Stabilizing selection on microsatellite allele length at arginine vasopressin 1a receptor and oxytocin receptor loci

The loci arginine vasopressin receptor 1a (avpr1a) and oxytocin receptor (oxtr) have evolutionarily conserved roles in vertebrate social and sexual behaviour. Allelic variation at a microsatellite locus in the 5′ regulatory region of these genes is associated with fitness in the bank vole Myodes glareolus. Given the low frequency of long and short alleles at these microsatellite loci in wild bank voles, we used breeding trials to determine whether selection acts against long and short alleles. Female bank voles with intermediate length avpr1a alleles had the highest probability of breeding, while male voles whose avpr1a alleles were very different in length had reduced probability of breeding. Moreover, there was a significant interaction between male and female oxtr genotypes, where potential breeding pairs with dissimilar length alleles had reduced probability of breeding. These data show how genetic variation at microsatellite loci associated with avpr1a and oxtr is associated with fitness, and highlight complex patterns of selection at these loci. More widely, these data show how stabilizing selection might act on allele length frequency distributions at gene-associated microsatellite loci.

. Model selection table for probability of breeding (proportional data) predicted by an individual bank vole's avpr1a microsatellite allele length characteristics, separately for (a) females and (c) males and for breeding success predicted by an individual's oxtr microsatellite allele characteristics, separately for (b) females and (d) males. Variables are: (i) MAL, mean (over the diploid genotype) allele length of an individual (centred over the average mean allele length of all females or males) and (ii) MAL^2, its polynomial term (i.e. mean allele length-squared), and (iii) DAL, the difference in allele length between the two alleles within an individual, and (iv) ORIG the origin of the individual (i.e. wild-caught or laboratory-born): in addition, (v) all two-way interactions (*) were included in the full model. The full model is underlined, the best model is highlighted bold. All models within 2 QAICc units of the model with the lowest QAICc value are provided. Detailed results of the best models (bold) are provided in the main text and tables therein. (over the diploid genotype) allele length of an individual (centred over the average mean allele length of all females or males) and (ii) MAL^2, its polynomial term (i.e. mean allele lengthsquared), and (iii) DAL, the difference in allele length between the two alleles within an individual, (iv) ORIG, the origin of the individual (i.e. wild-caught or laboratory-born) and (vi) all significant two-way interactions (*) that were identified in the analysis of individual breeding probability (Supplementary Table 1). Moreover, model selection included (vii) DAL_FM, the difference between the female and male mean allele length as a potential predictor of breeding. Random effects male identity (n=256) and female identity (n=220) are included in all models. The full model is underlined and the sex of the individual (i.e. F_ or M_) is indicated before the variable name. All models within 2 AICc units of the model with the lowest AICc value are provided. The best model is highlighted bold, and the detailed results of the best models (bold) are provided in the main text and tables therein.

Supplementary material -analyses of litter size Methods
We examined whether microsatellite allele length at avpr1a and oxtr affected an individual's litter size at birth or at weaning (pups at the age of 20 days). Both female and male microsatellite genotypes were examined simultaneously. Separate models were run for avpr1a and oxtr as they are independent loci. In these analyses we included only successful pairings, i.e. 410 litters, of which 402 litters were monitored until weaning.
We estimated the litter size (count variable) using generalized linear mixed model (GLMM) approach with negative binomial error distribution and log link function (glmmadmb command in glmmADMB package in R). The lack of zeros in the litter data (at birth) was taken into account using zero truncated models. To control for potential pseudoreplication (from repeated observations per individuals), male and female identities were included as random effects in the models. The explanatory variables included in the full model were: (i) MAL -mean (over the diploid genotype) allele length of an individual (centred over the average MAL of all females or all males) and (ii) its polynomial term (i.e. mean allele length 2 ), and (iii) DAL -the difference in allele length between the two alleles within an individual, (iv) the difference in the MAL the female and male of each pair, and (v) the origin of the individual (i.e. wild-caught or laboratory-born). In addition, (vi) the order of the litter (1 st , 2 nd , 3 rd etc) for the female and for the male and their polynomial terms as well as biologically relevant twoway interactions were included in the full model. The litter size at the age of 20 days (at weaning) was examined using the same explanatory variables, random effects and the model selection procedure as above, but, as some litters had no surviving pups, we used negative binomial distribution without zero truncation. The model selection was carried out manually by omitting one by one non-significant terms, starting from interactions, then polynomials and last non-significant main effects. The model selection procedure was guided by AIC that confirmed that omitting terms decreased the AIC values. The models were simplified until the final models included only significant terms and further simplifying would have increased the AIC value by more than 2 units.

Results
The final model for the litter size at birth, both for avpr1a and oxtr, included only the origin of the female, with wild captured females having 11 % bigger litters than lab-born females at the time of birth (Table S3). There was no evidence that litter size at birth was affected by microsatellite allele length or other microsatellite characteristics.
Similarly, litter size at the age of 20 days was not associated with avpr1 microsatellite characteristics but the origin of the female, with wild capture females having 18.2% bigger litters at weaning age than lab-born females (Table S3). The within individual difference in the female's oxtr microsatellite length had a significant negative effect on litter size with one base pair increase in the difference associated with a decrease in litter size by 0.8% and the field captured females having 17% bigger litters at weaning age than lab-born females (Table  S3). Supplementary Table S3. Final GLMMs (after model selection) that provide the litter size (log scale) of pairs of bank voles in relation to the origin of the animals (wild-caught or laboratory bred), DAL -the intra-individual difference in allele length, at two microsatellite loci (avpr1a and oxtr. The variance attributable to random effect (σ 2 ) and the standard deviation of σ 2 (SD). Intercept for avpr1a represents a laboratory born female. Intercept for oxtr represents a laboratory born female with the difference between the allele lengths=0 (i.e. no difference in length in the two alleles).