Study Population

The Soay sheep is a primitive breed of domestic sheep that was isolated on the island of Soay in the remote St. Kilda archipelago several millennia ago and has been living under unmanaged conditions and evolving under natural selection since then (Clutton-Brock and Pemberton 2004). In 1932, 107 Soay sheep were moved to the larger island of Hirta after the evacuation of all human residents. Approximately one-third of the population of these sheep live in the Village Bay area of Hirta, and these individuals have been the subject of a long-term study since 1985 (Clutton-Brock and Pemberton 2004). In April of each year, around 95% of all lambs born in this area are caught and individually tagged. Each August, as many sheep as possible from the study population are recaptured using temporary traps (Clutton-Brock and Pemberton 2004). At capture, animals are weighed and blood and fecal samples are collected. Whole blood samples are collected into heparin tubes and centrifuged at 1,008 g for 10 min, and plasma are removed and stored at −20°C. The majority of sheep deaths occur over winter, and the population dynamics of the Soay sheep are characterized by periods of low but rising population sizes followed by high mortality crash winters in which more than half the population can die (Clutton-Brock and Pemberton 2004). Regular censuses and mortality searches during the winter months result in the majority of carcasses being discovered, providing accurate death date information.

Breeding success for study individuals is known with high precision but is calculated differently depending on sex. For female annual breeding success, we used the number of offspring born to the female in the subsequent spring (derived from observational data), while for males we calculated the number of offspring sired by the ram and born in the subsequent spring (derived using genetic pedigree construction). The pedigree was constructed using 315 unlinked single-nucleotide polymorphisms (linkage disequilibrium $r^2<0.05$) with a minor allele frequency >0.4 using the R library sequoia (Huisman 2017). Total Village Bay population (referred to as population density) was used as a measure of environmental conditions because it captures annual variation in both food availability and nematode parasite exposure (Wilson et al. 2004; Hayward et al. 2014).

In this study, we included all animals that were caught and had plasma samples taken in August between 1990 and 2015. At the same time point, fecal samples are collected for the estimation of strongyle FEC. FEC is estimated from fecal samples as the number of eggs per gram using a modified McMaster technique (for full details, see Gulland and Fox 1992; Hayward et al. 2014).

Measurement of Antibodies

Antibody measurements were based on previously established protocols in this study population (Nussey et al. 2014). IgA, IgG, and IgE activity against antigens of the third larval stage of Teladorsagia circumcincta were measured using direct (IgA, IgG) and indirect (IgE) enzyme-linked immunosorbent assays (ELISAs; henceforth, anti-Tc antibodies). We used T. circumcincta L3 somatic antigen (Moredun Research Institute) as the capture antigen for all assays diluted to 2 µg/mL in 0.06 M carbonate buffer at pH 9.6. We added 50 µL of the diluted capture antigen to each well of a Nunc-Immuno 96-microwell plate, which was covered and incubated at 4°C overnight. After the wells were washed three times in Tris-buffered saline-Tween (TBST) using a plate washer, 50 µL of the Soay sheep plasma sample diluted 1∶50 for IgA and IgE and 1∶12,800 for IgG were added to each well. The plates were then covered and incubated at 37°C for 1 h. Plates were then washed five times with TBST, and 50 µL per well of rabbit anti-sheep IgA detection antibody conjugated to horseradish peroxidase (HRP; AbD Serotec AHP949P) diluted 1∶16,000 was added to the anti–T. circumcincta IgA assay, and 50 µL per well of rabbit anti-sheep IgG detection antibody conjugated to HRP (AbD Serotec 5184-2104) diluted 1∶16,000 was added to the anti–T. circumcincta IgG assay. For the anti–T. circumcincta IgE assay, 50 µL per well of anti-sheep IgE (mouse monoclonal IgG1, clone 2F1; Moredun Research Institute) diluted 1∶100 was added, followed by a 1-h incubation at 37°C and five washes with TBST, and then 50 µL per well of goat anti-mouse IgG1-HRP detection antibody (AbD Serotec STAR132P) were added diluted 1∶8,000 in TBST. All plates were then incubated at 37°C for 1 h. Plates were then washed five times with TBST, and 100 µL of SureBlue TMB 1-Component microwell peroxidase substrate was added per well and left to incubate for 5 min in the dark at 37°C. Reactions were stopped by adding 100 µL per well of 1 M hydrochloric acid, and optical densities (ODs) were read immediately at 450 nm using a Thermo Scientific GO Spectrophotometer.

All results were recorded as OD values. In order to minimize confounding of capture year and age effects with plate to plate variation, each plate included samples from 2 years paired at random with different age groups on each plate. All plates were run in duplicate, and duplicate sample ODs were removed if the coefficient of variation between duplicates was >0.2 or the absolute difference between the duplicates was >0.2 OD units. We also checked the correlation of ODs across duplicate plates and reran both plates if the correlation coefficient was <0.8. No duplicate plates failed this check, and the number of samples that failed quality control per assay was 13 for IgA (seven lambs and six adults), eight for IgE (six lambs and two adults), and 27 for IgG (five lambs and 22 adults).

To reduce error due to within-plate variation, we included two sample free wells per plate (50 µL TBST added instead of sheep plasma) as blanks and two wells of positive controls. Positive controls for the IgE assay were serum from ewes trickle infected with T. circumcincta and, for the IgA and IgG assay, plasma from normal healthy nonimmunized domestic sheep. For subsequent analyses, the mean optical density ratio of each sample was taken according to the formula

Statistical Analyses

The final data set used in the analyses encompassed 6,398 samples from 3,114 individuals whose age was known. Five samples from very late-born lambs caught in August that were within 50 days of birth were excluded from the data set because of the potential presence of maternal antibodies and differences in development stage to other lambs. The final data set is deposited in the Dryad Digital Repository: http://dx.doi.org/10.5061/dryad.9qr0401 (Sparks et al. 2018).

All analyses were conducted using linear and generalized linear mixed models in R v3.4.4 (R Development Core Team 2016). Because of the immaturity of the immune system at 4 months (Smith et al. 1985) and the large increase in antibody levels between 4 and 16 months, we ran three models per response variable for lambs (both sexes), adult females, and adult males. Adults were separated into females and males because of the differences in range and distribution of phenotypic traits, such as weight and FEC in adults, as well as differences in survival rates and breeding success (). We built models with August weight, August strongyle FEC, overwinter survival, and annual breeding success as response variables with anti–T. circumcincta IgA, IgE, and IgG levels as explanatory variables. Continuous explanatory variables were rescaled to mean 0 and 1 SD before inclusion in each model subset. We first fitted a model containing the fixed effects previously identified as important predictors of these variables. We then simplified the model by stepwise deletion, sequentially removing fixed effects with the lowest t values and determining statistical significance using likelihood ratio tests until a base model containing only significant ($P<.05$) fixed effects was left. All dropped terms were then retested against this base model using the same criteria. Population size and sex (where fitted) were kept in all models in order to test for interactions with antibody levels. To the minimal model containing only significant terms, we separately added linear and quadratic functions of anti-Tc IgA, IgE, and IgG and assessed them for significance using likelihood ratio tests. Next we tested whether there were sex- or density-dependent effects of antibody levels in each of the models by looking at interactions with sex and population density against the linear and, if significant, quadratic antibody terms.

August strongyle FEC was heavily right skewed and zero inflated. Since the accuracy of the method used to estimate FEC is limited to 100 eggs per gram (Gulland and Fox 1992), we binned August FEC into multiples of 100, with zeros counted as 0 and values >6,000 eggs grouped into a single final bin (fig. A1A, A1B). Associations between antibody measures and August strongyle FEC were modeled using generalized linear mixed models via the glmmADMB package v0.8.3.3 (Fournier et al. 2012; Skaug et al. 2016). To assess which distribution best fitted the data set, we tested Poisson distributions and two negative binomial distributions (NB2 parameterization where $\mathrm{variance}=\mu \left(1+\mu /k\right)$ and NB1 fit where $\mathrm{variance}=\phi \mu$) with and without zero inflation and compared the Akaike information criterion (AIC) between all six models. We found that the negative binomial NB2 distribution without zero inflation had the lowest AIC for all models of FEC: lambs, adult females, and adult males. We included fixed effects for sex (lamb model only), weight, age, and population density. Because previous studies of FEC in Soay sheep have identified curvilinear relationships between FEC and both age and weight (Hayward et al. 2009), we tested both linear and quadratic effects of these terms in our models. We included capture year as a random effect in both models and individual identity as a random effect in the adult models.

We investigated potential associations between August weight and anti-Tc antibody levels using linear mixed effects models via the lme4 package v1.1–17 (Bates et al. 2015), assuming a Gaussian distribution (fig. A1C, A1D). For the lamb models, we included sex, twin status, age in days, maternal age (quadratic), and population size as fixed effects and capture year and maternal identity as random effects. For adult weight models, we included age (linear and quadratic terms) and population size as fixed effects and individual identity and capture year as random effects.

We calculated overwinter survival as survival from capture in August to May 1 in the subsequent year using death date, capture, and census information. Survival probability varied with age and sex (fig. A1E) as well as among years, as previously documented (Clutton-Brock and Pemberton 2004). Analyses of survival were performed using generalized linear mixed models (GLMMs) in the lme4 package with a binomial error distribution and a logit link function. Binomial GLMMs were run with the built-in bobyqa optimizer, and the maximum number of iterations increased to 2E⁻⁶ to improve model convergence. For the lamb model, we included sex, twin status, weight (linear and quadratic), and population size as fixed effects and capture year as a random effect. For the adult models, we included age (linear and quadratic), weight (linear and quadratic), and population size as fixed effects, with individual identity and year as random effects.

For females that survived the winter, we calculated annual breeding success as the number of offspring born to a female in the subsequent spring from observational data. For male annual breeding success, we calculated the number of offspring sired by the ram and born in the subsequent spring using the pedigree. The male annual breeding success measure included males that died over the winter but excluded all males that were not seen to participate in the rut in the census records taken between October and December. The probability of producing a lamb was highly age and sex dependent (fig. A1F), with female and male lambs having particularly low probabilities of breeding (44% and 9%, respectively). Since the distributions of fecundity measures are very different for males and females (fig. A1F), sex-specific models were run for breeding success in both lambs and adults in this case. Female annual breeding success measures were treated as binary in both lambs and adults (lambed/did not lamb), while male fecundity measures were treated as binary only in lambs because of the low number of individuals siring lambs in their first year. All three models were run as a GLMM with a binomial error structure as for survival. For adult males, we analyzed annual breeding success in the glmmADMB package, where the best-fitting model was the negative binomial with the NB2 parameterization without zero inflation. For male and female lamb models, we included weight (linear and quadratic) and population size as fixed effects and year as a random effect. For adult male and female breeding success, we included age (linear and quadratic), weight (linear and quadratic), and population size as fixed effects and individual identity and year as random effects.

Associations between Antibody Levels and Strongyle FEC

In lambs, August strongyle FEC was weakly but significantly negatively associated with all three anti-Tc antibody isotype (IgA, IgG, and IgE) levels (table 1; fig. 1A–1C). For all three isotypes, August FEC decreased linearly with increasing antibody levels. In a model containing all three antibody isotypes, IgA levels were independently negatively associated with FEC ($b=\mathrm{-0.074}\pm 0.021$ SE, ${{\chi }^2}_{\left(1\right)}=11.860$, $P=.001$), while IgE and IgG were no longer significantly associated with FEC (IgE: $b=\mathrm{-0.030}\pm 0.020$ SE, ${{\chi }^2}_{\left(1\right)}=2.180$, $P=.140$; IgG: $b=\mathrm{-0.034}\pm 0.022$ SE, ${{\chi }^2}_{\left(1\right)}=2.420$, $P=.120$). We found no evidence for sex- or density-dependent effects of antibody levels on FEC in lambs (table 1).

In adults, there was a negative but weak linear association between anti-Tc IgG levels and August FEC for both males and females (table 1; fig. 1E, 1F). In females, there was also a weak curvilinear relationship of IgA with FEC (table 1; fig. 1D). A model with both significant antibody terms for females showed these two associations to be independent (IgG: $b=\mathrm{-0.116}\pm 0.029$ SE, ${{\chi }^2}_{\left(1\right)}=15.780$, $P<.001$; IgA: $b\left({\mathrm{IgA}}^2\right)=0.312\pm 0.120$ SE, $b\left(\mathrm{IgA}\right)=\mathrm{-0.339}\pm 0.119$ SE, ${{\chi }^2}_{\left(1\right)}=6.720$, $P=.010$). Model predictions suggested that FEC was lowest at intermediate anti-Tc IgA levels in adult females (fig. 1D), and without the quadratic term in the model there was not a significant linear association between anti-Tc IgA and FEC ($b=\mathrm{-0.050}\pm 0.029$ SE, ${{\chi }^2}_{\left(1\right)}=2.980$, $P=.084$). We found no evidence for density-dependent effects of antibody levels on FEC in adults (table 1).

Associations between Antibody Levels and Weight

There was a significant quadratic association between August weight and anti-Tc IgE levels in lambs (table 2). Model predictions suggested that weight increased and then plateaued with increasing IgE levels in lambs (fig. 2A). Without the quadratic in the model, there was a significant positive linear association between IgE and weight ($b=0.128\pm 0.042$ SE, ${{\chi }^2}_{\left(1\right)}=9.399$, $P=.002$). This association was not driven by a few individuals with high anti-Tc IgE levels because a quadratic association remained after removing the 40 individuals with scaled IgE levels >3 ($b\left({\mathrm{IgE}}^2\right)=\mathrm{-1.019}\pm 0.294$ SE, $b\left(\mathrm{IgE}\right)=0.704\pm 0.154$ SE, ${{\chi }^2}_{\left(1\right)}=11.996$, $P=.001$). We found no association of the other antibody isotypes and weight in lambs or antibody × sex or density interactions (table 2).

In adult females, there was a positive but weak linear association between August weight and anti-Tc IgE levels, while a significant curvilinear association was found between IgA and IgG levels and weight (table 2; fig. 2B–2D). While higher IgE levels were associated with heaviest August weights, it was the intermediate anti-Tc IgA and IgG levels that were associated with heaviest weights (fig. 2B–2D). These associations were not driven by a few individuals with high anti-Tc antibody levels because the association remained after removal of individuals with high antibody levels of each isotype (after removal of IgA levels >2.5 [$n=15$]: ${{\chi }^2}_{\left(1\right)}=10.235$, $P=.001$; after removal of IgE levels >2.5 [$n=61$]: ${{\chi }^2}_{\left(1\right)}=6.981$, $P=.008$; after removal of IgG levels >3.75 [$n=5$]: ${{\chi }^2}_{\left(1\right)}=26.621$, $P<.001$). Linear associations between weight and IgA and IgG were weaker and less significant than quadratic effects (IgA: $b=0.102\pm 0.052$ SE, ${{\chi }^2}_{\left(1\right)}=3.873$, $P=.049$; IgG: $b=0.085\pm 0.044$ SE, ${{\chi }^2}_{\left(1\right)}=3.628$, $P=.057$). When all the significant antibody terms were in the same model, all remained significant, suggesting that each antibody isotype had independent associations with August weight (IgA: $b\left({\mathrm{IgA}}^2\right)=\mathrm{-0.621}\pm 0.187$ SE, $b\left(\mathrm{IgA}\right)=0.667\pm 0.189$ SE, ${{\chi }^2}_{\left(1\right)}=11.075$, $P=.001$; IgE: $b=0.156\pm 0.053$ SE, ${{\chi }^2}_{\left(1\right)}=8.934$, $P=.003$; IgG: $b\left({\mathrm{IgG}}^2\right)=\mathrm{-0.605}\pm 0.145$ SE, $b\left(\mathrm{IgG}\right)=0.674\pm 0.154$ SE, ${{\chi }^2}_{\left(1\right)}=17.413$, $P<.001$). A marginal density-dependent association with IgA was found, indicating a slightly stronger increase with weight at high sheep densities (table 2). The interaction remained significant when the other antibody terms were included in the model ($b=0.077\pm 0.031$ SE, ${{\chi }^2}_{\left(1\right)}=6.262$, $P=.012$). In adult males we found a positive but weak linear association of IgA with weight but no association with the other antibody isotypes or density interactions (table 2; fig. 2E).

Associations between Antibody Levels and Overwinter Survival

We found little evidence for associations between any of the three anti-Tc antibody levels and survival in lambs or adult males (table 3). Although there were no significant interactions between antibody levels and sex for lambs, we found a marginally significant interaction between anti-Tc IgA level and population density. In adult females, we found that anti-Tc IgG and IgE levels were significant predictors of overwinter survival (table 3; fig. 3). Females with higher IgG and IgE levels were more likely to survive the winter, independent of age and weight (table 3; fig. 3). In a model with both IgG and IgE fitted, IgE became nonsignificant, but IgG remained significant (IgE: $b=0.149\pm 0.083$ SE, ${{\chi }^2}_{\left(1\right)}=3.301$, $P=.069$; IgG: $b=0.217\pm 0.084$ SE, ${{\chi }^2}_{\left(1\right)}=7.151$, $P=.007$). In addition, IgG remained significant if strongyle FEC was included in the model ($b=0.278\pm 0.095$ SE, ${{\chi }^2}_{\left(1\right)}=9.544$, $P=.002$) and if the one individual with unusually high IgG (>5) was dropped ($b=0.235\pm 0.085$ SE, ${{\chi }^2}_{\left(1\right)}=8.240$, $P=.004$). There was no evidence for environment-dependent selection on adult females (table 3).

Associations between Antibody Levels and Annual Breeding Success

In females, we found a negative association between anti-Tc IgE levels as lambs and the probability of females breeding in their first year (table 4). Although our initial model detected a significant quadratic association between IgE and female first year breeding probability (table 4), visual inspection of this relationship and the model predictions suggested that it could be driven by a handful of lambs with very high antibody levels. When the four individuals with high anti-Tc IgE values were dropped, the quadratic was no longer significant (after removal of IgE levels >4: ${{\chi }^2}_{\left(1\right)}=0.906$, $P=.341$), but there was a significant negative linear relationship between IgE levels and first-year breeding probability ($b=\mathrm{-0.294}\pm 0.118$ SE, ${{\chi }^2}_{\left(1\right)}=6.465$, $P=.011$; fig. A4A). We found no associations with any other antibody isotype or density interactions in female lambs. In males, we found a positive association between IgE levels as lambs and breeding success in their first year (table 4). Male lambs with higher IgE antibody levels were more likely to sire offspring in their first year (fig. A4B), and this association remained after dropping the six individuals with high IgE levels ($b=0.353\pm 0.113$ SE, ${{\chi }^2}_{\left(1\right)}=9.037$, $P=.003$). We found no associations with any other antibody isotype or environment-dependent interactions in lambs.

In adult females, we found a curvilinear association of both anti-Tc IgA and anti-Tc IgG on breeding probability, but these associations were extremely weak and marginally significant (table 4). There was no underlying linear association between either isotypes and annual breeding probability (IgA: $b=0.030\pm 0.153$ SE, ${{\chi }^2}_{\left(1\right)}=0.039$, $P=.843$; IgG: $b=\mathrm{-0.015}\pm 0.132$ SE, ${{\chi }^2}_{\left(1\right)}=0.013$, $P=.908$). There was a weak and marginally significant positive association between IgG levels and the breeding success of adult males the following spring (table 4). We found no significant associations between annual breeding success and any other antibody isotypes or density interactions in adults.

Age-Dependent Associations among Antibodies, Strongyle Egg Counts, and Fitness

Generally, we found significant negative associations between anti-Tc antibody levels and strongyle FEC and positive associations with survival and reproduction, which is consistent with levels of these antibodies reflecting, in part, host resistance to nematode parasites rather than simply variation in parasite exposure (Hayward et al. 2014). However, the associations between antibody levels and strongyle FEC were typically weak, highlighting that numerous other factors beyond antibodies influence strongyle FEC, some of which have previously been identified in this study population (Wilson et al. 2004; Craig et al. 2008; Hayward et al. 2009, 2014). We also found evidence of a striking shift in the relationships between different antibody isotypes, worm egg counts, and overwinter survival probabilities with host age. IgA antibodies were the best predictor of FEC in lambs, but none of the isotypes predicted lamb overwinter survival, while in adults IgG antibodies were the only isotype to predict FEC and also were positively associated with female overwinter survival and weakly with male annual breeding success. This pattern may reflect an important shift in selection on the immune response to parasitic helminths from initial exposure and the development of an immune response during early life to the mature adult immune response, which is thought to be more geared toward tolerating infection, wound healing, and limiting damage (Allen and Maizels 2011). Our results suggest that while antiparasite IgA levels may represent the best marker of the development of immunity to these parasites during early life, IgG represents the best marker of an adult’s long-term ability to cope with persistent exposure and infection with nematode parasites. To our knowledge, our results provide the first evidence of such an age-related change in selection on immunity from the wild and provide an important basis for further studies to explore age-dependent changes in the immune response and its consequences for health and fitness under natural conditions.

Although all three antibody isotypes were moderately positively correlated in lambs (fig. A2) and were all significantly negatively associated with FEC when fitted separately in models, only anti-Tc IgA levels remained significant when all three isotypes were included in the model. This suggests shared covariance between isotypes and FEC, with IgA providing the strongest independent association. Veterinary studies of domestic sheep have demonstrated an important role of the mucosal IgA and IgE responses in the development of immunity to Teladorsagia circumcincta in lambs (Stear et al. 1995; Murphy et al. 2010). The hypersensitivity response (allergic inflammation caused by IgE-mediated mast cell degranulation) associated with production of antiparasite IgE is generally thought to be involved in the expulsion of larval parasites (McNeilly et al. 2009), while mucosal antiparasite IgA, which is positively correlated with circulating levels of antiparasite IgA (Henderson and Stear 2006), is thought to be involved in reducing adult worm growth and fecundity and inhibiting the development of worm larvae (Stear et al. 1995, 2004; Strain et al. 2002). Negative associations between FEC and circulating anti-Tc IgA levels have been previously reported in both domestic and Soay sheep lambs (Stear et al. 1995, 2004; Coltman et al. 2001; Strain et al. 2002). Despite its negative association with FEC, we found little evidence that anti-Tc IgA levels in summer predicted first-winter survival or the probability of breeding the following year. This may be due to the immature, dynamic nature of the immune response at 4 months old, which may not reflect their anti-Tc antibody levels going into winter (Smith et al. 1985). Overall, the absence of relationships between anti-Tc antibody levels and first-winter survival in lambs was surprising, especially given that a previous study of Soay sheep found a negative association between strongyle FEC and annual fitness in lambs but not in adults (Hayward et al. 2011). This seems likely to reflect the greater importance of variation in exposure to helminths for health and fitness in immunologically immature lambs and contrasts with the patterns observed in immunologically mature adult sheep.

In adult Soay sheep there was a significant negative relationship between anti-Tc IgG and FEC in both sexes and significant positive associations between anti-Tc IgG and overwinter survival in females and annual breeding success in males. Anti-Tc IgA did independently predict FEC in adult females, but the relationship was not linear and implied slightly raised FEC at the extremes of the antibody distribution. Two previous smaller-scale, cross-sectional studies of this population investigating samples from different years have found that anti-Tc IgG levels predict overwinter survival in adult females, independent of other immune measures and body weight (Nussey et al. 2014; Watson et al. 2016). Observing the relationship again in our much larger-scale longitudinal study suggests that this association is observed only in adult females but is robust and independent of environmental conditions and individual body mass. Although studies on domestic ruminants have focused mainly on young animals (Stear et al. 2009), negative associations between anti-Tc IgG levels and FEC or worm number have been reported for adults (Williams et al. 2010; McBean et al. 2016), and there is a clear role for IgG in mediating protective immunity against helminths in laboratory mice (Blackwell and Else 2001; McCoy et al. 2008). It has been hypothesized that in early life, when animals are most vulnerable to helminth-associated morbidity, an aggressive immune response to limit worm establishment may develop, while in later life, once immune responses have matured and helminths have established infections, a more tolerant immune phenotype that limits worm fecundity, promotes healing of damaged tissues, and prevents immunopathology would be favored by natural selection (Allen and Maizels 2011; Medzhitov et al. 2012). Age-dependent selection on a tolerance-associated immune phenotype has been documented in wild voles, where increased GATA3 gene expression (a master regulator of the Th2 response) was positively associated with body condition and survival but only in older and larger males (Jackson et al. 2014). Our results provide support for age-dependent changes in the role of different antibody isotypes in resistance and tolerance to nematode parasites under natural infection. Further detailed immunological study is now required to understand why IgG antibodies provide the best marker of an individual’s ability to cope with helminth infection in adulthood but not during early life.

Sex- and Environment-Dependent Selection on Antibody Levels

We found little evidence for sex differences in the associations between antibody levels and FEC or body weight, but marginal significant trends in our analysis suggest that relationships with survival and fecundity might differ between males and females. Perhaps most strikingly, in lambs there was a positive association between anti-Tc IgE and the probability of males producing offspring in the first year of life but a negative association in females (fig. A4). IgE antibodies are associated with a hypersensitivity response and mast cell degranulation in the gut, which is an effective mediator of nematode expulsion. It is also associated with inappetence, diarrhea, and reduced weight gain in domestic lambs (Stear et al. 2009). Given that males grow more rapidly than females in this species and body size is under strong sexual selection in male Soay sheep (Clutton-Brock and Pemberton 2004), it is surprising that an immune measure potentially associated with a cost to weight gain should be associated with improved reproductive performance in males but not females. We also found evidence for a weak and marginally significant positive association between anti-Tc IgG levels and adult male annual breeding success, while this isotype predicted overwinter survival in adult females. In wild ruminants generally, males are thought to invest a large amount of energetic resources in the autumn rut at a clear cost to their overwinter survival prospects, while females are more conservative and balance investment between their developing fetus and their own maintenance over winter (Clutton-Brock and Pemberton 2004). If antibody levels in summer reflect an individual’s ability to resist nematode infections and limit the damage they cause, this may reduce the cost of infection in autumn and winter, freeing up resources that adult males would be expected to invest in the rut, while females invest in their own maintenance over winter. Overall, these results may reflect important and previously unknown differences in the costs and benefits between the sexes of mounting an aggressive hypersensitivity response to infection during early life and mounting a more tolerance-oriented response in adulthood. These trends warrant further investigation in both wild and domestic ruminant systems, involving more regular seasonal monitoring of reproductive hormones, immunity, and worm burdens across seasons in both sexes.

Given the considerable annual variation in environmental conditions and nematode parasite exposure experienced by Soay sheep on St. Kilda that is associated with often dramatic changes in population size (Clutton-Brock and Pemberton 2004), the general lack of evidence for interactions between antibody measures and sheep density in our models was surprising. We would predict the benefits, in terms of parasite burden and fitness, of improved immune responses to nematodes to be most strongly felt in years of high sheep density, when animals are in the poorest conditions and exposure to nematode larvae is highest (Hayward et al. 2014). However, there was very limited evidence for such relationships. This lack of environment-dependent associations may reflect the fact that anti-Tc antibody levels in Soay sheep appear highly repeatable in adulthood (Hayward et al. 2014; A. M. Sparks, unpublished data). Although there has been a far greater theoretical and empirical focus on environment-dependent selection on immune phenotypes as a force maintaining genetic variation in the immune response in natural systems (Svensson et al. 2001; Calsbeek et al. 2008; Graham et al. 2010; Seppälä 2015), our analyses provide far stronger support for age-dependent rather than environment-dependent selection. This argues for more theoretical and empirical effort being directed at investigating the evolutionary and ecological causes and consequences of such age-dependent variation in selection on immunity in the future.

Reproductive Costs and Condition Dependence

Unlike a number of experimental studies in wild birds, we found no strong evidence for the predicted negative association between parasite-specific antibodies and subsequent reproductive success, indicative of a reproductive cost of immunity (Ilmonen et al. 2000; Bonneaud et al. 2003; Marzal et al. 2007; Gasparini et al. 2009). There was a negative association between IgE levels and the probability of female sheep breeding in their first year; however, this result was only marginally significant. A previous study in Soay sheep found that high levels of self-reactive antinuclear antibodies were associated with reduced breeding success in both adult male and female Soay sheep (Graham et al. 2010). Furthermore, previous work using an anti-Tc antibody ELISA that simultaneously captures all isotypes (pan-isotype) found that adult males with high antibody levels were less likely to sire offspring the following year, and females of high body weight and high antibody levels were also less likely to have a lamb the subsequent year (Hayward et al. 2014). This study also found a marginal association between anti-Tc antibodies and survival in female lambs only (Hayward et al. 2014). This pan-isotype assay is expected to largely capture variation in IgG, which is the predominant isotype in circulation. However, these results contrast with the results of the current and previous isotype-specific studies (Nussey et al. 2014; Watson et al. 2016), suggesting that it is actually providing a different immunological signal to IgG when measured alone. One interesting possibility is that the pan-isotype measure better captures overall protein costs associated with antiparasite immune responses in general because it captures total immunoglobulin levels directed at these parasites. On the other hand, specific isotypes offer a clearer signal of specific immune responses directed at resisting or tolerating worm infection and thus are more strongly associated with survival. This difference among the current and previous study further highlights the complex and variable patterns of natural selection that can be observed even on apparently functionally closely related immune measures within the same study population.

There is a strong expectation for immune responses to be tightly linked with the condition of the host, with individuals in a better nutritional state more able to invest resources in immunity and therefore to resist infection (van Noordwijk and de Jong 1986; Kraaijeveld and Godfray 1997; Reznick et al. 2000). Interestingly, we found evidence for both positive linear and curvilinear relationships between body weight, depending on the age or sex group, and antibody isotype. However, these associations were typically weak, suggesting that anti-Tc antibody levels are not consistently or strongly positively associated with body weight of the host. A number of other studies have shown that body mass and food intake are closely related to immune responses (Siva-Jothy and Thompson 2002; Seppälä and Jokela 2010). However, in a previous study of a smaller sample of adult female Soay sheep, we did not find associations between anti-Tc antibody levels and weight (Nussey et al. 2014). Similarly, a recent study found that protein markers expected to be associated with nutritional state, total protein, and albumin were only weakly associated with body mass (Garnier et al. 2017). While body weight is often used as a proxy for condition in field studies, both our results and previous studies suggest that the relationship between body weight and levels of proteins in the blood that we expect to be intimately linked to nutritional state and physiological condition may be variable and weak (Hayward et al. 2014; Garnier et al. 2017). While our results highlight the point that using a few catch-all measures of immunocompetence is very unlikely to capture the complexity of the immune response and selection acting on it in wild systems (Demas et al. 2011), the complex relationships observed with body weight emphasize the need for biomarkers of nutritional state and condition within ecology (Garnier et al. 2017). Our study shows that closely related immune traits may have very different associations with health and fitness measures and suggests that multivariate immune phenotypes must be measured in order to fully understand how genetic variation is maintained in these traits in the wild.