Introduction
In aquaculture species, traditional selective breeding for complex traits based on estimated breeding values (EBVs) is often not feasible due to the costs and challenges associated with phenotyping. While genomic selection (GS) leveraging high-density SNP genotype data offers a more efficient approach to improve complex traits, whole-genome sequencing (WGS) has frequently yielded negligible accuracy gains in genomic prediction (Perez-Enciso et al., 2017). This is mainly caused by redundant, non-informative variants, generating noise that masks real informative variants. To "rescue" the WGS potential in GS, this study investigates, for the first time in an aquaculture species, a variant prioritization strategy based on the Functional-And-Evolutionary Trait Heritability (FAETH) score (Xiang et al., 2019). This approach was applied to the European sea bass (Dicentrarchus labrax) to refine genomic prediction for a complex trait, disease resistance, with viral nervous necrosis (VNN) resistance as a representative case study.
Materials and Methods
990 sea bass were phenotyped for body length (mm) and binary VNN mortality following a 29-day experimental challenge. Experimental fish were genotyped using the Med-Fish SNP array (27,740 SNPs; Peñaloza et al., 2021) and imputed to WGS (6,072,853 SNPs) using a reference panel of sequenced parents (25 sires and 25 dams) and 40 experimental fish. Six functional and evolutionary categories, namely ATAC/ChIP-seq data (from the AQUA-FAANG project; https://www.aqua-faang.eu/), Ensembl regulatory features, Variant Effect Predictor consequences, genomic islands of differentiation between the Atlantic and Mediterranean bass populations, linkage disequilibrium (LD) and minor allele frequency score quartiles, were used to partition the WGS dataset into 30 variant sets. Each set was converted into a genomic relationship matrix to estimate the SNP-based heritability () using GCTA (Yang et al., 2011). The estimate was divided by the number of variants in each set, yielding a mean per-variant heritability (). The FAETH score was calculated as the mean of across both traits and all six categories. Variant subsets including SNPs with the highest FAETH scores (top 1%, 5%, 10%, 25%, and 50%) were used in prediction of genomic estimated breeding values (GEBVs) for VNN resistance and compared with the 25% lowest-ranking SNPs. GEBVs were obtained using Bayesian threshold models (GIBBSf90+/BLUPf90+; Misztal et al., 2018). Eight scenarios were considered: four LD-pruning filters (no filtering, and thresholds of 0.99, 0.95 and 0.80), each combined with either the inclusion or exclusion of a major QTL for VNN resistance on chromosome 3 (Mukiibi et al., 2025). A 2-fold cross-validation, minimising genetic relatedness between training and testing sets, was employed, and prediction accuracy was calculated as , where corr(EBV, GEBV) is the Pearson correlation between EBV and GEBV, PEV is the prediction error variance and is the additive genetic variance of the trait. Genomic predictions were replicated 100 times for each variant subset, each time randomly sampling 2,000 SNPs from the target subset.
Results
For VNN mortality and body length, estimates ranged from 0.35 to 0.42 and from 0.40 to 0.56, respectively. Regarding , nine of the top ten sets belonged to the ATAC/ChIP seq functional category. Prediction accuracies consistently improved when using high FAETH-ranked variant subsets across all eight scenarios. In the "No LD pruning" scenario, accuracy increased by 1.18% (top 50%) to 6.90% (top 1%) relative to the subset consisting of 25% low-ranking SNPs. In the "LD pruning 0.99" and "LD pruning 0.95" scenarios, similar predictive gains were observed, ranging from +3.30% (top 25%) to +5.41% (top 10%), and from +3.94% (top 25%) to +6.93% (top 1%), respectively. Under the "LD pruning 0.80" scenario, gains were even larger, ranging from +4.05% (top 25%) to +8.10% (top 10%). Excluding chromosome 3 reduced overall prediction accuracy by approximately 3-4%; however, the trend of increased accuracy remained consistent.
Discussion
This study represents the first attempt in an aquaculture species to apply a variant prioritization strategy that leverages WGS data for genomic prediction of a complex trait using the FAETH score classification. Genomic prediction accuracies for VNN resistance improved consistently for high-ranking subsets, across all eight scenarios. Notably, we observed an increasing trend in accuracy as data was restricted to the highest-ranking variants and under more stringent LD pruning scenarios. We hypothesize that subsets consisting of top-ranking variants are less susceptible to redundancy-induced noise and exhibit a greater capacity to preserve performance compared to the low-ranking variant subset. These findings suggest that the FAETH score might be a reliable strategy to enhance genomic prediction accuracy for complex traits in aquaculture species.
Acknowledgment
This work was funded by the European Union – PNRR-CN5 Spoke 2 (CUP C93C22002810006), Horizon 2020 AQUAFAANG (grant agreement 817923) and Horizon Europe EUAqua.Org (grant agreement 101181589).
References
Misztal I., Tsuruta S., Lourenco D., et al. 2018. Manual for BLUPf90 Family of Programs. http://nce.ads.uga.edu/wiki/lib/exe/fetch.php?media=blupf90_all7.pdf.
Mukiibi R, Ferraresso S, Franch R, et al. 2025. Integrated functional genomic analysis identifies regulatory variants underlying a major QTL for disease resistance in European sea bass. BMC Biol. 23:75.
Peñaloza C, Manousaki T, Franch R, et al. 2021. Development and testing of a combined species SNP array for the European seabass (Dicentrarchus labrax) and gilthead seabream (Sparus aurata). Genomics. 113:2096-2107.
Pérez-Enciso M, Forneris N, De Los Campos G, et al. 2017. Evaluating Sequence-Based Genomic Prediction with an Efficient New Simulator. Genetics. 205:939–53.
Xiang R, Berg IVD, MacLeod IM, et al. 2019. Quantifying the contribution of sequence variants with regulatory and evolutionary significance to 34 bovine complex traits. Proc Natl Acad Sci USA. 116:19398–408.
Yang J, Hong Lee S, Goddard ME, et al. 2011. GCTA: a tool for Genome-wide Complex Trait Analysis. Am J Hum Genet. 88:76-82.