The Atlantic salmon used came from the same batch of eggs hatched in the facilities of Cermaq, Norway in March 2015. The fish were kept in two large aluminum tanks (depth: 2.8 m, volume: 450 m³) in Hopen, Norway, for 23 weeks before the seawater transfer. The freshwater was pumped directly from the a nearby lake approved by the Norwegian Food Safety Authorities into the tanks with a drum filter removing large particles. There was no mixing of water from other sources. The temperature of the freshwater followed natural fluctuations and ranged from 9.8 to 16.1°C. The dissolved oxygen level was above 10 mg/L. Each tank contained about 180 000 fish with an average initial body weight of 3.3 ± 0.1 g (Mean ± SD). In September 2015, the fish went through a photoperiod-induced parr-smolt transformation to produce out-of-season smolts before transportation to a sea location in Sommarbukt, near Alta in the far north of Norway. This photoperiod-induced treatment takes 12 weeks from continuous light LD24:0 to LD12:12 for 6 weeks, followed by the day-length from LD12:12 to LD24:0 for 6 more weeks (Duncan et al., 1998). Subsequently, one-day seawater challenge test was conducted to confirm that smolt were all ready for seawater transfer. In saltwater period, the salmon were farmed as is the standard setup for Norwegian salmon farming. This farming was open net pen with no mix of other water qualities than the naturally seasonal hydrodynamics that would occur in a fjord. The water exchange in the net pen would depend on the currents occurring in the period and the growth of fallowing organisms onto the nets. About 55 000 fish were kept in each net pen. A vertical device connected to an automatic winch (HF5000, Belitronics, Lunde, Sweden) was used to monitor the water parameters at 3 m depth in this study. The temperature of saltwater followed natural fluctuations ranged from 6.8 to 9.7°C. The dissolved oxygen level was above 10 mg/L and the salinity levels fluctuated from 20.8 to 40.1 ‰.

The diets were extruded. As Table 1 shows, the reference (Ref) diets were formulated varying in ingredient composition to meet the variation in nutritional requirements of Atlantic salmon in the freshwater (FW) and seawater (SW) phases. The Test diets (Test) were formulated from the respective reference diets by supplementing with nucleotides and immune stimulants (extracts from plants and yeast cell wall) to the FW-Ref formulation (FW-Test), and nucleotides and EPA and DHA to the SW-Ref formulation (SW-Test). These functional feed ingredients, as dry meal mix, were added to the Test diets according to recommendations from the producers. Their inclusion levels are not specified herein for the need to protect intellectual properties.

Fish were sampled four weeks before seawater transfer, in August 2015 and four weeks after seawater transfer in October 2015 as illustrated in Supplementary Figure 1A . The sampling was carried out four hours after the last feeding. Only fish with chyme present along the whole intestine were sampled to ensure exposure to the diets until the point of sampling. Supplementary Figure 1B shows the anatomy of the alimentary tract of Atlantic salmon and the workflow of sampling for analysis based on our previous description in Bakke-McKellep et al. (2000). At the freshwater sampling, 36 fish were taken from each tank which were split into 3 replicates of 12 fish, i.e., a total of 36 fish per treatment. Earlier study has indicated that the variation between the tanks in this facility is negligible making repeated samplings from the same tank acceptable (Wang et al., 2020). For the seawater samplings, triplicate net pens of fish were farmed for each treatment, i.e., a total of 36 fish per treatment. All fish were anesthetized and euthanized using the tricaine methanesulfonate (MS222, Argent Chemical Laboratories) before sampling. The body weight and fork length were recorded. The blood samples were collected from the caudal vein of 36 fish per treatment. Plasma was collected after centrifugation at 2000 g for 10 min at 4°C, and snap-frozen in liquid nitrogen, and then stored at -80°C until analysis. Subsequently, the whole intestine was dissected from each fish, then opened longitudinally after removing attached adipose tissue.

Six fish per replicate were used for collection of digesta samples as follows: the intestinal tract was divided in five sections i.e. the proximal (PI1) and distal (PI2) half of the pyloric intestine, the mid intestine (MI), and the proximal (DI1) and distal (DI2) half of the distal intestine. The digesta from the six fish per tank were pooled, frozen and stored as described above for plasma, until analyses of total bile acid levels and trypsin activities. After removing the digesta, the tissues from PI, MI, and DI were weighed for estimation of intestinal somatic indices, and then collected and stored at -80°C, until analysis of leucine aminopeptidase (LAP) capacity.

Another six fish per replicate were collected for the histological evaluation, gene expression analysis (qPCR), and microbiota profiling (lower left of Supplementary Figure 1 ). In order to avoid contamination, face masks were used and the samples were handled near a gas burner to reduce bacterial number in the air. All sampling tools were thoroughly cleaned, decontaminated with 70% ethanol sprays and flamed after the sampling of each fish. Tissues from pyloric caeca (PC) and DI were rinsed in sterile phosphate-buffered saline (PBS) three times to remove digesta and cut into pieces for RNA extraction (preserved in RNAlater solution, incubated at 4°C for one day, then stored at -20°C before RNA extraction) and histological evaluation (18 fish per treatment, fixed in 4% phosphate-buffered formaldehyde solution for one day, and transferred to 70% ethanol for storage). For the microbiota profiling, DI digesta was collected into a 10 mL skirted sterile centrifuge tube and mixed thoroughly using a spatula. An aliquot of the homogenate was then transferred into a 1.5 mL sterile Eppendorf tube and snap-frozen in liquid nitrogen, then stored at -80°C before DNA extraction. Between each sampled fish, all tools were decontaminated by an ethanol spray, then sterilized by flaming.

Plasma samples were analyzed at Central Laboratory in the Faculty of Veterinary Medicine, Norwegian University of life science according to the standard procedures (Advia 1800, Siemens Healthcare Diagnostics, Erlangen, Germany).

Trypsin activity was analyzed according to the method of Kakade et al. (1973) using benzoyl arginine p-nitroanilide (No. B-4875, Sigma Chemical Co., St. Louis, MO) as the substrate. Total bile acid levels were determined at the Central Laboratory of Faculty of Veterinary Medicine of NMBU according to the method of Krogdahl et al. (2020).

The capacity of brush border leucine aminopeptidase (LAP) was analyzed as the description of Bieth et al. (Bieth et al., 1974). The tissue homogenate was prepared using ice-cold Tris−mannitol buffer (1:20, w/v). The four-(two-aminoethyl)-benzene-sulfonyl fluoride hydrochloride (Pefabloc SC, Basel, Switzerland) was used as a serine proteinase inhibitor to prevent protein hydrolysis during sample preparation. The incubations were performed at 37°C according to the method of Krogdahl et al., 2003. The LAP capacity is expressed as mmol substrate hydrolyzed per intestinal section per hour per kg fish.

Tissues from DI and PC were cut to produce sections of 3-μm thickness and stained with hematoxylin and eosin (H&E) (Penn et al., 2011). The following observations related to typical inflammatory morphological changes were evaluated by light microscopy: the length of mucosal fold, the width of submucosa, cellular infiltration of lamina propria and submucosa, as well as the degree of enterocyte vacuolization. The histological sections were assigned a rank ranging from normal to severe as described in one of our earlier studies (Bakke-McKellep et al., 2007).

Three fish per replicate (n = 9 fish per treatment) were randomly selected for qPCR assays, which were carried out following the MIQE guidelines (Bustin et al., 2009). Briefly, total RNA was extracted from distal intestine in a randomized order using a custom made Reliaprep simplyRNA HT protocol (Promega) on the Biomek 4000 laboratory automation workstation (Beckman Coulter). The RNA extraction included a DNase treatment according to the manufacturer’s protocol. The RNA purity and concentration were evaluated by Epoch Microplate Spectrophotometer (BioTeK Instruments, Winooski, VT, USA). The average 260/280 ratio of samples was 2.2 ± 0.02 (Mean ± SD). The RNA integrity was measured by the 2100 Bioanalyzer in combination with an RNA Nano Chip (Agilent Technologies) and 6000 Nano LabChip kit (Agilent Technologies, Palo Alto, CA, USA). The average RNA integrity number (RIN value) of all samples was 9.1 ± 0.4 (Mean ± SD).

The cDNA synthesis and qPCR assays were conducted as described elsewhere (Kortner et al., 2013). Briefly, the cDNA synthesis was conducted using 1.0 μg total RNA from all samples using a Superscript^™ IV VILO^™ cDNA synthesis kit (Invitrogen, Carlsbad, CA, USA, catalog no., 11756050). The qPCR assays were carried out using the LightCycler 96 (Roche Applied Science, Basel, Switzerland) with a 10-μL DNA amplification reaction including 2 μL of PCR-grade water, 2 μL diluted cDNA template, 5 μL LightCycler 480 SYBR Green I Master (Roche Applied Science) and 0.5 μL (10 μM) of each forward and reverse primer. Samples were run in duplicates in addition to a no-template control for each gene. A three-step qPCR program was performed including an enzyme activation step at 95°C (5 min), 45 cycles of 95°C (10 s), 56–60°C (10 s), as well as 72°C (15 s). The plate pipetting was done using the Biomek^® 4000 automation workstation. Details of primer pairs used for the qPCR assays in DI and PC are shown in Supplementary Tables 1 , 2 , respectively. The RNA polymerase II (rnapoii) and hypoxanthine phosphoribosyl transferase 1 (hprt1) for DI and β-actin (actb) and glyceraldehyde-3-phosphate dehydrogenase (gapdh) for PC using as reference genes were selected based on their overall coefficient of variation (CV) and their interspecific variance (Kortner et al., 2011). The reference genes as an internal normalization factor, the raw quantification cycle (Cq) values were used to calculate the mean normalized levels of target genes in DI and PC (Muller et al., 2002).

DNA extraction: Three fish per replicate (n = 9 fish per treatment) were used for DNA extraction (same fish as qPCR). The sample size was decided here according to previous suggestions (Panteli et al., 2020). One fish per treatment was randomly selected to divide all fish into 9 batches for DNA extraction to avoid batch effect. About 100 mg of DI digesta from each sample was homogenized using the bead beating, then an additional heating step of 95°C for 7 min, was conducted according to the suggestion by Knudsen et al. (2016). Hereafter, the homogenate were used for DNA extraction based on the standard procedure provided by the manufacturer of QIAamp Fast DNA Stool Mini Kit (Qiagen, Hilden, Germany). For quality control purposes, a blank and a mock (Catalog No: D6300, ZymoBIOMICS Mock Community Standard, Zymo Research) were added to the DNA extraction step used as a negative and positive control, respectively.

PCR amplification: The PCR amplification of the 16S rRNA V1-V2 region was performed using the universal primer set 27F (5′-AGA GTT TGA TCM TGG CTC AG-3′) and 338R (5′-GCW GCC WCC CGT AGG WGT-3′) (Roeselers et al., 2011) based on the descriptions of Gajardo et al. (2016). Briefly, total 25μl reaction volume was used for PCR including 2 μl of DNA template, 22.4 μl PCR Master Mix (Thermo Scientific, CA, USA; catalog no: F531L), and 0.6 μl mixture of forward (27F) and reverse 338R primers (50 pM). The PCR was run in duplicate including the molecular grade water as negative PCR controls. The amplification program was set as follows: initial denaturation at 98°C for 3 min; 35 cycles of denaturation at 98°C for 15 s, annealing decreasing from 63°C to 53°C in 10 cycles for 30 s followed by 25 cycles at 53°C for 30 s, and extension at 72°C for 30 s; followed by a final extension at 72°C for 10 min. For samples with invisible bands in the agarose gel after PCR, the PCR condition was optimized by applying serial dilutions to the DNA templates to reduce the influence of PCR inhibitors. After the PCR amplification, the duplicate PCR products were pooled (10 µl) to evaluate the library preparation using 1.5% agarose gels. Samples with a bright band of size between 300bp and 350bp were considered suitable for further processing. As one batch of samples showed poor PCR amplification, we removed these samples and left 8 samples per treatment (n = 8) for final sequencing.

Quantification of 16S rRNA gene by qPCR: The 16S rRNA gene quantity of DNA templates used for the amplicon PCR was measured by qPCR as described in the section of quantitative real-time PCR (qPCR). The qPCR assays were performed using a universal primer set (forward, 5’-CCA TGA AGT CGG AAT CGC TAG-3’; reverse, 5’-GCT TGA CGG GCG GTG T-3’) used for bacterial DNA quantification as described in previous studies (Ramseier et al., 2009; Vandeputte et al., 2017).

Library preparation and sequencing: The PCR products cleanup, library preparation, and sequencing were performed following the Illumina 16S Metagenomic Sequencing Library Preparation Protocol (Illumina, 2013). Briefly, the PCR products were cleaned using the Agencourt AMPure XP system (Catalog No: A63881, Beckman Coulter) multiplexed by dual indexing using the Nextera XT Index Kit (Catalog No: FC-131-1096, Illumina) and purified again using the AMPure beads. The representative libraries were analyzed using the Agilent DNA 1000 Kit (Catalog No: 5067-1505, Agilent Technologies) to verify the library size. Cleaned libraries were quantified using the Invitrogen Qubit™ dsDNA HS Assay Kit (Catalog No: Q32854, Thermo Fisher Scientific), diluted to 4 nM in 10 mM Tris, and pooled in an equal volume. The library was loaded at 6 pM and sequenced using the Miseq Reagent Kit v3 (Catalog No: MS-102-3003, Illumina) according to the manufacturer’s instructions.

The demultiplexed raw sequence data were denoised by the DADA2 algorithm (Callahan et al., 2016) in QIIME 2 (version 2019.4) to generate amplicon sequence variants (ASVs) (Bolyen et al., 2019) as described elsewhere (Li et al., 2021). The taxonomy was assigned in QIIME2 against the SILVA database (version 132) (Quast et al., 2013) trained by a naive Bayes machine-learning classifier (Bokulich et al., 2018). The contaminant sequences, including Acinetobacter, Aeromonas, Cutibacterium, Flavobacterium, Leptothrix, Pseudomonas, and Streptophyta, as well as unclassified Chitinophagales (order) and Betaproteobacteriales (class), were removed based on the criteria suggested by Davis et al. (2018). The alpha diversity was evaluated by the Observed species index and Shannon’s index. The beta diversity was assessed by Unweighted UniFrac and Weighted UniFrac Bray-Curtis distances using Primer 7 (version, 7.0.13) followed permutation multivariate analysis of variance (PERMANOVA) (Clarke et al., 2006).

The microbial clades (genus level) were tested for the associations with metadata of interest (the metadata seen in Supplementary Table 3 ) using the MaAsLin2 (version, 0.99.12) (https://huttenhower.sph.harvard.edu/maaslin2) in R with the default parameters (Mallick et al., 2021). Bacterial taxa of more than 0.1% abundance and at least 25% prevalence of all samples were selected for association testing. The significant association was set at a q-value < 0.25. The metadata of health responses, i.e., plasma nutrients, DI supranuclear vacuolization (histology), and gut immune functions (qPCR in DI), were selected to run the multivariate association testing with fixed factor, i.e., treatment. Since plasma nutrients, including plasma cholesterol, triglyceride, free fatty acid, and glucose, were highly correlated ( Supplementary Figure 2 ), we ran a principal component analysis (PCA) and extracted the first principal component (PC1) for the association testing to avoid multicollinearity and reduce the number of association testing. Regarding the DI supranuclear vacuolization, the skewed five-category degree was collapsed into more balanced binary data, i.e., normal and abnormal (mild, moderate, marked, and severe), to make the data appropriate for the association testing. The gut immune function-related genes selected for the association testing included pro-inflammatory cytokines [interleukin-8 (il8) and interferon gamma (ifnγ)], anti-inflammatory cytokine [transforming growth factor-beta (tgfβ)], as well as T-cell markers [e.g. cluster of differentiation-4 alpha (cd4α), 8 beta (cd8β) and 3γδ (cd3γδ)]. Since the immune-related genes were highly correlated ( Supplementary Figure 3 ), the extracted PC1 of the PCA was also used for association testing.

Regardless of dietary treatment, transfer to seawater reduced growth rate of the fish, estimated as thermal growth coefficient (TGC) (p <.05, Figures 1A, B ). No clear diet effect on growth performance was observed either in FW- or SW-fish (p >.05, Figures 1A, B ). Transfer to seawater also decreased condition factor (p <.05, Figure 1C ) and this reduction was more pronounced for the Test-fed fish than for those fed the Ref diet (p <.05, Figure 1C ).

Regarding somatic indices of the intestinal segments, seawater transfer significantly increased the values for PI, MI, and DI (p <.05, Figures 1D–F ) Diet did not affect the indices either in FW- or in SW-fish (p > .05, Figures 1D–F ).

Seawater transfer decreased plasma cholesterol, triglyceride, free fatty acid, and glucose levels substantially, in the range from 40% to 70% (p <.05, Figures 2A–D ). For plasma chloride and sodium levels, an increase was observed (p <.05, Figures 2E, F ), as expected.

At the freshwater stage, but not after seawater transfer, fish fed the Test diet, showed significantly higher cholesterol levels than the Ref fed fish (p <.05, Figure 2A ). No significant diet effects were found for other plasma biomarkers (p >.05, Figure 2 ) whether in FW- or SW-fish.

Fish from the FW-Test treatment, compared to those fed the Ref diet, showed elevated digesta trypsin activity, but only in PI-2 and MI (p <.05, Figure 3A ). After sea transfer, no significant diet effects were observed (p >.05, Figure 3A ).

Regarding the chyme bile acid levels, decreased levels were found in PI-2, MI, and DI-1 after seawater transfer (p <.05, Figure 3B ), whereas no significant diet effects were observed either in fresh nor seawater (p >.05, Figure 3B ).

Compared to fish in FW, fish in SW had significantly higher LAP capacity in PI and DI (p <.05), but not in MI ( Figure 3C ). There was no significant diet effect on intestinal LAP capacity at any sampling time points or any intestinal segments (p >.05, Figure 3C ).

Most of the fish showed normal histological characteristics in the DI regarding mucosal fold height, submucosa cellularity and supranuclear vacuolization ( Figures 4A–C ). Enterocyte hyper-vacuolization, or the so-called steatosis, was observed in the PC in all the sampled fish. The steatosis became significantly less pronounced after the seawater transfer (p <.05, Figure 4D ). Moreover, when the fish were kept at FW, the steatosis was less severe in fish fed the Test-diet (p <.05) than in the Ref fed fish ( Figure 4D ).

Regarding goblet cell markers, lower mucin-2 (muc2) expression was observed in fish in SW compared to those in FW, whereas the mucin-13 (muc13) showed the opposite trend (p <.05, Figures 5A, B ).

A drop in the expression of several immune-related genes was observed after seawater transfer. These genes included pro-inflammatory cytokines (il8 and ifnγ), anti-inflammatory cytokine (tgfβ), and T-cell markers (cd4α, cd8β, and cd3γδ) (p <.05, Figures 5C–H ).

Compared to those sampled in FW, fish sampled in SW showed a lower expression in catalase (cat), while the expression of superoxide dismutase (sod) showed an opposite trend (p <.05, Figures 5I, J ).

For genes related to osmoregulation, the expression of aquaporin 8ab, aqp8ab, increased significantly after the seawater transfer (p <.05, Figure 5K ), whereas an opposite trend was found for aquaporin-1 b (aqp-1b) expression (p <.05, Figure 5L ). The claudin-15 (cldn15) and e-cadherin (cdh1) tended to have higher expression in fish at SW compared to those at FW (p <.05, Figures 5M, N ).

No significant differences between dietary treatments in FW or in SW were observed for any of the aforementioned genes (p >.05, Figure 5 ).

The results for other genes showing no clear seawater transfer or diet effect are presented in Supplementary Tables 4 , 5 .

Fish in SW showed lower perilipin-2 (plin2) expression than those in FW (p <.05, Figure 6A ). In contrast, higher expression of cytochrome P450 51 (cyp51), Niemann-Pick C1 like 1 (npc1l1), and peptide transporter (pept) was observed in SW compared to those in FW (p <.05, Figures 6B–D ). No significant diet effects were observed for these genes (p >.05, Figure 6 ).

The results for other genes showing no clear effect of seawater transfer or diet effect are presented in Supplementary Table 6 .

The mock from 8 different DNA extraction batches showed a similar microbiota profile indicating good reproducibility ( Supplementary Figure 4 ). Regarding quantification of bacterial 16S rRNA by qPCR, there were no significant differences among treatments (p >.05, Supplementary Figure 5 ).

Alpha diversity: Compared to fish in FW, significant increases in microbial richness (Observed species index) and diversity (Shannon’s index) were observed in fish in SW. No clear diet effects were observed ( Figures 7A, B ).

Beta diversity: PERMANOVA analysis of the gut microbiota showed clear differences in the bacterial composition between FW and SW as illustrated by the PCoA plots based on the UniFrac-Weighted and -Unweighted distances (p <.001, Supplementary Table 7 and Figures 7C, D ). A diet effect was observed for the SW samples based on the Pairwise test of UniFrac-Unweighted (p = 0.028, Supplementary Table 7 ) but not based on the UniFrac-Weighted. In FW, no diet effects were indicated.

Relative abundance of all ASVs for the samples is presented in Supplementary Table 8 . The taxonomic compositions of the gut microbiota in Atlantic salmon differed greatly between the FW and SW stage regardless of dietary treatment ( Figure 8 ). More specifically, the microbiota composition of fish in FW was strongly dominated by Firmicutes (98% ± 4%), mainly the unclassified Ruminococcaceae (85% ± 35%). After seawater transfer, Firmicutes (mainly genera Lactobacillus) and Proteobacteria (mainly genera Photobacterium) dominated the gut microbiota, and together they accounted for about 90% of the total abundance, i.e., Firmicutes (63% ± 10%) and Proteobacteria (27% ± 7%), respectively.

The multivariate association analysis identified one taxon, i.e., Clostridium sensu stricto 8, for which the relative abundance was negatively associated with the levels of plasma nutrients ( Figure 9 and Supplementary Figure 6 ), showing an increase as the PC1 of the PCA increased. The relative abundance of unclassified Cyanobiaceae (family) and unclassified SAR86 clade (order) was found to be increased in fish with abnormal DI vacuolization ( Figure 9 and Supplementary Figure 7 ). Twelve taxa, including the lactic acid bacteria, e.g. Lactobacillus, Lactococcus, and Leuconostoc, correlated significantly with immune-related gene expression in the DI. Except for the Synechococcus.CC9902, the relative abundance of these taxa was positively associated with expression of immune-related genes ( Figure 9 and Supplementary Figure 8 ), which increased as the PC1 of the PCA increase.

The reductions in growth rate and condition factor observed after seawater transfer in the present study were most likely due to a combination of reduction in feed intake and increased demand for nutrients and energy the first weeks after seawater transfer. The work of Usher et al. (1991), showing reduced appetite lasting more than one month after seawater transfer, supports this suggestion. Reduction in feed intake and increased energy demand may be related to the physiological alterations taking place in the fish during smoltification, as well as the stressful conditions due to handling, long transport, and environmental changes (Hoar, 1988; Bjornsson et al., 2011). The observation in our study of an increase in LAP activity and capacity may be related to changes in environmental, physiological, diet and nutritional conditions. The physiological transformation required to handle the change in salinity, the increase in workload regarding osmoregulation in saltwater compared to freshwater, and the accompanying increase in need for energy, may induce increase in LAP production. These changes occur in a situation of increased handling and transport, which may take many days, and greatly reduce feed intake. In such situation, the fish may increase its capacity for digestion and utilization of nutrients (Krogdahl et al., 2004). Moreover, the change in environmental temperature, in this case, from a higher to a lower, with inevitable reduction in enzyme activity, may also have triggered higher production of LAP, as well as of other enzymes. The observed reduction in plasma nutrient levels after seawater transfer support the suggestion of deficient nutrient and energy intake (Sheridan, 1989; Sissener et al., 2009), and is also in line with the results of our previous study with normal smolts (Wang et al., 2020).

The lower expression of immune genes including pro-inflammatory cytokines, anti-inflammatory cytokine and T-cell markers after the seawater transfer in our study is in line with earlier findings in both normal and out-of-seasonal smolts (Johansson et al., 2016; Karlsen et al., 2018; Wang et al., 2020) indicating frangible intestinal immune homeostasis after seawater transfer (Secombes et al., 2001; Secombes, 2016). This may be an unavoidable effect of the physiological alterations during smoltification and just after seawater transfer due to the stressful occasions. These effects on immune functions may be related to the state of nutrient deficiencies that the fish supposedly suffer from after transfer to the sea (Moore et al., 2018). Also, the downregulation of muc2 after seawater transfer might indicate the fish’s susceptibility to diseases, since muc2 is highly expressed in the intestinal segments of Atlantic salmon (Sveen et al., 2017) and acts as the first-line defense against pathogens (Van der Marel et al., 2012; Koshio, 2015; Salinas, 2015).

The noticeable hyper-vacuolization in the PC enterocytes independent of diet indicates a situation of limited lipid transport capacity across the gut mucosa, leading to steatosis, also called lipid malabsorption syndrome. The steatosis may be due to high fat intake concomitant with the deficient supply of choline which is necessary for efficient lipid transport (Hansen et al., 2020a; Hansen et al., 2020b). Slower rate of several physiological functions just before and after seawater transfer could be an additional cause of limited lipid transport capacity (Hoar, 1988; Tipsmark et al., 2010a; Tipsmark et al., 2010b; Bjornsson et al., 2011). Seawater transfer reduced PC hyper-vacuolization and decreased the expression of plin2, a marker for intestinal lipid droplets and steatosis (Heid et al., 1998), supposedly as a consequence of reduced feed intake with corresponding reduction in demand for lipid transport capacity. The increased expression of genes related to the synthesis and transport of cholesterol (cyp51 and npc1l1) supports this explanation. On the contrary, normal smolts showed opposite results in our previous study, i.e., exacerbating lipid malabsorption symptoms in fish after seawater transfer (Wang et al., 2020). The explanation for the apparent difference may be differences in the feed intake due to different environmental temperature. The normal smolts in the previous study were transferred from cold (4.5°C) to warmer (9.9°C) waters, which may have stimulated the feed intake (Wang et al., 2020). The situation was the opposite for the out-of-season smolt in the present study, which was transferred from relatively warm (15.6°C) to colder (7.3°C) waters, supposedly reducing feed intake.

In line with previous studies in out-of-season as well as in normal smolts (Lokesh and Kiron, 2016; Dehler et al., 2017b; Jaramillo Torres, 2017; Rudi et al., 2018; Jaramillo-Torres et al., 2019; Lokesh et al., 2019), we observed that seawater transfer had a substantial effect on the microbiota profile. When in FW the fish were strongly dominated by Ruminococcaceae, suggesting that Ruminococcaceae could be a core taxon in the freshwater life stage, at least under conditions similar to those in the present study. This consideration is supported by a previous study in Atlantic salmon parr showing Ruminococcaceae as one core taxon accounting for about 46% of the total relative abundance in most intestinal digesta samples (Dehler et al., 2017a). Ruminococcaceae can tolerate bile and degrade complex polysaccharides in the mammalian gut (La Reau and Suen, 2018). These traits might have allowed Ruminococcaceae to thrive in the salmon gut. After the seawater transfer, gut microbiota was mainly dominated by Lactobacillus and Photobacterium, which is in line with our previous findings in normal smolts (Wang et al., 2021). Independent of experimental conditions, Lactobacillus is particularly well-adapted to the Atlantic salmon gut environment, especially in the seawater stage (Dehler et al., 2017b; Rudi et al., 2018; Wang et al., 2021). Lactobacillus is resistant to bile acid (Cai et al., 1999; Buntin et al., 2008), which might facilitate their colonization in the salmon gut. Photobacterium has also been reported to be part of the core bacteria in the Atlantic salmon gut (Gajardo, 2016; Gajardo et al., 2017; Wang et al., 2021) and is commonly found in the surrounding water where the experiment was conducted (Rud et al., 2017).

The difference in the microbiota profile before and after seawater transfer may be due to exogenous and/or endogenous factors, which differ substantially between FW and SW. The possible exogenous factors include environmental microbiota, temperature, salinity, as well as the photoperiod under commercial Arctic conditions (Herlemann et al., 2011; Canfora et al., 2014; Dehler et al., 2017a; Dehler et al., 2017b; Rudi et al., 2018; Jin et al., 2019). For instance, salmon migrating from the freshwater to the sea must increase their drinking rate to prevent dehydration in a hyperosmotic environment. This behavioral change increases exposure to the environmental microbiome in the seawater which contains a higher abundance of Photobacterium than in the freshwater (Lorgen-Ritchie et al., 2021). Moreover, the diet compositions differ in protein and fat levels between FM and SW, a factor known to influence the gut microbiota profile (Ringø et al., 2016). Endogenous factors which vary between FW and SW, and which can affect microbiota composition include a range of metabolic, morphological, immunological, and osmoregulatory characteristics (Hoar, 1988; Bjornsson et al., 2011; Johansson et al., 2016; Karlsen et al., 2018), as discussed in more detail above.

The substantially higher microbial diversity and richness observed in fish in SW compared to those in FW, suggest a more diverse microbiota in the SW fish. In mammal, a richer and more diverse microbiota is commonly taken as an indicator of healthier and more normal physiological functions, while relatively low microbial diversity is generally associated with dysbiosis or certain diseases (Mosca et al., 2016). However, whether more diverse microbiota is inherently beneficial for salmon cannot be concluded upon, as our present knowledge basis is still limited regarding the interactions between the microbial diversity and physiological responses (Ray et al., 2012). However, one recent study indicates that relatively low microbial richness and diversity could maintain or even promote growth performance and health of Atlantic salmon (Bozzi et al., 2020), as also may be indicated by the results of the present study.

Our study showed that the relative abundance of twelve taxa, including lactic acid bacteria (e.g. Lactobacillus, Lactococcus, and Leuconostoc), were positively associated with the expression of several gut immune genes. This result agrees with previous observations made in salmonids showing that lactic acid bacteria may affect immune responses through modulation of expression of pro and anti-inflammatory cytokine genes (Panigrahi et al., 2007; Perez-Sanchez et al., 2011). Regarding fish, most previous papers suggest that increased abundance of LAB in the intestine or dietary probiotics which contains LAB strains could exert beneficial actions for the host (Balcázar et al., 2008; Gatesoupe, 2008). From studies of tilapia, it has also been reported that abrupt suspension of probiotic Lactobacillus strains administration could increase host pathogen susceptibility by inducing gut dysbiosis (Liu et al., 2016). However, the presumption that LAB is beneficial, is challenged as there are indications that an increased population of LAB, at least sometimes, accompanies enteritis, suggesting that high LAB may not always be beneficial for Atlantic salmon (Reveco et al., 2014; Schmidt et al., 2016; Gajardo et al., 2017; Krogdahl et al., 2020). Krogdahl et al. (2020) suggest that the increased populations of LAB in fish fed diets containing soybean meal may be due to soybean’s high fiber content, indigestible for the fish, but which can serves as substrate for the microbiota. However, the specific interaction between LAB abundance and host response needs to be further investigated as an increase in abundance of LAB may, under other circumstances, induce detrimental effects on host health. Concerns regarding the efficacy and safety of feeding certain lactic acid bacteria strains have been raised [See reviews by Lerner et al. (2019) and Martinez et al. (2015)]. These concerns are based on studies showing that Lactobacillus plantarum, which is a potent strain of probiotics, may disrupt healthy intestinal tissues in humans (Tsilingiri et al., 2012) and worsen colitis in mice (Mileti et al., 2009). Taken together, it is clear that present knowledge is not sufficient to conclude whether the increase in population of lactic acid bacteria is beneficial or detrimental to Atlantic salmon health. More investigation is required to understand if, when and how the lactic acid bacteria might modulate gut health for the benefit of the fish.

Our results, showing that the Test-diet fed fish in SW revealed less severe steatosis in the PC compared to those fed the reference diet in the FW, suggest that the mixture of nucleotides and immune stimulants improved lipid transport across the mucosa. The elevation in plasma cholesterol levels, and the similar trend shown for plasma fatty acids in the Test-fed fish support this suggestion (Field and Mathur, 1995)

As in our previous study with normal smolts (Wang et al., 2020), only minor effects of functional feed ingredients were observed regarding most growth and health endpoints, especially for the fish in SW. The same regards the gut microbiota. The explanation may be that effects of functional feed ingredients in general are weak, and that effects of variation in environmental factors under practical farming conditions have greater impact on the fish physiology and gut microbiota and are masking possible effects of functional feed ingredients (Ringø et al., 2010). On the other hand, it is unknown, but likely, that the suppression in feed intake and physiological functions just before and after seawater transfer weakened host responses to these selected functional feed ingredients. However, absence of the intended, beneficial effect of functional feed ingredients deserves attention and further investigation to strengthen the basis for making decisions regarding use of functional feed ingredients in salmon production, at least for economic reasons.