How to use molecular data to account for non random mating in quantitative genetic estimates derived from family structured experiments

Accurate estimates of heritability (h²) of fitness-related traits are necessary to assess theiradaptive response in changing environments. For plants, maternal families are often sampled toestimate h². Estimates thus rely on several assumptions about the mating system and the relatednessamong mates (basically offspring are assumed to be half-sibs or unrelated). In particular, whenmating events are impossible to observe and sample sizes are large, the paternal relatedness is oftenneglected, leading to bias in the h² estimates.We performed a simulation study to investigate how molecular markers can be used toanalyze more efficiently the family structured designs. We compared the simple method based on amaternal family model assumption with methods using marker-based relatedness coefficients,pedigree or hybrid genetic information (animal models) to account for paternal relatedness. Weassessed the effect of unequal male reproductive success, inbreeding and maternal effects on h²estimates.Results showed that in presence of unequal male reproductive success, neglecting thegenetic relatedness between families led to bias the h² estimates derived from the family model.Regarding the animal model, the method using mean marker-based relatedness coefficients wasefficient to deal with non-random mating system and ancestral inbreeding in the parentalpopulation. But this method was biased in presence of maternal effects. In such cases, only the useof finest genetic information, like pedigrees, allowed to improve the accuracy of the h² estimates.Using a hybrid approach, that uses both the pedigree and mean relatedness information, wasefficient to deal with any departures from the family assumptions.In conclusion, family structure designs are part of the experimental exceptions, with inbredsamples, where pedigree-free, or partial pedigree-free (also called hybrid approach), methods can beused to get accurate h² estimates, at the condition that mean relatedness coefficients are used(estimated at the family/population level).