Genetic parameters and accuracy of selection for resistance to White Spot Syndrome Virus (WSSV) in Penaeus (Litopenaeus) vannamei using different statistical models

Genetic parameters for resistance to White Spot Syndrome Virus (WSSV) in the shrimp species Penaeus vannamei were estimated by using five different statistical models to analyze challenge test data. Data were recorded on the offspring of 338 full-sib families experimentally infected with WSSV, corresponding to four consecutive generations. Both the linear model (LBM) and the threshold model (TBM) defined disease resistance based on whether or not the animal was alive when the population reached 50% total mortality. The Cox (CM) and the Weibull (WM) proportional hazard frailty models were based on time until death (days post infection) and took censored observations into account. Finally, the linear repeatability model (LRM), considered test-day survival and censoring; where for every animal a binary record was defined for each test day up till the day of death (0 if still alive or 1 if dies, at the specific day). LBM and TBM measured the probability of surviving, whereas the CM, WM and LRM measured the risk of dying. Heritability estimates ranged from 0.01 (CM and LRM) to 0.21 (WM). The rank correlations between full-sib estimated breeding values (EBVs) from the LBM and TBM was close to 1, but lower between EBVs of these two models and the other models (ranging from - 0.82 to - 0.89). We attempted to predict the accuracy of selection that would be obtained with each model by calculating Pearson correlation coefficients between the full-sib EBVs estimated with data from different tanks. The highest accuracy of selection was found in the CM (0.79) followed by the WB and LRM (0.77 and 0.75 respectively). However the WM did not model properly the mortality pattern of the test population. Lowest correlations were found in the LBM and TBM (both 0.67). Based on these results we suggest selection programs for white spot resistance should be based on EBVs using models taking time to death into account with proper model of the mortality pattern of the test population (CM and LRM), rather than the models that define survival as a binary trait at 50% mortality (LBM and TBM).