Genomic selection for heterobothriosis resistance concurrent with body size in the tiger pufferfish, Takifugu rubripes

Abstract

Parasite resistance traits in aquaculture species often have moderate heritability, indicating the potential for genetic improvements by selective breeding. However, parasite resistance is often synonymous with an undesirable negative correlation with body size. In this study, we first tested the feasibility of genomic selection (GS) on resistance to heterobothriosis, caused by the monogenean parasite Heterobothrium okamotoi, which leads to huge economic losses in aquaculture of the tiger pufferfish Takifugu rubripes. Then, using a simulation study, we tested the possibility of simultaneous improvement of parasite resistance, assessed by parasite counts on host fish (HC), and standard length (SL). Each trait showed moderate heritability (square-root transformed HC: h2 = 0.308 ± 0.123, S.E.; SL: h2 = 0.405 ± 0.131). The predictive abilities of genomic prediction among 12 models, including genomic Best Linear Unbiased Predictor (GBLUP), Bayesian regressions, and machine learning procedures, were also moderate for both transformed HC (0.248‒0.344) and SL (0.340‒0.481). These results confirmed the feasibility of GS for this trait. Although an undesirable genetic correlation was suggested between transformed HC and SL (rg = 0.228), the simulation study suggested the desired gains index can help achieve simultaneous genetic improvements in both traits.

Genomic prediction for testes weight of the tiger pufferfish, Takifugu rubripes, using medium to low density SNPs

“…In conclusion, our data revealed that GS is available for improvement of TW as is the case for SL and BW in selective breeding of the tiger pufferfish 22 . A medium size dataset of SNPs (4075 SNPs) is sufficient for accurate prediction, but the number of SNPs can be reduced to 1200 without much loss of accuracy.…”

“…Atlantic salmon ( Salmo salar ; length: 0.61, weight: 0.60) 29 , common carp ( Cyprinus carpio ; length: 0.33) 30 , Nile tilapia ( Oreochromis niloticus ; weight: 0.36) 31 , channel catfish ( Ictalurus punctatus ; weight: 0.34) 32 , large yellow croaker ( Larimichthys crocea ; length: 0.59, weight: 0.60, gonad weight: 0.37) 33 , 34 , yellowtail kingfish ( Seriola lalandi ; length: 0.43, weight: 0.47) 35 , and the tiger pufferfish (length: 0.41) 22 . Prediction accuracy was relatively high for all of the traits (prediction accuracy: 0.665‒0.792), but also within the range of previous studies: Atlantic salmon (weight: 0.70, length: 0.66) 29 , common carp (length: 0.71) 30 , Nile tilapia (weight: 0.60) 31 , channel catfish (weight: 0.37) 32 , large yellow croaker (length: 0.40, weight: 0.41) 33 , yellowtail kingfish (length: 0.67, weight: 0.69) 35 , and the tiger pufferfish (length: 0.73) 22 . The high heritability and prediction accuracy indicate that GS can be applied for selective breeding on TW in addition to SL and BW of the tiger pufferfish.…”

“…Aquaculture production using these methods now dominates (approximately 90%) overall production of this species (Japan Fisheries Research and Education Agency (FRA), Japan: http://abchan.fra.go.jp/digests2019/index.html , accessed 12 Nov 2020). Although a selective breeding program for this species is still in its infancy 17 , the possibility of using GS for SL, BW, and for resistance against the monogenean parasite, Heterobothrium okamotoi , has already been tested 22 . Precociousness is another important economic trait in this species 23 .…”

Research advances and future perspectives of genomics and genetic improvement in allotetraploid common carp

“…GS predicts GEBVs using whole‐genome molecular markers for individuals from test populations with only genotype data and for individuals from reference populations with both phenotypes and genotype data. This technology has been implemented in increasing numbers of species and has led to successful rapid genetic improvement, especially for traits with low heritability and traits affected by complex and multiple minor loci 214–218 . After several years of exploring the GS algorithms, there have been multiple genomic prediction models, and these can be divided into three main categories: direct models such as GBLUP, indirect models such as Bayesian models and machine learning models such as random forests.…”

“…This technology has been implemented in increasing numbers of species and has led to successful rapid genetic improvement, especially for traits with low heritability and traits affected by complex and multiple minor loci. [214][215][216][217][218] After several years of exploring the GS algorithms, there have been multiple genomic prediction models, and these can be divided into three main categories: direct models such as GBLUP, indirect models such as Bayesian models and machine learning models such as random forests. Which model is a better fit depends on their predictive abilities and correlations between the simulated and actual data sets.…”