Calculating genomic breeding values for growth using DNA pools within commercially reared black tiger shrimp (Penaeus monodon)

Genomic estimated breeding values (gEBVs) are routinely used in genomic selection; however, their practical application in shrimp aquaculture is impeded by high genotype costs and uneven family contributions. DNA pooling, as a means of reducing the genotyping costs, has been employed to generate hybrid Genomic Relationship Matrixes (h-GRMs) which contain allele frequencies of individual and pooled DNA samples. By using a h-GRM and phenotype data within breeding algorithms, parental gEBVs can be produced, however this remains untested for commercial shrimp data. Within, we evaluated a h-GRM approach using pooled data to predict genomic estimated breeding values (gEBVs) for body weight in Penaeus monodon. Reconstructed pools (RPs) of 2, 5, 10, 15, 20, and 25 individuals ranked by body weight across the population were generated in-silico using individual genotypes across 5,097 DArTCap SNPs. In addition, the in-silico predictions from pool size 10 (PS10) were tested using physical DNA pools (PDP) genotyped using a custom Axiom 70k SNP array. Across the RP and PDP pools, gEBVs were estimated using a GBLUP animal model and the h-GRM approach in three scenarios. First, parental gEBVs predicted from RPs achieved high accuracy in a small pool size (0.92 ± 0.005, PS2), but declined with larger pool size (0.82 ± 0.07, PS25). Second, sibling gEBV prediction accuracy was high with a small pool size (0.94 ± 0.15 for PS2) but decreased more substantially with larger pool sizes (0.58 ± 0.62 for PS25) across ponds. Within the top 20% of siblings, the accuracy ranged from 0.67 ± 0.57 (PS2) to 0.48 ± 1.59 (PS25). PS10 provided an optimum balance between cost and accuracy of the RP data and therefore was used for the PDP validations. Third, PDP prediction accuracy for PS10 in a moderate-diversity pond (high genetic variance, h2 = 0.29, accuracy = 0.53 ± 0.013) was compared with those in a low-diversity pond (low genetic variance, h2 = 0.19, accuracy = 0.43 ± 0.01). Prediction accuracies were 5% and 18% higher with moderate genetic diversity pool data in both the RP and PDP training datasets than with low genetic diversity pond data. Pooling DNA from ten ranked individuals reduced genotyping costs tenfold, however prediction accuracy decreased by 27% – 30% compared to calculating gEBVs using individual genotyping. Although pooling reduces accuracy, having a structured pool design and moderate genetic diversity in reference populations enables reliable gEBV estimation, validating DNA pooling as a cost-effective tool for genomic selection in commercial applications for P. monodon.