ages (agemale and agefemale respectively) and the interaction between age and pedigree-determined inbreeding for parents of both sexes (i.e. fpat : agemale and fmat : agefemale) in the global model. Because pairs contributed multiple broods to the data and because some females and males mated with multiple partners over the study period, we included pair ID, female ID and male ID as random intercepts in the models where relevant. The global model described above was similarly constructed for each of the life-history stages investigated.

A candidate model set was prepared using all possible combinations of the six predictor terms and two interaction terms described above. This resulted in 256 candidate models. GLMMs were fit to the data using the R package lme4 version 0.999375-42 (Bates et al., 2011; R Core Development Team, 2011). The suitability of the global model was examined by plotting binned residuals against predicted values (Gelman & Hill, 2007) and adjusted likelihood ratio-based R2 comparing the global model with a null model, that is a model containing only the intercept and random-effects terms (Nagelkerke, 1991; Anderson, 2008). Model selection was performed on the basis of AICC using the R package MuMIn (Bartoń, 2012). Model selection did not indicate a single best model based on ∆AICC, therefore a top model set was defined using models with substantial support (∆AICC ≤ 2; Burnham & Anderson, 2002; Grueber et al., 2011). We report individual model-averaged parameter estimates accompanied by unconditional standard errors (SE) and associated 95% confidence intervals (CI). Unconditional SE incorporates a term for the variance in parameter estimates across those models included in the averaged model in addition to the sampling variance conditional on individual models (Anderson, 2008). Unconditional SE is provided in the output for an averaged model by the R package MuMIn (Bartoń, 2012). Model details for the top model set, to accompany the averaged model parameters presented below, are included in the Supporting Information.

After obtaining the averaged model, we calculated LE with 95% CI using a method modified from Grueber et al. (2011). We obtained estimates of the probability of success at each life-history stage for outbred (fbrood = 0) and inbred (fbrood = 0.25) based on the fixed-effects portion of the averaged model using parametric bootstrapping (Faraway, 2006). We generated 100 000 values for the number of diploid LE by substituting survival probabilities into eqn (1), which calculates B, or the number of haploid LE (Keller & Waller, 2002; Grueber et al., 2011). The estimates for B were converted to 2B (to return a value relevant to diploid organisms) and reported as the median and 95% CI of the bootstrapped estimates obtained.

display math(1)

Incorporating genetic measures of relatedness

All birds included in the above calculation of LE based on the pedigree were genotyped for a panel of 35 microsatellite loci. Genotyping and DNA extraction followed the methods described in Townsend et al. (2012). The 35 loci used had an average of 3.31 alleles per locus and expected heterozygosity (He) of 0.480 ± 0.037 (mean ± SE). One locus, 4G9, was monomorphic for the set of individuals included in this analysis and subsequently dropped. Additional details on the loci used can be found in Table S1. Genotypes for individuals were converted to standardized multilocus heterozygosity (SH), using the add-on function for R, GenHet version 3 (Coltman et al., 1999; Coulon, 2010; R Core Development Team, 2011). We selected SH because it is commonly employed in studies of HFCs and standardizes individual heterozygosity by the expected heterozygosity of the loci typed in each individual, providing comparable measures of heterozygosity when all individuals are not scored at all loci (Coltman et al., 1999; Chapman et al., 2009). Pairwise relatedness values (RI; Ritland, 1996) were calculated for all breeding pairs using the GenAlEx (version 6.4) add-in for Excel (Peakall & Smouse, 2006). Pairwise relatedness was chosen since it provides a means of calculating individual-based estimates of relatedness conceptually comparable to the pedigree-derived parental kinship coefficients (fbrood) we have used, and because it performs well when used to reflect variance in true relatedness across a variety of data sets (Csilléry, 2006).

The relationship between RI and pedigree-determined inbreeding coefficients for the 159 breeding pairs included in the analysis was examined using a linear model with RI as a response, pair kinship as the only predictor variable and a Gaussian error structure. Some females and males occurred in the data set as a member of more than one breeding pair, however, the proportion was small (12.8%) relative to the overall number of breeding individuals (N = 36; 20 males and 16 females, of 282 total individuals; 137 males and 145 females), and a mixed-model yielded zero variances for male and female IDs, so a linear model without random effects was used to examine the relationship between RI and pair kinship.

The global models used to examine the effect of pedigree f described above were modified to generate two additional global model sets. We first replaced pedigree-determined maternal and paternal inbreeding coefficients and pair kinship coefficients with microsatellite-derived SHmat and SHpat and RI respectively. The purpose of this model was to investigate evidence for HFCs at the life-history stages examined. This analysis is comparable to studies of HFCs when a pedigree is unknown. We also constructed a global model including both molecular and pedigree estimates of relatedness. This allowed examination of evidence for HFCs with variation due to pedigree kinship and inbreeding controlled in the models. In both cases, model selection and averaging were conducted as described above. Because input predictor variables in all models were centred and standardized prior to model fitting, we report standardized effect sizes that are directly comparable between models even though the absolute values of the molecular and pedigree measures of relatedness we use are on different scales (Nakagawa & Cuthill, 2007; Schielzeth, 2010).

Results

There were 558 broods produced between 2005 and 2010 that met the inclusion criteria described in the 'Materials and methods' section. The presence of eggs, as confirmed by the observation of female incubation (Laws et al., 2010) could not be confirmed for ten nests, which were subsequently dropped from further analysis. Pedigree-determined kinship coefficients (fbrood) for the remaining 548 broods ranged from 0 to 0.293, with a mean and SE of 0.0863 ± 0.00257. Paternal (fpat = 0.0493 ± 0.00238) and maternal (fmat = 0.0425 ± 0.00182) inbreeding coefficients covered the same range but were lower on average than fbrood.

Estimating inbreeding depression using pedigree-based measures of relatedness

Hatching success

Due to the timing of nest checks, hatching success could not be determined for 113 broods that failed at an unidentified stage prior to fledging. These broods were excluded from the analysis of hatching success leaving 435 broods laid by 162 pairs (145 males and 137 females). Of the 435 included in the analysis 378 (86.9%) hatched at least one chick.

All main effects were included in the final averaged model but the 95% CIs for year and fbrood did not include zero (Table 1). The adjusted likelihood ratio-based R2 was 0.124. Hatching success decreased as the inbreeding coefficient of the brood increased, but the effect expressed as LE (2B), based on all fixed predictors in the averaged model was relatively low, 2.414 (95% CI = 0.198–6.708).

Table 1. Model-averaged parameter estimates for factors affecting (i) hatching and (ii) fledging success and (iii) juvenile survival in Stewart Island robins on Ulva Island when using (A) pedigree- or (B) molecular-based measures of relatedness. Top model sets were averaged using the natural averaging method (i.e. parameters and standard errors (SE) averaged over only those models containing the parameter of interest). Unconditional SEs include model uncertainty and upper and lower CI represent the 95% confidence intervals based on these SEs. The relative importance of parameters is based on the top model set selected using AICC (see Table S2 for details)
Model predictorParameter estimateSELower CIUpper CIRelative importance
(A) Pedigree-based relatedness
(i) Hatching success
(Intercept)1.9040.2611.3922.415 
year0.5340.1430.2540.8151.00
 f brood −0.8600.390−1.624−0.0961.00
 f pat −0.5390.426−1.3750.2960.40
 f mat −0.3870.442−1.2520.4780.19
agemale−0.6190.582−1.7590.5210.35
agefemale0.6510.660−0.6421.9450.25
(ii) Fledging success
(Intercept)1.3220.2130.9051.740 
agemale0.6410.3100.0341.2481.00
 f brood −0.4790.229−0.928−0.0300.89
year−0.1730.101−0.3710.0250.68
 f mat 0.4270.283−0.1280.9830.63
agefemale0.3000.328−0.3430.9420.14
(iii) Juvenile survival
(Intercept)2.4340.3371.7743.094 
year−0.4870.202−0.883−0.0911.00
agefemale1.5840.6220.3652.8021.00
 f mat 0.1040.587−1.0471.2561.00
agefemale : fmat3.4861.2501.0365.9361.00
 f pat 0.3180.531−0.7241.3590.22
 f brood −0.2360.408−1.0350.5640.22
(B) Molecular-based relatedness
(i) Hatching success
(Intercept)1.9550.2661.4322.477 
year0.4860.1430.2050.7661.00
RIpair−0.7920.425−1.6260.0410.85
agefemale0.8310.715−0.5692.2320.35
agemale−0.7050.598−1.8770.4660.37
SHmat−0.2950.440−1.1570.5660.11
SHpat0.3760.443−0.4921.2430.10
agemale : SHpat1.3760.710−0.0162.7690.10
(ii) Fledging success
(Intercept)1.2660.1850.9031.630 
SHmat−0.6390.251−1.131−0.1481.00
RIpair−0.4870.233−0.943−0.0310.94
agemale0.5290.295−0.0491.1070.74
year−0.1340.093−0.3170.0490.47
SHpat−0.2670.256−0.7680.2340.42
agemale : SHpat0.8330.531−0.2071.8740.19
agefemale0.3600.314−0.2560.9760.18
(iii) Juvenile survival
(Intercept)2.1900.2921.6172.763 
year−0.4450.192−0.821−0.0691.00
agefemale1.2760.5550.1882.3641.00
SHmat−0.5020.469−1.4210.4170.67
SHpat0.7330.449−0.1471.6140.56
agefemale : SHmat1.6720.973−0.2353.5790.39

Fledging success

Of the 378 broods that produced at least one chick, the fledging success could be determined for all but four broods, which were excluded from further analysis of fledging success. Of the remaining 374 broods, 282 (75.4%) fledged at least one fledgling. The adjusted likelihood ratio-based R2 was 0.0587. The 95% CIs around the parameter estimates for the effect of male age and fbrood did not include zero (Table 1). Therefore, there was evidence that fledging success declined as fbrood increased, but LE (2B) for fledging success were low, 1.936 (95% CI = 0.108–4.647).

Juvenile survival

We were able to determine the juvenile survival status of all 186 broods that produced at least one fledgling. All models included the parameter fmat and its interaction with female age, whereas only one model contained fbrood. The adjusted likelihood ratio-based R2 was 0.187. The 95% CIs for the parameter estimates for year and female age, and the interaction between female age and fmat, did not include zero (Table 1). Juvenile survival was strongly affected by the mother's age. In particular, nests laid by young inbred mothers had lower juvenile survival than those laid by older inbred mothers. Given that fbrood had the lowest relative importance in the model set, it is not surprising that LE (2B) for juvenile survival were close to zero (0.347), and that the 95% CI included zero (−0.716 to 2.665).

Estimating inbreeding depression using molecular-based measures of relatedness

Mean parental pairwise relatedness (RI) for the 548 broods was 0.00258 ± 0.00415 (range = −0.173 to 0.442). Paternal standardized heterozygosity (SHpat) was 1.0336 ± 0.00757 (range = 0.594–1.425) and maternal standardized heterozygosity (SHmat) was similar 1.0546 ± 0.00732 (range = 0.475–1.553). As expected, RI was strongly positively influenced by pedigree-derived kin coefficient (β = 0.980, SE = 0.0937, F1,157 = 109.5, P < 0.001, adj. R2 = 0.407; Fig. 1).

Figure 1.

The relationship between pairwise genetic relatedness (RI; Ritland, 1996) and pedigree-determined kinship coefficients (fbrood) for 159 pairs of Stewart Island robins (Petroica australis rakiura) breeding on Ulva Island between 2005 and 2010.

Hatching success

RIpair, SHmat and SHpat as well the interaction between agemale and SHpat were included in the averaged model (Table 1). The adjusted likelihood ratio-based R2 was 0.129. However, all of the 95% CIs of the effects of genetic measures included zero. Although the effect of RI on hatching success was small, and the 95% CI included zero, increased RI had a negative effects on hatching as expected. Similarly, SHpat had a positive effect on hatching success but the estimate for SHmat was not in the predicted direction; more heterozygous mothers had lower hatching success, although the 95% CI included zero.

Fledging success

RIpair, SHmat and SHpat, as well the interaction between agemale and SHpat, were included in the averaged model (Table 1). The adjusted likelihood ratio-based R2 was 0.0890. Pairs with higher RI had lower fledging success and the 95% CI did not include zero. Similarly, the 95% CI for the effect of SHmat did not include zero; however, more heterozygous mothers experienced decreased fledging success. More heterozygous fathers also experienced decreased fledging success, but the 95% CI for this effect included zero.

Juvenile Survival

For juvenile survival, RI was not supported for inclusion in the final averaged model (Table 1). The adjusted likelihood ratio-based R2 was 0.142. SHmat and SHpat were included in the averaged model, although in both cases the 95% CI for the effect size estimate included zero and more heterozygous mothers experienced reduced juvenile survival of their offspring.

Comparing the results of models using pedigree and molecular-based estimates of kinship and pairwise relatedness (RI), we observed that similar standardized effect sizes and SEs are observed for hatching and fledging success, although RI consistently had slightly wider CIs (Figure 2). In the case of juvenile survival, the inclusion of pairwise relatedness was not supported in the final model, while the estimate for kinship was negative but included zero.

Figure 2.

A comparison of standardized effect size estimates for the effect of kinship coefficient (fbrood) on hatching and fledging success, and juvenile survival based on averaged models, with standardized effect sizes for heterozygosity-fitness correlations based on a similar global model using molecular estimates of pairwise relatedness (RI). For a complete summary of other factors included in the averaged models see Table 1. In the case of juvenile survival, the inclusion of microsatellite pairwise relatedness was not supported in the final averaged model so no effect size is estimated.

The results above suggest that pedigree measures may have given slightly more robust estimates of inbreeding depression than molecular measures. This conclusion was further supported in an analysis that included both pedigree and molecular measures in the same global model (Tables S3 and S4). The analysis including molecular and pedigree measures in the same model indicated that for all life-history stages, there is little evidence to suggest that RI provides additional information about inbreeding or performs better than pedigree measures. The kin coefficient of the pair/inbreeding coefficient of the brood (fbrood) appears to explain slightly more variation in survival than RI, particularly for hatching and fledging success.

Discussion

Our results indicated that pedigree f strongly influenced molecular-based measures of standardized heterozygosity. Therefore, it was not surprising that the results from models using molecular estimates of pairwise relatedness and individual heterozygosity correspond well for all life-history stages with the results from estimating inbreeding depression using the pedigrees (Figure 2). Pedigree-based estimates were more robust, with slightly larger effect sizes and smaller CIs than molecular based estimates. We also performed model selection based on a global model including both pedigree and genetic measures of relatedness and inbreeding (Tables S3 and S4), but found no evidence to suggest that microsatellite heterozygosity, after controlling for pedigree-determined inbreeding and relatedness, explains additional variation in survival at any three of the life-history stages.

The levels of inbreeding we observed in this population are relatively low compared with other studies of wild birds. Laws & Jamieson (2011) reported a very low number of LE of inbreeding depression in the Ulva Island robin population based on a previous study which considered only juvenile survival and data up to and including 2006 when the mean level of inbreeding was lower (B = 0.24; 95% CI −1.92 to 1.04). Our findings for this life-history stage were similar. Laws & Jamieson (2011) included a comparison of other recent studies involving pedigreed passerines for inbreeding depression in early life-history stages, which illustrates a wide range of empirically derived effects of inbreeding. Haploid LE ranged from 7.47 in Collared Flycatchers (Ficedula albicollis; Kruuk et al., 2002) to 0 in the Medium ground finch (Geospiza fortis; Keller et al., 2002). In the closely related North Island robin (Petroica longipes) on Tiritiri Matangi Island, Armstrong & Cassey (2007) report a large number of haploid LE (6.71) but the 95% CI included zero (−0.66 to 14.08).

Studies that report 95% CIs, illustrate that studies of inbreeding depression in the wild often suffer from a lack of precision, resulting in estimates for LE that include zero (e.g. Armstrong & Cassey, 2007; Grueber et al., 2010; Laws & Jamieson, 2011). However, it is important to note that the effects of inbreeding depression considered for a limited life-history window may not be representative of realized fitness effects due to inbreeding over the entire lifespan, which may accumulate a high number of LE with significant effects (Szulkin et al., 2007; Grueber et al., 2010). Thus, even when levels for individual life-history stages are low the effects over a lifetime should not be discounted. In a number of recent empirical studies illustrate this including low reported LE (B = 0.41) for hatching success in Great tits (Parus major; Szulkin et al., 2007) and Takahe (B = −0.691; Porphyrio hochstetteri; Grueber et al., 2010) and embryo to fledging survival in male Hihi (B = 1.03; Notiomystis cincta; Brekke et al., 2010). In these studies levels accumulated and over complete life histories a much higher number of LEs were reported (B = 2.12, 8.03 and 6.91 respectively). These values based on individual empirical studies illustrate that levels of inbreeding observed in wild populations cover the range suggested by a number of meta-analytical studies of both captive bred (diploid LE, 2B = 3.1; Ralls et al., 1988) and wild populations (diploid LE, 2B = 12; O'Grady et al., 2006).

Although pedigree-derived measures of relatedness might provide slightly more robust estimates of inbreeding depression, the difficulty of obtaining accurate pedigrees for wild populations makes the use of molecular methods for estimating inbreeding appealing. The past decade has offered relatively prolific application of HFC studies in wild populations. Taken as a whole, studies support the existence of weak but significant correlations between heterozygosity and fitness (Coltman & Slate, 2003; Chapman et al., 2009). Interpreting the results of individual empirical studies of HFCs in terms of within-population inbreeding depression is not always straightforward (for a comprehensive review see; Grueber et al., 2008). The evolutionary theoretical background upon which the study of HFCs is based supports a role for inbreeding in the general sense whenever HFCs occur (Szulkin et al., 2010). Since small, recently bottlenecked, or admixed populations may be more affected by linkage disequilibrium, HFCs in these populations may be stronger than expected (Hansson, 2004). It is still under-appreciated how well theoretical expectations will extend to empirical situations involving nonequilibrium populations. Here, we found that HFCs in a bottlenecked population of Stewart Island robins strongly corresponded with the results of our investigation of inbreeding depression using the same fitness data and individuals.

It is also interesting to note that while more closely related pairs had lower fledging success, more heterozygous mothers also experienced decreased fledging success, a result that was unexpected. More heterozygous mothers are expected to have lower pedigree-determined inbreeding coefficients, on average, and lower proportions of loci that are IBD. Olano-Marin et al. (2011) report negative HFCs, also in females, when using microsatellites located in putatively functional areas of the genome in blue tits (Cyanistes caeruleus), which they interpret as potentially due to locus-specific effects. Although confirming the reason for negative HFC we observed here is beyond the scope of the current study, this trend may deserve further investigation.

The role of recent population demographics (i.e. bottleneck of 12 individuals in 2000/2001) and its potential influence on the ability to detect loci that are IBD with relatively few markers should not be overlooked. Recent findings comparing higher density SNP markers with only 11 microsatellite markers in zebra finches (Taeniopygia guttata) showed that microsatellites performed much better than expected in a moderately inbred zebra finch population (Forstmeier et al., 2012). This may be partially explained by the fact that the zebra finch genome segregates in abnormally large blocks; however, Forstmeier et al. (2012) also demonstrated that, among inbred zebra finch, a greater proportion of the genome occurred as large homozygous stretches, most likely due to inheritance of extensive IBD loci. The extent of this phenomenon throughout animal taxa and among non model organisms is poorly understood. Linkage disequilibrium (LD) is expected to be influenced by population demographic processes such as genetic drift and population admixture which result in inbreeding in the broad sense (Szulkin et al., 2010). The role of LD may be partly responsible for the somewhat confounding nature of overall trends in HFCs research to date and for confusion in the field over the role of ‘local-effects’ and linkage (Chapman et al., 2009; Szulkin et al., 2010). The interpretation of HFCs as inbreeding depression of some form, as encouraged by Szulkin et al. (2010), is theoretically supported and encouraged by our findings in a small and isolated population recently bottlenecked from a population remnant.

In our particular study population, averaged models using pedigree-based measures of inbreeding support weak inbreeding depression in hatching and fledging success. For juvenile survival, the effect of brood inbreeding coefficient was negative, but the CI included zero. Maternal age had a relatively strong positive effect on probability of juvenile survival, and maternal age interacted with maternal inbreeding level, indicating that juveniles produced by young inbred females experienced reduced survival. Laws & Jamieson (2011) found a similar overall result, but with a smaller data set. Thus, the findings of this study add further support to the conclusion that inbreeding depression, while not absent, is likely weak in Stewart Island robins relative to other inbred wild populations (Laws & Jamieson, 2011). Evidence of weak inbreeding depression is consistent with a population with low genetic load as a result of a historical population bottleneck and genetic purging (Laws & Jamieson, 2011). However, Laws & Jamieson (2011) could not rule out that the weak (and nonsignificant) inbreeding depression was due to bottleneck-induced fixation of deleterious alleles leading to both inbred and noninbred individuals having similarly reduced fitness. However, the evidence of HFCs we present here plus in a concurrent study that experimentally controlled for within-nest differences in inbreeding and the environment between sibling nest-mates (S. Townsend & I. Jamieson, unpublished data) indicate that this hypothesis is unlikely.

As a whole, the results presented here provide a potentially optimistic future for the use of microsatellites in conservation when information regarding inbreeding depression is desired. It should be noted, however, that population structure and genome-specific characteristics (i.e. existence of large linkage blocks observed in zebra finches; Forstmeier et al., 2012), will likely influence the ability of molecular markers to reflect inbreeding depression. The extent to which the recent bottleneck experienced by the Stewart Island population of robins influenced our results cannot be determined here. This does, however, highlight the need for an increased focus on studies involving data from populations that deviate from typical demographic histories involving large effective population sizes and outbreeding, and on factors influencing the abilities of genetic markers to reflect loci that are IBD due to pedigree inbreeding.

Acknowledgments

This research was carried out under Environmental Risk Management Authority (ERMA) approval number GMD002754 and Otago University Animal Ethics Committee permits 87/05 and 79/08. Funding was provided by Landcare Research, University of Otago, Marsden Fund Council, Department of Conservation, Royal Forest Bird Protection Society and S. Townsend was supported by a Commonwealth Scholarship. The authors thank M. Szulkin, C. Grueber, B. Robertson and one anonymous reviewer for helpful comments and input, which improved the manuscript.

Ancillary