A cost-effective approach for harnesing genetic selection with the power of epigeneteic assessment in animal production

In animal production, genetic selection (GS) has been used to improve the desired traits for decades, without taking into account the full contribution of environmental factors and epigenetic mechanisms, which now are known also to contribute to phenotypic variability. The development of epigenetic biomarkers (EBs) using machine learning (ML) algorithms requires data from hundreds to thousands of individuals. However, in the context of animal production this is a limiting factor due to costs, time and space. Here we propose a new method for developing EBs, which allows the integration of both genetic and epigenetic selection strategies in a cost-effective manner. We used several families of European sea bass, a major aquaculture in the Mediterranean region, already genetically-selected for higher growth as a model to test our approach. By screening hundreds of thousands of CpGs in a small sample set (n = 20 pools made of 64 individual samples), the study aimed to identify a few candidate features holding the potential to further improve sea bass production. To overcome the use of small sample sizes we used a combination of strategies such as careful sample selection, statistical filtering, combinations of feature selection methods and ML algorithms. We reveal that by examining the methylation pattern of three particular CpGs, we could predict which fish were the best among the genetically-selected based on attained biomass at a given time (a combination of survival and growth), proportion of females (the desired sex) and resistance to the masculinizing effects of environmental high temperature. Conversely, the fish whose CpGs did not follow that pattern were consistently classified as worst by the algorithm. Additionally, such CpGs showed high heritability both in male and female offspring. Finally, we further validated them in three independent ways using: 1) a larger set of samples from additional families (accuracy = 90%), 2) sperm samples of the future broodstock (accuracy = 100%), and 3) a set of males of a completely independent origin and without being subjected to GS (accuracy = 88%). Our approach allows incorporating epigenetic selection into existing breeding programs to speed up the production of animals with not only the desired genetics but also the proper epigenetics to deal with a particular environment. Furthermore, we argue that this approach could be applied in other situations where sequencing hundreds of items may not be feasible.