This experiment evaluated effects of diet composition on digestive functions in lumpfish with an initial body weight of 15 g growing to 50 g. The experiment was conducted based on a three-component mixture design (Cornell, 2011), set up with Design Expert ver. 8.0.4. (Stat-Ease Inc. MN, USA). As it is a special type of response surface methodology for experimental design, for which replicates are not necessary (Hamre et al., 2003; Hamre and Mangor-Jensen, 2006; Hamre et al., 2013), we set one tank for most diets. In total, twelve different diets were formulated and fed to fish in one tank for each one of 11 diets. The 12^th diet was fed to fish in 3 tanks to obtain a measure of tank variation. The experimental diets were produced at the Aquafeed Technology Centre of Nofima in Bergen, Norway, in the same production series, using a Wenger TX-52 co-rotating twin-screw extruder with 150 kg/h capacity. The dietary oil was added in the different feed mixes prior to extrusion. The settings of the extruder were “normal”; i.e., the production can be upscaled to a feed factory (extruder settings considered: screw configuration (D), die opening (1.5 mm), knife speed (2671-3108 rpm), feed rate (110-150 kg/h), at the DDC the amount of added steam was 10-12 kg/h, and water 0.16-0.18 kg/min, and in the extruder there was added 0.24-0.45 kg/min water and no steam. The ingoing temperature of the feed mass in the extruder was 79-84°C and the outgoing 103-118°C. The produced pellets were air-dried in a carousel dryer (Model 200.2, Paul Klöckner GmbH, Nisteral, Germany) at 85°C for 10-12 min to a final moisture level between 6.36 and 8.26%. The size of the dried pellets was approx.: 1:5 − 1:7 × approx.: 2:3mm with bulk density between 443 and 582 g/L. Diet recipes and proximate composition are shown in Table 1 and Figure 1 illustrates the dietary design of experiment E1. The following ranges of macronutrients were covered: Protein 43-68%, lipid 4-17%, and carbohydrate 6-17%. For most observed biomarkers, samples from all treatments were investigated. For some, due to resource restrictions, it was possible to investigate only selected macronutrient combinations.

Details of the experimental conditions are published in (Hamre et al., 2022). In brief, the fish were produced by a commercial hatchery and transported to Nofima’s research facility at Sunndalsøra, Norway. Fourteen 150 L conical tanks were used, each with 90 fish. Dead fish were removed daily, counted, and weighed. Feed was distributed continuously (15 sec feeding every 5 minutes) using small belt automatic feeders above each tank. The fish were fed to satiation, and the feed rations increased from 30 to 68 g/tank in tanks according to appetite. The fish were fed 1-1.6mm pellets until 20 g size and 1.6-2.3mm until the end of the experimental period. Due to the small pellet sizes, feed intake could not be recorded by our system. The temperature was set to a mean of 9.8°C (min 8.9°C, max 10.6°C). The water flow was set to 4 l/min and oxygen was adjusted to 80-100% by adding oxygen to the water holding tank when needed. Temperature was recorded daily, and oxygen was measured and adjusted 2-3 times per week. As the trial was run in a flow-through system, ammonia nitrogen and nitrate nitrogen were not recorded.

An additional experiment, E2, was conducted to gain knowledge on macronutrient digestibility. Starch digestibility was of particular interest as great variations have been observed between other species (Krogdahl et al., 2004). Feed formulations are shown in Table 2 . The diets were formulated to contain 55% crude protein with low (Diet 1), medium (Diet 2) and high (Diet 3) levels of lipid, and with carbohydrate level of contrary variation to lipid level. The same ingredients were used for all three diets, and the different lipid and carbohydrate levels were balanced by differing the dietary inclusion levels of lipid and wheat meal, respectively. The diets were produced at Aquafeed Technology Centre of Nofima in Bergen, Norway. Each of the three diets were fed to fish in triplicate tanks for a period of 14 days.

Lumpfish with an average body weight of around 100 g were distributed to 9 tanks, 100 individuals per tank, providing a biomass of around 10 kg in each tank. The tanks were cylindrical, with flat bottom and black inner surface, and an efficient water volume of 350 L. All tanks were equipped with automatic belt feeders for distribution of feed, and a separate light source mounted on each tank. Inlet water was run through a 10 μm filter and UV treated. Water flow was set at 12 L/min and oxygen saturation was kept within 80-100%. Average water temperature was 11.6°C (10.9-12.4°C).

The trial was terminated after 42 days. At termination of the experiment, 6 fish from each tank were sampled at random, weighed, and intestinal tissues collected for analyses. The intestine was cleaned of mesenteric fat and divided into three segments, i.e., pyloric intestine including the pyloric caeca (PI), mid intestine (MI) and distal intestine (DI) (See Supplementary Figure S1 ). Each segment was opened longitudinally. Limited by fish size, gut content was only collected from MI, where the amount was enough for analysis. The collected gut content of MI was snap frozen in liquid nitrogen and stored at -80°C before further processing and analysis of activity of digestive enzymes and concentration of bile salts. Samples of tissue from all sections, located in the middle of the sections were collected for RNA extraction (submerged in RNAlater solution, incubated at 4°C for 24 h and stored at -20°C) and histomorphological evaluation (fixed in 4% phosphate-buffered formaldehyde solution for 24 h and transferred to 70% ethanol for storage). The remaining tissue was collected for brush border digestive enzyme assessment, snap frozen in liquid nitrogen and stored at -80°C.

The trial was terminated after 14 days, when fecal samples were collected as pooled samples per tank. Because of the very short length of DI of lumpfish, it cannot be stripped for feces like salmonids, therefore fish were euthanized by an overdose of anesthetic (Tricaine mesylate, MS-222) and the total content of the DI was collected by dissecting the intestine.

The intestinal tissue samples were thawed and homogenized (1:20 w/v) in ice-cold 2 mM Tris/50 mM mannitol, pH 7.1, containing phenyl-methyl-sulphonyl fluoride (P-7626, Sigma, Norway) as serine protease inhibitor. The homogenates were sonicated, aliquoted and stored at -80°C until analysis.

Leucine aminopeptidase (LAP) and maltase were the brush border digestive enzymes assessed. The activity of LAP was measured employing the Sigma procedure no. 251 (Krogdahl et al., 2003). L-leucyl-b-naphthylamide is used as the substrate and is reacted with diluted homogenates at 37 °C for 1 h. The reaction is terminated by HCl. A subsequent color reaction is done by adding sodium nitrite, ammonium sulfate and N-(1-Naphthyl)ethylenediamine dihydrochloride in order at room temperature. A standard curve is made for calculation by measuring serial diluted 2-Naphthylamine. The color absorbance is read at 580nm. To measure maltase activity, the method described by Dahlqvist (Dahlqvist, 1970) was applied. Maltose is used as substrate and is reacted with diluted homogenates at 37 °C for 1 h. The reaction is terminated and colorized by TGO solution (mixture of Trizma buffer, detergent, o-dianisidine and peroxidase). A standard curve is made for calculation by measuring serial glucose dilution. The color absorbance is read at 480nm.

Activity of trypsin, amylase and lipase, and total bile salt level were measured in pooled freeze-dried content from MI. Trypsin activity was determined colorimetrically (Kakade et al., 1973). Benzoylarginine p-nitroanilide (Sigma no. B-4875, Sigma Chemical Co., St. Louis, MO) is used as substrate and is reacted to freeze-dried content solution at 37 °C for 10 min. The reaction is terminated by 30% acetic acid. The color absorbance is read at 410nm. Absorbance variance between sample and blank is defined as enzyme unit instead of international enzyme unit. Lipase activity was determined as described in (Brockman, 1981). 4-Nitrophenyl myristate (Sigma 70124, Sigma Chemical Co., St. Louis, MO) is used as substrate and the reaction temperature is 37 °C. The absorbance is read at 405nm every 30 sec for 7 times. The enzyme unit is defined as absorbance variance per minute. Amylase was determined as described by (Froystad et al., 2006), using a Randox amylase assay kit (AY3805, Randox Laboratories Ltd., Crumlin, UK). The kit employs Ethylidene PNPG7 method (Kruse-Jarres et al., 1989) for α-amylase measurement. Ethylidene-blocked p-nitrophenyl-maltoheptaoside is the substrate in the enzyme assay reagent and reaction temperature is 25°C. Absorbance is read at 405nm after 20, 30 and 40 min of reaction and the enzyme unit is defined as absorbance variance per minute. Bile salt level was determined using the enzyme cycling amplification/Thio – NAD method (Inverness Medical, Cheshire, UK) in the ADVIA^®1650 Chemistry System (Siemens Healthcare Diagnostics Inc.) at the Central Laboratory of the Faculty of Veterinary Medicine, Norwegian University of Life Sciences, Oslo. The assay measures total 3αOH of cholic acid, whereas the reported results indicate the corresponding level of bile salt as taurocholate.

Total RNA was extracted from tissue samples of PI, MI, DI, and liver. (~20 mg) of all fish using Trizol reagent (PureLink™ RNA Mini Kit, Thermo Fisher Scientific). RNA was purified by an on-column DNase kit (PureLink™ DNase Set, Thermo Fisher Scientific) according to the manufacturer’s protocol. RNA purity and concentration were measured using the Epoch Microplate Spectrophotometer (BioTeK Instruments, Winooski, USA). The RNA integrity was verified by the 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA, USA) in combination with RNA Nano Chip (Agilent Technologies, Santa Clara, CA, USA). First-strand complementary DNA (cDNA) was synthesized using 1.0 μg RNA from two fish of the same tank, namely 0.5 μg RNA of each two fish were combined as a unit of RNA sample for cDNA synthesis. The reaction volume was 20 μl, including 4 μl mastermix of from the kit SuperScript™ IV VILO™ Master Mix (Thermo Fisher Scientific). Negative controls were performed in parallel by omitting RNA or enzyme.

Lumpfish mRNA sequences were derived from NCBI database. The selected genes include elongation factor 1-alpha (ef1a), beta-actin (bactin), hypoxanthine-guanine phosphoribosyltransferase 1 (hprt1), tubulin beta chain (tubb), sucrase-isomaltase (si), solute carrier family 27 member 4 (slc27a4), solute carrier family 15 member 1 (slc15a1), niemann-pick C1-like 1 (npc1l1), solute carrier family 12 member 1 (slc12a1), tight junction protein 1a (tjp1a1), occludin (occludin), cyclooxygenase-2 (cox2), immunogobulin M (igm), inhibitor of nuclear factor kappa B kinase subunit beta (ikbkb), complement component 5 (c5), chemokine (C-X-C motif) ligand 19 (cxcl19), tumor necrosis factor alpha (tnfa), nuclear factor kappa-light-chain-enhancer of activated B cells (nfkb), transcription factor p65 (rela), major histocompatibility complex II (MHCII), matrix metallopeptidase 13 (mmp13) and proliferating cell nuclear antigen (pcna). The selected genes cover the functions of disaccharide digestion (si), nutrient transport (slc27a4, slc15a1, npc1l1), ion-exchange (slc12a1), tight junction forming (tjp1a1, occludin), immune regulation (cox2, igm, ikbkb, c5, cxcl19, tnfa, nfkb, rela, MHCII, mmp13) and cell proliferation (pcna). Reference gene candidates include ef1a, tubb, hprt1 and bactin.

Primer information are shown in Supplementary Table S2 . The qPCR primers were designed using the Primer-BLAST tool of NCBI. All primer pairs were first used in gradient reactions to determine optimal annealing temperatures. To confirm amplification specificity, the PCR products from each primer pair were subjected to melting curve analysis and visual inspection of the PCR products by agarose gel electrophoresis. PCR efficiency for each gene assay was determined using 2-fold serial dilutions of randomly pooled complementary DNA.

Expression of target genes were analyzed using the LightCycler 96 (Roche Diagnostics, Basel, Switzerland) with a 10-μl DNA amplification reaction. Each 10-μl DNA amplification reaction contained 2 μl PCR grade water, 2 μl of 1:10 diluted cDNA template, 5 μl LightCycler 480 SYBR Green I Master Mix (Roche Diagnostics) and 0.5 μl (10 μM) of each forward and reverse primer. Each sample was assayed in duplicate, including a no-template control. The three-step qPCR run included an enzyme activation step at 95°C (5 min), 40 cycles at 95°C (10 s), annealing temperature (10 s), and 72°C (15 s) and a melting curve step. The mean normalized expression of the target genes was calculated from raw Cq values (Muller et al., 2002). Reference genes selected was done in terms of its stability among different fish (Kortner et al., 2011). The chosen reference genes are: ef1a, tubb and hprt1 for PI and liver, ef1a, bactin and hprt1 for MI and tubb, bactin and hprt1 for DI.

The samples fixed for histological evaluation from all intestinal segments collected, i.e., PI, MI, and DI, were processed using standard histological methods and stained with hematoxylin and eosin (H&E). The histological sections were estimated to describe the general structure as well as assessing for any morphological changes in the intestinal mucosa such as inflammation. Since the relative weights (OSI) of intestinal sections and liver seemed to vary more specifically to dietary lipid content, histological examination was conducted for the six sampled fish fed the following four diets spanning from low to high lipid content: i.e., 6, 21, 28, and 38 g lipid/100g protein (Diet 1, 5, 3 and 12, respectively). The evaluated morphological characteristics included changes in cellularity of submucosa and lamina propria, enterocyte supranuclear vacuolization and numbers of intra-epithelial lymphocytes. The degree of changes was graded as normal, mild, moderate, marked and severe, and was scored from 0 to 4, respectively.

Samples of the feed and feces were analyzed at the Department of Animal and Aquacultural Sciences, Norwegian University of Life Sciences, Ås, Norway. Dry matter was determined by oven drying at 105°C, 16–18 h, to constant weight. Nitrogen was determined using the Kjeldahl method (Kjeltech Auto Analyser, Tecator, Höganäs, Sweden) (crude protein: N×6.25), fatty acid composition by FAME analysis (O’Fallon et al., 2007) in a Trace GC Ultra with auto injector (Thermo Scientific), and starch was measured as glucose after hydrolysis by α-amylase and amyloglucosidase, followed by glucose determination by the ‘Glut-DH method’. Yttrium was analyzed by dissolution of ashed samples with hydrochloric acid and nitric acid by heating, then dissolved in 5% nitric acid. Yttrium was then detected with an ICP-AEF, Optima 3000 V (Perkin Elmer, USA).

Variation of diet composition significantly affected the intestinal weight of PI and MI ( Figure 2 and Table 3 ). In PI, OSI increased with decreasing protein level, increased with increasing dietary lipid level, and had no response with carbohydrate variation. In MI, OSI increased with decreasing protein level, increased with increasing dietary lipid level, and increased with increasing carbohydrate level.

Specific activity of both LAP and maltase was highest in MI, lowest in the PI, and intermediate in DI. Also, the value for LAP capacity, i.e., per kg of fish, was highest in the MI, whereas for maltase, the highest value was observed for PI. The DI showed the lowest capacity values for both LAP and maltase.

In PI, specific activity of LAP and maltase showed significant relationships with diet macronutrient composition (p = 0.032 and 0.034, Table 3 and Figure 3 ). The specific LAP and maltase activities (U/mg protein) decreased with decreasing protein level, decreased with increasing dietary lipid level and, to lesser extent, increased with increasing carbohydrate level. On the other hand, the enzyme capacity (U/kg fish) of LAP and maltase (p = 0.090 and p > 0.10, Table 3 and Figure 3 ) showed no significant correlations with diet macronutrient composition in this section.

In MI, both specific activity and capacity of LAP and maltase showed clear relationships with diet composition (p = 0.005, 0.016, 0.001 and 0.002, Table 3 and Figure 3 ), i.e., a decrease with decreasing protein level, a decrease with increasing dietary lipid level, and an increase with increasing carbohydrate level.In DI, there were no significant effects for either specific activity or capacity and the analysis by software cannot illustrate a clear trend neither (p = 0.080, 0.1, 0.060 and 0.070, Table 3 ).

Lipase activity in digesta from MI showed a significant, linear, relationship with dietary protein level, decreasing with decreasing protein level (p = 0.020, Table 3 and Figure 3 ). Lipase activity decreased with increasing lipid level as well as with increasing carbohydrate level. The trypsin and amylase activities in digesta from MI did not show significant relationships (p > 0.10, Table 3 ) with diet macronutrient composition. Bile salt concentration showed a tendency towards a similar response pattern as for the lipase (p = 0.077, Table 3 and Figure 3 ), i.e., decrease with decreasing protein, increase with increasing lipid and carbohydrate level.

In PI, si and slc15a1, coding for a disaccharidase and a peptide transporter, respectively, were the two genes showing significant responses to diet macronutrient composition (p = 0.039 and 0.050, respectively, Table 4 and Figure 4 ). Their relationships with diet composition were similar, i.e., decreasing in value with decreasing protein content, decreasing with increasing lipid level and increasing, although only slightly, with increasing carbohydrate. Expression of rela, igm and cxcl19, all involved in immune functions, tended to vary with diet composition (p = 0.080, 0.073 and 0.094, respectively, Table 4 and Figure 4 ). The trend observed for expression of rela was a decrease with decreasing protein level, decrease with increasing lipid level and increase with carbohydrate level, while igm expression decreased with decreasing protein level, decreased with increasing lipid level, but increased slightly with increasing carbohydrate level. The trend of cxcl19 expression fitted a quadratic model, where the minimum was at medium protein, high lipid, and low carbohydrate level, and the maximum was at medium protein and lipid level, and high carbohydrate.

In MI, more genes responded significantly to variation in diet composition. As in PI, significant response was observed for si and slc15a, and the pattern of responses was quite similar, with a decrease with decreasing dietary protein, a decrease with increasing lipid and an increase with dietary carbohydrate level (p = 0.026 and 0.001, Table 4 and Figure 5 ). Also, slc12a1, the gene involved in ion exchange expression decreased with decreasing protein level, decreased with increasing lipid (p = 0.031, Table 4 and Figure 5 ) and increased with increasing carbohydrate level. Regarding genes involved in immune functions, linear effects of diet composition were seen for igm and mmp13, and quadratic responses were seen for rela and cxcl19 (for p values see Table 4 and Figure 5 ). For the former four, expression decreased with decreasing protein level, most pronounced for mmp13, decreased with increasing lipid level as well as with increasing carbohydrate level.

For rela, which fitted a quadratic model, the maximum expression was found near the center of the design, i.e., at medium levels of all the three macronutrients. Two low values were seen, i.e., at medium protein, high lipid, and low carbohydrates, and at low protein, high lipid, and high carbohydrates. The highest value for cxcl19 was observed at the highest protein level, at which both lipid and carbohydrates were low. Lowest value was found at medium protein and lipid level and high carbohydrate level. For some genes showing fewer clear relationships with diet composition, trends (0.05 < p <0.10) were indicated, i.e., for npc1l1, cox2, and pcna (for p values see Table 4 ). The relationship of expression of npc1l1, the cholesterol transporter, showed decreased level with decreasing protein level, decreased level with increasing lipid level and increased level with increasing carbohydrate level. The expression cox2 and the cell proliferating related gene pcna fitted quadratic models. The maximum expression of cox2 was at high protein, low lipid, and low carbohydrate level, with a lower peak at low protein level, high lipid, and high carbohydrate level. The minimum was shown at medium protein level, medium lipid, and high carbohydrate level. Maximum expression of pcna was at medium lipid and protein level and low carbohydrate level.

In DI, only expression of the tight junction related gene occludin and the immune related gene igm were significantly influenced by diet composition (p = 0.003 and 0.004 respectively, Table 4 ). Expression of igm was best explained by a quadratic model ( Figure 6 ), with two maxima, one at medium protein, high lipid, and low carbohydrate level, and the other at medium protein level, low lipid, and high carbohydrate level. The minimum was observed at low protein, high lipid, and high carbohydrate level. The occludin expression, following a linear model, showed highest values at high protein a low lipid level, with minor, but positive dependency on carbohydrate level.

Among the genes observed in the liver, only c5 and pcna, coding for proteins participating in immune regulation and cell proliferation, respectively, showed significant responses to the dietary variation (p = 0.020 and 0.025, respectively, Table 4 ). The expression of c5 was affected only to a minor degree by protein level, decreased with increasing lipid, and increased with increasing carbohydrate level ( Figure 7 ). The pcna expression data fitted a quadratic model ( Figure 7 ) with two maxima, one at medium protein, high lipid, and low carbohydrate level, the other at low protein, high lipid, and high carbohydrate level. One minimum was apparent, i.e., at high protein, low lipid and medium carbohydrate and high protein level.

The general histological appearance of the gut mucosa is presented in Figures 8A–D . The stomach structure was quite similar to that observed in other fish species with simple columnar epithelium, gastric glands, and a thick smooth muscle layer with two layers, the inner circular and the outer longitudinal ( Figure 8A ). The structure of the pyloric caeca ( Figure 8B ) also appeared similar to what has been observed in other fish species. The caeca were surrounded by pancreatic tissue and were of thin wall with simple unbranched mucosal folds. Lamina propria and submucosa were thin. Fused caeca, due to branching of pyloric caeca were a unique finding for the lumpfish with shared muscular and serosal wall ( Figure 8B , blue arrow). The mid intestine showed structures comparable to those observed in Atlantic salmon with short mucosal folds and short branching, thin lamina propria and submucosa with sparse cellularity, but thick muscle layer ( Figure 8C ). The structure of the distal intestine ( Figure 8D ) appeared similar to that in Atlantic salmon with simple (Baeverfjord and Krogdahl, 1996; Knudsen et al., 2007), and complex (branched) mucosal folds which were taller than in mid intestine, and lamina propria and submucosa with fibrous tissue and little cellular composition. The enterocytes showed little or no vacuolization, a feature which may distinguish the lumpfish from Atlantic salmon which, in the fed state, show high vacuolation of the distal intestinal enterocytes.

Figures 8E–H show histological features of the cells of the PI and liver in fish fed diets with high protein and low lipid and low protein high lipid levels. Hyper-vacuolization, a symptom of excessive lipid accumulation, so-called steatosis, was observed in both PI and liver. The steatosis level was scored and analyzed for PI, MI, and liver ( Figure 9 ), clearly showing increasing steatosis with increasing lipid level.

The results of the additional digestibility experiment, E2, are illustrated in Figure 10 , showing effects of dietary lipid level on starch, protein, and fatty acid digestibility. Protein digestibility did not change as lipid or starch level varied. Fatty acid digestibility increased as dietary lipid level increased. For saturated fatty acids this effect was not significant, but the results showed the same trend as for other fatty acids (p = 0.06). The digestibility of ω-6 fatty acids was relatively low, between 60% and 68%. Noticeably, starch digestibility decreased greatly as starch level increased, from 84.3% at 6.6% inclusion to 50% at 16% inclusion. This means a mean partial digestibility of the starch increment from 6.6% to 16.0% of about 27% and increase in available starch level from 55-80 g/kg diet.