Genotype-specific responses in Atlantic salmon (Salmo salar) subject to dietary fish oil replacement by vegetable oil: a liver transcriptomic analysis

Microarray results

Two-way ANOVA of the cDNA microarray dataset returned a high number of features showing evidence of differential expression for each factor - 713 for diet and 788 for genotype - and hence a more detailed analysis was restricted to the top 100 most significant hits for each factor, which were then categorised according to function (excluding 33-35% non-annotated features) (Figure 1). The functional category most affected by diet was that of metabolism (mainly lipid and carbohydrate metabolism), while immune response and intracellular trafficking were also affected. Within lipid metabolism, the affected genes are involved in PUFA, fatty acid and cholesterol biosynthesis (fatty acyl desaturases - Δ5 fad and Δ6 fad, fatty acid synthase - FAS, squalene monooxygenase and possibly cytochrome P450 reductase), glycerophospholipid metabolism (phospholipase D3) and acylglycerol homeostasis (angiopoietin-like 3). Some genes related to carbohydrate metabolism, implicated in glycolysis, glutamine/fructose 6-phosphate and glycerol-3-phosphate metabolism, such as alpha-enolase, glutamine-fructose-6-phosphate transaminase 1(GFPT1) and glycerol kinase, respectively, were also identified as being significantly affected by diet. Genotype had a lower impact on metabolism-related genes (primarily lipid and protein metabolism) and affected mostly genes involved in signalling. Regarding lipid metabolism, primary roles of affected genes are in glycerophospholipid metabolism (N-acylethanolamine-hydrolyzing acid amidase precursor, lipid phosphate phosphohydrolase 2 - LPP2 and 1-acyl-sn-glycerol-3-phosphate acyltransferase - AGPAT), fatty acid transport (intestinal fatty acid binding protein) and lipoprotein metabolism (apolipoprotein B - ApoB and endothelial lipase - EL). In addition, both factors had an effect on a relatively high number of transcription-related genes. Detailed lists of the top 100 most significant genes for diet and genotype, organised by biological function and including the normalised expression ratio between treatments, are shown in Tables 1 and 2, respectively. Gene Ontology enrichment analysis, which enables the identification of GO terms significantly enriched in the input entity list when compared to the whole array dataset, was performed for both factors, providing evidence for which biological processes may be particularly altered in the experimental conditions being compared. For diet, seven significant GO terms, all interrelated, were identified: oxidoreductase activity, stearoyl-CoA 9-desaturase activity, unsaturated fatty acid biosynthetic and metabolic processes, very long chain fatty acid (VLCFA) biosynthetic and metabolic processes. This is explained by the high number of Δ5 fad and Δ6 fad features that were significantly altered when dietary FO was replaced by VO (Table 1). In contrast, no GO terms were significantly enriched in the genotype list.

RT-qPCR

In terms of regulation of gene expression by the factor diet, the qPCR results confirmed the significant up-regulation of Δ5 fad and Δ6 fad in fish fed VO, with a higher fold change being measured for Δ6 fad. In addition, the expression ratio was higher in the Lean family group than in Fat fish, as had also been indicated in the microarray analysis. Of the elongase genes, only elovl2 was significantly up-regulated by the VO diet, but just in the Lean family group. Furthermore, quantification of PPAR genes revealed that only PPARα was down-regulated significantly when salmon were fed the VO diet, but only in the Lean family group. On the other hand, expression of SREBP-1 was only significantly affected in Fat fish, being up-regulated in fish fed the VO diet. Other genes which were significantly and consistently regulated were FAS and EL (both up-regulated when VO replaced FO in the diet), while GST, HOX and AGPAT only showed significant regulation in Fat fish. Finally, comparison between the two family groups showed a significantly lower expression of Δ5 fad, Δ6 fad, PPARα, PPARβ, SREBP-1 and GST in the Lean group but only when fish were fed FO, in the case of fads, or when fed the VO diet, in the case of PPARs, SREBP-1 and GST. In addition, FAS was also significantly down-regulated in the Lean group, independent of diet.

Liver fatty acid composition

Effects of diet on lipid metabolism

Within the list of genes affected by diet, those involved in fatty acyl desaturation were prominent, leading to the identification, through GO enrichment analysis, of several terms related to LC-PUFA biosynthetic and metabolic processes. The up-regulation of Δ5 fad and Δ6 fad in both family groups when dietary FO was replaced by VO was confirmed by RT-qPCR. Several studies have previously demonstrated up-regulation of genes involved in LC-PUFA biosynthesis in salmon when FO is replaced by VO [10, 14, 15]. RT-qPCR also confirmed previous work showing that elovl2 is responsive to dietary n-3 LC-PUFA levels [15], being the only elongase whose expression was up-regulated when FO was replaced by VO. However, a significant effect was only observed in the Lean family group. In addition, both microarray and RT-qPCR analyses indicated that the up-regulation of Δ5 fad and Δ6 fad showed a considerably higher fold-change in the Lean fish, due mainly to lower basal expression of fads in Lean salmon, compared to Fat, when fed FO. These results indicate that the activity of this biosynthetic pathway may be dependent on the genetics of the fish, with different family groups showing differences in the magnitude of response. The liver fatty acid composition revealed that differences in EPA and DHA levels between fish fed either diet were smaller in the Lean fish, due to higher n-3 LC-PUFA in fish fed VO and lower n-3 LC-PUFA in fish fed FO, compared to the equivalent treatments in the Fat group. In addition, intermediates in the biosynthetic pathway, such as 20:4n-3 and 22:5n-3, tended to be present at higher levels in the Lean family group, suggesting that differences observed in the levels of mRNA of LC-PUFA biosynthesis genes, which have been shown to correlate with the enzymatic activity of this pathway in salmon [16, 17], were reflected in biochemical composition.

Another lipid metabolism gene significantly affected by diet was FAS, which was up-regulated in both family groups when fed VO. A well demonstrated effect of dietary FO supplementation in mammals is hypotriglyceridemia, resulting from a coordinated effect of n-3 LC-PUFA in suppressing hepatic lipogenesis and enhancing fatty acid oxidation in liver and muscle [18]. Furthermore, this gene also appears to be regulated at a pre-translational level and hence changes in FAS transcription are likely to result in important effects in terms of enzyme activity [19]. Similar mechanisms are believed to operate in fish but, although reduced hepatic lipogenic activity modulated by LC-PUFA has been demonstrated in vitro[20], a direct relationship with dietary FO and VO has not always been clear in vivo[21, 22]. The regulation of FAS in response to FO replacement by VO did not show an interaction with the flesh leanness/fatness phenotype in this study, as might have been expected. This was because genotype also had a significant effect, with the Lean group having lower levels of FAS expression than the Fat fish, with a similar fold-change in both diets.

Regulation of lipid metabolism is complex and controlled by several transcription factors and nuclear receptors, including PPARs and SREBPs. SREBP-1c is a major regulator of lipogenesis in mammals [18]. Here we measured the expression of SREBP-1 as there is no evidence for the existence of alternatively spliced isoforms in salmon, and primers corresponded to an identical region in mammalian SREBP-1a and SREBP-1c [23]. Our results agree with Minghetti et al. [23], who showed SREBP-1 was increased by cholesterol and decreased by EPA and DHA supplementation in a salmon cell line, denoting a similar nutritional regulation to mammals [18]. However, there was a clear genetic effect as expression of SREBP-1 was 3-fold higher in Fat salmon fed VO, containing lower EPA, DHA and cholesterol, than in fish fed FO, whereas no regulation was observed in the Lean group.

PPARs have been less studied in fish than in mammals but present evidence suggests PPARα and PPARβ have similar ligands and functions to their mammalian homologues, while PPARγ may present some functional differences [24, 25]. LC-PUFA are well recognised enhancers of PPARα activity in fish, and while the response of PPARβ to LC-PUFA might be variable between fish species, an enhancement of activity in sea bass, plaice and sea bream [24–26] and of expression in Atlantic salmon [27] has been observed. In addition, and unlike rodents, PPARα and PPARβ have a similar pattern of expression in response to fasting and feeding in sea bream liver, indicating that they may be regulated similarly [25]. In the present study, PPARα was down-regulated when VO replaced FO but only in the Lean family group and, although not statistically significant, PPARβ showed a similar trend, suggesting similar transcriptional regulation of these nuclear receptors by dietary fatty acid composition. These results thus indicate that the genetic background of the fish might affect PPAR transcriptional responses to LC-PUFA. In contrast, no nutritional regulation was observed for PPARγ transcription in liver, in accordance with previous studies in fish, including salmon, and its predominant role in adipocytes [24, 28].

The hypotriglyceridemic effects of n-3 LC-PUFA in mammals involve activation of PPARα, leading to up-regulation of β-oxidation genes (including carnitine palmitoyltransferase I - CPT1 and acyl-CoA oxidase - ACO) and suppression of SREBP-1c transcription that down-regulates lipogenic enzymes [29, 30]. As previously reported, FAS expression was up-regulated in both family groups fed the VO diet but neither CPT1 nor ACO expression, was affected [12]. As elovl2 expression was only altered in the Lean fish and both Δ5 fad and Δ6 fad showed greater up-regulation in Lean salmon fed VO, we may speculate that PPARα (and potentially also PPARβ) expression may be involved in down-regulation of LC-PUFA biosynthesis. Paradoxically, fatty acyl desaturases are regulated by both SREBPs and PPARs in mammals [31]. In addition, PPARα agonists regulate the transcriptional activity of elongases in rat, although only elovl5 and not elovl2 [32]. However, in mammals, PPARα ligands induce the transcription of elongases and desaturases while we observed an up-regulation of elovl2 and a stronger stimulation of Δ5 fad and Δ6 fad transcription when PPARα expression was lower. In the rat and human Δ6 fad gene promoters, both PUFA and PPARα response regions have been identified which suppress and induce, respectively, Δ6 fad expression [33]. The molecular mechanisms of transcriptional regulation of these genes are complex and will require further investigation in salmon [34]. In contrast, target genes of SREBP-1 remain elusive and, although it may regulate FAS expression [23], this was only observed in Fat fish whereas, in the Lean group, another mechanism is required to explain up-regulation of FAS in VO-fed fish as expression of SREBP-1 was unaffected. Nonetheless, the action of SREBP-1 is under the regulation of liver X receptor (LXR) and these complex pathways have only recently started to be investigated in fish [23].

Another gene affected by diet was squalene epoxidase (SQLE), which was up-regulated by VO but only markedly in the Lean family group. This enzyme catalyses the first oxygenation step in sterol biosynthesis, a pathway identified earlier as presenting a diet × genotype interaction [12]. In contrast, cytochrome P450 reductase (CPR) was down-regulated in salmon fed VO, particularly in Lean fish. This enzyme has multiple roles as the electron donor for several oxygenase enzymes, such as cytochrome P450 (involved in drug and xenobiotic metabolism, and sterol and bile acid synthesis), HOX and cytochrome b5 (which supports both sterol and LC-PUFA biosynthesis pathways). In addition, it has key roles in the biosynthesis of several signalling factors and the regulation of oxidative response genes [reviewed by [35]]. CPR is transcriptionally regulated by PPARα in mouse and, given the comparable PPARα and CPR expression in Lean salmon fed VO, similar regulation likely occurs in salmon. However, changes in CPR expression can be related to several processes that were affected by FO replacement. Thus, CPR expression could be linked to changes in both cholesterol and LC-PUFA biosynthesis, both more marked in Lean fish, although this is unlikely because VO induced up-regulation of these pathways. A more likely association is with cell oxidant metabolism, also suggested by the microarray results as being possibly down-regulated in VO-fed fish. In particular, down-regulation of HOX in salmon fed VO, more marked for Lean fish correlating with CPR expression, might be an indication of this.

Effect of diet on carbohydrate and intermediate metabolism

Within the metabolism genes that were identified by the microarray analysis as being significantly affected by dietary oil substitution, a few relate to carbohydrate metabolism, particularly glucose and intermediary metabolism. Given that similar effects were observed in previous salmonid studies, and that a few signal transduction genes present in the list of diet significant effects are also potentially implicated in these pathways, these results warrant further discussion, even if the observed fold changes were low. An association between lipid and carbohydrate metabolism in salmon is not surprising given that the pathways of lipogenesis, lipolysis, glycolysis, gluconeogenesis and pentose phosphate shunt are all interrelated in the regulation of body energy homeostasis. In mammals, the role of LC-PUFA as "fuel partitioners" involves both directing fatty acids away from anabolic and towards catabolic routes as well as enhancing glucose flux to glycogen, mediated by effects on SREBP-1 and transcription factors that regulate key genes of lipid metabolism and glycolysis [30]. Similar mechanisms may operate in fish but differences are likely given that carnivorous fish like salmon have low capacity to use carbohydrate and appear to show features of glucose intolerance [36, 37]. Nonetheless, dietary n-3/n-6 ratio has been shown to influence mRNA levels of the glucose transporter GLUT4 in Atlantic salmon muscle, with some reflection in plasma glucose [38]. In addition to a decreased hexokinase and phosphoenolpyruvate carboxykinase expression, complete replacement of FM and FO by vegetable alternatives in rainbow trout resulted in a slightly increased expression of glycerol kinase, as observed here [11]. This enzyme is at the intersection of lipid-carbohydrate metabolism and over-expression of this gene in human muscle and rat hepatoma cells resulted in higher TAG synthesis and up-regulation of the pentose phosphate pathway providing reducing power for lipogenesis [39]. Panserat et al. [11] hypothesised that the up-regulation of glycerol kinase may be related to higher lipid biosynthesis in liver when trout were fed plant-based diets. Similarly, our results, associated with the observed changes in FAS mRNA when VO replaced FO, suggest a possible relationship with lipogenesis. Also possibly related with this was the up-regulation of two different biotinidase clones with the potential to increase availability of substrates for FAS and/or gluconeogenesis in VO-fed fish. This gene, besides being involved in the regulation of gene expression, including genes of glucose metabolism, codes for an enzyme that recycles biotin, which is a co-factor for several carboxylases responsible for production of substrates for lipogenesis and gluconeogenesis [40].

Another gene affected by diet was alpha-enolase, which was slightly down regulated in Lean fish fed VO. A similar effect was observed in liver of salmon fed rapeseed oil in comparison to FO [9]. This glycolytic enzyme participates in the conversion of glucose to pyruvate, a key intermediate at the intersection of multiple metabolic pathways, including lipogenesis. Thus, this might result in lower levels of pyruvate for conversion to acetyl-CoA in VO-fed fish. This result does not necessarily conflict with an increase in lipogenesis given that, in fish, carbon skeletons for de novo fatty acid production are mainly derived from amino acid catabolism rather than from carbohydrates, whose main contribution towards lipogenesis is to supply NADPH via the pentose-phosphate pathway [37].

Finally, a few signalling genes that were significantly affected by diet might also have an effect on glucose metabolism, assuming that similar cascades exist in fish. One of these is phosphoinositide 3-protein kinase (PI3K), which mediates insulin's effects on glucose, lipid and protein metabolism, and that was significantly down regulated in VO-fed fish. Among other roles, it regulates glucose cellular uptake in mammals, recruiting GLUT4 transporters to the cell surface [41]. In addition, it is found upstream of a signal transduction cascade regulating glycogen synthesis through glycogen synthase, by inactivating glycogen synthase kinase-3 (GSK3) [41, 42]. In our study, expression of GSK3-binding protein (GBP) was significantly increased in VO-fed Lean fish. GBP is a protein that blocks GSK3, which in turn inactivates glycogen synthase [43]. Hence, it is possible that the oil composition of the diet might also affect glucose metabolism and glycogen storage.

Effect of diet on oxidative stress and immune response

Increased oxidative stress associated with the consumption of FO has been typically reported in fish and mammals [27, 44, 45]. Accordingly, genes related to oxidant metabolism were found in the significant list for diet. A thioredoxin domain-containing protein, possessing an antioxidant role [46], and GST, which detoxifies peroxidised lipids and xenobiotics [47], were down-regulated in salmon fed VO, consistent with the higher auto-oxidative potential of LC-PUFA in FO. However, quantification of GST by RT-qPCR was not consistent with the microarray result, although the possibility exists that different GST genes with differential regulation exist in salmon and this requires clarification. In addition, the observed down-regulation of HOX in VO-fed fish, validated by RT-qPCR, might be related to a decrease in oxidative stress in these fish. This enzyme catalyses the degradation of heme and can be induced by oxidative stress [48] and may be increased during pro-inflammatory states, being thought to increase resistance to oxidative injury and ameliorate inflammation [49]. The n-3 LC-PUFA in FO have important anti-inflammatory actions in mammals [50], which does not correlate with the expression of HOX and its putative role in inflammation in this case. Inflammation is an important mechanism in immune defence but, in fish, the demonstrated effects of LC-PUFA on immune and inflammatory mechanisms have been inconsistent [45]. However, a recent study has clearly shown an effect of dietary oil composition on the progression of a myxosporean parasite infection in Gilthead sea bream, with fish fed the VO diet showing higher signs of the disease and faster course of infection in comparison with those on a FO diet [51]. On the other hand, the synthesis of pro-inflammatory eicosanoids was increased in the intestine of salmon fed vegetable-based diets in response to acute stress [52]. In the present study immune response was the second highest category of genes affected by diet, after metabolism. Whether this is due to the potential anti-inflammatory role of dietary FO or whether VO diets can have detrimental health effects is not clear as the fold-changes were subtle, as expected in unchallenged animals. Nonetheless, the majority of genes related to processes of both innate and adaptive immunity were up-regulated in fish fed VO. Only T-cell and leukotriene B₄ (LTB₄) receptors, that are reduced after antigen and LTB₄ exposure, respectively, and, in the case of LTB₄ receptor, increased after EPA administration [53–55], were down-regulated in salmon fed VO.

Differences in gene expression between Lean and Fat genotypes

Muscle adiposity is a trait of great importance in animal production, aquaculture included, and hence physiological changes induced by genetic selection for this phenotype have been examined in various animals, including rainbow trout [7, 8]. In the present study the main differences between family groups were associated with signal transduction pathways, followed by metabolism. Only a small number of lipid metabolism genes varied in relation to muscle adiposity, as reported previously in rainbow trout, where the main differences were related to lipogenesis and mitochondrial oxidative metabolism [7, 8]. In our study glycerophospholipid metabolism may have been down-regulated in the Lean family group through AGPAT and LPP2, two enzymes acting consecutively on de novo TAG and phospholipid biosynthesis [56, 57]. Quantification of AGPAT and LPP2 expression by RT-qPCR confirmed this down-regulation but fold-changes were too subtle to be significant. AGPAT converts lysophosphatidic acid into phosphatidic acid (PA), while LPP2 then catalyzes the conversion of PA to diacylglycerol. All these molecules can function as second messengers and are involved in the regulation of multiple signalling pathways. Therefore, down-regulation of this pathway in the Lean group has the potential to lower lipid biosynthesis, at least partly explaining the flesh lipid phenotype, but may also alter levels of lipid signalling molecules. On the other hand, differences in muscle adiposity might also be caused by higher hepatic "de novo" fatty acid synthesis in the Fat family group, as indicated by the expression of FAS. In a previous study, no differences were found in the expression of ACO and CPT1, which suggested that the phenotypes could not be explained by differences in β-oxidation [12]. By contrast, in rainbow trout Fat and Lean families, β-oxidation and mitochondrial oxidative metabolism, but not lipogenesis, were affected by genetic selection [7], although another study using the same trout lines suggested differences related to lipogenesis rather than fatty acid oxidation [8]. Thus, both metabolic processes are likely involved and discrepancies in the data are likely due to lack of methodological sensitivity to detect the small fold-changes that are possibly characteristic of these biological processes and typical in this type of experiment.

PPARα, PPARβ and SREBP-1 were also regulated in response to genotype, being down-regulated in Lean fish, but only when fed the VO diet. In cobia, Rachycentron canadum, a negative correlation was found between PPARα mRNA levels in liver and body lipid deposition [58]. Furthermore, PPARβ appears to play a similar role in fish to that in mammals, as a ubiquitous regulator of fat burning and with a role in energy mobilisation during early development [24, 25]. Therefore, both PPARα and PPARβ might have a role in the control of adipogenesis in fish and it may be the case that, similarly to chickens [59], Fat salmon might have higher lipid turnover than their Lean counterparts when fed a diet that predisposes for hepatic fat deposition, even though the end result is higher lipid accumulation in liver [60]. To explain this, Collin et al. [59] suggested that a fat chicken family is better "equipped" to deal with higher circulating levels of TAG when fed a high fat diet, compared to lean chicken. On the other hand, we observed a direct relationship between SREBP-1 and FAS expression in the Fat family group in response to diet, as well as in VO-fed fish in response to genotype. It thus appears that SREBP-1 may be partly responsible for higher lipogenesis in Fat fish, compared to Lean, when fed VO.

Feeding trial and sampling

A dietary trial was conducted using two genetically characterised and contrasting groups of farmed Atlantic salmon post-smolts, comprising full-sib families selected from the Landcatch Natural Selection Ltd (LNS) breeding program (Argyll, Scotland). The choice of the two family groups was based on estimated breeding values (EBVs) of the parents for high or low flesh adiposity, assessed by Torry Fatmeter (Distell Industries, West Lothian, UK), a trait that was found to have a heritability ranging from 0.17 to 0.39 in this dataset. The two groups were created from four unrelated full-sib families; two families from the extreme lower end of the EBV distribution for flesh lipid content ('Lean') and two families from the extreme upper end of the distribution ('Fat'). The average EBV for the lipid content of the two Fat families was 2.00 percentage units higher than that of the two selected Lean families, representing a standardised selection differential of 2.33 standard deviations. Assessment of the flesh and viscera lipid content at the end of the feeding trial confirmed differences in adiposity between the two genotypes, in spite of an interaction with diet being also found [12].

Two thousand fish of each group were stocked into eight 12 × 5 m³ net pens at the Ardnish Fish Trials Unit (Marine Harvest Scotland, Lochailort, Highland; 500 fish pen^-1). Duplicate pens from each group of fish were fed one of two experimental diets (Skretting ARC, Stavanger, Norway) containing 32-25% fish meal, 40-45% plant meals and 27.5-30% oil supplied either as northern fish oil (FO) or as a vegetable oil (VO) blend comprising rapeseed, palm and Camelina oils in a ratio of 5:3:2 [12]. Diets were formulated to fully satisfy the nutritional requirements of salmonid fish [61] and contained similar levels of PUFA (around 31%) but different n-3 and n-6 PUFA contents, 25.3% and 4.6% in the FO diet and 13.4% and 17.1% in the VO diet, respectively. Further details including full diet formulations, proximate and fatty acid compositions of the feeds can be found in Bell et al. [60].

After 55 weeks on the experimental diets 25 fish were sampled per pen. The fish were killed by a blow to the head following anaesthesia using MS222, 24 h after the last meal. Samples of liver were immediately frozen on dry ice and stored at -70°C for molecular and fatty acid analyses.

RNA extraction and purification

Liver tissue (0.2 g) from six individuals per experimental group was rapidly homogenised in 2 mL of TRI Reagent (Ambion, Applied Biosystems, Warrington, U.K.) using an Ultra-Turrax tissue disrupter (Fisher Scientific, Loughborough, U.K.) and stored at -70°C. Total RNA was later isolated, following manufacturer's instructions, and RNA quality (integrity and purity) and quantity was assessed by gel electrophoresis and spectrophotometry (NanoDrop ND-1000, Thermo Scientific, Wilmington, U.S.A.). One hundred micrograms of total RNA from each individual sample was further cleaned by mini spin-column purification (RNeasy Mini Kit, Qiagen, Crawley, UK), and then re-quantified and quality assessed as above.

Microarray hybridizations and image analysis

The TRAITS/SGP (v.2.1) salmon 17 k cDNA microarray, described in detail by Taggart et al. [10], was used in this experiment (ArrayExpress accession: A-MEXP-1930). A dual-label experimental design was employed for the microarray hybridisations. Each experimental sample was competitively hybridised against a common pooled-reference sample, which comprised equal amounts of all samples used in the study. This design permits valid statistical comparisons across all treatments to be made. The entire experiment comprised 24 hybridisations - 2 genotypes (Lean/Fat) × 2 diets (FO/VO) × 6 biological replicates.

An indirect labelling methodology was employed in preparing the microarray targets. Antisense amplified RNA (aRNA) was produced from 500 ng of purified total RNA per sample using the Amino Allyl MessageAmpTM II aRNA Amplification Kit (Ambion, Applied Biosystems), as per manufacturer's instructions, followed by Cy3 or Cy5 fluor incorporation mediated by a dye-coupling reaction, as previously described in detail [12]. Experimental samples and the pooled reference sample (batch reaction) were labelled with Cy3 and Cy5 dye suspension stocks (PA23001 or PA25001, GE HealthCare, Little Chalfont, UK), respectively. Unincorporated dye was removed by column purification (Illustra AutoSeq G-50 spin columns; GE Healthcare). Dye incorporation and aRNA yield were quantified by spectrophotometry (NanoDrop ND-1000) and further quality controlled by separating 0.4 μL of the sample through a thin mini-agarose gel and visualising products on a fluorescence scanner (Typhoon Trio, GE Healthcare).

Microarray hybridisations were performed in a Lucidea semi-automated system (GE Healthcare), without a pre-hybridisation step. For hybridisation of each array, each labelled biological replicate and corresponding pooled reference (40 pmol each dye, c. 150 ng aRNA) were combined and added to the hybridisation solution, comprising 185 μL 0.7X UltraHyb buffer (Ambion), 20 μL poly(A) at l0 mg/mL (Sigma-Aldrich, Dorset, UK), 10 μL herring sperm at c. 10 mg/mL (Sigma-Aldrich) and 10 μL ultra pure BSA at 10 mg/mL (Sigma-Aldrich), as detailed previously [12]. Two post-hybridisation automatic washes followed by six manual washes to a final stringency of 0.1× SSC (EasyDipTM Slide staining system; Canemco Inc., Quebec, Canada) were performed before scanning.

Scanning was performed at 10 μm resolution using an Axon GenePix 4200AL Scanner (MDS Analytical Technologies, Wokingham, Berkshire, U.K.) with laser power constant (80%) and "auto PMT" enabled to adjust PMT for each channel such that less than 0.1% of features were saturated and that the mean intensity ratio of the Cy3 and Cy5 signals was close to one. BlueFuse software (BlueGnome, Cambridge, U.K.) was then used to identify features and extract fluorescence intensity values from the resultant TIF images. Following a manual spot removal procedure and fusion of duplicate spot data (BlueFuse proprietary algorithm), the resulting fluorescence intensity data and quality annotations for the 17,102 gene features, were exported into the GeneSpring GX version 10.0.2 analysis platform (Agilent Technologies, Wokingham, Berkshire, U.K.) after undergoing a block Lowess normalisation. Data transformation and quality filtering were then performed and all control features were excluded from subsequent analyses [12]. This returned a list of 14,772 genes eligible for statistical analysis. Experimental annotation complied fully with minimum information about a microarray experiment (MIAME) guidelines [62]. The experimental hybridisations and further methodological details are archived on the EBI ArrayExpress database (http://www.ebi.ac.uk/arrayexpress/) under accession number E-TABM-1089.

RT-qPCR

For RT-qPCR, 1 μg of column-purified total RNA per sample was reverse transcribed into cDNA using the VersoTM cDNA kit (ABgene, Surrey, U.K.), following manufacturer's instructions, using a mixture of random hexamers (400 ng/μL) and anchored oligo-dT (500 ng/μL) at 3:1 (v/v). Negative controls (containing no enzyme) were performed to check for genomic DNA contamination. A similar amount of cDNA was pooled from all samples and the remaining cDNA was then diluted 20-fold with water. RT-qPCR analysis used relative quantification with the amplification efficiency of the primer pairs being assessed by serial dilutions of the cDNA pool. qPCR amplifications were carried out in duplicate (Quantica, Techne, Cambridge, U.K.) in a final volume of 20 μL containing either 5 μL or 2 μL (for the reference genes and HOX) diluted (1/20) cDNA, 0.5 μM of each primer and 10 μL AbsoluteTM QPCR SYBR^® Green mix (ABgene). Amplifications were carried out with a systematic negative control (NTC-non template control). The qPCR profiles contained an initial activation step at 95°C for 15 min, followed by 30 to 40 cycles (depending on target): 15 s at 95°C, 15 s at the specific primer pair annealing temperature (Ta; Table 5) and 15 s at 72°C. After the amplification phase, a melt curve of 0.5°C increments from 75°C to 90°C was performed, enabling confirmation of the amplification of a single product in each reaction. RT-qPCR product sizes were checked by agarose gel electrophoresis and the identity of amplicons of newly designed primers (FAS, GFPT1, HOX, LPP2 and AGPAT) was confirmed by sequencing.

Lipid extraction and fatty acid analyses

Total lipids from six fish per treatment were extracted and determined gravimetrically from 1-2 g of liver by Ultra Turrax homogenisation in 20 volumes of chloroform/methanol (2:1 v/v) [63]. Fatty acid methyl esters (FAME) were prepared by acid-catalysed transesterification of total lipids [64]. Following purification, FAME were separated and quantified by gas-liquid chromatography using a Thermo Fisher Trace GC 2000 (Thermo Fisher, Hemel Hempstead, UK) equipped with a fused silica capillary column (ZB wax, 30 m×0.32 mmi.d.; Phenomenex, Macclesfield, UK) with hydrogen as carrier gas and using on-column injection. The temperature gradient was from 50 to 150°C at 40°C/min and then to 195°C at 1.5°C/min and finally to 220°C at 2°C/min. Individual methyl esters were identified by comparison with known standards. Data were collected and processed using the Chromcard for Windows (version 2.00) computer package (Thermoquest Italia S.p.A., Milan, Italy).

Statistical analysis

Microarray hybridisation data were analysed in GeneSpring GX version 10.0.2 (Agilent Technologies) by two-way ANOVA, which examined the explanatory power of the variables 'diet' and 'genotype' (diet×genotype interaction presented in [12]), followed by Gene Ontology (GO) enrichment analysis, at a significance level of 0.05. No multiple test correction was employed as previous analyses, confirmed by RT-qPCR, indicate that such corrections are over-conservative for this type of data [14]. Gene expression results assessed by RT-qPCR were analysed by the ΔΔCt method using the relative expression software tool (REST 2008, http://www.gene-quantification.info/), employing a pair wise fixed reallocation randomisation test (10,000 randomisations) with efficiency correction [65], to determine the statistical significance of expression ratios between two treatments. Finally, significant differences in liver fatty acid composition were determined by means of two-way ANOVA, at a significance level of p < 0.05, using the Graphpad Prism™ (version 4.0) statistical package (Graphpad Software, San Diego, CA).