Results and discussion Genetic parameter estimates Heritability estimates (h2) for FY, FW, BW10, BW13, and CAR were moderate to high (0.31–0.62, Table 2) suggesting that these traits can be improved through selective breeding. The heritability estimates for body weight measures (BW10 and BW13) and FW were in the range of 0.19–0.50 (Elvingson and Johansson, 1993; Neira et al., 2004; Kause et al., 2007) and 0.22–0.52 (Kause et al., 2002, 2007; Powell et al., 2008; Haffray et al., 2012), respectively, as previously reported for salmonids. Similarly, the heritability estimate for FY was in the interval previously reported for salmonids (0.03–0.38), and CV was in the upper range of estimates (0.12–6.5%) (Neira et al., 2004; Kause et al., 2007; Powell et al., 2008; Haffray et al., 2012). The heritability estimate for CAR (h2 = 0.62) was higher than previous estimates of 0.36–0.53 reported in the literature (Powell et al., 2008; Haffray et al., 2012). Table 2 Genetic parameters of the traits body weight at 10 (BW10) and 13 (BW13) months post-hatching, carcass weight (CAR), fillet weight (FW), and fillet yield (FY) without (WO_GI) and with (W_GI) genomic information. Parameters BW10, g BW13, g CAR, g FW, g FY, % WO_GI W_GI WO_GI W_GI WO_GI W_GI WO_GI W_GI WO_GI W_GI σ a 2 3399.5 4108.9 15,044 17,939 679.9 684.5 363.0 412.8 1.8 1.9 σ w 2 1427.4 1346.1 6344.1 6316.7 46.0 19.7 10.6 7.6 0.6 0.3 σ e 2 6031.5 5659.4 24,979 23,482 362.7 355.5 533.8 510.8 3.0 3.1 σ p 2 10,858.4 11,114.4 46,367.1 47,737.7 1088.6 1059.6 907.4 931.2 5.4 5.4 h2 0.31 0.37 0.32 0.38 0.62 0.65 0.40 0.44 0.34 0.36 Acc 0.66 0.70 0.65 0.69 0.28 0.55 0.25 0.50 0.13 0.55 r(EBV,GEBV) 0.99 0.99 0.76 0.79 0.72 σa2, additive variance; σw2, common environment variance; σe2, residual variance; σp2, phenotypic variance; h2, heritability (additive variance/phenotypic variance); Acc, average accuracy; r(EBV,GEBV), correlation between estimated breeding values (EBV) and genomic estimated breeding values (GEBV). The correlations between estimated breeding values (EBVs) and genomic breeding values (GEBVs) were near unity (0.99) for traits measured directly on breeding candidates (BW10 and BW13) and smaller (0.72–0.79) for lethally-measured traits (CAR, FW, and FY; Table 2). In the absence of progeny performance data or a correlated trait that can be measured directly on breeding candidates, traditional BLUP-based EBVs for CAR, FW, and FY are necessarily identical among non-phenotyped siblings, whereas the use of genomic information (GI) in GBLUP enables fish-specific GEBVs despite the absence of phenotypic data for the fish, its progeny, or for a correlated trait. Thus, the smaller correlations between EBVs and GEBVs for lethally-measured traits was expected because the correlation is between family-specific and fish-specific estimates of genetic merit, respectively. When GI was added to the aforementioned analysis, a slight increase in heritabilities was observed for all traits (Table 2). However, the accuracies of GEBVs were increased by ~100% for CAR and FW (from 0.28 and 0.25 to 0.55 and 0.50, respectively) and by ~420% for FY (from 0.13 to 0.55) compared traditional pedigree-based EBVs. For lethally-measured traits like FW, FY, and CAR, traditional selective breeding programs rely on sib-testing with limited reliability (Odegård et al., 2014). Therefore, methods that increase accuracy of predictions and expedite genetic progress by exploiting within-family genetic variation for economically-important traits in aquaculture species are important for continued development of the aquaculture industry (Yáñez et al., 2015). Models that include GI from numerous SNP markers in addition to the phenotypic and pedigree information without previous knowledge of the underlying QTL outperformed models without GI (Nielsen et al., 2009; Odegård et al., 2014). This improved performance of GI models is expected based on the definition of the accuracy as a function of heritability and amount of information used (Chen et al., 2011). Examples of traits for which inclusion of GI resulted in an increase in accuracy include growth in broiler chickens (Wang et al., 2014); lice resistance and fillet color in Atlantic salmon (Odegård et al., 2014); and weight and length traits in Atlantic salmon (Tsai et al., 2015). GWAS The Manhattan plots from GWAS results at iteration 5 for FY, FW, BW10, BW13, and CAR are shown in Figures 1–5, respectively. In total, 1906 non-overlapping, non-repetitive windows of 20 successive SNPs were used. Of these windows, two windows located on chromosome Omy9 explained 1.5 and 1.0% of the genetic variance for FY. The same windows explained 1.2 and 1.1% of the genetic variance for FW. Only one window, located on Omy5, was responsible for 1.38 and 0.95% of the genetic variance for BW10 and BW13, respectively. Three windows located on Omy27, Omy17, and Omy9 were responsible for 1.7, 1.7, and 1.0%, of the genetic variance in CAR, respectively. Figure 1 The proportion of genetic variance explained by 20-SNP regions for fillet yield. Figure 2 The proportion of genetic variance explained by 20-SNP regions for fillet weight. Figure 3 The proportion of genetic variance explained by 20-SNP regions for 10-month body weight. Figure 4 The proportion of genetic variance explained by 20-SNP regions for 13-month body weight. Figure 5 The proportion of genetic variance explained by 20-SNP regions for carcass weight. No major QTL was detected for FY, suggesting that this trait has a polygenic architecture affected by multiple loci with small effects in this rainbow trout population. The SNP markers from the two windows that explained at least 1.0% of the proportion of variance for FY and harboring or neighboring genes from the same genome scaffold (Berthelot et al., 2014) are listed in Table 3. An extended list of all the SNPs markers is provided in Table S2. Only 13 of the 40 SNPs identified by the wssGWAS on Omy9 explained a proportion of the genetic variance equal to or greater than 0.1%. Of them, nine were located within a gene (exon or intron) and the other four had known neighboring genes in the same genome scaffold. Four of those 13 SNPs were located in scaffold_516, of which one SNP was mapped to exon 6 of Properdin, another to exon 3 of Transmembrane protein 201-like, and two SNPs were mapped in the intron regions of Calsyntenin-1-like isoform x2 (Table 3). Briefly, Properdin is one of the proteins that participates in the complement system, which also may interfere with fatty acid uptake and esterification in adipocytes (Gauvreau et al., 2012). Transmembrane protein 201-like is a component of transmembrane actin-associated nuclear lines with a role in centrosome orientation and nuclear movement prior to cell migration (Borrego-Pinto et al., 2012). Calsyntenin-1 is associated with kinesin-1-mediated transport of vesicles and tubulovesicular organelles (Konecna et al., 2006), appears to participate in intracellular transport and endosomal trafficking, and is necessary for the formation of peripheral sensory axons (Ponomareva et al., 2014). However, the biological functions of calsyntenins are not well understood yet (Ortiz-Medina et al., 2015). Table 3 The SNP markers that explained the largest proportion of variance for fillet yield using 20-SNP windows. Marker Chr Position (cM) Alleles VE (%) Scaffold Scaffold position Scaffold size Loc Description WINDOW 1 TOTAL PROPORTION 1.5% AX-89976492 9 125.19 T/C 0.13 Scaffold_8612 10,780 26,627 Intron Beta-catenin-interacting protein 1 isoform x1 AX-89970327 9 125.19 A/C 0.11 Scaffold_516 399,414 728,099 Intron Calsyntenin-1-like isoform x2 AX-89940136 9 125.19 A/G 0.12 Scaffold_516 407,663 728,099 Intron Calsyntenin-1-like isoform x2 AX-89936139 9 125.67 A/C 0.13 Near* None AX-89944669 9 125.67 G/T 0.13 Scaffold_32707 921 3946 Exon 1 Nudix hydrolase chloroplastic-like AX-89940514 9 125.67 T/C 0.12 Scaffold_10308 15,430 21,663 Near* None AX-89937961 9 125.67 G/T 0.13 Scaffold_516 586,177 728,099 Exon 6 Properdin AX-89951506 9 125.67 G/A 0.12 Scaffold_516 555,991 728,099 Exon 3 Transmembrane protein 201-like AX-89938525 9 125.86 G/A 0.10 Scaffold_52 1,795,215 2,128,772 Intron Atpase family aaa domain-containing protein 3-like AX-89957923 9 126.04 A/G 0.13 Scaffold_19674 6079 8261 Near* None WINDOW 2 TOTAL PROPORTION 1.0% AX-89951447 9 117.12 G/A 0.10 Scaffold_43535 864 3113 Near* None AX-89940159 9 117.12 T/C 0.11 Scaffold_347 148,757 935,129 Intron Serine threonine-protein kinase sbk1-like AX-89924961 9 117.91 C/T 0.10 Scaffold_347 414,782 935,129 Exon 6 Phd finger protein 20-like isoform x3 Chr, chromosome; VE, percentage of the genetic variance explained by the SNP; Loc, location in the scaffold with three possibilities: intron, exon, or near an exon. The same two windows found in the FY GWAS were also associated with FW. However, from the 40 SNPs markers identified by the wssGWAS, only 12 explained a proportion equal to or greater than 0.1% and of them, nine were in a gene (exon or intron, Table 4). An extended list of all the SNP markers is provided in Table S3. Similar to FY, no major QTL was detected for FW which supports the polygenic architecture of FW. Three of the 12 SNPs were also located in scaffold_516 within the Calsyntenin-1-like isoform x2, Properdin and Transmembrane protein 201-like genes (Table 4). Three additional SNPs were located in scaffold_347 in the Src-like-adapter 2; Serine/ threonine-protein kinase sbk1-like; and Phd finger protein 20-like isoform x3 genes. Briefly, Src-like-adapter 2 is an adaptor protein that regulates T and B cell maturation and development, and it is a critical component regulating signal transduction in immune and malignant cells (Sosinowski et al., 2001; Dragone et al., 2006; Kazi et al., 2015). Additionally, differentially-expressed transcripts in response to handling and confinement stress in rainbow trout were mapped to a serine/threonine-protein kinase, SBK1, homologous in zebrafish that participates in signal transduction pathways related to brain development (Chou et al., 2006; Liu et al., 2015a). Table 4 The SNP markers that explained the largest proportion of variance for fillet weight using 20-SNP windows. Marker Chr Position (cM) Alleles VE (%) Scaffold Scaffold position Scaffold size Loc Description WINDOW 1 TOTAL PROPORTION 1.2% AX-89976492 9 125.19 T/C 0.12 Scaffold_8612 10,780 26,627 Intron Beta-catenin-interacting protein 1 isoform x1 AX-89970327 9 125.19 A/C 0.10 Scaffold_516 399,414 728,099 Intron Calsyntenin-1-like isoform x2 AX-89936139 9 125.67 A/C 0.11 Near* None AX-89944669 9 125.67 G/T 0.11 Scaffold_32707 921 3946 Exon 1 Nudix hydrolase chloroplastic-like AX-89940514 9 125.67 T/C 0.10 Scaffold_10308 15,430 21,663 Near* None AX-89937961 9 125.67 G/T 0.11 Scaffold_516 586,177 728,099 Exon 6 Properdin AX-89951506 9 125.67 G/A 0.11 Scaffold_516 555,991 728,099 Exon 3 Transmembrane protein 201-like WINDOW 2 TOTAL PROPORTION 1.1% AX-89953042 9 116.09 G/A 0.10 Scaffold_1609 166,168 200,099 Exon 4 Kelch domain-containing protein 8b AX-89951447 9 117.12 G/A 0.13 Scaffold_43535 864 3113 Near* None AX-89975284 9 117.12 A/C 0.11 Scaffold_347 348,224 935,129 Exon 7 Src-like-adapter 2 AX-89940159 9 117.12 T/C 0.12 Scaffold_347 148,757 935,129 Intron Serine threonine-protein kinase sbk1-like AX-89924961 9 117.91 C/T 0.13 Scaffold_347 414,782 935,129 Exon 6 Phd finger protein 20-like isoform x3 Chr, chromosome; VE, percentage of the genetic variance explained by the SNP; Loc, location in the scaffold with three possibilities: intron, exon, or near an exon. Genome regions associated with growth have been detected on most of the 29 chromosomes in Atlantic salmon (Baranski et al., 2010; Gutierrez et al., 2012, 2015; Tsai et al., 2015) and in rainbow trout (O'Malley et al., 2003; Perry et al., 2005; Wringe et al., 2010). The heterogeneity in the results makes it hard to compare our results to previous studies. Many factors contribute to this observed heterogeneity between studies including: (1) the highly polygenic architecture of growth and growth-related traits; (2) differences in marker segregation that may be affected by the strain genetic background, and different types and densities of markers used in each study; (3) different algorithms used in the QTL detection analyses; (4) large variation in sample size (Baranski et al., 2010; Wringe et al., 2010; Tsai et al., 2015), and (5) possible false positives. None of the 20 SNPs identified by the wssGWAS for BW10 and BW13 on chromosome Omy5 (Figures 3, 4) were able to surpass the threshold of 0.1%. The complete list of the SNPs is provided in Tables S4, S5 for BW10 and BW13, respectively. Our findings of the polygenic architecture of growth traits in fish is consistent with previous reports in the literature (Devlin et al., 2009; Dai et al., 2015; Tsai et al., 2015), and, congruent with our GWAS results, several markers were associated with weight in a GWAS for Atlantic salmon, but the proportion of variance explained by each marker was less than 0.1% (Tsai et al., 2015). Lastly, SNP markers that explained more than 0.1% of the proportion of genetic variance for CAR on Omy27, 17, and 9, and harboring or neighboring genes from the same genome scaffold (Berthelot et al., 2014) are listed in Table 5. No major QTL was detected for this trait. An extended list of all the SNP markers is provided in Table S6. From the 60 SNP markers identified by wssGWAS, only 18 explained a proportion equal to or greater than 0.1%, of which 10 were in a gene (exon or intron). Four of the 18 SNPs were located in scaffold_173; one was in the exon of ubiquitin-conjugating enzyme e2 variant 1, and three were near this gene and near the histone h2b 1 2-like gene. One of the SNPs was mapped to the Calsyntenin-1-like isoform x2 intron region in scaffold_516 that also affected FY and FW. Table 5 The SNP markers that explained the largest proportion of variance for carcass weight using 20-SNP windows. Marker Chr Position (cM) Alleles VE (%) Scaffold Scaffold position Scaffold size Loc Description WINDOW 1 TOTAL PROPORTION 1.7% AX-89952551 27 75.09 A/G 0.12 Scaffold_1006 143,453 383,627 Near nitric oxide inducible/serine threonine-protein kinase nlk AX-89954149 27 75.09 C/A 0.12 Scaffold_147 921,922 1,497,438 Near atp-sensitive inward rectifier potassium channel 1-like/cmp-n-acetylneuraminate-beta-galactosamide-alpha-sialyltransferase 4-like isoform x1 AX-89938133 27 75.09 A/G 0.13 Scaffold_1006 46,265 383,627 Exon3 nitric oxide inducible AX-89948564 27 74.78 G/A 0.12 Scaffold_8798 5532 26,005 Near Undetermined/None AX-89974542 27 74.58 G/T 0.11 Scaffold_842 38,757 463,739 Intron kinase suppressor of ras 1-like isoform x2 AX-89926230 27 74.58 A/G 0.12 Scaffold_1952 116,665 147,230 Near neurofibromin isoform x2/oligodendrocyte-myelin glyco AX-89938965 27 74.43 G/T 0.12 Scaffold_842 38,933 463,739 Intron kinase suppressor of ras 1-like isoform x2 AX-89928353 27 73.96 G/A 0.10 Scaffold_1675 177,158 191,261 Intron vascular endothelial zinc finger 1-like isoform x2 AX-89968747 27 73.96 A/G 0.11 Scaffold_3611 60,104 62,580 Intron unconventional myosin-xviiia-like isoform x1 AX-89947091 27 73.96 na 0.10 na Na AX-89942611 27 73.42 C/A 0.16 Scaffold_3980 33,385 57,002 Intron unconventional myosin-xviiia-like isoform x2 WINDOW 2 TOTAL PROPORTION 1.7% AX-89973675 17 115.87 G/T 0.10 Scaffold_26752 3396 4948 Intron spectrin beta non-erythrocytic 1-like AX-89969602 17 114.85 C/T 0.10 Scaffold_24 197,378 2,579,057 Intron calpain-2 catalytic subunit-like AX-89935000 17 114.02 A/G 0.17 Scaffold_173 119,365 1,392,108 Near histone h2b 1 2-like/ubiquitin-conjugating enzyme e2 variant 1 AX-89918454 17 114.02 C/A 0.12 Scaffold_173 123,912 1,392,108 Near histone h2b 1 2-like/ubiquitin-conjugating enzyme e2 variant 1 AX-89923840 17 113.51 C/T 0.18 Scaffold_173 195,297 1,392,108 Exon4 ubiquitin-conjugating enzyme e2 variant 1 AX-89925576 17 113.51 A/G 0.18 Scaffold_173 160,398 1,392,108 Near histone h2b 1 2-like/ubiquitin-conjugating enzyme e2 variant 1 WINDOW 3 TOTAL PROPORTION 1.0% AX-89940136 9 125.19 A/G 0.10 Scaffold_516 407,663 728,099 Intron Calsyntenin-1-like isoform x2 Chr, chromosome; VE, percentage of the genetic variance explained by the SNP; Loc, location in the scaffold with three possibilities: intron, exon, or near an exon. Putative genes Networks of genes offer insight to the molecular relationships among genes. The network was visualized considering genes and neighboring genes located in the same genome scaffold as the SNPs we identified in the wssGWAS in 20-SNP windows responsible for ~1.0% or more of the total genetic variance for the analyzed traits. Genes included in the network are described in Tables S2–S6. The network includes 115 gene nodes, of which 21 nodes were genes detected by the wssGWAS analysis (green nodes) while the rest were connecting neighbors (pink nodes, Figure 6). In this network, SRY (sex determining region Y)-box 2 (Sox2), Kinase suppressor of ras 1 (Ksr1), Tripartite motif-containing 33 (Trim33), and Nitric oxide synthase 2 inducible (Nos2) were well-connected gene nodes linking to 29, 11, 7, and 6 other gene nodes, respectively. Figure 6 Network of genes and neighboring genes in the same scaffold as the SNPs identified in the wssGWAS in windows responsible for ~1.0% or more of the total genetic variance of the analyzed traits. Green nodes denote genes and near genes detected by the wssGWAS analysis and pink nodes denote connecting neighbors. Edges denote known relationships between genes in the SysBiomics repository. Framed genes (red square) are discussed in the manuscript. A number of genes detected by wssGWAS analysis within windows responsible for 1.0% or greater of the trait variance have been identified in mammalian systems as significant for muscle development, particularly with respect to the maintenance of hyperplastic capacity. Rainbow trout exhibit indeterminate growth potential as they have the ability increase both body length and muscle mass throughout adulthood (Johnston et al., 2011). A physiological mechanism unique to indeterminate growers is the ability to maintain a population of myogenic precursor cells with proliferative capacity that then differentiate into myotubes or contribute to nuclear accretion (Mommsen, 2001; Froehlich et al., 2013). Therefore, the continued ability for hyperplasia and hypertrophy in skeletal muscle contributes to the indeterminate growth phenotype (Johnston et al., 2011). In this regard, SNPs affecting genes related to myogenic and cell proliferation mechanisms are expected to affect growth and fillet yield. One gene detected by wssGWAS analysis that was the most connected node was Sox2. The Sox2 gene plays a critical role in maintenance and proliferation of pluripotent and neural progenitor stem cells (Takahashi and Yamanaka, 2006; Zhang and Cui, 2014) through its interaction with transforming Growth Factor β (TGFb) signaling (Gaarenstroom and Hill, 2014), although little is known about the role of Sox2 in muscle. Albeit, TGFb ligands like myostatin inhibit muscle growth (Lee et al., 2005; Phelps et al., 2013), partially through reductions in myogenic precursor cell proliferation (Garikipati and Rodgers, 2012; Seiliez et al., 2012). Trim33 is another gene detected by wssGWAS analysis that is up-regulated during muscle regeneration in mice and appears to play a role in myoblast proliferation (Mohassel et al., 2015). Therefore, similar to Sox2, Trim33 may also contribute to maintaining a proliferating population of myogenic precursor cells throughout development in the rainbow trout. Trim33 also inhibits Smad4 (Xi et al., 2011), a transcription factor activated by TGFb signaling that inhibits muscle regeneration and maintenance of myogenesis with age (Lee et al., 2005). A third gene identified by wssGWAS analysis was Fragile X mental retardation gene 1 (Fxr1), an autosomal gene that is highly expressed in muscle (Blonden et al., 2005). Depletion of this gene during early development of the zebrafish leads to cardiomyopathy and muscular dystrophy (Van't Padje et al., 2009). In cultured muscle cells, depletion of Fxr1 reduced myoblast abundance, suggesting an evolutionarily-conserved role for Fxr1 protein in myogenesis (Davidovic et al., 2013). A fourth gene detected by wssGWAS analysis that may affect myogenic capacity was Ksr1. This gene is a scaffold protein for the Raf/MEK/ERK kinase cascade (Ory et al., 2003). Although, a role of for Ksr1 in muscle has not been demonstrated, activation of the Raf/Mek/ERK kinase cascade promotes proliferation of myogenic cells (Knight and Kothary, 2011). A final gene that may affect myogenesis was Proviral integration site 1 (Pim1), a gene that, when overexpressed in cardiomyocytes, causes increases in regenerative capacity and increases the pool of progenitor cells (Del Re and Sadoshima, 2012), although it is unknown if this gene has significance in skeletal muscle. However, expression of Pim1 has been positively correlated with intramuscular fat content in steers (Sadkowski et al., 2014). Another pair of genes that were detected by wssGWAS and are known to encode proteins with functional importance to muscle physiology are the cysteine proteases Calpain-1 (Capn1) and Calpain-2 (Capn2). Calpains are calcium activated cysteine proteases regulated by myogenic factors like myoD and myogenin (Dedieu et al., 2003). Calpains have a relevant role in signal transduction (Glading et al., 2001; Sato and Kawashima, 2001), cell-cycle regulation, apoptosis (Atencio et al., 2000; Patel and Lane, 2000), cell spreading and migration (Dourdin et al., 1997; Huttenlocher et al., 1997; Potter et al., 1998), and myogenesis (Barnoy et al., 2000). Calpains are also involved with myofibrillar protein disassembly and degradation, contributing to loss of the Z disk (Busch et al., 1972). Increased calpain activity in muscle occurs during periods of muscle reorganization and restructuring (Dedieu et al., 2004), such as during weight loss and rapid growth (Salem et al., 2005; Johnston et al., 2011; Salmerón et al., 2015), therefore it is feasible that SNPs affecting calpain-related proteolysis contributes to differences in muscle growth capacity. Calpain-induced protein degradation is also associated with post-mortem proteolysis so genetic variation in the calpain system may result in differences in fillet quality (Koohmaraie, 1992; Delbarre-Ladrat et al., 2006).