Computational protocol: Kinship and familiarity mitigate costs of social conflict between Seychelles warbler neighbors

Similar protocols

Protocol publication

[…] DNA was extracted from blood samples with a Qiagen DNeasy Blood and Tissue Kit and was used to determine sex following Griffiths et al. () and individual genotypes at a panel of 30 microsatellite markers already developed for the Seychelles warbler (, ). We compared the suitability of two pairwise relatedness estimators, Queller and Goodnight (QG) () and Lynch and Ritland (LR) (), in the R package related () and determined that QG was more suitable in our microsatellite panel ( and ). QG relatedness estimates between all dominant breeders in the population over the study period (including all focal individuals and all neighboring breeders) were calculated in GenAlEx 6 (). Pairwise relatedness has previously been shown to reflect pedigree relatedness in the Seychelles warbler (), and heterozygosity across our microsatellite panel is also known to reflect genome-wide heterozygosity ().We measured telomere length in each blood sample according to the protocol described in detail elsewhere (–). Briefly, we calculated a relative measure of telomere length as the concentration of telomeric DNA relative to that of a normalizer gene, GAPDH, using quantitative real-time PCR. We then calculated each individual’s ∆RTL over the sampling period as the difference between telomere length at first and final sampling, such that positive values indicate increases in telomere length and negative values indicate decreases in telomere length. [...] We used a model-averaging approach to determine which properties of the social neighborhood influenced the likelihood of territorial fights, residual body mass, telomere dynamics, and survival. Exploring the relative influence of kinship and familiarity on territorial costs under varying social circumstances requires testing interactions between different neighborhood properties. To investigate whether such interactions are equally important for male and female territory owners while simultaneously avoiding the need to model numerous complex, three-way interactions, we created separate models for focal males and females throughout. Collinearity between all variables was checked before modeling using variance inflation factors (VIF). In no case was the VIF large enough to cause issues in the analysis (all VIF <4) (). Using the package MuMIn () in R (version 3.3.1) (), we created a global mixed-model that contained all variables, plus selected interactions of interest, as standardized predictors so that both main effects and interactions could be interpreted (). We report natural averages of each parameter across the top model set, which contained all models in which the change in Akaike's second-order information criterion (ΔAICc) was ≤2 (). Natural averaging can inflate effect sizes and P values but is well suited to determining the effect of a given variable when the additional variance explained by that parameter is likely to be small (, ), as is the case with biological markers such as body mass and telomere dynamics. Complete outputs for top model sets are presented in . Since our global models contain a large number of parameters relative to the sample size (which increases the risk of over-fitting and the production of spurious results). (, ), we repeated our analyses using reduced models that contained only parameters that were deemed important (P < 0.1) in the full analysis. The results of this second analysis were qualitatively identical to those we report and can be found in .We first tested whether focal individuals were more likely to engage in territorial fights with unrelated or unfamiliar neighbors by considering each focal–neighbor dyad (consisting of a focal individual and one of its neighbors) in our dataset as an individual data point at which a territorial fight has the potential to occur. Separately for focal males and females, we tested whether fight occurrence for a given focal–neighbor dyad (whether we observed a fight at the boundary or not in a given season; binomial response) was predicted by relatedness to and familiarity with the male and female neighboring territory owners. We included two random effects: focal individual identity, to account for repeat sampling of focal individuals across the years of the study, and neighboring territory identity, to account for similarity in neighbor identity between focal individuals in the same area of the island.To measure the immediate influence of the social neighborhood on individual costs, we tested whether the ∆Mass between two sampling points was related to the corresponding change in each of the five social neighborhood properties (modeled as continuous variables) as measured at the same two sampling points. To determine whether the influence of kinship and familiarity depends on the intensity of territorial interactions, we also tested whether the change in these components interacted with the change in neighbor density between the two sampling points (change in relatedness or change in number of new neighbors × change in neighbor density). To investigate the interplay between kinship and familiarity, we tested whether change in the number of new neighbors had a varying effect on ∆Mass according to the change in relatedness (change in relatedness × change in number of new neighbors, separately for male and female neighbors). We also included change in group size between the two sampling points to account for between-sample variation in per-capita territorial resources (). In addition, we included random effects of age at first sampling and time between the two samples. In males, we also included individual identity as a random effect, as some males had multiple measurements of ∆Mass.To measure the long-term influence of the social neighborhood on individual costs, we tested whether (i) ∆RTL and (ii) survival to the year following the final sample (binomial response) were related to the mean of each of the five social neighborhood properties across the sampling period. We included interactions between neighbor density and each of the other social neighborhood properties and also tested whether the number of new neighbors had a differential effect on ∆RTL and survival depending on mean neighbor relatedness (mean relatedness × number of new neighbors, separately for male and female neighbors). We included three covariates: (i) the focal individual’s mean territory quality [accounting for environmentally induced differences in physiological costs ()], (ii) mean group size [the number of independent resident birds in the territory, accounting for variation in per-capita territorial resources ()] across the sampling period, and (iii) age at first sampling to account for age-related differences in telomere dynamics (). In the analysis of ∆RTL we also included the time (in years) between first and last sampling.The spatial nature of our data posed a risk of nonindependence: If two focal individuals lived in adjacent territories, they would be included in each other’s neighborhood and hence have the potential to influence each other. The correlation between neighborhood properties of neighboring focal individuals was consistently low (ca. 0.2) and therefore was unlikely to influence our results. However, there is also the potential for spatial autocorrelation between the social neighborhood and undetected environmental factors. We therefore tested whether similarity in individual body mass and telomere dynamics was related to the spatial proximity of two individuals. Using ArcMap 10.3, we calculated the center point of each territory using the spatial map of 2006 as a template and calculated the distance in meters between each of these center points. Using the ncf package in R, we calculated Moran’s I () and the significance values of the residuals of regression models of each response variable on all neighborhood properties. Moran’s I was not significantly different from zero for the residuals of any of the predictor variables (), and a visual inspection of the distribution of neighborhood properties across the island did not reveal any spatial grouping of neighborhood relatedness, familiarity, or neighbor density (), so we conclude that spatial structure is unlikely to have influenced the results of our analyses. […]

Pipeline specifications

Software tools GenAlEx, kinship, MuMIn
Applications Phylogenetics, Population genetic analysis
Organisms Homo sapiens