The gilthead sea bream (Sparus aurata) has long been considered resistant to viral nervous necrosis (VNN), until recently, when significant mortalities caused by a reassortant nervous necrosis virus (NNV) strain were reported. Selective breeding to enhance resistance against NNV might be a preventive action. In this study, 972 sea bream larvae were subjected to a NNV challenge test and the symptomatology was recorded. All the experimental fish and their parents were genotyped using a genome-wide single nucleotide polymorphism (SNP) array consisting of over 26,000 markers. Estimates of pedigree-based and genomic heritabilities of VNN symptomatology were consistent with each other (0.21, highest posterior density interval at 95% (HPD95%): 0.1–0.4; 0.19, HPD95%: 0.1–0.3, respectively). The genome-wide association study suggested one genomic region, i.e., in linkage group (LG) 23 that might be involved in sea bream VNN resistance, although it was far from the genome-wide significance threshold. The accuracies (r) of the predicted estimated breeding values (EBV) provided by three Bayesian genomic regression models (Bayes B, Bayes C, and Ridge Regression) were consistent and on average were equal to 0.90 when assessed in a set of cross-validation (CV) procedures. When genomic relationships between training and testing sets were minimized, accuracy decreased greatly (r = 0.53 for a validation based on genomic clustering, r = 0.12 for a validation based on a leave-one-family-out approach focused on the parents of the challenged fish). Classification of the phenotype using the genomic predictions of the phenotype or using the genomic predictions of the pedigree-based, all data included, EBV as classifiers was moderately accurate (area under the ROC curve 0.60 and 0.66, respectively). The estimate of the heritability for VNN symptomatology indicates that it is feasible to implement selective breeding programs for increased resistance to VNN of sea bream larvae/juveniles. Exploiting genomic information offers the opportunity of developing prediction tools for VNN resistance, and genomic models can be trained on EBV using all data or phenotypes, with minimal differences in classification performance of the trait phenotype. In a long-term view, the weakening of the genomic ties between animals in the training and test sets leads to decreased genomic prediction accuracies, thus periodical update of the reference population with new data is mandatory.

中文翻译：

---

金头鲷 (Sparus aurata) 幼体阶段的病毒性神经坏死抗性：不同训练和测试设置下基因组预测的遗传性和准确性

长期以来，金头鲷 (Sparus aurata) 一直被认为对病毒性神经坏死 (VNN) 具有抵抗力，直到最近才报道了重配神经坏死病毒 (NNV) 毒株导致的大量死亡事件。选择性育种以增强对 NNV 的抵抗力可能是一种预防措施。在这项研究中，对 972 只海鲷幼虫进行了 NNV 攻击试验，并记录了症状。使用由超过 26,000 个标记组成的全基因组单核苷酸多态性 (SNP) 阵列对所有实验鱼及其亲本进行基因分型。VNN 症状学的基于谱系和基因组遗传力的估计值相互一致（0.21，最高后验密度区间为 95%（HPD95%）：0.1-0.4；0.19，HPD95%：分别为 0.1-0.3）。全基因组关联研究提出了一个基因组区域，即。例如，在可能与鲷鱼 VNN 抗性有关的连锁群 (LG) 23 中，尽管它远未达到全基因组显着性阈值。三个贝叶斯基因组回归模型（贝叶斯 B、贝叶斯 C 和岭回归）提供的预测估计育种值 (EBV) 的准确度 (r) 是一致的，在一组交叉验证中评估时平均等于 0.90 （简历）程序。当训练集和测试集之间的基因组关系最小化时，准确性会大大降低（对于基于基因组聚类的验证，r = 0.53，对于基于受挑战父母的留一法的验证，r = 0.12鱼）。使用表型的基因组预测或基于谱系的基因组预测对表型进行分类，包括所有数据，EBV 作为分类器的准确度适中（ROC 曲线下面积分别为 0.60 和 0.66）。对 VNN 症状学遗传力的估计表明，实施选择性育种计划以增加鲷鱼幼虫/幼体对 VNN 的抗性是可行的。利用基因组信息提供了开发 VNN 抗性预测工具的机会，并且可以使用所有数据或表型在 EBV 上训练基因组模型，并且性状表型的分类性能差异最小。从长远来看，训练和测试集中动物之间基因组联系的减弱会导致基因组预测准确性下降，因此必须使用新数据定期更新参考种群。对 VNN 症状学遗传力的估计表明，实施选择性育种计划以增加鲷鱼幼虫/幼体对 VNN 的抗性是可行的。利用基因组信息提供了开发 VNN 抗性预测工具的机会，并且可以使用所有数据或表型在 EBV 上训练基因组模型，并且性状表型的分类性能差异最小。从长远来看，训练和测试集中动物之间基因组联系的减弱会导致基因组预测准确性下降，因此必须使用新数据定期更新参考种群。对 VNN 症状学遗传力的估计表明，实施选择性育种计划以增加鲷鱼幼虫/幼体对 VNN 的抗性是可行的。利用基因组信息提供了开发 VNN 抗性预测工具的机会，并且可以使用所有数据或表型在 EBV 上训练基因组模型，并且性状表型的分类性能差异最小。从长远来看，训练和测试集中动物之间基因组联系的减弱会导致基因组预测准确性下降，因此必须使用新数据定期更新参考种群。