Six isonitrogenous and isoenergetic (on dry matter basis; see Table 1 ) experimental diets were formulated based on the nutrient requirements of Atlantic salmon (National Research Council, 2011). The diets were designed to include one of the following microalgae: Chlorella vulgaris, Tetraselmis chuii (TC), Arthrospira platensis (AP) or Schizochytrium limacinum (SL) at an inclusion level of 8%. Two different Chlorella vulgaris were included in the experimental design, one had an intact (CVI) and the other a broken cell wall (CVB). The microalgae were obtained from a commercial supplier and were cultivated in both open and closed bioreactors under commercial settings ( Supplementary Table 1 ). All microalgae were spray-dried after harvesting. Inclusion of the microalgae was done in exchange for wheat starch, wheat gluten and canola oil. In contrast to other studies basal feed components (e.g. fish meal and fish oil) were kept constant in the diet formulation and hence allowed to evaluate the direct effect of every microalgae ingredient. The experimental diets were pelletized (Type 14U175, Amandus Kahl, Hamburg, Germany) at temperatures below 60°C to pellets with 4 mm diameter and stored at 4°C in the dark before and during the trial.

The experiment was conducted at the facilities of the Fraunhofer IMTE, Büsum, Germany. Atlantic salmon smolts were obtained from Jurassic Salmon, Poland and were acclimated for two months in a recirculating aquaculture system. During acclimation, the fish were fed a commercial salmon diet (Aller Aqua, Denmark). Water treatment of the RAS (7.6 m³, turnover rate 4 times h⁻¹) consisted of a moving bed biofilter, a bead filter (PolyGeyser, Model DF-6, Aquaculture Systems Technologies, L.L.C., New Orleans, LA, USA), a protein skimmer and UV-light disinfection. Water quality parameters were measured on a daily basis (NH₄ ⁺ and NO₂ ^- biweekly) and kept in a suitable range for Atlantic salmon (13.5 ± 0.4°C, 7.3 ± 0.1 pH, 10.3 ± 0.2 mg/L O₂, 0.2 ± 0.1 mg/L NH₄ ⁺, 0.2 ± 0.04 mg/L NO₂ ^-, (Microquant test kit for NH₄ ⁺ and NO₂ ^-, Merck, Darmstadt, Germany). Salinity was set to 13.0 ± 0.8 psu (HI 96822 Seawater Refractometer, Hanna Instruments Inc., Woonsocket-RI, USA) by mixing freshwater and seawater. Light was provided for 24 h throughout the experimental period. Prior to the start of the experiment Atlantic salmon smolts (mean body weight 82.32 ± 1.96 g) were randomly divided into six different groups in triplicate, each consisting of 28 fish and stocked into 18 tanks (300 L) of the RAS ( Figure 1 ). Following rearing the fish for eight weeks in brackish water, all of the fish were transferred into a new RAS system, which was identical to the other, but operated with full strength seawater (salinity of 31.8 ± 0.5 psu). Water parameters for this system were as followed: 13.4 ± 0.3°C, 7.2 ± 0.1 pH, 10.4 ± 0.3 mg/L O₂, 0.2 ± 0.1 mg/L NH₄ ⁺, 0.2 ± 0.07 mg/L NO₂ ^-. The fish were kept under these conditions for additional two weeks before the experiment was terminated. The fish were fed manually twice per day (8 a.m. and 2 p.m.) until apparent satiation during the entire experiment. Leftover pellets were collected, counted and used to calculate feed intake.

Samples were collected before the onset of the experiment (T0), after two weeks (T1) and eight weeks of feeding the experimental diets in brackish water (T2) and two weeks following transfer into seawater (T3; Figure 1 ). At each sampling nine fish per treatment (three per tank) were randomly sampled. The fish were quickly netted from the experimental tanks and euthanized by an overdose of buffered MS-222 (0.3 mg/L). For each fish total length and total weight was recorded. 2 ml of blood was collected in heparinized syringes by caudal vein puncture. The blood was transferred into 2 ml Eppendorf tubes and centrifuged at 4000 g for 8 min. Aliquots of the plasma were flash-frozen on dry ice and stored at −80°C for the determination of plasma metabolites, total carotenoid content, and enzyme activities.

The liver and spleen were carefully removed and weighed for the calculation of organ specific indices. At the end of the brackish water phase (T2) a piece of the liver, anterior intestine and the spleen was placed in an RNase free tube and flash frozen in liquid nitrogen for gene expression analysis. In addition, after two (T1) and eight weeks (T2) of feeding the experimental diets a piece of the liver was flash-frozen on dry ice for later protein analysis using western blots.

At T0, T2 and T3 both fillets from every fish were taken, de-skinned, homogenized by means of a knife-mill (Grindomix GM200, Retsch GmbH, Haan, Germany) and stored at −40°C for later analysis of proximate composition, fatty acid profile, as well as carotenoid content. At T0 and T2 three additional fish per tank were sampled and pooled for the analysis of whole-body proximate composition.

Proximate composition was analyzed in microalgae ( Supplementary Table 2 ), diets ( Table 2 ) and whole-body homogenates in duplicates using the same methods. Whole body samples were freeze-dried (Alpha 1-2 LD plus and Alpha 1-4 LSC, Martin Christ Gefriertrocknungsanlagen GmbH, Osterode am Harz, Germany) until a stable weight was achieved and homogenized using a knife mill (GM 200, Retsch GmbH). Nutrients and gross energy were analyzed according to EU guideline (EC) 152/2009. Dry matter content was determined following drying of samples at 103°C in a drying oven for 4 h (ED 53, Binder GmbH, Tuttlingen, Germany). Ash content was determined after combustion in a combustion oven at 550°C (P300, Nabertherm, Lilienthal, Germany). Crude protein content was analyzed following the Kjeldahl method (InKjel 1225M, WD30, Behr, Düsseldorf, Germany). Crude lipid content was extracted with petroleum ether in a Soxhlet extraction system (Soxtherm, Hydrotherm, Gerhardt Königswinter, Germany) and quantified gravimetrically. Gross energy was determined using a bomb calorimeter (C 200, IKA, Staufen, Germany).

To physically break down the material, 10 g of each diet and microalgae ( Supplementary Table 3 ) was ground with a mortar, mixed with 20 ml of distilled water and homogenised for 2 min with an Ultra-Turrax disperser (IKA). The slurry was lyophilised and mortared again. To determine the fatty acid composition, 20 mg sample (weighed to the nearest 0.1 mg) was then subjected to direct transesterification according to Griffiths et al. (2010). However, boron trifluoride-methanol was replaced by 3 M methanolic HCl (Sigma-Aldrich, Taufkirchen, Germany) and distilled water by 1 M aqueous NaCl (Merck, Germany). All other reagents were from Sigma-Aldrich. Each diet was transesterified in triplicate and the fatty acid methyl ester (FAME) extracts used for separate gas chromatography (GC) analysis. Fatty acid contents were calculated as mg FAME/100 g dry matter ( Table 2 ). For GC conditions, see the section on muscle proximate and fatty acid composition.

Moisture and ash content were determined by drying samples of around 5 g for 12 h at 105°C, followed by ashing at 550°C. The nitrogen content was measured by Dumas using a LECO TruSpecN (Leco Instruments GmbH, Mönchengladbach, Germany). Lipids were extracted according to Smedes (1999) using cyclohexane and 2-propanol (VWR, Darmstadt, Germany) with modifications by Karl et al. (2012). For GC analysis of fatty acids, lipids were obtained in a separate extraction without final drying at 105°C and transesterified into FAME using methanolic potassium hydroxide (ISO-IDF, 2002).

Fatty acid analysis was performed using a 7890A gas chromatograph (Agilent Technologies, Santa Clara, CA, USA) equipped with a 7683B autosampler, a split injection port (injection volume 1 µL, split 1:100), flame ionisation detection and a 100 m x 0.25 mm i.d. x 0.20 µm - CP-Sil 88 column (Agilent Technologies). Hydrogen was used as the carrier gas with a constant flow of 1.6 mL min^-1. Two min after injection, the initial oven temperature of 175°C was increased by 1°C per minute to 190°C, held constant for 1 min, then increased by 5°C per minute to 225°C, held constant for 7 min, and finally increased by 1°C per minute to 237°C (1 min constant). Chromatograms were evaluated using EZChrom Elite 3.3.2 (Agilent Technologies). Identification of individual FAME was achieved by comparison to known standards (Supelco™ 37 Component FAME mix, PUFA No. 1, PUFA No. 3; all obtained from Sigma-Aldrich) in the range from C14:0 to C22:6n3. Fatty acid contents for fillet samples were calculated as weight percentage (g FA/100 g FA) and are given as means of duplicate analyses.

Carotenoids were extracted from the diets and microalgae with methanol using an Ultra-Turrax (IKA) at 24,000 rpm for two cycles of 45 s (Schüler et al., 2020). Samples were centrifuged and the supernatants of three repetitions were combined. During extraction the samples were kept on ice. The extraction of the fish fillet homogenate was done according to Ostermeyer and Schmidt (2004). After an initial evaluation of carotenoids in individual fish samples (n = 18), muscle samples were pooled on a tank level (n = 3).

For the quantitative determination of the carotenoids an aliquot of the extract was evaporated and the residue dissolved in mobile phase. HPLC was carried out using a C30 analytical column (5 µm, 250 x 4.6 mm i.d., YMC Europe, Dinslaken, Germany) preceding by a C30 guard column (5 µm, 10 x 4.0 mm i.d.) and a gradient of methyl tert-butyl ether, methanol with a small amount of an ammonium acetate buffer (pH 4.6), similar to (Rasmussen et al., 2012). A flow rate of 1.0 mL min^-1 at 25°C with an injection volume of 100 µL was used. The detection was performed with a photodiode array detector (UV 6000 LP; Thermo Finnigan, San Jose, CA, USA) at 450 nm and 470 nm. Peaks were identified by comparison of the retention times and the absorption spectra (between 380 nm and 700 nm) with those of synthetic standards. The carotenoids were quantified using an external standard containing lutein, fucoxanthin, violaxanthin, neoxanthin, astaxanthin, zeaxanthin and canthaxanthin (ChromaDex, Irvine, CA, USA; Sigma-Aldrich; Dr. Ehrenstorfer, Augsburg, Germany).

Plasma glucose, triglycerides, total protein, alkaline phosphatase (ALP), aspartate aminotransferase (AST) and alanine amino-transferase (ALT) activity were measured on a Fuji Dry Chem NX500i (Fujifilm, Ratingen, Germany) using commercial kits and following the manufacturer´s instructions.

Total carotenoid content in plasma samples ( Supplementary Figure 1 ) was measured after Donaldson (2012) with slight modifications. Briefly, 100 µL of plasma was mixed with 100 µL of 70% ethanol in a 1.5 mL reaction tube wrapped with aluminium foil and vortexed for one minute to precipitate the proteins. Then 300 µL of n-heptane (Roth, Karlsruhe, Germany) was added and the mixture was vortexed for four minutes at maximum speed. The mixture was centrifuged at 2000 g for two minutes. Following separation of both layers 290 µL of the heptane layer including the dissolved carotenoids were decanted and added to a 10 mm quartz cuvette (type 104-QS, Hellma, Müllheim, Germany) and further diluted with 410 µL of heptane. The absorbance was measured at 448 nm using a spectrophotometer (SPECORD210, Analytik Jena GmbH, Jena, Germany). The carotenoid content was then calculated according to Donaldson (2012). Spectral profiles of every sample (300 – 600 nm) confirmed the presence of a carotenoid peak at ~ 448 nm ( Supplementary Figure 1B ). Preparation of extracts and measurements were performed under reduced light conditions to minimize pigment degradation in the samples.

Total protein from liver samples (n = 3 pool per tank) was extracted with Radioimmunoprecipitation (RIPA) lysis buffer according to the manufacturer’s protocol (RIPA Lysis Buffer System, Santa Cruz Biotechnology, Dallas, Texas, USA). The proteins Cu, Zn superoxide dismutase (Sod1) and myeloperoxidase (Mpo) were analyzed in the salmon liver per SDS-PAGE and Western Blot. The protein ß-actin served as loading control. A no template control and one positive control per antibody were included, Danio rerio liver for Mpo, Bovine liver for Sod1 and HEK-293 cells for ß-actin. 20 µg total protein was processed in reducing conditions with SDS sample and reducing buffer (both TruPAGE, Sigma-Aldrich, Schnelldorf, Germany) at 70°C for 10 min. SDS-PAGE was performed in a Xcell SureLock Mini-Cell (Thermo Fisher Scientific, Waltham, Massachusetts, USA) using precast 4-12% gradient gels, TruPAGE running buffer and antioxidant (Sigma-Aldrich). Proteins were electro-transferred to a PVDF membrane. For parallel protein detection of Mpo and Sod1 the membrane was horizontally cut. Primary antibody incubation, with Mpo antibody (ab210563, Abcam, Cambridge., UK) in 1:5000 dilution in PBS-T containing 2.5% skim milk and Sod1 antibody (NBP2-24915, Novus Biologicals, Bio-Techne Ltd., Abingdon, UK) in 1:500 dilution in PBS-T containing 2.5% skim milk, was performed at 4°C overnight. Secondary antibody anti-rabbit IgG conjugated HRP (sc-2357, Santa Cruz Biotechnology) was incubated in a 1:5000 dilution for 90 min at room temperature. Detection was performed using ECL detection reagents (Amersham, Global Life Sciences Solutions USA LLC, Marlborough, MA, USA) and chemiluminescence film (Amersham, GE Healthcare Ltd, Little Chalfont, UK) with 40 sec exposure time for both Mpo and Sod1. For the subsequent detection of ß-actin the antibodies were stripped using 100 mM Glycin buffer (pH 2.5). The membrane was incubated in 1:5000 dilution of β-actin antibody (NB600-503, Novus Biologicals, Bio-Techne Ltd., Abingdon, UK) at 4°C overnight. Quantification of protein expression was done following densitometric analysis of the protein bands using GIMP and normalized to housekeeping (ß-actin) protein expression.

Total RNA was extracted using TRIzol (ThermoFisher Scientific, Waltham, MA, USA) and further purified with the ISOLATE II RNA Micro Kit (Meridian Bioscience Inc., Cincinnati, OH, USA). The concentration and integrity of the extracted RNA was measured by NanoDrop One (Thermo Fisher Scientific). Subsequently, cDNA synthesis was performed using Reverse Transcription Master Mix (Fluidigm, San Francisco, CA, USA). The samples were preamplified by the PreAmp Master Mix (Fluidigm) and at last treated with exonuclease I (New England BioLabs, Frankfurt/Main, Germany). All steps have been carried out according to the manufacturer’s instructions.

45 genes with tissue specific regulation were selected from an established gene set composed of key immune and stress regulated genes in Atlantic salmon (Krasnov et al., 2020; Lund et al., 2022; Supplementary Table 4 ). We extended this set by three immunogene-specific primers derived for hamp, saa5 and sod1 ( Supplementary Table 4 ). The same 48 genes (45 target and 3 reference genes) were measured in the tissue anterior intestine, liver and spleen.

The 48.48 gene expression biochips were primed in the MX IFC Controller (Standard BioTools, San Francisco, CA, USA). The pre-amplified cDNA samples were pipetted to the sample inlets and the primers were loaded on the assay and finally, analyzed with the Biomark HD using the manufacturer’s thermal protocol “GE Fast 48 × 48 PCR+Melt v2.pcl” (application type: gene expression; passive reference: ROX; assay: single probe).

The raw qPCR Ct values were obtained using the Fluidigm real-time PCR analysis software v. 3.0.2 (Munich, Germany). Relative expression was calculated based on ΔΔCt where three reference genes coding for b-actin, ribosomal protein L4 and ribosomal protein S20 (actb, rpl4, rps20) were used as internal normalisers. The mean Ct per gene for all samples was used as a calibrator during the calculation. Relative expression values were log2 transformed prior to statistical analysis. Four individuals with abnormal phenotypic signs and subsequently abnormally high gene expression profiles were removed from the dataset. The genes cxcl8 and il1b in the liver and cxcl8 in the spleen were removed from the dataset, as to many missing values hampered analysis of the data.

Statistical analysis and data visualization were conducted using the software R (R version 4.1.0) in the environment RStudio. For all test α = 0.05 was used as the level of significance. Data is presented as mean ± standard error of mean (SEM). For the performance parameters, protein concentrations in the liver as well as carotenoid concentrations in the muscle an appropriate statistical model based on generalized least squares was defined (Carroll and Ruppert, 1988) which included the factor diet as well as timepoint for the latter two. The residuals were assumed to be normally distributed and to be heteroscedastic, which was based on a graphical residual analysis. Analysis of variance (ANOVA) was conducted, followed by multiple contrast tests for heteroscedastic data (Hasler and Hothorn, 2008) in order to compare the several diets with the control diet. For plasma parameters, proximate and fatty acid composition of the muscle, as well as gene expression data, mixed effect models (Laird and Ware, 1982; Carroll and Ruppert, 1988; Pinheiro and Bates, 2000) were used. The model included diet, timepoint and their interaction as fixed factors and tank as a random factor. The residuals were assumed to be normally distributed and to be heteroscedastic. Based on this model, a Pseudo R² was calculated (Nakagawa and Schielzeth, 2013) and an ANOVA was conducted, followed by multiple contrast tests in order to compare the several diets with the control diet, and the timepoints, respectively. If the factors diet and timepoint had no significant interaction, then corresponding multiple contrast tests were pooled over the levels of the remaining factor. Spearman correlation analysis was employed to relate muscle lutein and plasma carotenoid concentrations, since muscle lutein concentrations were not normally distributed. Pearson correlation analysis was applied to relate the increase in DHA with a decrease in fat content in the muscle.

Feeding the experimental diets for eight weeks to the salmon in brackish water revealed no difference in growth and feed intake among groups ( Table 3 ). Feed conversion ratio (FCR) was highest in the control group (CD), and including broken C. vulgaris (CVB) in the feed significantly improved the feed conversion ratio (p = 0.03). Furthermore, feeding CVB slightly improved the protein efficiency ratio (p = 0.11) but reduced body condition (p = 0.01). Hepatosomatic index and spleen somatic index were not affected by the diet ( Table 3 ). Mortality was low and not different among diet groups. Furthermore, the diet did not affect proximate whole-body composition of the salmon ( Table 4 ) but feeding CVB slightly increased ash content (p = 0.06). After the fish were transferred to full strength seawater voluntary feed intake decreased to one third of the levels prior transfer but was not different among groups ( Table 3 ).

The diet did not affect protein, water and ash content of the muscle in brackish water (T2) and following transfer to seawater (T3; Tables 5 ; S5 ). Fat content was significantly reduced in brackish water in fish fed A. platensis (AP; 12% reduction) and S. limacinum (SL; 13.5% reduction) compared to CD ( Table 5 ). Fat content decreased on average by 29% after transfer to seawater across all groups ( Table 5 ). It decreased most in groups receiving intact and broken C. vulgaris, CVI (38%) and CVB (32.9%), but markedly less in fish fed AP (18.5%).

Generally, fatty acid composition in the muscle lipids reflected that of the diet ( Table 5 ; Table 2 ). Both diet and timepoint significantly influenced the fatty acid composition, however, an interaction of both factors was absent in most cases ( Tables 5 ; S5 ). Alpha-linolenic acid (LA) was significantly enriched in muscle lipids of CVB, CVI and T. chuii (TC) fed fish in brackish water and in CVB and CVI fed fish after transfer to seawater ( Table 5 ). Steraidonic acid (SDA) content was higher in TC compared to CD at both timepoints. Eicosapentaenoic acid (EPA) levels were affected by an interaction of diet and timepoint and levels increased in CD by 11.3%, CVB by 16.2% and CVI 18.1% following transfer to seawater ( Tables 5 ; S5 ). Docosahexaenoic acid (DHA) levels were significantly increased in SL compared to CD fed fish at both timepoints and relative levels increased following transfer to seawater in all groups by 23.8% ( Figure 2A , Table 5 ). A significant negative correlation between the relative reduction in total fat content in the muscle and the relative increase in DHA based on diet group means was detected (R = 0.89, p < 0.001; Figure 2B ).

Lutein was the main carotenoid present in all groups and significantly enriched in the muscle at both timepoints of fish fed CVI, CVB and TC ( Table 5 ) reflecting the content of lutein in the respective algae ( Supplementary Table 3 ). Lutein content in fish fed CVB was more than two times higher than in fish fed CVI and TC. Zeaxanthin was detected in muscle of fish fed AP and TC and violaxanthin was only detected in fish fed TC ( Table 5 ).

The total carotenoid content in plasma of CVB fed fish was two times higher than in fish fed CVI (p = 0.08) and TC (p = 0.05; Figure 3A ), which was also visible when comparing plasma samples of the respective groups directly ( Supplementary Figure 1A ). No carotenoid content was detectable in groups fed CD and SL ( Supplementary Figures 1A, B ), while values for AP were below the calculated standard curve and were subsequently excluded from further analysis. Muscle lutein concentrations significantly correlated with total carotenoid concentration in plasma samples (Spearman R = 0.52, p = 0.03; Figure 3B ).

CVI inclusion significantly lowered aspartate aminotransferase and alanine aminotransferase activity in microalgae fed groups compared to CD (p ≤ 0.046; Figures 4A, B ). Furthermore, SL lowered aspartate aminotransferase activity (p = 0.06; Figure 4A ). The highest variance in aspartate aminotransferase and alanine aminotransferase activity was found among fish fed CD, while it was lowest among fish fed CVI. Total plasma protein and alkaline phosphatase activity were not influenced by the diet but showed an overall increase between T1 and T2 (p < 0.001; Supplementary Figure 2 ). Total cholesterol increased significantly between the first two samplings for fish fed CVI, CVB and TC (p < 0.05; Supplementary Figure 2 ). Glucose furthermore increased over time only in fish fed SL (p = 0.02; Supplementary Figure 2 ).

Abundance of myeloperoxidase (Mpo) in the liver of Atlantic salmon fed a microalgae-enriched diet was lower in most cases at both timepoints compared to CD ( Figure 5A ). This effect was however only significant for fish fed AP at T1 (p = 0.049) with a 35% reduction compared to CD due to a large within group variation ( Figure 5A ). Though, AP at both timepoints showed the lowest variation. Sod1 protein level was induced by 3-fold in SL at T1 compared to CD (p = 0.10; Figure 5B ).

Only few genes were significantly differentially expressed between microalgae fed groups compared to CD fish, due to a large overall variability in expression ( Figures 6 – 8 ) although some large fold-changes were evident ( Supplementary Figure 3 ). In the anterior intestine, increased drtp1-transcript levels were found in CVI- (2.8-fold, p = 0.065; Figure 6A ) and CVB-fed salmon (7-fold; p = 0.023; Figure 6A ). isg15 transcript levels were induced in all microalgae fed groups except for CVI ( Figure 6B ), although not statistically significant. il1r2 transcript levels were reduced by 2.7-fold in fish fed CVI (p = 0.039; Figure 6C ) and il10rb levels were reduced in fish fed CVI (p = 0.028; Figure 6D ) and TC (p = 0.039).

In the liver, the transcript levels of the acute-phase gene saa5 were significantly reduced in fish fed CVB (0.37-fold; p = 0.016; Figure 7A ) and SL (0.29-fold; p = 0.012, Figure 7A ). c1ql2 transcripts were significantly induced in fish fed CVI (p = 0.032) and SL (p = 0.03; Figure 7B ). hamp transcripts were 2.2-fold higher concentrated in the liver of fish fed CVB (p = 0.066; Figure 7C ) and lyzc2 transcripts were even 12-fold increased in fish fed TC (p = 0.04; Figure 7D ) compared to the control group.

In the spleen, c1ql2 transcripts were slightly 1.5-fold induced in fish fed CVB (p = 0.095) and TC (p = 0.097; Figure 8A ). ikba transcripts were 0.59-fold reduced in the spleen of fish fed CVI (p = 0.06; Figure 8B ). Furthermore, transcript abundance of saa5 was 2.1-fold increased in fish fed CVI (p = 0.034; Figure 8C ). The transcript level of cd209d was reduced across all microalgae-fed fish, although not statistically significant ( Figure 8D ).

Since the levels of plasma markers and selected transcripts varied largely, we conducted a correlation analysis to identify connections and validate the overall utility of the used health parameters. However only alkaline phosphatase (ALP) activity in plasma significantly correlated with clra (R = 0.46, p = 0.008; Supplementary Figure 4A ) and c4b transcript levels in the liver (R = 0.45, p = 0.008; Supplementary Figure 4B ).