Direct fitness benefits explain mate preference, but not choice, for similarity in heterozygosity levels

Abstract Under sexual selection, mate preferences can evolve for traits advertising fitness benefits. Observed mating patterns (mate choice) are often assumed to represent preference, even though they result from the interaction between preference, sampling strategy and environmental factors. Correlating fitness with mate choice instead of preference will therefore lead to confounded conclusions about the role of preference in sexual selection. Here we show that direct fitness benefits underlie mate preferences for genetic characteristics in a unique experiment on wild great tits. In repeated mate preference tests, both sexes preferred mates that had similar heterozygosity levels to themselves, and not those with which they would optimise offspring heterozygosity. In a subsequent field experiment where we cross fostered offspring, foster parents with more similar heterozygosity levels had higher reproductive success, despite the absence of assortative mating patterns. These results support the idea that selection for preference persists despite constraints on mate choice.

before a blood sample could be taken, and were therefore not genotyped. Using known mother-offspring dyads it was possible to detect the occurrence of null alleles and other irregularities. On the basis of this analysis the following three microsatellite loci were excluded from further analyses because of the non-reliability of their results: PmaD130, PmaGAn40 and Pma196 (Kawano 2003;Saladin et al. 2003). See table S1 for the properties of the markers used in this study.  (Marshall et al. 1998 Extrapair paternity Using the microsatellite data from 17 loci, paternity of chicks was assigned using a likelihood approach in the software program Cervus 3.07 (Marshall et al. 1998). These loci had a combined second-parent exclusion probability (Pre) of 0.9999999945. We calculated critical values of LOD (log likelihood ratio) and delta (difference in LOD scores between the most likely candidate parent and the second most likely candidate parent) using the following parameters in CERVUS: 10000 cycles, 98% of loci typed, error rate 0.01%, two candidate parents.
Offspring were assigned to be extra-pair when these critical values were exceeded in the comparison of the genotypes of the mother, the putative father and the offspring.

Relatedness
We estimated marker-based relatedness by calculating the pairwise r following the method of Wang (2002) in the program Coancestry (Wang 2011). By calculating r for full sibling pairs (extra-pair chicks were excluded) using different methods we determined that, for this population and these microsatellite markers, the relatedness measure using the method of In the breeding season pairs within the experimental area had a relatedness of between -0.17 and 0.32 with a mean of 0.02 ± 0.01 (N=70).
C. Statistical analysis mate preferences (As described in the main paper, with more detail added) To analyse the proportion of time that a focal bird spent associating with each stimulus bird we used a binomial generalized linear mixed model (GLMM) with a logit link function. The fixed part of the model contained as explanatory variables heterozygosity of both the focal and the stimulus birds, relatedness between each focal and stimulus dyad, offspring heterozygosity for each focal and stimulus dyad and sex of the focal bird. We also added the square of relatedness since a preference for moderately related individuals can be expected (Bateson 1983). To test for differences in preference depending on the chooser's traits we added the interaction between heterozygosity of the focal and the stimulus bird, and the interaction between the focal heterozygosity and relatedness. To test for sex differences, we also included interaction effects between sex and the previously mentioned explanatory variables.
Modelling the combined effect of continuous explanatory variables like the focal bird's heterozygosity, stimulus bird's heterozygosity and their interaction is necessarily sparse: only three parameters are used to describe the combined effect. To check whether the systematic trend captured in this way is not too restrictive, we also modelled the effect of these variables after categorization of each into three groups, based upon tertiles. Replacing the two regressors and their product by the categorized versions and their interaction, leads to a model with eight parameters replacing the earlier three. This model is more flexible than the original one, although it has its own shortcomings (Altman 2005).
For the random part of the GLMM we followed the experimental design as closely as possible, specifying the next random terms (on the logit scale): 1) random effects for stimulus birds, since each stimulus bird was tested repeatedly; 2) random slopes for focal birds with respect to the stimulus bird's heterozygosity, relatedness, and offspring heterozygosity, as each focal bird was tested multiple times. Together these random effects define the G-side covariance structure. Furthermore for the R-side covariance structure we allowed the six proportions per test to be negatively correlated (as they sum to one per six-choice test), by introducing a compound symmetric correlation structure at the proportion scale and we introduced an extra scale parameter, because we analysed a continuous proportion, for which the binomial variance-mean relationship only holds up to a scale factor. The statistical analysis was performed using procedure PROC GLIMMIX from the SAS software system (version 9.3; SAS Institute Inc., Cary, NC). We fitted the model using backward elimination for the fixed part of the model, removing first higher order terms and later lower order terms if not significant (P>0.01). The reported P-value for an explanatory variable is the P-value in the last model in which it still occurred, or in the final model if not removed (see table 1).

B. Accounting for spatial structure in mating patterns
Although the study site was relatively small, there is the possibility that individuals were constrained in their choice for a mate by the locally available potential mates. To check this we also compared the existing mating pattern to a different null model of random mating; one which considers the local mate availability. Using the 'nearest neighbour scenario' as used in Szulkin (2009), we paired the male and female of every pair to a known individual of the opposite sex breeding in the nest box nearest to the focal pair and compared the observed mating pattern with the simulated pattern. We tested for differences in the relatedness and heterozygosity similarity between the observed and simulated scenario of random mating using a non-parametric Wilcoxon matched-pair test. We found no differences between the observed and the simulated mating patterns, not for heterozygosity (WSR: females, T+=829, n1=n2=63, p=0.53; males, T+=969, n1=n2=63, p=0.87) or for relatedness (WSR: females, T+=3616, n1=n2=63, p=0.43; males, T+=900.5, n1=n2=63, p=0.91).

C. Accounting for effects of captures on mating patterns
By capturing and bringing the focal birds into captivity we may have unintentionally affected pair bonds and in extreme cases even split up pairs. To check whether testing the birds had any effect on mate choice we also compared mating patterns between untested pairs and pairs in which one or both were brought to the lab for testing. However, the mating patterns in these two groups did not differ. Both tested and untested birds did not mate differently from random mating in both heterozygosity and relatedness.

D. Alternative parametrization of mixed models
Results mate preference based on categorized heterozygosities As described in Appendix S1-C we performed an extra analysis to check whether the specification of the interaction of continuous variables is not too restrictive, we also modelled the effect of these variables after categorization of each into three groups, based upon tertiles.

B. A.
Results offspring fledging probability based on categorized heterozygosities First, the foster mother, foster father and biological mother heterozygosities and biological relatedness were categorized into three groups based on tertiles. For the foster mother heterozygosity we took the cut points -0.02536803 and 0.03791166. For the foster father heterozygosity we took the cut points -0.03623277 and 0.06356268. For the biological mother heterozygosity we took the cut points -0.01245976 and 0.03742855. And for the biological relatedness we took the cut points -0.06911875 and 0.04628325. Next, in the GLMM we replaced the continuous heterozygosities and relatedness values and their interaction by main effects and interaction of the grouped heterozygosities and relatedness. Results after removal of non-significant terms are given in table S4. The estimated mean response (on the logit scale; mean  se) for the combination of foster mother and foster father heterozygosities. We have also given the back transformed mean responses, which indicate the fledging probability for the combination of foster mother and foster father heterozygosity. The estimated mean response (on the logit scale; mean  se) for the combination of biological mother heterozygosity and her relatedness with the biological father. We also give the back transformed mean responses, which indicate the fledging probability for the combination of foster mother and foster father heterozygosity.