Integrating functional annotation data into genomic prediction models for viral nervous necrosis resistance in European sea bass (Dicentrarchus labrax)

Viral nervous necrosis (VNN) represents a major threat for European sea bass (Dicentrarchus labrax) aquaculture, and selective breeding for VNN resistance represents a sustainable long-term control strategy. In particular, genomic prediction enables the evaluation of disease resistance by estimating genetic merit (genomic estimated breeding values, GEBVs) without the need for routine large-scale challenge tests. A key limitation of genomic prediction models is the progressive loss of predictive accuracy as genetic relatedness between the training population and selection candidates decreases, particularly after several generations. Because resistance to viral diseases is often influenced by regulatory variation, incorporating functional annotation data into genomic prediction models may improve their predictive performance. However, this hypothesis remains largely untested in non-model aquaculture species with limited functional genomic resources. In this study, we assessed whether functional annotation of SNPs improves genomic prediction accuracy of genetic merit for VNN resistance in juvenile European sea bass. An experimental population (N=990) generated from an NNV-free commercial broodstock using a full-factorial mating design was subjected to a 29-day challenge test using the red-spotted grouper nervous necrosis virus (RGNNV) strain. Resistance phenotypes were recorded as a binary trait. Experimental fish were genotyped (27,740 SNPs) and then imputed to whole-genome using the whole-genome sequences of their parents (~ 6 millions SNPs). SNPs were classified into functional categories based on genomic annotation and chromatin accessibility data, distinguishing variants located in open chromatin, promoter and enhancer regions from non-functional SNPs located in quiescent regions. Genomic prediction of genetic merit for VNN resistance was then performed using functionally filtered SNP datasets on the basis of functional information and Bayesian threshold models. A 2-fold cross-validation minimizing genetic relatedness between training and testing sets was used to assess model accuracies, whereas the performance of GEBVs in classifying the binary phenotype was evaluated using receiver operating characteristic (ROC) curves and area under the curve (AUC) metrics. Models incorporating functionally annotated SNPs consistently showed improved discrimination between resistant and susceptible individuals compared with models relying on non-functional SNPs or treating all variants equally. In particular, SNPs located in open chromatin and regulatory regions contributed disproportionately to predictive ability, indicating that regulatory variation plays an important role in VNN resistance. These results demonstrate that integrating functional genomic information into genomic prediction models can enhance prediction accuracy for VNN resistance in European sea bass. More broadly, this study provides empirical evidence that SNP prioritization based on functional annotation data may be a viable strategy to improve genomic selection accuracy for complex traits in non-model aquaculture species across multiple generations, when genetic relatedness between the reference population and selection candidates is reduced.