Transverse sections of skin from all the experimental groups at both time points (days 3 and 7) of the experiment were analysed (Figure 1). Sea bream skin had the typical organisation of teleost skin and was composed of three well defined layers: the epidermis, dermis and hypodermis which overlaid a fat layer that varied in thickness. The scales were each enclosed within a scale pocket and were composed of a mineralized external layer and a partially mineralized basal plate. The scale pocket was localized in the superficial dermis and projected into and was covered by a thin layer of epidermis (Figure 1B, H and 1I). Removal of the scales damaged the epidermis, dermis and scale pocket, the latter two tissues became exposed to the ambient water and the epidermis which remained attached to the dermis hung loose (Figure 1A). The ontogeny of the regenerative response was similar in all sea bream. Histology of the day 3 samples revealed a rapid repair process, with the epidermis already re-established and the enclosed scale pocket without a scale was visible in the dermis (Figure 1C and 1F). Hence within 3 days, the animals had re-established their external barrier and protection to the environment and 7 days after scale removal a thin regenerated scale was visible (Figure 1D, E and 1G). From a morphological perspective the regeneration process in sea bream was similar to that described in the cichlid Hemichromis bimaculatus 1 30 and also in zebrafish and goldfish 8 18 .

No significant difference in plasma glucose or plasma lactate (frequently used as indicators of physiological stress) was observed between any of the experimental groups analyzed on day 3 of the experiment (Table 1). Plasma glucose and lactate levels were both within the normal range for fish with normal metabolism and not suffering from stress 31 32 33 . No significant difference in total plasma calcium was found between any of the experimental groups and the control (Table 1), with concentrations in the normal range for intact animals of this species 34 . Plasma phosphorus levels also varied within normal levels (Table 1). However, a significant (p < 0.05) reduction in plasma phosphorus was measured in animals without scales which were fasted (2.685 mmol/L ± 0.225) in relation to the fed animals without scales (3.755 mmol/L ± 0.234). Nutrient depletion will amplify the effects of scale removal as both will cause increased mineral requirements by the fish in order to maintain whole body calcium and phosphorus homeostasis. Phosphorus is mainly obtained via the diet, whilst calcium can be obtained from both the diet and seawater 35 . Hence when fish are deprived of food the requirement for these minerals will be evidenced first via the phosphorus measurements (as observed here) that probably acts as an indicator of enhanced calcium mobilization from sea water by the fish.

Although the sea bream oligo-array had been previously annotated 36 , the sequences of the oligos used in the microarray were reanalysed in order to take advantage of the recent large increase in molecular data available for teleosts. Of the 39,379 oligo probes on the array, 16,025 (40.6%) showed significant match similarity to a known protein in uniprot database (1e^-10) 37 . To facilitate the understanding of the underlying cellular processes of epidermis and scale regeneration, a number of comparisons were carried out at days 3 and 7 after scale removal. Control animals (N) were compared with fed animals without scales (WS), fasted animals (ST) and fasted animals without scales (STWS). To specifically dissect out the enhanced effects of scale removal under conditions of nutrient depletion, an additional comparison of fasted animals (ST) with fasted animals which had had scales removed (STWS) was made. Table 2 shows the numbers of differentially regulated genes under these comparisons, with the major effect shown for the fasted vs. fasted without scales analysis (additional file 1). It is clear that within the skin, the response to scale removal is rapid and of short duration (even under food deprivation). To obtain a clear overview of the transcripts with a conserved response between the different comparisons Venn diagrams were generated for the up-regulated genes at both day 3 and 7 (Figure 2). For example of the 53 up-regulated genes in fish without scales compared to the control, 27 were also significantly up-regulated in unfed fish without scales compared to control and fasted fish.

By day 7 there were much reduced levels of differential expression between groups with only 49 up-regulated probes compared to 729 up-regulated probes over the four comparisons at day 3. The considerable difference in gene expression found between day 3 and 7 reflects the rapid repair response and temporally different processes that accompany skin repair. Since most of the expression changes took place within three days and resulted in the differential expression of a considerable number of probes, an initial overview of the major processes involved was conducted before a more detailed gene-by-gene analysis.

Microarray probes were classified according to their Gene Ontology terms (GO ID) in order to determine whether particular biological processes were enriched in response to the different treatments. Overall, 25.3% of the probes were associated with at least a GO term. The most represented Biological Processes on the microarray were "cellular processes" (GO:0009987) (24%), "regulation of biological process" (GO:0050789) (16%), "response to stimulus" (GO:0050896) (13%) and "multicellular organismal development" (GO:0007275) (12%) (Additional file 2). Interestingly, when GO enrichment analysis was performed on differentially expressed genes from different comparisons, no particular Biological Process term was enriched amongst the up-regulated gene lists with the exception of "cellular processes" between the STWS and the ST groups. The down-regulated gene lists revealed a significant reduction in metabolic processes, indicating that the animals were repartitioning their translation machinery away from normal housekeeping functions towards repair and regeneration (data not shown). This conclusion is substantiated by the Ingenuity pathway analysis software (IPA) results which identified the main molecular and cellular biological functions that were significantly affected (Table 3) and also which physiological systems with regard to development and function were involved (Additional file 3). The IPA top networks for all the comparisons which included animals with scales removed produced matches to cancer, indicating that genes which have been implicated in non-controlled cell proliferation in human may be involved in normal cellular proliferation in fish skin. Not surprisingly the top networks also included those involved in the cell cycle, cellular growth and proliferation, and biological processes included tissue/organ development and morphology and haematopoiesis. Lipid metabolism was one of the most significant functions represented in IPA in the fasted fish indicating the effects of nutrient depletion on the general metabolism of the animals. This finding was substantiated in the STWS comparisons, which also included networks involved in vitamin and mineral metabolism (Table 3, Additional file 3). The Ingenuity results, whilst providing an overview of the main cellular processes affected in the experiments, provide more detail than the simplified GO enrichment analyses and link in far more directly to analysis of individual genes and their putative functional identification.

Analysis of the differentially expressed genes in each of the comparisons was restricted to up-regulated genes and those which could be assigned a putative function via the Uniprot/Swissprot and Uniprot/Trembl databases 37 . Nonetheless, one aspect to bear in mind is that approximately 50% of the oligoarray transcripts had no known match to any transcripts with functional annotation which limits the overall analysis and therefore the pathways invoked could only be inferred from those genes with a functional annotation. The unknowns will form an important aspect of future investigations, particularly those that are differentially expressed in more than one comparison (see Figure 2 and additional file 1). The following discussion focuses on the results of the samplings at 3 days.

This initial analysis compared the most differentially expressed probes in the group of animals which had scales removed with control animals (Table 4). These probes shared high sequence similarity with genes involved in cell cycle regulation, cell proliferation and adhesion, immune response and antioxidant activities. Whilst many of the putative functions have been ascribed from human or mammalian research, both the receptor-transporting protein 2 and IFI56 have been identified in salmon and carp, respectively as interferon responsive genes induced in response to viral infections 38 39 . In addition Galectin 3 and LOC406638 have putative roles in the immune response, whilst methionine sulfoxide reductase and cytochrome p450 2W1 have antioxidant activities, indicating that removal of scales provoked an inflammatory response, with activation of cell defence mechanisms to protect the animal against the breach in external protection.

During regeneration, the immune system is important for immune surveillance and control of pathogens, but there is increasing awareness of the importance of immune physiology. The latter term refers to the role of the immune system in tissue homeostasis and it is increasingly recognised that complement, lymphocytes and monocyte derived cells (MDC) promote tissue growth and regeneration in mammals (reviewed by 40 ). This aspect has received little attention in fish, as research about immune functioning is generally focussed on infection or disease control, an important priority for aquaculture (see reviews by 41 42 43 ). Immunological diseases and lymphoid tissue structure and development are an enriched category (IPA) for transcripts in fish with regenerating skin and scales. In addition to the probes listed in the tables for the different treatments, transcripts for chemokines (CCL3 and CCL4) associated with monocytes and macrophages were also identified and it remains to be established if their presence is associated with immune surveillance or immune physiology and tissue regeneration 44 . Recent investigations in stem cell biology have linked several molecules traditionally associated with the immune cells/function to stem cells. For example, immune associated transcripts differentially expressed in skin/scale from fasted fish 3 days after scale removal included CD55 (delay accelerating factor), a modulator of complement activity and a recently identified candidate surface marker for early and late definitive endoderm 45 . Similarly, CD200, a novel immunosuppressive molecule, has been identified as a biomarker of the hair follicle bulge in human and dog skin 46 47 48 . Future work will be required to establish the role of immune physiology and cellular defence mechanisms in the regenerating fish skin and also the involvement of stem cells in this process.

At the same time as the immune response, there is a clear requirement to rapidly re-construct this external barrier with various genes involved in metabolic processes such as amino acid biosynthesis and also cell division and proliferation. Interestingly, in a link with the IPA results, several of these genes have been described in cancer studies. Cyclin-dependant kinase inhibitor is involved in haematopoietic cell cycle regulation and has been shown to be over-expressed in breast and prostate cancer 49 ; S-phase kinase-associated protein interacts with c-myc during the G1-S phase transition of the cell and is a co-factor of c-myc which is a known transcriptional regulator of oncoproteins and involved in cell growth, apoptosis and oncogenesis 50 ; whilst the mitotic check point serine/threonine protein kinase has been shown to be preferentially expressed in cells with a high mitotic index 51 . Adaptation to new conditions involves an element of cytoskeletal re-modelling 52 , as evidenced by the up-regulation of cytokeratin which has been associated with epidermis development, fibrinolysis and also regulation of angiogenesis. It is tempting to speculate that the up-regulation of cytokeratin in response to scale removal may represent a keratinization-like phenotype provoked by the osmotic shock. There was also up-regulation of genes involved in apoptosis such as Galectin-3 and the multifunctional S-phase kinase-associated protein (described earlier) and the somewhat confusingly named cation transport regulator-like protein 50 53 . Hence competing interests between infection/inflammation control and cellular proliferation/tissue repair in fish with scales removed appear to be ongoing.

Skin tissue metabolism is clearly being redirected as the animals cope with adaptation to food deprivation. One aspect of this is a reduction in cell division with the up-regulation of angiopoietin-related protein 4 (ANGPTL4), MYND and KIAA0711, all of which have been shown to play roles in the inhibition of proliferation (Table 5) 54 55 56 . Another regulator of a proto-oncogene (MYC) is present in the form of ubiquitin carboxyl-terminal hydrolase and levels of MYC decline is response to intracellular stress signals 57 . Molecular signals of cell stress are also present with up-regulation of antioxidants (glutathione-S-transferase, methionine-R-sulphoxide reductase, alcohol dehydrogenase and cytochrome p450 2W1).

During periods of food deprivation fish seem to maintain energy homeostasis, at least during the initial stages of fasting, by mobilizing energy reserves such as lipids and hepatic glycogen and reduction in the rate of glucose utilization and enhancement of lipid metabolism 31 58 59 60 . In fact, the genes differentially expressed in the array from the groups in which food was withheld suggests that lipid metabolism and angiogenesis are the main processes induced in the treated fish. Pyruvate dehydrogenase is involved in production of energy via glucose metabolism and ANGPTL4, in addition to a role in non-proliferation, and has also been shown to be a regulator of glucose homeostasis, lipid metabolism and angiogenesis 61 62 , but this more conventional pathway may be supplemented by the actions of serine/threonine kinase ULK1 and a patatin-like gene. The former has been shown to be involved in autophagy induced by nutrient depletion to provide essential amino acids within cells, whilst the latter may enhance hydrolysis of triglycerides to provide free fatty acids to other tissues to be oxidised in situations of energy depletion 63 64 . Taken together the results appear to indicate that fish under food deprivation "slow down" their metabolism to save energy and break down macromolecules to release energy.

Interestingly, two of the genes putatively identified here play roles in human diseases, which may be of relevance to the condition of the fish in this experiment. Myospryn has been shown to be up-regulated in hypertrophy inducing conditions in humans and is involved in maintaining muscle integrity 65 and the phenotype of mutants of the CD151 antigen include fragility of the skin and mucus membranes 66 . Starvation directly affects muscle wastage in mammals and fish 67 68 . Hence these genes may be playing a similar structural role in fish as they do in humans, and represent novel candidates for understanding this physiological response in fish.

The most differentially regulated genes in this group of animals display a gene expression profile, which is intermediate between the previous two (Table 6 and Figure 2) with representatives of cell proliferation and cell cycle control genes, energy homeostasis, antioxidant repair enzymes and the immune response. The results of the gene expression profiles in this group clearly represent the whole organism trade-offs that are occurring within the fish for several competing essential cellular processes. Food deprivation leads to a reduction in metabolism, but if the animal is challenged, then there is the question of what predominates in terms of the minimal requirements for survival. Trade-offs occur 69 and a recent study in salmon clearly documents the competing transcriptomic responses to food deprivation and immune challenge 70 . Which requirements predominate in this study is difficult to determine and entail further studies. Perhaps, not surprisingly, there is an indication that repair processes are slowed under food deprivation with the enhanced presence of genes involved in blood coagulation and wound healing (Beta-2-glycoprotein I and lymphatic vessel endothelial hyaluronic acid receptor). To verify this hypothesis, further experimentation will be required with a more detailed sampling regime over the same or a slightly elongated time course with the same treatments.

Curiously, one of the genes up-regulated in this group of animals, cytosolic sulfotransferase 2, which is involved in detoxification reactions, and participates in the activation and deactivation of hormones, neurotransmitters, steroids and bile acids 71 , has been correlated with low plasma P levels in trout 72 . The preceding study developed several intestinal mRNA biomarkers for P depletion in the rainbow trout although only sulfotransferase 2 was modified in the present study presumably because the target tissue was different. The results of the present study in the sea bream suggest that cytosolic sulfotransferase 2 may be a promising general marker of P depletion in fish and certainly merits further investigation.

There was not such a pronounced effect as had been expected with scale removal and food deprivation and so an additional comparison was carried out between fasted animals and fasted animals with scales removed. Indeed, this comparison did produce the highest number of differentially expressed genes (429 up-regulated probes, compared with 53 (WS), 181 (ST) and 66 (STWS) when the latter three treatments were compared with control animals. Surprisingly little overlap in significantly up-regulated transcripts or modified networks were found between animals without scales and fasted animals without scales (see Tables, 3, 4, 7 and Figure 2A). The reason for this lack of overlap is difficult to explain but may result from asynchronous regeneration associated with the slow-down in cellular metabolism and activation of alternative pathways to ensure barrier function when food is in short supply. This comparison provides the clearest signal of the sea bream response to scale removal with half of the 20 most up-regulated annotated probes involved in cell division and mitosis (Table 7). These probes are ideal candidates for the monitoring of cell division processes related to the regeneration of scales. Of the remaining annotations, there are representatives of cell growth and metabolism (Interleukin enhancer binding factor, phosphoserine aminotransferase, D-3-phosphoglycerate dehydrogenase), cell proliferation (Macrophage stimulating 1 hepatocyte growth factor-like), and cell signaling (Inositol monophosphatase). The function of the multifunctional ubiquitin in the present experiments remains to be elucidated as this gene has a large number of roles 73 including cell cycle regulation, DNA repair, embryogenesis, regulation of transcription and apoptosis. Interestingly, this comparative analysis may reveal the first hint of the start of mineralization processes. In particular mutations in the gene 3-beta-hydroxysteroid-delta8, delta7-isomerase which catalyses the conversion of delta(8)-sterols to their corresponding delta (7) isomers is linked to chondrodysplasia punctata in humans (Conradi-Hunermann-Happle syndrome) that causes punctiform calcification of cartilage 74 . It remains to be established if this gene also influences the calcification process in fish but if it does it may represent a useful biomarker. Moreover, it suggests that up-regulation of transcripts involved in calcification occurs early in regeneration well before the most active phase of this process (estimated at 14-28 days in goldfish) 8 .

The overexpression of developmental genes is already known to be involved in stem cell activation and in epidermal-dermal interactions. Examples such as FGFs, Wnts and SHH were not observed to be differentially regulated in the experiments described here, which could be for several reasons, including 1) some of those genes were not represented in the oligo array e.g. sonic hedgehog (SHH) and 2) the timescale of the experiment was not ideal to identify changes in gene expression of developmental genes associated with skin healing since by day 3 histological analysis revealed that epidermis is already re-established. Furthermore, metalloproteinases 2 (MMP-2) and 9 (MMP-9) have been recently suggested to have a role in scale regeneration in zebrafish 21 . However, MMP-2 is not represented in the sea bream microarray and although MMP-9 is represented (SAPD23115) it did not change significantly between the groups analyzed. Of the metalloproteinases present on the microarray (MMP-9, MMP-7.MMP-15 and MMP-28) only matrilysin (MMP-7) was modified and it was down-regulated in unfed fish without scales compared to the unfed fish on day 3 of the experiment. The sea bream oligo-array results were also queried for known calcitropic factors, for example, PTH (parathyroid hormone) and PTH related peptide (PTHrP), which are related with calcium and phosphorus homeostasis in fish 35 75 , and calcitonin, whose hypo or hypercalcemic role in fish is not yet clarified 11 76 77 , no significant differences in expression were observed. The same was observed for other calcitropic hormones represented in the array (such as prolactin, prolactin-receptor and PTH receptor 1), but to a certain extent this was to be expected given the previously estimated timescales for these processes, albeit in a different species 8 . Moreover, the target tissue in the present study, the skin, is not recognised as an important source of these hormones which tend to be produced in appreciable levels by specific endocrine tissues.

The biggest changes in the skin/scale transcriptome amongst the treated groups occurred at day 3. By day 7, when re-epithelisation had occurred and a thin regenerated scale was visible (Figure 1), relatively few differences were found when expression analyses were carried out. Over the four comparisons, a total of 49 probes were up-regulated only 21 of which had associated annotation, representing 17 putative unique transcripts (Table 8). It was also difficult to make generalisations about the on-going cellular processes, but the differentially expressed genes indicate a continued requirement for cell division and proliferation. The putative identification of the transforming acidic coiled coil 3 (TACC3) indicates the continuance of scale cell proliferation, as this gene in humans was shown to be involved in the control of cell growth and differentiation 78 79 . As in day 3, some genes are up-regulated in more than one of the comparisons made (Figure 2B). The GINS complex subunit 1 (PSF1 homolog) reported to be involved in regulating proliferation of stem cells, for example in response to acute bone marrow regeneration in mammals 80 , is up-regulated in 3 of the four comparisons performed for both days 3 and 7 after scale removal where the factor analysed is the skin/scale regeneration. Taken together, the up-regulation of the GINS complex transcripts in all the groups where scales were removed at both day 3 and 7 of regeneration may indicate that, in teleosts, this transcript and its protein product may be regulating the process of skin/scale regeneration by the induction of cell proliferation.

There is also an element of cytoskeletal remodelling via the up-regulation of structural proteins (e.g. cytokeratin) and protein degradation. Similarly, as was noted with the fasted group (when compared with controls), there are a couple of genes up-regulated in sea bream, that in humans produce structural problems when defective. Mutations in Dynamin 2 produce abnormally large nuclei in skeletal muscle cells, resulting in muscle weakness 81 , whilst serpin H1 is an essential chaperone in collagen synthesis, with deficiencies in humans resulting in the premature rupture of placental membranes 82 . As with the 3 days fasted vs. fasted without scales comparisons, at day 7 there are up-regulated transcripts potentially linked with mineralization. Such as, caldecrin precursor that exerts a hypocalcaemic activity, decreasing serum calcium, and may indicate that the fish is now actively mobilising calcium for scale mineralization.

To corroborate the microarray data gene-specific qPCRs were performed. In general, the direction of change in expression was concordant between qPCR and the microarray probes and a positive Pearson correlation was obtained between qPCR and probe_1 (r² = 0.895, p < 0.0001 n = 40) and between qPCR and probe_2 (r² = 0.901, p < 0.0001 n = 40) (Table 9). Despite the good correlation observed between the gene expression analyzed by qPCR and the microarray data, a few exceptions were observed, the qPCR fold-change for SPP1 transcript in group 3ST; for ColVA2 transcript in group 3WS; for Col1A1 transcript for group 3ST and for p22phox transcript in group 3STWS vs. starved group was not correlated with the microarray fold-change. The latter is explained by the high variability found for this gene in different individuals in the experimental groups. The best concordance between qPCR and microarray data is achieved when the observed microarray fold change is between 2 and 7 36 , and in the present experiments the genes analyzed by qPCR which had fold changes closest to the lower limit had lower correlations with the microarray data.

Juvenile sea bream (Sparus auratus) were maintained at the Centre of Marine Science (CCMAR) field station (Ramalhete, Faro, Portugal) in through-flow seawater tanks (2000L) at 17-21°C, 36% salinity and 12 h light and 12 h dark photoperiod for several weeks prior to the start of the experiments. The maintenance of fish and subsequent experiments complied with the Guidelines of the European Union Council (86/609/EU) and was covered by a group 1 licence (Direcção-Geral de Veterinária, Portugal). Behaviour and health of animals was monitored visually each day and no mortality occurred during the experiment.

Sea bream (n = 8 fish/tank, 93.4 ± 3.1 g) were acclimatised for one week to the experimental circuit which consisted of 8 through-flow seawater tanks (500L), with water maintained at 19-21°C, 36% salinity and a 12 h light and 12 h dark photoperiod. Food was withheld from the fasted experimental groups for 1 week (-7 days) prior to removal of the scales which was considered day 0 of the trial (Figure 3). The experiment had 3 treatment groups: ST = fasted for duration of experiment; WS = scales removed at time 0; STWS = fasted for duration of the experiment with scales removed at time 0 and the control group (N) with no treatment but subjected to the same anaesthesia/handling as the treatments groups.

Duplicate tanks for each treatment were prepared (day 3 and day 7) and 8 fish were sampled from one tank 3 days after the scales had been removed and from the second tank 7 days after the scales had been removed. Two tanks contained the control fish and were sampled at the same time as the experimental groups at day 3 and 7. To remove the scales, fish were lightly anaesthetised in 2-phenoxyethanol (1:5000, Sigma-Aldrich, Madrid, Spain) and scales were removed (approximately 60-70%) by wiping fish with a wet paper towel in order to minimize damage. For sampling fish were anaesthetised in 2-phenoxyethanol, as described above, weight and length was measured, blood collected and centrifuged (10,000 rpm for 5 minutes) and the plasma stored at -20°C. Fish were sacrificed by sectioning the spinal cord and skin was collected (approximately 1 cm²) from below the dorsal fin and carefully dissected free of muscle and snap frozen in liquid nitrogen and stored at -80°C. For histology, skin with some adhering muscle was fixed overnight at 4°C in 4% PFA.

To determine the possible effect of treatments on plasma composition but also to evaluate the physiological condition of the experimental and control animals, calcium, phosphorus, glucose and lactate were determined. Duplicate samples of sea bream plasma (10 μl diluted 1/4 in milliQ water), collected from 8 fish/experimental group, 3 days after the removal of scales were measured. Colourimetric tests were performed according to the manufacturers' instructions and absorbance measured with a microplate reader:

• Calcium: Calcium-ο-C v/v kit (ο-cresolphtalein, v/v, colorimetric; Spinreact ref. 1001061, Spain), absorbance at 570 nm.

• Phosphorus: Phosphorus - UV kit (Phosphomolybdate, uv; Spinreact ref. 1001155, Spain), absorbance at 340 nm.

• Glucose: Glucose-TR kit (GOD-POD; Spinreact ref. 1001190/1001191/1001192, Spain), absorbance at 505 nm.

• Lactate: Lactate kit (LO-POD Enzymatic colorimetric; Spinreact ref. 1001330, Spain), absorbance at 505 nm.

To characterize sea bream skin and scale morphology, samples fixed in 4% PFA were decalcified overnight in 0.5 M EDTA, pH 8, dehydrated through a graded ethanol series, saturated in xylene and impregnated and embedded in paraffin wax (Merck, Germany). Serial sections (5 μm) of skin were mounted on 3-aminopropyltriethoxysilane (APES; Sigma-Aldrich, Madrid, Spain) coated glass slides. The sections were dried overnight at 37°C, cooled to room temperature and stored or stained. To distinguish between collagen rich and/or mineralized and non-mineralized tissue, sections were stained with Masson's trichrome 83 . Skin sections were rapidly dehydrated through a graded series of alcohols, cleared in xylene and mounted in DPX mountant (BioChemika, Sigma-Aldrich, Madrid, Spain). Stained sections were analyzed using a microscope (Leica DM2000) coupled to a digital camera (Leica DFC480) and linked to a computer for digital image analysis.

A 4 × 44 k oligo-array developed and validated for the gilthead sea bream by Ferraresso et al. 36 was used in this study. The array contained 39,379 sea bream oligonucleotide probes covering 19,715 unique transcripts and of these, 19,650 were represented by two non-overlapping probes and 65 were present as a single probe. Owing to the expansion of teleost sequences in the public databases since the publication of Ferraresso et al. 36 , a re-annotation of probes was carried out with Blastx similarity searches against the Uniprot/Swissprot and Uniprot/Trembl databases 37 . Annotation was assigned for probes with an expected score in excess of 1e^-10.

Total RNA was extracted from five individual fish using an RNeasy Mini kit (Qiagen, Hilden, Germany) according to the manufacturer's instructions. RNA quality and integrity were checked using an Agilent 2100 Bioanalyser (Agilent technologies, Palo Alto, CA) and only samples with an RNA integrity index number (RIN) >7.5 were processed for use in microarray analysis. Samples from each treatment group (5 replicates) were labelled with Cy3-dCTP and hybridizations were performed using the Agilent One-Colour Microarray-Based Gene Expression Analysis protocol with the modifications described by Ferraresso et al. 36 . The arrays were scanned on an Agilent G2565BA DNA microarray scanner, at a resolution of 5 μm, and at two different sensitivity levels (XDR Hi 100% and XDR Lo 10%). The XDR Hi and XDR Lo images generated per array were analysed together and the data extracted. Background subtraction was performed using the standard procedure in the Agilent Feature Extraction Software 9.5.1. Spike-In Viral RNAs were used to control array hybridization intensities and ensure normalization gave a uniform signal across all microarray slides. The R 84 limma package 85 was used for microarray analysis. A factorial design of the treatments were compared by fitting a linear model 86 with differentially expressed clones selected by a Benjamini and Hochberg 87 globally adjusted p-value of 0.05 and a minimum two-fold change. The transcripts represented by two non-overlapping probes were only selected when both probes were differentially expressed.

The data discussed in this publication have been deposited in NCBI's Gene Expression Omnibus 88 and are accessible through GEO Series accession number GSE30717 (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE30717).

Accession numbers associated with the probe annotations were used to assign GO and GOSLIM terms 89 . GO enrichment was determined by a proportion test between the number of clones representing a GO term on the array compared to the number of differentially expressed clones representing the same GO term in a given comparison with a p-value cut-off of 0.05.

Cellular networks arising from the gene expression data were identified and established through the use of IPA (Ingenuity Systems, http://www.ingenuity.com). The sequences of differentially expressed genes from treatments collected at 3 days were all submitted to BLASTN in order to identify human orthologues. The accession numbers were extracted and used as identifiers in IPA together with the fold-changes of the corresponding differentially expressed genes. The Ingenuity knowledge base was used as a reference and direct and indirect relationships were included and no filters were applied. Bio functions, namely molecular and cellular functions and physiological system development and function significantly (p < 0.05) related with the input dataset were identified. Networks were then algorithmically generated based on their connectivity and a score was assigned. The score was used to rank networks according to how relevant they were to the genes in the input dataset.

Array results were corroborated by real-time RT-PCR using when possible RNA extracted from the same individuals used for array analysis from all the different treatments at the 3 day time point. Ten genes were analysed and primers were designed using Beacon Design software (Premier Biosoft Int., Palo Alto, CA) (Table 10). For cDNA synthesis, 1 μg of total RNA was pre-treated with DNA-free Kit (Ambion, UK) to remove genomic DNA and then cDNA synthesis carried out using 250 ng of DNAse-treated total RNA, 200 ng of random hexamers (GE Healthcare, Little Chalfont, UK), 40 U of MMLV reverse transcriptase (RT) (Promega, Southampton, UK) and 5 U of RNAguard Rnase inhibitor (GE Healthcare, Little Chalfont, UK) in a final reaction volume of 20 μl. Q-PCR was performed in duplicate reactions using SYBRgreen chemistry (Power SYBR^® Green PCR Master Mix, Applied Biosystems, UK) and the relative standard curve method, using a StepOnePlus qPCR thermocycler and StepOne software v2.0 (Applied Biosystems, UK). PCR cycling conditions were 10 min at 95°C, followed by 55 cycles of 10 sec at 95°C, 20 sec at the optimal temperature for each primer pair (Table 10), and 30 seconds at 72°C. A final melting curve was carried out between 60 and 95°C for all genes and each produced single products/dissociation curves. Standard curves relating initial template quantity to amplification cycle were generated using serial dilutions of linearized plasmid DNA containing the gene of interest or of RT-PCR specific product obtained from the same specie and tissue, and the efficiency of qPCR reactions ranged between 82-100%, with the exception of SAPD20351 and SAPD13946 that had efficiencies of 73.1 and 78.6%, respectively, and all gave R² > 0.985. All amplicons were sequenced to confirm specificity of the PCR reaction. Ribosomal protein S18 (S18) expression was quantified using the same conditions as the other genes. No statistically significant differences were found between experimental groups so it was chosen as an endogenous reference gene to normalize qPCR data as it had a low inter-group variation and a similar level of expression to the analyzed genes. Statistical significance of relative gene expression between groups was analysed by one-way ANOVA using the software SigmaStat v.3.1 (SPSS Inc, Chicago, USA). Pearson correlations between the qPCR relative expression and microarray expression of both probes (average of individuals in each experimental group) were calculated for each gene. Statistical significance was established at p < 0.05.