Sea lice (L. salmonis) is an ecto-parasite that occurs naturally on wild and farmed Atlantic salmon. This parasite is one of the major threats for the farmed A. salmon industry in Norway, causing huge economic losses due to frequent treatment costs and increased mortality due to these treatments. The parasite load trait has shown low to moderate heritability with polygenic architecture of the trait reported in multiple studies 1-4. There is a scarcity of studies showing significant signal(s) of quantitative trait loci (QTL) for lice count possibly due to lack of sufficient statistical power with low sample sizes and low lice counts.
The aim of the current study was to estimate genetic variation for lice counts obtained from three lice challenge tests belonging to the two year-classes, perform genome-wide association analyses of these data, and test the accuracy of genomic predictions from across and within population validation schemes.
Material and Methods
MOWI GENETICS routinely performs controlled challenge tests for lice on full- and half-sibs of breeding candidates and that smoltify at ˂ one year (S0) and/or ˃ one year (S1) of age. The recorded individuals were from two year-classes (YC); the parent YC-2018 (2825 “S1” individuals from 316 full-sib families) and the offspring YC-2022 (2329 “S0” and 2319 “S1” individuals representing 248 and 238 families, respectively) with family size ranging from 1 to 19. All 7473 individuals were genotyped using a custom developed Affymetrix axiom ~65K SNPs genotyping array. Moreover, tank (two or three per test), lice counter (3-6 for each of the three tests) and the body weight of the fish at the lice counting were also recorded. The phenotypic distribution of the lice count data was positively skewed and therefore the lice counts were log transformed to make them more normally distributed.
Analyses: The estimates of genetic parameters were obtained from the following linear mixed animal model(s) using genomic relationship matrix implemented in ASREML v4.2.
where is the vector of log transformed lice count values ; is the overall mean; and are a design matrix to relate the animal records to fixed effects and genetic values, respectively; is a vector of fixed effects of tank, lice counter and body weight of the fish as a covariate; is a vector of random additive genetic effects , where is the genetic variance, is the genomic relationship matrix computed using VanRaden’s method 1; and is the vector of random residual effects with . The three datasets (“S0” of YC-2018; “S0” of YC-2022, and the “S1” data of YC-2022) were analyzed separately, and as a single trait by combining datasets, and with a bivariate model considering the two year-class specific values as two different traits. Additional fixed effects of year-class and smolt type (“S0” and “S1”) were also added when both year classes were analyzed together.
Genomic breeding values (GEBVs) were computed using two different matrices derived from two sets of SNPs; G1) all SNPs regardless of linkage disequilibrium (LD) phase among parent and progeny year-classes, or G2) a subset of the SNPs selected based on consistent LD phase among the two year-classes. A fivefold cross validation scheme was designed and used to evaluate the accuracy of predictions across the year-classes with multiple scenarios i.e., across and within year-class predictions.
The GWAS analysis was performed with the ) of each dataset separately and by combining datasets from both year-classes as the same trait with GCTA software using the “--mlma-loco” function 5.
Results and Discussion
The heritability estimates for was 0.25±0.03 (YC-2018) and 0.20±0.02 (YC-2022). The genetic correlation for recorded on “S0” and “S1” populations of YC-2022 was high with estimates of 0.88±0.07 indicating possibility to analyze “S0” and “S1” of YC-2022 as a single trait. However, the genetic correlation of for the two year-classes was medium (0.69±0.06). The GWAS analysis revealed consistent strong signals of multiple QTLs across the two year-classes located at chromosomes 2, 5, 11 and 25. There were also inconsistent signals of QTLs detected, e.g., single SNP at chromosome 14 crossing significant line (Figure 1). Overall, low proportions of the genetic variances were captured by the different QTLs (calculated as , Falconer and Mackay,19966); e.g. the highest significant SNP at chromosome 2 explained 4.66% of the genetic variance.
The mean accuracy of the across year-class predictions of the GEBVs using either of the two types of G-matrices was generally low with magnitude of 0.22 and 0.42 using G1 and G2, respectively. The gain in accuracy for the across year-class prediction using G2 was approximately doubled as compared to using G1, possibly due to refining trait specific hidden relationship between the two year-classes. The mean accuracy of the within year-class prediction was 0.60 and 0.58 using G1 and G2 matrices, respectively. Hence, in the case of within year-class predictions the use of G2 was not beneficial resulting in slightly lower accuracy of prediction as compared to the use of G1.
In conclusion, lice count in A. salmon showed moderate heritability and a polygenic trait architecture, the accuracy of across-year class prediction was not advantageous compared to within year-class predictions (the latter of which is currently used in aquaculture breeding programs). GWAS revealed consistent signals of QTLs across the parent and the progeny year-classes.
The study is a part of the NOLICE project funded by the Research Council of Norway with grant agreement number of 320619.
1. Kolstad, K., et al., Aquaculture 247, 145-151 (2005). 2. Gjerde, B., et al., Aquaculture 314, 66-72 (2011). 3. Tsai, H.-Y. et al. GSE 48, 47 (2016). 4. Odegård, J. et al., Front Genet. 5, 402 (2014). 5. Yang, J., et al., Am. J. Hum. Genet. 88, 76-82 (2011). 6. Falconer, D.S. & Mackay, T.F.C. Introduction to Quantitative Genetics, (Longman Green, Harlow, Essex, UK, 1996).