Jukuri, open repository of the Natural Resources Institute Finland (Luke) All material supplied via Jukuri is protected by copyright and other intellectual property rights. Duplication or sale, in electronic or print form, of any part of the repository collections is prohibited. Making electronic or print copies of the material is permitted only for your own personal use or for educational purposes. For other purposes, this article may be used in accordance with the publisher’s terms. There may be differences between this version and the publisher’s version. You are advised to cite the publisher’s version. This is an electronic reprint of the original article. This reprint may differ from the original in pagination and typographic detail. Author(s): Federico C. F. Calboli, Terhi Iso-Touru, Oliver Bitz, Daniel Fischer, Antti Nousiainen, Heikki Koskinen, Miika Tapio, Ilma Tapio & Antti Kause Title: Genomic selection for survival under naturally occurring Saprolegnia oomycete infection in farmed European whitefish Coregonus lavaretus Year: 2023 Version: Published version Copyright: The Author(s) 2023 Rights: CC BY 4.0 Rights url: http://creativecommons.org/licenses/by/4.0/ Please cite the original version: Federico C F Calboli, Terhi Iso-Touru, Oliver Bitz, Daniel Fischer, Antti Nousiainen, Heikki Koskinen, Miika Tapio, Ilma Tapio, Antti Kause, Genomic selection for survival under naturally occurring Saprolegnia oomycete infection in farmed European whitefish Coregonus lavaretus, Journal of Animal Science, Volume 101, 2023, skad333, https://doi.org/10.1093/jas/skad333 Journal of Animal Science, 2023, 101, 1–16 https://doi.org/10.1093/jas/skad333 Advance access publication 1 October 2023 Animal Genetics and Genomics Genomic selection for survival under naturally occurring Saprolegnia oomycete infection in farmed European whitefish Coregonus lavaretus Federico C. F. Calboli,† Terhi Iso-Touru,† Oliver Bitz,† Daniel Fischer,† Antti Nousiainen‡, Heikki Koskinen‡, Miika Tapio,† Ilma Tapio,† and Antti Kause†,1 †Natural Resources Institute Finland (LUKE), FI-31600 Jokioinen, Finland ‡Natural Resources Institute Finland (LUKE), FI-70210 Kuopio, Finland 1Corresponding author: antti.kause@luke.fi Abstract Saprolegnia oomycete infection causes serious economic losses and reduces fish health in aquaculture. Genomic selection based on thousands of DNA markers is a powerful tool to improve fish traits in selective breeding programs. Our goal was to develop a single nucleotide polymor- phism (SNP) marker panel and to test its use in genomic selection for improved survival against Saprolegnia infection in European whitefish Coregonus lavaretus, the second most important farmed fish species in Finland. We used a double digest restriction site associated DNA (ddRAD) genotyping by sequencing method to produce a SNP panel, and we tested it analyzing data from a cohort of 1,335 fish, which were measured at different times for mortality to Saprolegnia oomycete infection and weight traits. We calculated the genetic relationship matrix (GRM) from the genome-wide genetic data, integrating it in multivariate mixed models used for the estimation of variance components and genomic breeding values (GEBVs), and to carry out Genome-Wide Association Studies for the presence of quantitative trait loci (QTL) affect- ing the phenotypes in analysis. We identified one major QTL on chromosome 6 affecting mortality to Saprolegnia infection, explaining 7.7% to 51.3% of genetic variance, and a QTL for weight on chromosome 4, explaining 1.8% to 5.4% of genetic variance. Heritability for mortality was 0.20 to 0.43 on the liability scale, and heritability for weight was 0.44 to 0.53. The QTL for mortality showed an additive allelic effect. We tested whether integrating the QTL for mortality as a fixed factor, together with a new GRM calculated excluding the QTL from the genetic data, would improve the accuracy estimation of GEBVs. This test was done through a cross-validation approach, which indicated that the inclusion of the QTL increased the mean accuracy of the GEBVs by 0.28 points, from 0.33 to 0.61, relative to the use of full GRM only. The area under the curve of the receiver–operator curve for mortality increased from 0.58 to 0.67 when the QTL was included in the model. The inclusion of the QTL as a fixed effect in the model increased the correlation between the GEBVs of early mortality with the late mortality, compared to a model that did not include the QTL. These results validate the usability of the produced SNP panel for genomic selection in European whitefish and highlight the opportunity for modeling QTLs in genomic evaluation of mortality due to Saprolegnia infection. Lay Summary Saprolegnia infection causes serious economic losses and reduces fish health in aquaculture. We created a novel set of genetic markers to use in the selective breeding of European whitefish to reduce mortality due to the fungus. Using genetic markers, we estimated how much differ- ent fish traits are determined by genetic variation, and thus what potential traits have to be selected. We observed that resistance to infection was controlled by both a genetic variant with a major effect on mortality and by many other variants with a small effect distributed across the genome. We tested whether we could increase the precision of genomic breeding values used in the selective breeding by explicitly adding the major genetic variant to the analysis, and we observed an increase in precision in our results. We conclude that directly including information about the major genetic variant increases the precision of our predictions, rather than assuming that all genetic variants each explain a small amount of the genetic variation. Key words: ddRAD, genomic selection, infection resistance, Saprolegnia, whitefish Abbreviations: AUC, area under the curve; ddRAD, double digest restriction site associated DNA, GRM, genetic relationship matrix; h2, heritability; GEBV, genomic estimated breeding value; GWAS, Genome-Wide Association Study; MAS, marker-assisted selection; NGS, next generation sequencing; QTL, quantitative trait loci Introduction The advent of high throughput genetic technologies has allowed to genotype animals for hundreds, thousands, or even millions of different DNA markers, and to integrate this data in breeding programs. The availability of genome-wide data has allowed to extension of the genetic-based selection techniques, such as marker-assisted selection (MAS; Lande and Thomson, 1990; You et al., 2020) to more powerful approaches, such as genomic selection (Meuwissen et al., 2001; Goddard and Hayes 2007). Genomic evaluation with tens of thousands of DNA markers has been proven to be an effective selective tool in animal breeding because quantita- tive traits are typically determined by multiple genes with Received May 16, 2023 Accepted September 29, 2023. © The Author(s) 2023. Published by Oxford University Press on behalf of the American Society of Animal Science. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. D ow nloaded from https://academ ic.oup.com /jas/article/doi/10.1093/jas/skad333/7287507 by N atural R esources Institute Finland (Luke) user on 24 O ctober 2023 mailto:antti.kause@luke.fi https://creativecommons.org/licenses/by/4.0/ 2 Journal of Animal Science, 2023, Vol. 101 minor effects on the phenotype. Hence, MAS that specifi- cally modeled only a few major quantitative trait loci (QTL) of quantitative traits in a breeding value evaluation has been in fact largely abandoned in practice (Misztal et al., 2020). For aquaculture species, where the domestication period is short (Houston et al., 2020), the presence of QTL of large effect is surprisingly common for disease resistance traits (Fuji et al., 2007; Moen et al., 2009; Houston et al., 2010; Aslam et al., 2020; Karami et al., 2020; Calboli et al., 2022, Fraslin et al., 2022a; Vallejo et al 2022; Vela-Avitúa et al., 2022, 2023). This observation and all the instances where QTLs of large effects have been observed have induced a renewed interest in the explicit integration of QTL informa- tion in genetic evaluations (Li et al., 2017; Lopes et al., 2017; Nani et al., 2019; Ren et al., 2021; Fraslin et al., 2022a). Saprolegnia is a naturally occurring oomycete pathogen affecting aquaculture species, causing large losses, especially in freshwater growing conditions (Van Den Berg et al., 2013). At present, work has been carried out to identify the presence of genetically different Saprolegnia strains (Sarowar et al., 2019) in different hosts and growing conditions (Sandoval-Sierra et al., 2014), but far less information is available on the genetic resis- tance to Saprolegnia infection in host species (Misk et al., 2022). Our aim in this work was manyfold. First, we aimed to test the feasibility of using next generation sequencing (NGS) to identify a novel panel of single nucleotide polymorphisms (SNPs) for genomic selection for commercial traits in white- fish, an aquaculture species that currently lacks commer- cial/off-the-shelf genetic marker chips that can be used for selection. We developed a SNP panel for European whitefish (Coregonus lavaretus L.) through double digest restriction site associated DNA (ddRAD) sequencing, using the recently published reference genome (De-Kayne et al., 2020). Sec- ondly, we aimed to understand the genetic of mortality Saprolegnia infection, a common, commercially relevant infection. We were especially interested in understanding whether the genetic architecture of resistance is completely polygenic, or is based on one or a few QTLs explaining most of the genetic variance, or a mixture of these architectures. We genotyped a cohort of fish kept in freshwater tanks in the Natural Resources Institute Finland’s breeding program, and we followed their mortality due to Saprolegnia infection to sexual maturity. Finally, we were interested in whether an NGS-based genomic evaluation would provide accurate genomic estimated breeding values (GEBVs) and whether the explicit modeling of any major QTL would be useful in genomic selection-based breeding programs. Materials and Methods Animal care The establishment of progeny families at Natural Resources Institute Finland’s fish breeding facilities followed the proto- cols approved by the Natural Resources Institute Finland’s Animal Care Committee, Helsinki, Finland. Fish stock We analyzed data from the Natural Resources Institute Fin- land’s breeding program, housed in freshwater facilities at Enonkoski research station. Saprolegnia is present in the water source and thus fish were naturally exposed to the pathogen. In November 2018, a cross was carried out between 26 sires and 100 dams that resulted in a cohort of 1,671 fish (all off- spring of the same age), whose data we present here. Mating was carried out for 1 wk, and the eggs from all mating were pooled after the fertilization. The eggs were incubated in a single large incubator. At eyed stage, the eggs were placed in a single indoor tank in which the hatching occurred. These fish are the safe reserve of the main breeding program stock in which the families are typically incubated separately, reared separately in family tanks, individually tagged without a need for genotyping and only then pooled together (Kause et al., 2011). Phenotype data collection Between January 18 and February 1, 2021, a total of 1,671 fish were individually tagged with passive integrated tran- sponders (Biomark GPT12 pit tags) and a fin clip was sampled for genotyping, and the fish were then moved to an outdoor tank to initiate the experiment. The tank was an concreate, flow through tank of size of 63 m2. The farm uses water from an upstream lake, and Saprolegnia occurs naturally in this catchment area and has caused disease outbreaks annually in recent years. At tagging, all fish were weighted – this weight is the “Weight2” trait in our data. The choice of numbering follows the growth seasons, hence “Weight2” is the weight at the second growth season. Following tagging, mortality to Saprolegnia infection was collected daily during weekdays. A dying fish with the growth of fungus on the surface of a fish was captured with a hand net and its id-tag was recorded. If fish died and it did not have fungal growth on it, it was not considered to be dying due to Saprolegnia (Table 1). On May 11, 2022, all fish were measured for body weight “Weight3” and body height to body length ratio “Height/ Table 1. Traits summary. Sample size and mean values for recorded traits, and number of fish alive at measuring date, number of fish found dead to Saprolegnia up to a measuring date, and number of missing fish. Because some fish were lost (likely due to predation) the sample size indicates the actual number of fish that could be scored for mortality at any given time Sample size Mean Weight2 (g) 1,335 279.87 Weight3 (g) 620 756.2 Height/Length3 (dimensionless ratio) 620 0.23 Sample size Incidence # of fish alive # of fish dead due to fungus # of fish missing Mortality3 (count) 836 0.31 576 260 499 Mortality4 (count) 473 0.52 228 245 103 Mortalitytot (count) 733 0.69 228 505 602 D ow nloaded from https://academ ic.oup.com /jas/article/doi/10.1093/jas/skad333/7287507 by N atural R esources Institute Finland (Luke) user on 24 O ctober 2023 Calboli et al. 3 Length3”. Height/Length3 is one of the selected traits in the breeding program, because an elongated shape is preferred (Kause et al., 2011). Mortality due to Saprolegnia up to that day is termed “Mortality3”. Mortality to Saprolegnia were coded as 1 and survival was coded as 0. Mortality to Saprolegnia infec- tion was again collected as observed until November 8, 2022— mortality from time 3 up to that day is the “Mortality4” trait in our data. Mortality to Saprolegnia between time 3 and time 4 was coded as 1 and survival up to time 4 was coded as 0. Finally, we created a synthetic trait, Mortalitytot, as the sum of Mortality3 and Mortality4. All the fish that died due to Saprolegnia during the study were coded as 1 and the fish that survived until time 4 were coded as 0. The fish that were not recaptured after tagging were coded as missing observations for mortality. Fish disappear because of predators such otters and herons. ddRAD library preparation Adapting the protocol of Sala-Lizana and Oono (2018), DNA was extracted from the finclips using DNeasy 96 Blood & Tis- sue kit (Qiagen) following the manufacturer’s protocol. DNA concentration was measured by Nanodrop and normalized to 30 ng/µL concentration. About 500 ng of high-quality DNA per sample (17 µL; 30 ng/µL) was double-digested with two rare cutting restriction enzymes EcoRI-HF and SphI. The restriction was done in a volume of 20 to 17 µL DNA, 0.1 µL EcoRI-HF, SphI0.5 µL, 2 µL cutsmart buffer (10×), and 0.4 µL of molecular grade water at 37 °C for 15 min or 4 h, following heat-inacti- vating for 15 min at 65 °C. The subsequent ligation (add 2.5 µL ligation buffer, 1 µL of each adapter (100 nM), and 0.5 µL T4 ligase (NEB, New England, USA) of non-barcoded adapter was done at 16 °C for 14 min, following heat-inactivation at 65 °C for 15 min. Not incorporated adapters and other small DNA-fragments were purified/removed using SPRIselect mag- netic beads (Beckman Coulter). First, the volume of each sample was adjusted with molecular-grade water to 50 µL and then 40 µL of magnetic beads were added (0.8× to select for fragments longer than 200 bp) following the manufacturer protocol. The purified DNA was resuspended in 25 µL of molecular-grade water. The quantity of each DNA was measured using the Qubitflex dsDNA HS (high sensitive) assay. Each sample was individually barcoded with the Illumina Nextera v2 combinato- rial dual-indexed barcodes (i7 and i5). The indices were attached with 18 cycles of PCR using Phusion polymerase (Fisher Scien- tific) in 20 µL volume and an optimized annealing temperature of 61 °C. The amount of template was 5 µL. Subsequently, the PCR-products were quantified using Qubitflex with a dsDNA HS assay. Only products with a significantly higher amount than the NTC (no template control) were used for sequencing >3 ng/ µL. Sample with a low amount of product was repeated if appli- cable with 1 or 2 more cycles. Sequencing Single libraries were pooled in equimolar amounts and run on a 1.5% TAE gel. Fragments between 400 and 700 bps were cut out of the gel and purified with Qiagen Gel extraction Kit. The quality and size of the final, pooled sequencing library were checked on the Bioanalyzer or TapeStation using DNA HS assays. For sequencing the pooled library was adjusted to a concentration of 1 to 4 nM. Following the guidelines from Illumina, the libraries were diluted to a final concentration of 2.0 pM. The paired-end sequencing (2 × 150 bp) was done on the NextSeq 550 (Illumina). Basecalling and demultiplexing were done with Illumina bcl- 2fastq2 Conversion Software (Illumina) with one index mis- match allowed. For each sample individually, the paired-end fragments were aligned to the reference genome (De-Kayne et al., 2020) using Bowtie (Langmead and Salzberg, 2012), retaining only those fragments that had a unique match to the genome. The selected fragments were sorted with Samtools (Li et al., 2009), which was used to generate a population-wide consensus call for each location of the genome covered by individual sample fragments. The genotyping by sequencing (GBS) pipeline was implemented in Snakemake (Mölder et al., 2021) and is an extension of the GBS SNP-Calling Refer- ence Optional Pipeline (GBS-SNP-CROP) (Melo et al., 2016). This pipeline is publicly available (Fischer, 2023). The raw ddRAD genotyping produced 65,318 SNPs. Following De-Kayne (2022), and after analysis of our data coverage patterns, all SNPs mapping to poor-quality refer- ence assembly chromosomes 22, 28, 32, 35, 38, and 40 were removed. Using a quality control of at least 5% minor allele frequency, and a threshold of 80% genotyping success, 5,242 high-quality SNPs remained for 1,671 samples. Samples miss- ing 50% or more of the genotypes of their SNPs were also removed, leaving us with a final dataset of 1,335 samples genotyped for 5,242 SNPs. In all analyses, each SNP geno- type was coded as the standard 0/1/2 to reflect the number of minor alleles in the genotype (0, homozygote for the major allele; 1, heterozygote; 2, homozygote for the rare allele). In analyses where genotype (for any SNP) was used as a fixed effect in a liner model we transformed the 0/1/2 value as 1/2/3 to use genotype as a qualitative parameter with 3 levels with the software DMU (Madsen et al., 2014; see below). Estimation of heritability and genetic correlations Phenotypic, genetic, and residual (co)variances of fish traits were estimated using the Average Information Restricted Maximum Likelihood module of DMU (DMUAI; Madsen et al., 2014). Heritabilities were estimated with a 2-trait multi- variate model in which Weight2 was always included. Weight2 was available for all fish, whereas the other traits had some missing data, and hence this approach accounts for potential selection bias (Henderson, 1975; Ouweltje et al., 1988). Cor- relations were estimated with a series of 3-trait analyses. The model used for all traits was: Model 1 : yij = µj + aij + eij in which yij is the phenotype of a trait j for the ith individ- ual, µj is the mean value of a trait j, aij is the random genetic effect explained by the realized genomic relationship matrix between genotyped individuals (genetic relationship matrix [GRM]), and eij the random residual error. The GRM was calculated using htginv module of the MiX99 software pack- age (Stranden and Lindauer, 1999) on the full SNP dataset of the offspring fish (‘full GRM’), using the first VanRaden method (VanRaden et al., 2009), without the need of using any correction to obtain an invertible matrix. Missing geno- types were imputed by htginv. Heritability was computed as: h2 = σ2 G/(σ 2 G + σ2 E) with h2 the heritability, σ2 G the genetic variance for a traitj, and σ2 E the residual variance. The sum σ2 G + σ2 E corresponds D ow nloaded from https://academ ic.oup.com /jas/article/doi/10.1093/jas/skad333/7287507 by N atural R esources Institute Finland (Luke) user on 24 O ctober 2023 4 Journal of Animal Science, 2023, Vol. 101 to the overall phenotypic variance (σ2 P). The genetic correla- tion was: corjk = covjk/sqrt Ä σ2 gjxσ2 gk ä with corjk and covjk the genetic correlation and covariance between traitj and traitk, and σ2 gj and σ2 gk the genetic vari- ances for traitj and traitk. The phenotypic correlations were calculated accordingly. Phenotypic correlations were tested for significance using a Pearson’s correlation test. Genetic cor- relations were considered significant when the interval corjk ± 1.96 × SEjk did not include 0. Similar to our analysis here, linear multitrait models are routinely used for binary traits in practical commercial breed- ing value evaluations. Both heritability estimates and selection accuracies from linear models were transformed to account for the binary nature of the traits (see details below). Genetic correlations of binary traits estimated using linear models are expected to be unbiased, whereas phenotypic correlations are biased downwards (Mäntysaari et al., 1991). For mortality traits, the formula of Dempster and Learner (1950) was used to transform the h2 on the observed scale to the heritability on the underlying liability scale. Phenotypic and residual correlations of Mortality3 and Mortality4 can- not be estimated because of the limited number of samples in common between the two phenotypes. Genome-wide association study The statistical software DMU was used for a genome-wide association study (GWAS), using the leave one chromo- some out approach described by Yang et al. (2014). In this approach, the association between genotype and phenotype is carried out using a linear fixed regression model fitting one SNP at a time, but rather than using the full GRM as the random factor for all SNPs, a new partial GRM is calcu- lated every time by removing all the other SNPs on the same chromosome as the SNP in the analysis, calculating a partial GRM on the fly and fitting it as the random effect (Yang et al., 2014). Thus, for GWAS analyses, the following 2-trait model, in which Weight2 was always included, was used: Model 2 : yij = µj + bxin + aij + eij in which b is the fixed regression slope, xin is the genotype of the nth SNP marker of an individual i, a’ij is the random genetic effect explained by the partial GRM, and all the other terms as described before. Weight2 was used again to account for poten- tial selection bias. Significance of the SNP effects was deter- mined using Bonferroni correction (0.05/number of SNPs) for genome-wide significance, and as 0.05/(mean number of SNPs per chromosome) for chromosome-wide significance. The percentage of genetic variance explained by a SNP was calculated following (Lynch and Walsh, 1998), as: σ2 SNP = 2p (1 p) b2/σ2 G in which b is the fixed regression slope from Model 2, p is the frequency of the major allele, σ2 G is the total genetic variance explained by the full GRM (from Model 1). We tested the effect of the presence of a QTL for mor- tality by specifically modeling the fixed effect of SNP LR664349.1_1710215 (the only SNP associated with all mortality traits) along with the remaining GRM effects in the successive analyses. Estimation of genomic breeding values To assess the influence of a large QTL on breeding values of Mortality3, genomic breeding values were estimated in two ways, either excluding or including the top QTL SNP as a fixed effect. We focused only on Weight2, Weight3, and Mor- tality3 to illustrate the modeling approaches. GEBVs were estimated to solve the mix models with MiX99 (Pitkänen et al., 2022). First, GEBVs of all three traits were estimated using Model 1 and genomic best linear unbiased prediction (GBLUP). In this approach, the GEBVs are the result of mod- eling the realized genomic relationship matrix. Second, in the GBLUP+QTL model, also the top SNP was included as a separate fixed factor: Model 3 : yij = µj + QTLi + anoqtl + eij in which QTLi is the individual genotype of the top QTL SNP fitted as a fixed categorical variable, and anoqtl is the polygen- etic genetic effect explained by the partial GRM calculated without the QTL SNP, by first deleting from the genotype data the top QTL SNP in chromosome 6, and then calcu- lating a new partial GRM. To obtain the total GEBV in this approach, the solution of the fixed QTL effect was added to the polygenetic GEBV of an individual obtained via the partial GRM. For both approaches, the (co)variance compo- nents needed as input were estimated using the same models in DMU. The models used were the bivariate models, having Weight2 always as one of the traits in the analysis. The results of Mortality3 are shown for both GBLUP Model 1 and GBLUP+QTL Model 3, and for body weights, only Model 1 is given because the QTL used in Model 3 was not QTL affecting any weight trait. Validation of genomic evaluation with and without explicit QTL modeling The predictive ability of the GEBVs of Mortality3, Weight2, and Weight3 was validated in two different ways. Both val- idation steps were run both on GBLUP model and on the GBLUP+QTL model that included the genotype of the top SNP as a fixed effect. First, a cross-validation approach was used to randomly mask 20% of phenotypes of a trait, and then GEBVs of all fish were estimated. Then, for the masked individuals only, the correlation of their GEBVs to their actual phenotypes was calculated. Accuracy was calculated as the Pearson correlation coefficient between the GEBVs and the observed phenotypic values divided by the square root of the heritability of the trait on the observed scale (Legarra et al., 2008). For Mortality3 we also calculated the area under the curve (AUC) of the corre- sponding receiver–operator curve, that is, we assessed the clas- sification performance of the GEBV in predicting the actual mortality phenotypes. For a completely random performance (no classification power) the AUC is 0.5, whereas a perfect clas- sification corresponds to a value of 1. For cross-validation and calculation of AUCs, we ran 1,000 resampling steps, and the means and standard deviation are presented. When sampling individual fish randomly into the two groups, the reference group and the validation group maintain their close relationship. This is a typical setup when applying D ow nloaded from https://academ ic.oup.com /jas/article/doi/10.1093/jas/skad333/7287507 by N atural R esources Institute Finland (Luke) user on 24 O ctober 2023 Calboli et al. 5 genomic selection in commercial aquaculture breeding pro- grams in which all the families are held in the nucleus and their sibs are recorded for hard-to-record traits such as dis- ease resistance, carcass traits, or product quality. Such close relatedness is used to improve the power of genomic selection in commercial breeding, and our estimates of selection accu- racy are expected to be higher than if the reference group and the validation group were less related (Fraslin et al., 2022b). Secondly, we tested the power of the GEBVs of Mortality3 to predict the phenotype of Mortality4, using all data and the correlation between the GEBVs and phenotypes. Results Heritability estimates For all traits, we observed significant heritability estimates. Heritability was 0.44 to 0.58 for Weight2 and Weight3 (Table 2). Heritability on the observed scale for mortality traits ranged between 0.13 and 0.25, and on the liability scale, it ranged between 0.20 and 0.43 (Table 2). Phenotypic correlation The correlation test allowed us to assess whether traits were significantly correlated at the phenotypic level. For the pro- duction traits, we observed that Weight2 and Weight3 were positively and significantly phenotypically correlated (Table 3) and that there was also a positive correlation of Weight2 and Weight3 with Height/Length3. Therefore, all three produc- tion-related traits showed a strong positive phenotypic cor- relation among themselves. Testing for a correlation between weight traits and mortal- ity, we observed that Weight2 was positively and significantly phenotypically correlated with Mortality3, whereas the cor- relation between Weight2 and Mortality4 was not significant. In contrast, the correlation between Weight2 and Mortalitytot was once again positive and significant. We also observed that Weight3 was not significantly correlated with Mortality3, but it was positive and significant with Mortality4 and with Mor- talitytot. In all cases where the correlation test was significant, the phenotypic correlation between weight and mortality was positive, indicating that at the phenotypic level, heavier fish had a higher mortality. For Height/Length3, we did not observe a significant cor- relation with Mortality3, though we did observe a positive and significant correlation with Mortality4, and then again, we did not observe a significant correlation with Mortalitytot. Finally, all mortality traits were strongly correlated. Mor- tality3 was positively and significantly correlated with Mor- talitytot, and Mortality4 and Mortalitytot have a correlation of 1. Mortality3 and Mortality4 could not have a direct pheno- typic correlation because the samples that reached Mortality4 were all alive at Mortality3, making it impossible to calculate a phenotypic correlation between the two traits. Genetic correlation Genetic correlations broadly followed the phenotypic correla- tions (Table 3), but all the genetic correlations of mortality traits with the weight traits had very large standard errors, thus indicating that the correlations are not significant. On the other hand, Weight2 was positively and significantly genet- ically correlated with Weight3 and Height/Length3 (Table 3). Weight3 was also positively and significantly genetically correlated with Height/Length3. Mortality3 was significantly genetically correlated with Mortality4 (0.90) and Mortalitytot (0.99). The correlation between Mortality4 and Mortalitytot was estimated as 1 with no standard error estimation. These high positive correlations imply that mortality has similar genetic determination across ages. GWAS results Multiple traits and DNA variants showed a genome-wide or chromosome-wide association (Table 4 and Figures 1–6, the linkage disequilibrium between markers is presented in Supplementary Figure S1). Weight2 showed two genome- wide significant peaks: one, significant, on chromosome 4, and a second on chromosome 20 (Figure 1 and Table 4). The amount of genetic variance explained by these genome- wide SNPs ranges between 2.76% and 5.40%. Weight3 does not have any genome-wide significant SNPs, but we observe chromosome-wide significant peaks on chromosome 4, chro- mosome 8, and chromosome 16, with the genetic variance explained by any of these SNPs ranging between 1.77% and 6.30% (Figure 2 and Table 4). Height/Length3 also only has chromosome-wide level asso- ciation, with peaks on chromosome 1, chromosome 4, chro- mosome 20, and chromosome 27, with the genetic variance explained by any of these SNPs ranging between 1.65% and 3.19% (Figure 3 and Table 4). All body weight traits involve chromosome 4, and, to a smaller extent, chromosome 20. For Mortality3, there is a strong genome-wide significant peak on chromosome 6, with 20.3% to 51.3% of the genetic variance explained by the significant SNPs (Figure 4 and Table Table 2. Variance components (σ2 g: genetic variance; σ2 e: environmental variance; σ2 p: phenotypic variance) and heritability for recorded traits (h2, both on the observed and on the liability scales for mortality traits) Trait σ2 g σ2 e σ2 p h2 on the observed scale SE on the observed scale h2 on the liability scale SE on the liability scale Weight2 1,432.14 1,857.12 3,289.26 0.44 0.04 Weight3 15,500.57 13,537.16 29,037.73 0.53 0.05 Height/Length3 1.47E−04 1.05E−04 2.53E−04 0.58 0.05 Mortality3 0.04 0.19 0.23 0.19 0.06 0.33 0.10 Mortality4 0.03 0.23 0.26 0.13 0.07 0.20 0.11 Mortalitytot 0.06 0.17 0.23 0.25 0.06 0.43 0.11 D ow nloaded from https://academ ic.oup.com /jas/article/doi/10.1093/jas/skad333/7287507 by N atural R esources Institute Finland (Luke) user on 24 O ctober 2023 http://academic.oup.com/jas/article-lookup/doi/10.1093/jas/skad333#supplementary-data 6 Journal of Animal Science, 2023, Vol. 101 4). Mortality4 on the other hand does not show any genome- wide significant association, but only 3 chromosome-wide significant SNPs on chromosome 6 and chromosome 8, explaining 18.4% to 30.0% of the genetic variance (Figure 5 and Table 4). Mortalitytot shows one single genome-wide peak on chromosome 6 (the genome-wide significant SNP for Mortality4 on Chromosome 6 is also genome-wide significant for Mortality3 and Mortalitytot), explaining 7.65% to 15.82% of the genetic variance (Figure 6 and Table 4). In all cases, the mortality traits involve chromosome 6, with no overlap with the QTL found for the weight traits. For Mortality3, the regression slope of the most associated locus against mortality (b in QTL+GBLUP model) is −0.18. When the top SNP for Mortality3 is fitted as a categorical fixed effect, the result does not indicate any signs of dom- inance (Figure 7), and the difference between the extreme genotypes is 32% in mortality between the two homozygote genotypes. Validation of genomic evaluation with GBLUP and GBLUP+QTL Genomic breeding values of Mortality3 from the GBLUP and GBLUP+QTL models are shown in Figure 8. In GBLUP model, the GEBVs range between −0.22 and 0.26, and in the GBLUP+QTL model between −0.21 and 0.24. In the cross-validation, for the GBLUP model we observe a mean accuracy of 0.33 for Mortality3 (Figure 9). For the GBLUP+QTL model, the accuracy for mortality at time 3 is much higher, 0.61 (Figure 9). When calculating the AUC val- ues, we observe a mean of 0.58 AUC for the GBLUP model, whereas the GBLUP+QTL resulted in a 0.67 mean AUC (Fig- ure 10). As a comparison, for Weight2 the mean accuracy was 0.56, and for Weight3 the mean accuracy was 0.79 (both GBLUP models). The cross-validation accuracies showed large variation, especially in mortality traits (Figure 9). For binary traits such as mortality, the varying incidence of surviving fish in the reference group (or validation group) due to random sam- pling of the fish into the two groups has a major impact on the accuracy. Moreover, due to the random sampling, the relationships between reference and validation groups may change, impacting accuracies. When we test the predictive ability of GEBVs of Mortal- ity3 for Mortality4, we observe a positive correlation between Mortality3 GEBV with Mortality4 phenotype, of 0.18 for the GBLUP model, and of 0.22 for the GBLUP+QTL model (Table 5). When we assess the predictive ability of Mortal- ity3 GEBV for Mortality4 GEBV, calculated together using one single multivariate model, we observe a positive cor- relation between Mortality3 GEBV and Mortality4 GEBV, of 0.77 for GBLUP model, and 0.91 for the GBLUP+QTL model (Table 5). Discussion Feasibility of using NGS for genomic selection Our first goal, the assessment of the use of NGS for genomic selection in European whitefish was broadly successful. We were able to capture a total of 5,242 SNP markers, an aver- age of 154 SNPs for each chromosome, with an average SNP spacing of 351 kilobases. This result provided sufficient Ta b le 3 . P he no ty pi c co rr el at io ns (l ow er t ria ng ul ar m at rix ) a nd g en et ic c or re la tio ns ± s ta nd ar d er ro r (u pp er t ria ng ul ar m at rix ) W ei gh t 2 W ei gh t 3 H ei gh t/ L en gt h 3 M or ta lit y 3 M or ta lit y 4 M or ta lit y to t W ei gh t 2 – 0. 77 ± 0 .0 5 0. 72 ± 0 .0 6 0. 16 ± 0 13 −0 .1 0 ± 0. 19 0. 05 ± 0 .1 2 W ei gh t 3 0. 63 a – 0. 81 ± 0 .0 4 −0 .2 0 ± 0. 14 0. 17 ± 0 .2 0 0. 12 ± 0 .1 3 H ei gh t/ L en gt h 3 0. 5a 0. 78 a – −0 .1 9 ± 0. 14 0. 05 ± 0 .2 1 −0 .0 1 ± 0. 14 M or ta lit y 3 0. 22 a −0 .0 8d −0 .0 8d – 0. 90 ± 0 .1 8 0. 99 ± 0 .0 4 M or ta lit y 4 −0 .0 3d 0. 17 a 0. 08 c N A – 1 M or ta lit y to t 0. 11 b 0. 20 a 0. 08 d 0. 5a 1 – a P -v al ue < 0 .0 01 . b P -v al ue < 0 .0 1 an d P -v al ue > 0 .0 01 . c P -v al ue < 0 .0 5 an d P -v al ue > 0 .0 1. d N ot s ig ni fic an t. N A , n ot a va ila bl e. D ow nloaded from https://academ ic.oup.com /jas/article/doi/10.1093/jas/skad333/7287507 by N atural R esources Institute Finland (Luke) user on 24 O ctober 2023 Calboli et al. 7 Table 4. Genome-wide significant SNPs associated with traits. P-values shown are Bonferroni corrected, unless otherwise indicated. SNP names reflect the internal nomenclature of our RAD sequencing, chromosome and base-pair position are based on the assembly of De-Kayne et al. (2022) SNP name chr bp Beta SE Sample size t P −log10(p) MAF % genetic variance explained Weight2  � LR664347.1_2201425 4 2,201,425 15.73 2.95 1,095 5.33 1.21E−07 6.92 0.38 4.05  � LR664347.1_2693802 4 2,693,802 13.68 2.61 1,130 5.25 1.83 E−07 6.74 0.47 3.26  � LR664347.1_3146138 4 3,146,138 −15.67 2.55 1,108 6.15 1.09 E−09 8.96 0.49 4.28  � LR664347.1_10995075 4 10,995,075 13.96 2.69 1,120 5.18 2.60 E−07 6.58 0.28 2.76  � LR664347.1_14268077 4 14,268,077 15.88 2.82 1,185 5.62 2.36 E−08 7.63 0.30 3.71  � LR664347.1_14836448 4 14,836,448 14.53 2.38 1,127 6.11 1.33 E−09 8.88 0.34 3.31  � LR664347.1_20091990 4 20,091,990 −18.01 3.84 1,115 4.69 3.12 E−06 5.51 0.39 5.40  � LR664363.1_38855216 20 38,855,216 15.01 3.24 1,166 4.64 3.92 E−06 5.41 0.48 3.93 Weight3  � LR664347.1_3568737 4 3,568,737 −52.35 14.07 498 3.72 2.21 E−04 3.65 0.30 3.72  � LR664347.1_8489919 4 8,489,919 −62.48 16.61 518 3.76 1.88 E−04 3.73 0.42 6.12  � LR664347.1_8489934 4 8,489,934 −63.02 16.67 546 3.78 1.73 E−04 3.76 0.44 6.30  � LR664347.1_8489940 4 8,489,940 −60.67 15.35 544 3.95 8.76 E−05 4.06 0.42 5.80  � LR664347.1_10995075 4 10,995,075 40.77 10.48 527 3.89 1.12 E−04 3.95 0.28 2.17  � LR664347.1_14268077 4 14,268,077 46.80 10.56 562 4.43 1.13 E−05 4.95 0.30 2.98  � LR664347.1_14836448 4 14,836,448 34.93 9.18 518 3.81 1.59 E−04 3.80 0.34 1.77  � LR664347.1_14962376 4 14,962,376 54.65 12.95 521 4.22 2.90 E−05 4.54 0.42 4.70  � LR664351.1_49728274 8 49,728,274 −51.05 13.77 540 3.71 2.31 E−04 3.64 0.24 3.06  � LR664359.1_50667396 16 50,667,396 41.52 10.80 521 3.84 1.37 E−04 3.86 0.27 2.18  � LR664359.1_50667662 16 50,667,662 40.96 10.90 511 3.76 1.92 E−04 3.72 0.26 2.10 Height/Length3  � LR664344.1_44836701 1 44,836,701 0.00 0.00 539 3.63 3.07 E−04 3.51 0.31 2.08  � LR664347.1_3146138 4 3,146,138 0.00 0.00 513 3.70 2.40 E−04 3.62 0.49 1.94  � LR664347.1_8489940 4 8,489,940 −0.01 0.00 544 3.88 1.18 E−04 3.93 0.42 5.48  � LR664347.1_14268077 4 14,268,077 0.00 0.00 562 3.95 8.86 E−05 4.05 0.30 2.32  � LR664347.1_14836448 4 14,836,448 0.00 0.00 518 3.69 2.44 E−04 3.61 0.34 1.65  � LR664363.1_31261053 20 31,261,053 0.00 0.00 543 3.72 2.16 E−04 3.67 0.30 2.30  � LR664363.1_31261188 20 31,261,188 0.00 0.00 521 3.62 3.18 E−04 3.50 0.30 2.01  � LR664363.1_34941325 20 34,941,325 0.00 0.00 514 4.45 1.07 E−05 4.97 0.26 2.95  � LR664363.1_36174451 20 36,174,451 0.00 0.00 538 3.83 1.41 E−04 3.85 0.40 2.60  � LR664363.1_36909591 20 36,909,591 0.00 0.00 507 3.87 1.22 E−04 3.91 0.28 2.57  � LR664363.1_37414088 20 37,414,088 0.00 0.00 553 3.87 1.22 E−04 3.91 0.30 2.51  � LR664363.1_37414306 20 37,414,306 0.00 0.00 541 3.66 2.81 E−04 3.55 0.41 2.48  � LR664370.1_1484511 27 1,484,511 0.01 0.00 525 3.83 1.46 E−04 3.84 0.21 3.19 Mortality3  � LR664349.1_1710215 6 1,710,215 −0.18 0.03 725 6.58 9.22 E−11 10.04 0.39 36.21  � LR664349.1_2924390 6 2,924,390 0.16 0.03 673 5.37 1.08 E−07 6.97 0.30 22.67  � LR664349.1_2924638 6 2,924,638 0.15 0.03 683 5.09 4.57 E−07 6.34 0.30 20.32  � LR664349.1_5024842 6 5,024,842 0.15 0.03 759 5.00 7.30 E−07 6.14 0.43 26.39  � LR664349.1_6154414 6 6,154,414 −0.22 0.03 731 6.85 1.55 E−11 10.81 0.38 51.34  � LR664349.1_11957954 6 11,957,954 −0.21 0.03 728 6.00 3.12 E−09 8.51 0.23 33.92  � LR664349.1_14128348 6 14,128,348 0.15 0.03 710 5.47 6.21 E−08 7.21 0.44 23.83  � LR664349.1_16037751 6 16,037,751 0.15 0.03 733 5.44 7.14 E−08 7.15 0.44 25.66  � LR664349.1_16570835 6 16,570,835 −0.19 0.04 743 5.23 2.25 E−07 6.65 0.23 27.53  � LR664349.1_16619105 6 16,619,105 −0.18 0.04 710 5.14 3.62 E−07 6.44 0.19 23.52  � LR664349.1_21146341 6 21,146,341 0.15 0.03 726 5.06 5.27 E−07 6.28 0.35 24.11 Mortality4  � LR664349.1_1710215 6 1,710,215 −0.16 0.04 419 3.95 9.18 E−05 4.04 0.39 18.39  � LR664351.1_49261152 8 49,261,152 0.22 0.05 431 4.28 2.26 E−05 4.65 0.26 29.97  � LR664351.1_49261323 8 49,261,323 0.20 0.05 449 3.91 1.06 E−04 3.98 0.27 24.73 D ow nloaded from https://academ ic.oup.com /jas/article/doi/10.1093/jas/skad333/7287507 by N atural R esources Institute Finland (Luke) user on 24 O ctober 2023 8 Journal of Animal Science, 2023, Vol. 101 information for the estimation of the pairwise genomic rela- tionship between fish, and thus the GRM. Recent studies based on cross-validations in multiple aquaculture species and different traits imply that around 2,000 to 7,000 SNPs are needed for accurate genomic evaluation (Kriaridou et al., 2020; Fraslin et al., 2022b). The use of a GRM is obviously key for modern genomic evaluations, and our NGS approach allowed us to both fulfill this requirement for genomic selec- tion and to extend our analysis to a GWAS. Yet, for more fine-grained QTL mapping, a higher-density SNP panel would be needed. NGS offers great potential for aquaculture, especially for those species that do not have prior genomic informa- tion and no off-the-shelf SNP chips (Robledo et al., 2017). NGS can be used to identify enough markers to calculate the genomic relationship matrix, bypassing the need to use a pedigree record. The ability to produce genome-wide SNPs is especially valuable because it allows to capture much of the genetic variance, irrespective of what is known of the actual genetic architecture of the target trait(s). The successful use of NGS for this goal has been observed, for instance, in olive flounder (using 12,712 SNPs, Shao et al., 2015), gilthead sea bream (using 12,085 SNPs, Palaiokostas et al., 2016), barra- mundi (using 3,321 SNPs, Wang et al., 2015), bighead carp (using 323 SNPs, Fu et al., 2016), scallop (using 2,364 SNPs, Dou et al., 2016), Japanese sea cucumber (using 5,517 SNPs, Tian et al., 2015), and abalone (using 3,717 SNPs, Ren et al., 2016). The most obvious limitations of NGS are that it is much more computationally and labor intensive than SNP chips, especially in the absence of a reference genome, and that a good overlap between SNP sets between different genotyp- ing runs is never guaranteed, making comparing or combin- ing datasets complex. To obviate these two drawbacks, once a set of SNPs of the desired genome-wide density has been achieved, and a reliable reference genome is available, it is possible to develop custom SNP arrays and/or genotyping panels. SNP name chr bp Beta SE Sample size t P −log10(p) MAF % genetic variance explained Mortalitytot  � LR664349.1_1710215 6 1,710,215 −0.20 0.03 631 6.54 1.28 E−10 9.89 0.39 15.82  � LR664349.1_6154414 6 6,154,414 −0.18 0.03 637 5.06 5.46 E−07 6.26 0.38 12.40  � LR664349.1_14128348 6 14,128,348 0.14 0.03 624 4.75 2.51 E−06 5.60 0.44 7.65  � LR664349.1_21146341 6 21,146,341 0.15 0.03 635 4.68 3.48 E−06 5.46 0.35 9.13 SNP name: provisional SNP id; chr: chromosome; bp; base pair; beta: regression coefficient; SE: regression coefficient’ standard error; Sample size: number of sample in each linear model; t: t-value; P: P-value; −log10(P): −log10(P-value); MAF: minor allele frequency; % genetic variance explained: genetic variance explained by the SNP for the trait. Figure 1. Manhattan plot of the −log10(P-value) of Weight2. A clear, genome-wide significant, QTL peak is visible on Chromosome 4, with a second peak on chromosome 20. Table 4. Continued D ow nloaded from https://academ ic.oup.com /jas/article/doi/10.1093/jas/skad333/7287507 by N atural R esources Institute Finland (Luke) user on 24 O ctober 2023 Calboli et al. 9 Figure 2. Manhattan plot of the −log10(P-value) of Weight3. A chromosome-wide significant QTL peak is visible on Chromosome 4, but the peak does not reach genome-wide significance. Figure 3. Manhattan plot of the −log10(P-value) of Height/Length3. A chromosome-wide significant QTL peak is visible on Chromosome 4, with a second chromosome-wide peak on chromosome 20. Neither peak reaches genome-wide significance. D ow nloaded from https://academ ic.oup.com /jas/article/doi/10.1093/jas/skad333/7287507 by N atural R esources Institute Finland (Luke) user on 24 O ctober 2023 10 Journal of Animal Science, 2023, Vol. 101 Figure 4. Manhattan plot of the −log10(P-value) of Mortality3. A clear, genome-wide significant, QTL peak is visible on Chromosome 6. Figure 5. Manhattan plot of the −log10(P-value) of Mortality4. No QTL peaks are visible. D ow nloaded from https://academ ic.oup.com /jas/article/doi/10.1093/jas/skad333/7287507 by N atural R esources Institute Finland (Luke) user on 24 O ctober 2023 Calboli et al. 11 Genetic characteristics of mortality to Saprolegnia To our knowledge, there are no previous reports on the genomic characterization of mortality for Saprolegnia in fish. Our results show that for each mortality trait, h2 was low to moderate on the observed scale (0.13 to 0.25), and h2 on the underlying liability scale showed a substantial increase (0.20 to 0.43), suggesting that the genetic com- ponent for the mortality traits is actually bigger than what can be directly observed. These estimates are clearly higher than the h2 estimates of general mortality whose causes are unknown (h2 range 0.08 to 0.17 on the liability scale in rainbow trout, Vehviläinen et al., 2008; h2 range 0.07 to 0.20 on the liability scale, also in rainbow trout, Veh- viläinen et al., 2010). We found a major QTL for survival to infection to this oomycete in whitefish, with the SNPs involved in the main QTL each explaining an average of 21% of the genetic vari- ance for mortality. This was in strong contrast to the polygen- etic nature of body weight. Immunity to infection often shows a monogenic or oligogenic architecture in fish (e.g., Fuji et al., 2007; Moen et al., 2009; Houston et al., 2010; Calboli et al., 2022; Fraslin et al., 2022a). This may be because immunity can have a simple mechanism that changes the level of resis- tance, such as the case of infectious pancreas necrosis in Euro- pean Atlantic salmon, which is almost completely determined by the nedd-8 locus (Pavelin et al., 2021), and potentially because aquaculture species have been only recently domes- ticated and thus QTLs of large effects have not been brought to fixation across commercial stocks. In our study, the top SNP shows an additive pattern of effect with 32% difference in mortality between the most resistant homozygote genotype and the most susceptible homozygote genotype, providing a valuable target for selection in aquaculture settings. This is the first major Saprolegnia infection experienced by this fish stock, but the fungus causes major economic losses and reduces fish health in aquaculture. Figure 6. Manhattan plot of the −log10(P-value) of Mortalitytot. A clear, genome-wide significant, QTL peak is visible on Chromosome 6. Figure 7. Allele substitution effects for the top SNP associated with Mortality at time 3. The pattern is consistent with an additive effect of each allele substitution. D ow nloaded from https://academ ic.oup.com /jas/article/doi/10.1093/jas/skad333/7287507 by N atural R esources Institute Finland (Luke) user on 24 O ctober 2023 12 Journal of Animal Science, 2023, Vol. 101 Based on the annotation available on National Center for Biotechnology Information (NCBI), the QTL on chromosome 6 overlays six different putative genes (Supplementary Table S1). Blasting the amino acid sequences we identified the clos- est homologs in other species. Only for two of the putative genes the best homologs match has an amino acid sequence of a matching length, the tRNA methyltransferase and the breast cancer type 2 susceptibility protein, whereas the other proteins in the Coregonus lavaretus ‘Balchen’ assembly match only a fragment of the possible homologue proteins. Both the tRNA methyltransferase and the breast cancer type 2 susceptibility protein are involved in tumorigenesis in humans, which might suggest they play a role in the immunity to infections in Euro- pean whitefish, though these results are observational, and substantial more work would be needed to infer any func- tional effect—such as the functional genetic approach used by Palvelin et al. (2021) to show that is nedd-8, rather than the adjacent cdh1 locus controlling resistance to IPNV infection. One of the important considerations stemming from our work is the importance of choosing the most informative time-period to collect mortality data in a breeding program. Because the number of fish surviving over time decreases over time, our ability to calculate precise GEBVs for all traits, not just mortality, decreases over time, as the sample size shrinks. In our data, we observed a strong genetic correlation between the mortality traits over time (Table 3), and that the genet- ics of survival to Saprolegnia is nearly identical at all time points. Thus, our data suggest that using an early mortality to Saprolegnia measure would provide a robust proxy for all mortality to Saprolegnia at all times during the breeding pro- gram. Mortality data collected before the maturity will allow to estimate GEBVs and make selection well in advance before the fish are spawning. Importantly, while the genetic correlation between size- based traits and mortality traits is low with large standard errors, the phenotypic correlations are either not significant, or positive and significant, suggesting that phenotypically larger fish at any time point are more likely to have died by the time the next mortality census is carried out. The low genetic correlations mean that both weight and mortality Figure 8. Genomic breeding values estimated with GBLUP and GBLUP+QTL models. In GBLUP+QTL model, the top SNP effect with genotypes AA, AT and TT were fitted as a fixed effect. D ow nloaded from https://academ ic.oup.com /jas/article/doi/10.1093/jas/skad333/7287507 by N atural R esources Institute Finland (Luke) user on 24 O ctober 2023 http://academic.oup.com/jas/article-lookup/doi/10.1093/jas/skad333#supplementary-data http://academic.oup.com/jas/article-lookup/doi/10.1093/jas/skad333#supplementary-data Calboli et al. 13 can be improved simultaneously, yet care should be taken to track potentially correlated genetic changes in multiple traits. In addition, the phenotypic correlation between weight and survival indicates that the phenotypically the smallest fish tend to die during growth, creating a selection bias in the data. When such selection bias occurs, it needs to be taken into account by including the pre-selected trait(s) recorded from all the individuals in a multivariate animal model (Hen- derson, 1975; Ouweltjes et al., 1988). Hence, in the estima- tion of genetic parameters and GEBVs as well as in GWAS, we always included the initial body Weight2 as a trait in all (multivariate) analyses. Genetic characteristics of body size Heritability for body weight was moderate (0.44 to 0.53) and of similar magnitude to what has been observed previ- ously for this species (Kause et al., 2011), yet these estimates are at the higher end of the range typically observed for sal- monids (Carlson and Seamons, 2008; Kause et al., 2022). It is interesting to notice that for body weight we also observe a QTL of high statistical significance, but for this trait, the amount of genetic variance explained by the SNPs included in the QTL is much smaller than for mortality: the SNPs involved in the main QTL explain an average of 3% genetic variance for body weight. The main QTL for body weight maps on the same chromosome, but roughly 25 megabases apart from a previously reported sex-determining QTL in fish in the same species complex (De-Kayne et al., 2022). In our GWAS results, this region is not associated with any fish trait we recorded. Genomic selection for mortality to Saprolegnia Our results highlight that, irrespective of any other consider- ation, the benefit of testing for the presence of QTLs before estimating genomic breeding values is worth the extra effort in breeding settings. The inclusion of the QTL in genomic breeding value evaluation increased the accuracy of GEBVs, with a clear change in GEBV distribution pattern matching the actual QTL genotype. Estimation of GEBVs based on gen- otype data would require only a minimum amount of extra computational effort to test for QTLs, because the required phenotypic data would need to be ascertained in any case, with obvious benefits in terms of potentially increased selec- tion accuracy due to the more detailed modeling of the mode of inheritance. The assumption of genomic evaluation with a genomic relationship matrix (GBLUP) is that SNP effects are small and normally distributed, but these assumptions do not necessarily hold in practice. When these assumptions do not hold, other sophisticated statistical approaches have been developed. Examples of these approaches are Bayes- ian methods that can model non-normal distribution of SNP effects (Habier et al., 2011), weighted single step in which SNP effects are used to weight markers in GEBV estima- tion (Zhang et al., 2016), and trait-specific weighted GRM matrices (Fragomeni et al., 2017). The major limitation of these approaches is that currently, these methods cannot be effectively implemented in multivariate analyses where potentially many tens of traits are analyzed together, a practice that is standard in routine breeding value evalua- tion. For strongly oligogenic traits, genomic evaluation with MAS offers a practical solution from a breeding standpoint, Figure 9: Violin plot of cross-validation accuracies of mortality and weight traits in GBLUP and GBLUP+QTL models. Dark area shows the distribution of accuracies in 1,000 resampling runs. D ow nloaded from https://academ ic.oup.com /jas/article/doi/10.1093/jas/skad333/7287507 by N atural R esources Institute Finland (Luke) user on 24 O ctober 2023 14 Journal of Animal Science, 2023, Vol. 101 preserving the ability to use the multivariate analysis when the focus is purely on breeding. For strongly oligogenic traits, such as for Saprolegnia resis- tance here, the use of GBLUP+QTL is currently in fact an appealing approach. For instance, we validated the GEBV prediction improvement using a cross-validation approach. We observed an average of 0.33 in the accuracy for mortality calculated using the GBLUP model, but the accuracy increased to a mean of 0.61 for the GBLUP+QTL model, almost a dou- bling of the predictive accuracy. Additionally, when we used the AUC approach, to account for the fact that mortality is binary trait, we observed that the mean accuracy for the GBLUP+QTL (0.67) is not only higher than the mean accu- racy for the GBLUP (0.59), but it is very close to the maximum theoretical accuracy we can expect for this dataset, which is 0.7 (Wray et al., 2010). Furthermore, the ability of Mortal- ity3 GEBVs estimated with the GBLUP and GBLUP+QTL models to predict the Mortality4 phenotype improves from 0.46 in the GBLUP model to 0.61 for the GBLUP+QTL. All these observations indicate that the GBLUP+QTL improves our predictive ability compared to the simpler GBLUP model. Of course, these results are currently limited to the present subset of data, and the effect of the QTL estimated in a sin- gle fish group. The results would have to be validated over different generations of the breeding program to assess the overall, general effect of the GEBVs and QTL on survival to Saprolegnia. Overall, our results indicate the value of using NGS tech- niques for genomic selection on European whitefish aquacul- ture, and highlight the benefit that genome-wide markers offer, not just for genomic selection, but for exploring the genetic architecture of traits under selection, and to increase the pre- cision of the estimation of GEBVs. Additionally, we identified the presence of a major QTL for the resistance to Saprolegnia, Figure 10. Violin plot of the area under the curve (AUC) of mortality at time 3 in GBLUP and GBLUP+QTL models. Dark area shows the distribution of AUC in 1,000 resampling runs. Table 5. Correlations of GEBVs of mortality 3 with mortality at time 4 Mortality3 Mortality4 GBLUP GBLUP+QTL Correlation GEBV GEBV 0.77 0.91 Correlation GEBV Phenotype 0.18 0.22 D ow nloaded from https://academ ic.oup.com /jas/article/doi/10.1093/jas/skad333/7287507 by N atural R esources Institute Finland (Luke) user on 24 O ctober 2023 Calboli et al. 15 a common disease affecting this and many other fish species in aquaculture, indicating that there are opportunities to use genomic selection to improve resistance to Saprolegnia. Supplementary Data Supplementary data are available at Journal of Animal Science online. Acknowledgements We would like to thank the Natural Resource Institute Finland’s Enonkoski fish facility and the Natural Resource Institute Finland’s Jokioinen lab staff. This work was fund- ed by ‘ArctAqua - Cross-Border Innovations in Arctic Aquaculture’ project, co-funded by Kolarctic Cross-Boarder- Cooperation Programme 2014-2020, with a grant contract number 4/2018/095/KO4058, and the Statutory Services of Natural Resources Institute Finland. Conflict of interest statement The authors declare no conflict of interest. Literature Cited Aslam, M. L., R. Carraro, A. K. Sonesson, T. Meuwissen, C. S. Tsig- enopoulos, G. Rigos, L. Bargelloni, and K. Tzokas. 2020. Genetic variation, GWAS and accuracy of prediction for host resistance to Sparicotyle chrysophrii in farmed gilthead sea bream (Sparus au- rata). Front. Genet. 11:594770. doi:10.3389/fgene.2020.594770 Calboli, F. C. F., H. Koskinen, A. Nousianen, C. Fraslin, R. D. Hous- ton, and A. Kause. 2022. Conserved QTL and chromosomal in- version affect resistance to columnaris disease in 2 rainbow trout (Oncorhyncus mykiss) populations. G3 (Bethesda). 12:jkac137. doi:10.1093/g3journal/jkac137 Carlson, S. M., and T. R. Seamons. 2008. A review of quantitative genetic components of fitness in salmonids: implications for ad- aptation to future change. Evol. Appl. 1:222–238. doi:10.1111/ j.1752-4571.2008.00025.x De-Kayne, R., S. Zoller, and P. G. D. Feulner. 2020. A de novo chro- mosome-level genome assembly of Coregonus sp. “Balchen”: one representative of the Swiss Alpine whitefish radiation. Mol. Ecol. Resour. 20:1093–1109. doi:10.1111/1755-0998.13187 De-Kayne, R., O. M. Selz, D. A. Marques, D. Frei, O. Seehausen, and P. G. D. Feulner. 2022. Genomic architecture of adaptive radiation and hybridization in Alpine whitefish. Nat. Commun. 13:4479. doi:10.1038/s41467-022-32181-8 Dempster, E. R., and I. M. Lerner. 1950. Heritability of threshold char- acters. Genetics. 35:212–236. doi:10.1093/genetics/35.2.212 Dou, J., X. Li, Q. Fu, W. Jiao, Y. Li, T. Li, Y. Wang, X. Hu, S. Wang, and Z. Bao. 2016. Evaluation of the 2b-RAD method for genomic selec- tion in scallop breeding. Sci. Rep. 6:19244. doi:10.1038/srep19244 Fischer, D. 2023. fischuu/Snakebite-GBS: pipeline release version 0.18.3 (0.18.3). Zenodo. doi:10.5281/zenodo.7550722 Fragomeni, B. O., D. A. L. Lourenco, Y. Masuda, A. Legarra, and I. Misztal. 2017. Incorporation of causative quantitative trait nucle- otides in single-step GBLUP. Genet. Sel. Evol. 49:59. doi:10.1186/ s12711-017-0335-0 Fraslin, C., H. Koskinen, A. Nousianen, R. D. Houston, and A. Kause. 2022a. Genome-wide association and genomic prediction of re- sistance to Flavobacterium columnare in a farmed rainbow trout population. Aquaculture. 557:738332. doi:10.1016/j.aquacul- ture.2022.738332 Fraslin, C., J. M. Yáñez, D. Robledo, and R. D. Houston. 2022b. The impact of genetic relationship between training and validation populations on genomic prediction accuracy in Atlantic salmon. Aquacult. Rep. 23:101033. doi:10.1016/j.aqrep.2022.101033 Fu, B., H. Liu, X. Yu, and J. Tong. 2016. A high-density genetic map and growth related QTL mapping in bighead carp (Hypophthalmich- thys nobilis). Sci. Rep. 6:28679. doi:10.1038/srep28679 Fuji, K., O. Hasegawa, K. Honda, K. Kumasaka, T. Sakamoto, and N. Okamoto. 2007. Marker-assisted breeding of a lymphocystis dis- ease-resistant Japanese flounder (Paralichthys olivaceus). Aquacul- ture. 272:291–295. doi:10.1016/j.aquaculture.2007.07.210 Goddard, M. E., and B. J. Hayes. 2007. Genomic selection. J. Anim. Breed. Genet. 124:323–330. doi:10.1111/j.1439-0388.2007.00702.x Habier, D., R. L. Fernando, K. Kizilkaya, and D. J. Garrick. 2011. Ex- tension of the Bayesian alphabet for genomic selection. BMC Bio- inf. 12:186. doi:10.1186/1471-2105-12-186 Henderson, C. R. 1975. Best linear unbiased estimation and prediction un- der a selection model. Biometrics. 31:423–447. doi:10.2307/2529430 Houston, R. D., C. S. Haley, A. Hamilton, D. R. Guy, J. C. Mota-Velas- co, A. A. Gheyas, A. E. Tinch, J. B. Taggart, J. E. Bron, W. G. Starkey, et al. 2010. The susceptibility of Atlantic salmon fry to freshwater infectious pancreatic necrosis is largely explained by a major QTL. Heredity. 105:318–327. doi:10.1038/hdy.2009.171 Houston, R. D., T. P. Bean, D. J. Macqueen, M.K. Gundappa, Y. H. Jin, T.L. Jenkins, and D. Robledo. 2020. Harnessing genomics to fast-track genetic improvement in aquaculture. Nat. Rev. Genet. 21:389–409. doi: 10.1038/s41576-020-0227-y Karami, A. M., J. Ødegård, M. H. Marana, S. Zuo, R. Jaafar, H. Mathiessen, L. Von Gersdorff Jørgensen, P. W. Kania, I. Dalsgaard, T. Nielsen, et al. 2020. A major QTL for resistance to Vibrio an- guillarum in rainbow trout. Front. Genet. 11:607558. doi:10.3389/ fgene.2020.607558 Kause, A., C. Quinton, S. Airaksinen, K. Ruohonen, and J. Koske- la. 2011. Quality and production trait genetics of farmed Euro- pean whitefish, Coregonus lavaretus. J. Anim. Sci. 89:959–971. doi:10.2527/jas.2010-2981 Kause, A., A. Nousiainen, and H. Koskinen. 2022. Improvement in feed efficiency and reduction in nutrient loading from rainbow trout farms: the role of selective breeding. J. Anim. Sci. 100:skac214. doi:10.1093/jas/skac214 Kriaridou, C., S. Tsairidou, R. D. Houston, and D. Robledo. 2020. Genomic prediction using low density marker panels in aquacul- ture: performance across species, traits, and genotyping platforms. Front. Genet. 11:124. doi:10.3389/fgene.2020.00124 Lande, R., and R. Thompson. 1990. Efficiency of marker-assisted selec- tion in the improvement of quantitative traits. Genetics. 124:743– 756. doi:10.1093/genetics/124.3.743 Langmead, B., and S. L. Salzberg. 2012. Fast gapped-read alignment with Bowtie 2. Nat. Methods. 9:357–359. doi:10.1038/nmeth.1923 Legarra, A., C. Robert-Granié, E. Manfredi, and J. M. Elsen. 2008. Performance of genomic selection in mice. Genetics. 180:611–618. doi:10.1534/genetics.108.088575 Li, H., B. Handsaker, A. Wysoker, T. Fennell, J. Ruan, N. Homer, G. Marth, G. Abecasis, R. Durbin, and. 1000. Genome Project Data Processing Subgroup. 2009. The sequence alignment/map format and SAMtools. Bioinformatics. 25:2078–2079. doi:10.1093/bioin- formatics/btp352 Li, H., G. Su, L. Jiang, and Z. Bao. 2017. An efficient unified model for genome-wide association studies and genomic selection. Genet. Sel. Evol. 49:64. doi:10.1186/s12711-017-0338-x Lopes, M. S., H. Bovenhuis, M. Van Son, O. Nordbø, E. H. Grindflek, E. F. Knol, and J. W. M. Bastiaansen. 2017. Using markers with large effect in genetic and genomic predictions. J. Anim. Sci. 95:59–71. doi:10.2527/jas.2016.0754 Madsen, P., J. Jensen, R. Labouriau, O. F. Christensen, and G. Sahana. 2014. DMU—a package for analyzing multivariate mixed mod- els in quantitative genetics and genomics. Proceedings of the 10th World Congress of Genetics Applied to Livestock Production. p. 18–22. doi:10.1080/02664763.2013.868416 Mäntysaari, E., R. L. Quaas, and Y. T. Gröhn. 1991. Simulation study on covariance component estimation for two binary traits in an D ow nloaded from https://academ ic.oup.com /jas/article/doi/10.1093/jas/skad333/7287507 by N atural R esources Institute Finland (Luke) user on 24 O ctober 2023 https://doi.org/10.3389/fgene.2020.594770 https://doi.org/10.1093/g3journal/jkac137 https://doi.org/10.1111/j.1752-4571.2008.00025.x https://doi.org/10.1111/j.1752-4571.2008.00025.x https://doi.org/10.1111/1755-0998.13187 https://doi.org/10.1038/s41467-022-32181-8 https://doi.org/10.1093/genetics/35.2.212 https://doi.org/10.1038/srep19244 https://doi.org/10.5281/zenodo.7550722 https://doi.org/10.1186/s12711-017-0335-0 https://doi.org/10.1186/s12711-017-0335-0 https://doi.org/10.1016/j.aquaculture.2022.738332 https://doi.org/10.1016/j.aquaculture.2022.738332 https://doi.org/10.1016/j.aqrep.2022.101033 https://doi.org/10.1038/srep28679 https://doi.org/10.1016/j.aquaculture.2007.07.210 https://doi.org/10.1111/j.1439-0388.2007.00702.x https://doi.org/10.1186/1471-2105-12-186 https://doi.org/10.2307/2529430 https://doi.org/10.1038/hdy.2009.171 https://doi.org/10.1038/s41576-020-0227-y https://doi.org/10.3389/fgene.2020.607558 https://doi.org/10.3389/fgene.2020.607558 https://doi.org/10.2527/jas.2010-2981 https://doi.org/10.1093/jas/skac214 https://doi.org/10.3389/fgene.2020.00124 https://doi.org/10.1093/genetics/124.3.743 https://doi.org/10.1038/nmeth.1923 https://doi.org/10.1534/genetics.108.088575 https://doi.org/10.1093/bioinformatics/btp352 https://doi.org/10.1093/bioinformatics/btp352 https://doi.org/10.1186/s12711-017-0338-x https://doi.org/10.2527/jas.2016.0754 https://doi.org/10.1080/02664763.2013.868416 16 Journal of Animal Science, 2023, Vol. 101 underlying continuous scale. J. Dairy Sci. 74:580–591. doi:10.3168/ jds.s0022-0302(91)78205-2 Melo, A. T., R. Bartaula, and I. Hale. 2016. GBS-SNP-CROP: a refer- ence-optional pipeline for SNP discovery and plant germplasm characterization using variable length, paired-end genotyping-by-se- quencing data. BMC Bioinf. 17:29. doi:10.1186/s12859-016-0879-y Meuwissen, T. H. E., B. J. Hayes, and M. E. Goddard. 2001. Prediction of total genetic value using genome-wide dense marker maps. Ge- netics. 157:1819–1829. doi:10.1093/genetics/157.4.1819 Misk, E., S. Gonen, and A. F. Garber. 2022. Resistance to Saprolegnia parasitica infection: a heritable trait in Atlantic salmon. J. Fish Dis. 45:1333–1342. doi:10.1111/jfd.13664 Misztal, I., D. Lourenco, and A. Legarra. 2020. Current status of ge- nomic evaluation. J. Anim. Sci. 98:1–14. doi:10.1093/jas/skaa101 Moen, T., M. Baranski, A. K. Sonesson, and S. Kjøglum. 2009. Confirmation and fine-mapping of a major QTL for resistance to infectious pancreatic necrosis in Atlantic salmon (Salmo salar): population-level associations between markers and trait. BMC Ge- nomics. 10:368. doi:10.1186/1471-2164-10-368 Mölder, F., K. P. Jablonski, B. Letcher, M. B. Hall, C. H. Tomkins-Tinch, V. Sochat, J. Forster, S. Lee, S. O. Twardziok, A. Kanitz, et al. 2021. Sustainable data analysis with Snakemake. F1000Research. 10:33. doi:10.12688/f1000research.29032.2 Nani, J. P., F. M. Rezende, and F. Peñagaricano. 2019. Predicting male fertility in dairy cattle using markers with large effect and function- al annotation data. BMC Genomics. 20:258. doi:10.1186/s12864- 019-5644-y Ouweltjes, W., L. R. Schaeffer, and B. W. Kennedy. 1988. Sensitivity of methods of variance component estimation to culling type of selection. J. Dairy Sci. 71:773–779. doi:10.3168/jds.s0022-0302(88)79617-4 Palaiokostas, C., S. Ferraresso, R. Franch, R. D. Houston, and L. Barg- elloni. 2016. Genomic prediction of resistance to pasteurellosis in gilthead sea bream (Sparus aurata) using 2b-RAD sequencing. G3 (Bethesda). 6:3693–3700. doi:10.1534/g3.116.035220 Pavelin, J., Y. H. Jin, R. L. Gratacap, J. B. Taggart, A. Hamilton, D. W. Verner-Jeffreys, R. K. Paley, C. J. Rubin, S. C. Bishop, J. E. Bron, et al. 2021. The nedd-8 activating enzyme gene underlies genetic resistance to infectious pancreatic necrosis virus in Atlantic salmon. Genomics. 113:3842–3850. doi:10.1016/j.ygeno.2021.09.012 Pitkänen, T. J., H. Gao, A. Kudinov, M. Taskinen, E. A. Mäntysaari, M. H. Lidauer, and I. Strandén. 2022. From data to genomic breeding values with the mix99 software suite. Proceedings of 12th World Congress on Genetics Applied to Livestock Production (WCGALP). p. 1534–1537. doi:10.3920/978-90-8686-940-4_367 Ren, P., W. Peng, W. You, Z. Huang, Q. Guo, N. Chen, P. He, J. Ke, J. -C. Gwo, and C. Ke. 2016. Genetic mapping and quantitative trait loci analysis of growth-related traits in the small abalone Haliotis diversicolor using restriction-site-associated DNA se- quencing. Aquaculture. 454:163–170. doi:10.1016/j.aquacul- ture.2015.12.026 Ren, D., L. An, B. Li, L. Qiao, and W. Liu. 2021. Efficient weighting methods for genomic best linear-unbiased prediction (BLUP) adapted to the genetic architectures of quantitative traits. Heredity 126:320–334. doi:10.1038/s41437-020-00372-y Robledo, D., C. Palaiokostas, L. Bargelloni, P. Martínez, and R. Hous- ton. 2017. Applications of genotyping by sequencing in aquacul- ture breeding and genetics. Rev. Aquac. 10:670–682. doi:10.1111/ raq.12193 Salas-Lizana, R., and R. Oono. 2018. Double-digest RADseq loci using standard Illumina indexes improve deep and shallow phylogenetic resolution of Lophodermium, a widespread fungal endophyte of pine needles. Ecol. Evol. 8:6638–6651. doi:10.1002/ece3.4147 Sandoval-Sierra, J. V., F. Latif-Eugenin, M. P. Martín, L. Zaror, and J. Diéguez-Uribeondo. 2014. Saprolegnia species affecting the salmonid aquaculture in Chile and their associations with fish developmental stage. Aquaculture. 434:462–469. doi:10.1016/j. aquaculture.2014.09.005 Sarowar, M. N., R. Cusack, and J. Duston. 2019. Saprolegnia molecular phylogeny among farmed teleosts in Nova Scotia, Canada. J. Fish Dis. 42:1745–1760. doi:10.1111/jfd.13090 Shao, C., Y. Niu, P. Rastas, Y. Liu, Z. Xie, H. Li, L. Wang, Y. Jiang, S. Tai, Y. Tian, et al. 2015. Genome-wide SNP identification for the construction of a high-resolution genetic map of Japanese flounder (Paralichthys olivaceus): applications to QTL mapping of Vibrio anguillarum disease resistance and comparative genomic analysis. DNA Res. 22:161–170. doi:10.1093/dnares/dsv001 Strandén, I., and M. Lidauer. 1999. Solving large mixed linear mod- els using preconditioned conjugate gradient iteration. J. Dairy Sci. 82:2779–2787. doi:10.3168/jds.S0022-0302(99)75535-9 Tian, M., Y. Li, J. Jing, C. Mu, H. Du, J. Dou, J. Mao, X. Li, W. Jiao, Y. Wang, et al. 2015. Construction of a high-density genetic map and quantitative trait locus mapping in the sea cucumber Apostichopus japonicus. Sci. Rep. 5:14852. doi:10.1038/srep14852 Vallejo, R. L., J. P. Evenhuis, H. Cheng, B. O. Fragomeni, G. Gao, S. Liu, R. L. Long, K. L. Shewbridge, R. M. O. Silva, G. D. Wiens, et al. 2022. Genome-wide mapping of quantitative trait loci that can be used in marker-assisted selection for resistance to bacterial cold wa- ter disease in two commercial rainbow trout breeding populations. Aquaculture. 560:738574. doi:10.1016/j.aquaculture.2022.738574 Van Den Berg, A. H., D. Mclaggan, J. Diéguez-Uribeondo, and P. Van West. 2013. The impact of the water moulds Saprolegnia diclina and Saprolegnia parasitica on natural ecosystems and the aquaculture industry. Fungal Biol. Rev. 27:33–42. doi:10.1016/j.fbr.2013.05.001 Vanraden, P. M., C. P. Van Tassell, G. R. Wiggans, T. S. Sonstegard, R. D. Schnabel, J. F. Taylor, and F. S. Schenkel. 2009. Invited review: reli- ability of genomic predictions for North American Holstein bulls. J. Dairy Sci. 92:16–24. doi:10.3168/jds.2008-1514 Vehviläinen, H., A. Kause, C. Quinton, H. Koskinen, and T. Paananen. 2008. Survival of the currently fittest: genetics of rainbow trout sur- vival across time and space. Genetics. 180:507–516. doi:10.1534/ genetics.108.089896 Vehviläinen, H., A. Kause, H. Koskinen, and T. Paananen. 2010. Genet- ic architecture of rainbow trout survival from egg to adult. Genet- ics Res. 92:1–11. doi:10.1017/S0016672310000017 Vela-Avitúa, S., I. Thorland, V. Bakopoulos, K. Papanna, A. Dimitro- glou, E. Kottaras, P. Leonidas, B. Guinand, C. S. Tsigenopoulos, and M. L. Aslam. 2022. Genetic basis for resistance against viral ner- vous necrosis: GWAS and potential of genomic prediction explored in farmed European Sea Bass (Dicentrarchus labrax). Front. Genet. 13:804584. doi:10.3389/fgene.2022.804584 Vela-Avitúa, S., B. R. Lafrentz, C. A. Lozano, C. A. Shoemaker, J. F. Ospina-Arango, B. H. Beck, and M. Rye. 2023. Genome-wide asso- ciation study for Streptococcus iniae in Nile tilapia (Oreochromis niloticus) identifies a significant QTL for disease resistance. Front. Genet. 14:1078381. doi:10.3389/fgene.2023.1078381 Wang, L., Z. Y. Wan, B. Bai, S. Q. Huang, E. Chua, M. Lee, H. Y. Pang, Y. F. Wen, P. Liu, F. Liu, et al. 2015. Construction of a high-density linkage map and fine mapping of QTL for growth in Asian seabass. Sci. Rep. 5:16358. doi:10.1038/srep16358 Wray, N. R., J. Yang, M. E. Goddard, and P. M. Visscher. 2010. The ge- netic interpretation of area under the ROC curve in genomic profil- ing. PLoS Genet. 6:e1000864. doi:10.1371/journal.pgen.1000864 Yang, J., N. A. Zaitlen, M. E. Goddard, P. M. Visscher, and A. L. Price. 2014. Advantages and pitfalls in the application of mixed-model association methods. Nat. Gen. 46:100–106. doi:10.1038/ng.2876 You, X., X. Shan, and Q. Shi. 2020. Research advances in the genom- ics and applications for molecular breeding of aquaculture animals. Aquaculture. 526:735357. doi:10.1016/j.aquaculture.2020.735357 Zhang, X., D. Lourenco, I. Aguilar, A. Legarra, and I. Misztal. 2016. Weighting strategies for single-step genomic BLUP: an iterative ap- proach for accurate calculation of GEBV and GWAS. Front. Genet. 7:151. doi:10.3389/fgene.2016.00151 Lynch, M. and Walsh, B. 1998. Genetics and Analysis of Quantitative Traits. Sinauer Associates, Inc., Sunderland. D ow nloaded from https://academ ic.oup.com /jas/article/doi/10.1093/jas/skad333/7287507 by N atural R esources Institute Finland (Luke) user on 24 O ctober 2023 https://doi.org/10.3168/jds.s0022-0302(91)78205-2 https://doi.org/10.3168/jds.s0022-0302(91)78205-2 https://doi.org/10.1186/s12859-016-0879-y https://doi.org/10.1093/genetics/157.4.1819 https://doi.org/10.1111/jfd.13664 https://doi.org/10.1093/jas/skaa101 https://doi.org/10.1186/1471-2164-10-368 https://doi.org/10.12688/f1000research.29032.2 https://doi.org/10.1186/s12864-019-5644-y https://doi.org/10.1186/s12864-019-5644-y https://doi.org/10.3168/jds.s0022-0302(88)79617-4 https://doi.org/10.1534/g3.116.035220 https://doi.org/10.1016/j.ygeno.2021.09.012 https://doi.org/10.3920/978-90-8686-940-4_367 https://doi.org/10.1016/j.aquaculture.2015.12.026 https://doi.org/10.1016/j.aquaculture.2015.12.026 https://doi.org/10.1038/s41437-020-00372-y https://doi.org/10.1111/raq.12193 https://doi.org/10.1111/raq.12193 https://doi.org/10.1002/ece3.4147 https://doi.org/10.1016/j.aquaculture.2014.09.005 https://doi.org/10.1016/j.aquaculture.2014.09.005 https://doi.org/10.1111/jfd.13090 https://doi.org/10.1093/dnares/dsv001 https://doi.org/10.3168/jds.S0022-0302(99)75535-9 https://doi.org/10.1038/srep14852 https://doi.org/10.1016/j.aquaculture.2022.738574 https://doi.org/10.1016/j.fbr.2013.05.001 https://doi.org/10.3168/jds.2008-1514 https://doi.org/10.1534/genetics.108.089896 https://doi.org/10.1534/genetics.108.089896 https://doi.org/10.1017/S0016672310000017 https://doi.org/10.3389/fgene.2022.804584 https://doi.org/10.3389/fgene.2023.1078381 https://doi.org/10.1038/srep16358 https://doi.org/10.1371/journal.pgen.1000864 https://doi.org/10.1038/ng.2876 https://doi.org/10.1016/j.aquaculture.2020.735357 https://doi.org/10.3389/fgene.2016.00151 Calboli et al 2023.pdf skad333