Prebiotics Modify Host Metabolism in Rainbow Trout (Oncorhynchus Mykiss) Fed with a Total Plant-Based Diet: Potential Implications for Microbiome-Mediated Diet Optimization

However, plant-based cause certain complications in carnivorous species such as rainbow trout. Here, we examined whether prebiotics have the potential to affect the metabolism of juvenile trout (Average weight: 25.88±0.91 g) via microbially derived short-chain fatty acids (SCFAs). Fructo-oligosaccharides (FOS), inulin or mannan-oligosaccharides (MOS) were used at 1% or 2% in a 12-week feeding experiment. We measured changes in the intestinal microbiome, SCFA levels and metabolic responses in the intestine, liver, muscle, and adipose tissue.


Results
In the intestine, gene expression and SCFA production did not change signi cantly with prebiotics, although the MOS fed groups were clustered differently. Prebiotics had a signi cant effect on the abundance of Bacillus, Lactobacillus and Weissella, although posterior intestinal microbial diversity and composition did not change signi cantly after feeding prebiotics. Two operational taxonomic units (OTUs) belonging to Mycoplasma. dominated all samples with an average relative abundance of >95% per group. Intestinal microvillar structures were signi cantly improved in length in the inulin-fed groups.
Systemically, overall hepatic gene expression was signi cantly different from control with inulin-fed groups showing upregulation of several metabolic and the fatty acid receptor genes. MOS fed groups showed a dose-dependent but contrasting response in liver and muscle. In addition, a signi cant reduction in nal weight and SGR was observed in MOS fed at 1%. The relative abundance of OTUs belonging to Lactobacillus and Bacillus correlated with hepatic gene expression and nal weight of the sh.

Conclusions
Inulin and MOS appear to differentially affect host metabolism, mainly in the liver and muscle.
Differential abundance of Lactobacillus and Bacillus in the prebiotic-fed groups and their correlations with hepatic gene expression could indicate a prebiotic-microbiome-host axis, although this was not conclusively shown through the levels of SCFAs. In combination with a total plant-based diet, inulin could be a promising prebiotic for trout but need to be further investigated. These ndings could implicate in microbiome-mediated dietary optimization of rainbow trout.

Background
Rainbow trout (Oncorhynchus mykiss) is an important carnivorous species in aquaculture, with a total worldwide production of 861,000 metric tonnes in 2018 [1]. Traditionally, aquaculture diets have relied heavily on the protein and oil from marine sh. However, this practice is environmentally unsustainable as natural sh stocks are subject to severe shing pressure. In addition, the constant shortage and rising cost of shmeal and oil negatively affect the economic viability of the industry [2]. This necessitates a sustainable alternative diet devoid of sh meal and oil for the continued growth of the industry [2].
Signi cant research effort has been devoted to the development of alternative diets for rainbow trout by completely replacing marine protein and oil with vegetal protein and oil [3][4][5][6][7]. Although this approach has been successful in reducing the use of sh meal and oil in aquafeeds, culture-related traits, such as growth rate and gut health, have been negatively affected in some cases [8]. Furthermore, these alternative diets may also have an obesogenic effect on sh [9]. Therefore, it is necessary to improve and optimize the total plant-based diets in terms of use for growth and mitigation of adipogenic effects.
One of the strategies to improve the total plant-based diets would be the use of dietary supplements such as the prebiotics. Prebiotics are non-digestible mono-or polysaccharides used in diets of humans [10], livestock [11], and sh and shell sh [12][13][14]. Prebiotics are generally indigestible to the host and are speci cally fermented by anaerobic bacteria in the distal intestine, resulting in various secondary metabolites, mainly short chain fatty acids (SCFAs) [10,15]. SCFAs such as butyrate, acetate and propionate are a signi cant source of energy for intestinal epithelial cells [16] and have local and systemic immunomodulatory effects [17]. In addition, these SCFAs are transported to various metabolic organs such as the liver [18], skeletal muscle [19], and adipose tissue [20,21] and exert their effects on various aspects of metabolism that are bene cial to the host. These SCFAs may be useful in overcoming some of the detrimental effects of total plant-based diets on carnivorous species such as rainbow trout.
Prebiotics have been used in aquatic animal diets for decades. Several studies have demonstrated the bene cial effects of prebiotics on the growth and health of various aquatic species [12,14]. The mechanisms of utilization of prebiotics and the role of the resulting metabolites (mainly SCFAs) in host metabolism are not well understood in aquatic organisms, especially in carnivorous species such as rainbow trout. Studies concerning the effects of prebiotics on trout are few and far between [22][23][24][25][26].
Most of these studies have been conducted in the context of growth and disease resistance [24][25][26] using sh meal and oil based basal diet.
The e cacy of a prebiotic in aquatic animals may vary depending on the species, feeding habits, composition of the intestinal microbiome, type of basal diet ( sh meal or plant-based), and type of the prebiotic [13,14]. Knowledge of tissue-speci c responses to these molecules is also limited. Thus, it is important to test different prebiotics by measuring both local effects (responses in the intestine) and systemic effects (responses in different metabolic organs).
Therefore, the objective of this study is to understand the effects (both in the intestine and in other metabolic organs) of different types of prebiotics in rainbow trout fed a 100 % plant-based basal diet. For this purpose, rainbow trout were fed with three different prebiotics, namely fructo-oligosaccharides (FOS), mannan-oligosaccharides (MOS) and inulin at two different doses (1g − 100g and 2g − 100g feed) for 12 weeks and then intestinal microbiome, metabolites and structural changes were studied to understand the local effects of the prebiotics and their metabolites. Furthermore, the expression of metabolism-and in ammation-related genes in the intestine was analysed. In order to assess the systemic responses, we analysed the expression of genes involved in different metabolic pathways and putative SCFA receptors in metabolic tissues such as liver, skeletal muscle and adipose tissue.

Results
Whole body composition, zootechnical, hepatic and plasma parameters At the end of the feeding trial, whole-body protein, lipid, energy and ash contents did not differ signi cantly between groups (p>0.05; supplementary table 3). We measured different zootechnical parameters such as nal weight, speci c growth rate (SGR) and feed e ciency (FE). Plasma parameters such as glucose, triglycerides, free fatty acids, cholesterol and total amino acids were also measured. A one-way ANOVA revealed that the nal weight was signi cantly affected by inclusion of prebiotics in the diet (p<0.001). Post-hoc comparison using Tukey's HSD revealed that body weight was signi cantly lower in the MOS1 group compared to the control (Table 2). Similarly, SGR was also affected by prebiotic feeding (p<0.05; Table 2). The highest mean SGR was observed in the inulin2 group (2.21) and the lowest in the MOS1 group (1.95). Post-hoc tests revealed no signi cant difference in the mean values compared to the control group (Table 2). FE and protein e ciency ratio were not signi cantly affected by the prebiotics (p>0.05; Table 2), although the mean values of all prebiotic groups were relatively higher than those of control group. The hepatosomatic index (HIS) was signi cantly affected by the prebiotic diet (p<0.05; Table 2), although group-speci c differences were not detected compared to the control group.
The viscerosomatic index (VSI) and all plasma parameters were not affected by the inclusion of the prebiotics in the diet (p>0.05; Table 2). None of the liver and plasma parameters measured in the current study were affected by dietary inclusions of the prebiotics (p>0.05; Table 3).

Microbial diversity and composition
Sequence data were rare ed to 24000 sequences per sample. Alpha diversity was calculated using the observed OTUs ( Figure 1a) and the Shannon index ( Figure 1b). Both measures were not signi cantly different between the control and prebiotic groups (ANOVA; p>0.05). Beta diversity was calculated using the Bray-Curtis dissimilarity index and visualised using NMDS ordination (Figure 1c). Homogeneity of group dispersions was checked using the PERMDISP (p>0.05). Groups were compared using a pairwise PERMANOVA, which showed no signi cant difference between feeding groups (p>0.05).
We identi ed 1790 OTUs belonging to 16 phyla. Among all OTUs, the top 20 most abundant OTUs occupied almost 100% of the reads in all feeding groups (Figure 2a). These 20 most abundant OTUs belonged to only 4 phyla, namely Fusobacteria, Firmicutes, Proteobacteria and Tenericutes. These included the genera Aeromonas, Cetobacterium, Moraxella, Mycoplasma, Pseudomonas, Ralstonia, Streptococcus and Weissella (Figure 2a). Two OTUs (OTU1 and OTU2) belonging to the genus Mycoplasma were the most abundant, with more than 96% of reads assigned to these OTUs in each group (Figure 2a). Between these two OTUs, OTU1 was signi cantly higher (p<0.001) in each group except inulin2 (Figure 2b).

Firmicutes in the intestine of rainbow trout
Because OTUs belonging to Mycoplasma overshadowed the abundance of other OTUs, we additionally sequenced the V3-V4 region of the 16s rRNA of members of the phylum Firmicutes using a phylumspeci c primer pair. The phylum Firmicutes and Bacteroidetes are known to harbour several species of polysaccharide-utilising bacteria in intestinal microbiome of animals [27]. Alpha (p>0.05) and beta (PERMANOVA; p>0.05) diversity did not differ signi cantly between dietary groups (Figure 2c). Although these primers were unable to avoid ampli cation of Mycoplasma and some members of the phylum Proteobacteria, we were able to detect several OTUs belonging to the phylum Firmicutes. More than 95% of the sequencing reads were assigned to the 20 most abundant OTUs (Figure 2d). These OTUs belong to the genera Bacillus (16.6%), Gamella (0.1%), Lactobacillus (10.5%), Lactococcus (0.2%), Mycoplasma (31.5%), Staphylococcus (0.7%), Streptococcus (5.6%) and Weissella (28.3%).
We performed supervised partial least square discriminant analysis (PLS-DA) to understand the dissimilarity of the diet groups in terms of their microbial composition. Only the OTUs, which were present at >0.1% of the whole dataset, were retained for analysis. Although the clusters of the different dietary groups were not clearly separated by PLS-DA, the group means of ordination were clearly separated ( Figure 3a). The regression coe cients of the discriminatory OTUs on component 1 of PLS-DA are shown ( Figure 3b). Some of the most discriminatory OTUs belonged to Bacillus, Lactobacillus and Weissella. The correlation between these discriminatory OTUs and hepatic gene expression was evaluated using regularised canonical correlation analysis (rCCA). There was a positive correlation between the 3 OTUs belonging to the genus Bacillus (OTU3, OTU8 and OTU56)and genes involved in gluconeogenesis (g6pcb1b, g6pca, pck1 and fbp1b1), glycolysis (pfkla and pfklb), energy metabolism (sdhb) and fatty acid beta-oxidation (cpt1b and hoad). On the other hand, two OTUs (OTU7 and OTU11) belonging to the genus Lactobacillus and one OTU each belonging to Weissella (OTU2) and Alphaproteobacteria (OTU5) were found to be negatively correlated with the above metabolic pathways ( Figure 3c). Moreover, there was a signi cant effect of diet on the relative abundance of all these OTUs except OTU5 and the OTU2 (p<0.05). Two OTUs belonging to the genus Bacillus (OTU8 and OTU56) were signi cantly higher (p<0.05) in the MOS1 group than in the control (Figure 3d). In contrast, the relative abundance of OTU7 (Lactobacillus) was signi cantly lower in the MOS1 group ( Figure 3d). It was also found that Lactobacillus (OTU11) and Weissella (OTU2) were positively correlated with nal sh weight, while Bacillus (OTU3) and one OTU each belonging to Streptococcus and Lactococcus were negatively correlated with nal weight (supplementary gure 1).

Short-chain fatty acids
The quantities of SCFAs did not differ signi cantly between the different groups (p>0.05), although the MOS2 group had a higher quantity of all SCFA measured ( Intestinal morphology In the current study, FOS and MOS groups did not show a drastic change in microvillar structure compared to the control. On the other hand, the two inulin groups showed signi cantly higher microvillar length compared to the control ( Figure 5, supplementary gure 2). The attachment of the Mycoplasmalike organisms to the intestinal epithelial cell surface and the potential damage to the villous structure are shown in Figure 5h, Figure 5i, respectively.

Hepatic gene expression
We quanti ed the expression of several genes involved in amino acid metabolism, energy metabolism, fatty acid oxidation, fatty acid transformation, gluconeogenesis, glucose transport, glycolysis, lipogenesis and putative receptors for SCFAs in the liver.
We also compared global hepatic gene expression using PERMANOVA between the control and the inulin and MOS groups. This showed signi cantly different clustering in the inulin1 group compared to the control (p=0.01; Figure 6b). A similar comparison between the control and the MOS1 and MOS2 groups was not signi cant (p>0.05), although the global expression pattern in the MOS1 group was similar to that of the control group, it was divergent from the MOS2 group (p=0.021; Figure 6c).

Gene expression in muscle, intestine and adipose tissue
The expression of various genes in muscle was largely unaffected, with the exception of gdh2 (amino acid catabolism), which was signi cantly lower in the inulin2 group (p<0.05). Other genes affected by prebiotic feeding included atp5a and qcr2 (energy metabolism) and pfkmba, which is involved in glycolysis (ANOVA; p<0.05; Figure 7a). The global expression pattern in the MOS1 group was similar to the control, but signi cantly different from the MOS2 group (PERMANOVA; p=0.021; Figure 7b).

Discussion
Prebiotics are widely used in aquaculture. Given the potential bene cial effects of prebiotics on growth, health and the intestinal microbiome [12,14], there is a need to demonstrate their mechanism of action in aquatic organisms, particularly in carnivorous species such as rainbow trout. In mammals, speci c members of the phylum Bacteroidetes, Firmicutes and Actinobacteria have been shown to produce SCFAs by degrading the complex polysaccharides [27]. These SCFAs have been shown to affect host functions both in the intestine and systemically in liver, muscle and adipose tissue [15]. We adopted a similar approach in the current study, where we analysed the effect of different prebiotics both at the local level by examining the intestinal gene expression, gut morphology and metabolic output, and at the systemic level in terms of plasma metabolites and gene expressions in liver, muscle and adipose tissue.
The effect of prebiotics, possibly mediated via the intestinal microbiome on the intestinal health, has been noted in many teleost species [14]. In the present study, although there was an effect of prebiotics on a few genes in the energy metabolism pathway, there were no drastic changes in overall gene expression in the intestine. Diet-speci c changes in SCFA content in the digesta were also not signi cant. Although, in the MOS2 group all the SCFAs seemed to be higher in general, it's hard to conclude about their functional importance due to high interindividual variation (speci cally in the MOS2) and the small sample number (n = 3). On the other hand, the differences in abundance of OTUs belonging to Bacillus and Lactobacillus between dietary groups may be indicative of the microbial responses to prebiotics and interactions between microbial metabolites other than SCFAs and intestinal cells. Although species belonging to Bacillus and Lactobacillus are known to be widely used as probiotics in various teleosts [28], an in-depth analysis of their metabolic output at strain level and interaction with intestinal cells needs to be performed. Interestingly, the abundance of Bacillus and Lactobacillus was signi cantly altered only in the MOS1 group, which could be an indication of the type and dose speci city of the prebiotics, since these changes were not observed in the other types of prebiotics and the MOS fed at 2%. The dosedependent differences in the effects of mannanoligosaccharides depending on dietary habits and life stages have been previously reported in teleosts [29]. The different host responses based on the variations in the molecular structures of different prebiotics and polysaccharide utilising enzyme repertoire (in the microbiome) are also plausible [30].
It has been shown that prebiotic feeding affects microvilli length and density in different teleosts [31,32]. We observed a signi cant effect of inulin on microvilli length. These structural changes are known to be mediated by the various microbial metabolites [33], although such relationships between microbial metabolites and intestinal morphology have not yet been demonstrated in teleosts. Damage to the tight junctions between cells has been observed in Arctic charr (Salvelinus alpinus) fed inulin at a dose of 15% [34] and in gilthead sea bream (Sparus aurata) [31]. Normally, this damage occurs due to in ammatory responses [35], and such responses were not evident in the present study, as most in ammatory cytokines were either not expressed or the expression levels in the prebiotic groups were similar to the control.
In contrast to the intestine, at the systemic level, hepatic gene expression was signi cantly affected by inulin1 and a similar pro le was observed in the inulin2 group. It is noteworthy that the effect of inulin was only observed in the liver, showing higher expression of several genes involved in the different metabolic pathways. It is interesting because these responses were observed with the concomitant higher expression of the potential fatty acid receptors (ffar31 and ffar32), indicating a possible interaction between the liver and SCFA (or other similar microbial metabolites). Inulin has been implicated in glucose and lipid metabolism in rats [36] and mice [37], respectively. A recent study on tilapia showed that inulin resolved the metabolic syndrome induced by high carbohydrate [38]. The involvement of inulin in cholesterol metabolism by enhancing the activity of bile salt hydrolase (BSH), is well known. This enhancement leads to an increase in the biosynthesis of bile salts from cholesterol in the liver, resulting in a decrease in serum cholesterol levels [39]. In the current study, we did not observe any drastic change in serum cholesterol level.
On the other hand, it is interesting to note that these inulin-speci c responses were not observed in intestine or muscle warranting further investigation of the metabolic output from inulin degradation and the mode of entry into the host, as well as the tissue speci city of the metabolites. Moreover, these hepatic gene expression changes observed in the inulin groups were not accompanied by the concomitant changes in the intestinal microbiome in the same group. Some previous studies have also noted the absence of changes in the intestinal bacterial composition in response to inulin in juvenile beluga (Huso huso) [40], red drum (Sciaenops ocellatus) [41] and hybrid striped bass (Morone chrysops × Morone saxatilis) [42]. This highlights the importance of application of meta-omic approaches to both host and the microbiome to understand the functional changes and molecular interactions that can occur without a drastic change in the composition.
Furthermore, the two MOS groups showed an opposite trend in gene expression with a signi cant difference in the overall expression pro le between 1% and 2%. A similar contrast between MOS1 and MOS2 was also observed in muscle, but opposite to that of liver. The SCFA pro le also clustered MOS1 and MOS2 separately. Overall, this may indicate a tissue-speci c effect of these microbial metabolites between MOS1 and MOS2. The target metabolic organs of the microbial metabolites could vary [19]. In humans, the most abundant SCFA (butyrate) are generally thought to be utilized by colonocytes and the remaining SCFAs (propionate and acetate) are subject to hepatic metabolism. The class of metabolites affecting muscle metabolism could be complex, as both intestinal-derived metabolites and metabolites resulting from assimilation in the liver could interact with muscle [19]. While inulin had an effect on gene expression in the liver, absorption and transport of SCFA could not be con rmed in this study. The lack of differential expression of putative SCFA receptors (ffar) in the intestine could indicate a direct diffusion into cells or active uptake by previously unknown receptors. Apart from a few studies in zebra sh showing interaction of microbial proteins with host beta cells [43] and neutrophils [44,45], receptormediated uptake of microbial proteins or metabolites has not been extensively demonstrated in teleosts.
To understand the role of the microbiome in utilisation of prebiotics, we analysed the intestinal bacterial composition. Interestingly, there was a high abundance (> 95%) of Mycoplasma in most of the samples.
There are more than 100 species under the genus Mycoplasma associated with different organs in various animals and include both obligate symbionts and pathogens [46,47]. High abundance and prevalence of Mycoplasma has already been reported in salmonids [48][49][50]. A recent metagenomederived characterization of intestinal Mycoplasmas from three salmonid species clearly indicates a strong coevolutionary mechanism between the host and Mycoplasma strains [51]. Moreover, the strains found in salmon and trout are quite closely related and clusters with a previously reported intracellular human respiratory and urogenital pathogen Mycoplasma penetrans [51]. Indeed, uorescent in situ hybridization suggested that salmonid strains of Mycoplasma are intracellular [52]. The absence of pathogenicity genes in the salmonid strains [51] suggests a mutually bene cial relationship, as observed in the hadal sail sh, where the Mycoplasma is bene cial to the host by providing biotin in a nutrient-poor environment [53] and aiding in host defence [53,54]. In the present study, we observed 2 OTUs belonging to the genus Mycoplasma and they were differentially abundant in all feeding groups (except the inulin2 group). It needs to be con rmed whether these 2 OTUs actually belong to 2 different species of Mycoplasma and whether they have different potential to utilise different prebiotic molecules. Furthermore, we observed morphological structures resembling Mycoplasma on the microvilli of intestinal epithelial cells with a ask shape and a tip organelle, a typical feature of an intracellular Mycoplasma [55]. Similar structures were previously observed in another member of the salmonid family, Arctic charr [34]. The adhesion tendency of intracellular Mycoplasma to ciliated cells compared to secretory cells has been shown previously [55]. The damage to the ciliary structure and vacuolization inside the cells observed in the present study has been previously described in the case of infection of respiratory epithelial cells with Mycoplasma pneumoniae [56]. The consequences of such morphological changes on intestinal cell function and the fate of these Mycoplasma after entering the host cells need to be further investigated, either in an in vitro or in vivo setup.
Due to the overwhelming presence of Mycoplasma (> 99%) in most samples and the resulting extremely low read depth of other OTUs, we used Firmicutes-speci c primers to investigate the abundance of OTUs under this phylum. This group is known to harbour several members involved in polysaccharide utilisation [27]. These primers were not able to fully discriminate OTUs belonging to Mycoplasma as they are known to be phylogenetically similar [46].
OTUs belonging to the Bacillus group were positively correlated with the expression of genes in glycolysis, gluconeogenesis, energy metabolism and fatty acid beta oxidation pathways. In contrast, OTUs belonging to the genus Lactobacillus and Weissella were negatively correlated with the expression of genes in the same metabolic pathways. The most striking correlation was with genes involved in gluconeogenesis. The association between glucose metabolism and microbiota has not been clearly established in teleosts. A recent study in zebra sh has shown the effect of a microbial (Cetobacterium somerae) metabolite (acetate) on glucose homeostasis in zebra sh via the regulation of insulin secretion [57]. In addition, a microbial protein (BefA) has been shown to be involved in the differentiation of pancreatic B cells at early stages of zebra sh development [43]. In mammals, certain Lactobacillus species are known to inhibit gluconeogenesis by reducing the expression of phosphoenolpyruvate carboxykinase [58]. This effect is thought to be caused by different SCFAs (butyrate and propionate) via activation of the hepatic AMPK pathway [59]. On the other hand, the positive correlation between Lactobacillus and nal weight could be due to the higher visceral fat, as there was also a positive correlation between Lactobacillus and visceral fat index. This could be an indication of the reduced entry of lipolysis related metabolites such as free fatty acid and glycerol into the gluconeogenic pathway [60]. Bacillus and its positive correlation with the gluconeogenic pathway and negative correlation with nal sh weight is intriguing and needs further investigation. It should be mentioned that several members of Lactobacillus and Bacillus are used as probiotics in teleost aquaculture [28], with a wide range of bene cial effects on the host. Therefore, it is important to understand the genomes of OTUs belonging to Bacillus and Lactobacillus in the current study to describe the basis of the observed correlations with host metabolism. The signi cantly lower abundance of Lactobacillus and concomitant higher abundance of Bacillus in MOS1, and their opposite correlation with the expression of genes in hepatic metabolism collectively suggest a mechanism involving these two members of the Firmicutes and possibly their metabolites that alters hepatic metabolism and ultimately the sh growth in MOS1.

Conclusion
The present study showed that the polysaccharide inulin and mannan-oligosaccharides may have an effect on the expression of genes involved in various metabolic pathways in the metabolic organs, liver and muscle. The local effect of inulin in the intestine was limited in terms of gene expression, microbial composition and SCFAs, although the effect on microvillar structure was evident. On the other hand, MOS showed a dose-dependent difference in the abundance of bacterial groups (Lactobacillus and Bacillus).
Systemically, the effect of inulin and MOS was seen in the liver, whereas in muscle the response was dose-dependently restricted to MOS. The high abundance of Mycoplasma sp. in the trout intestine is an indication that it is an obligate symbiont of trout, and this host-Mycoplasma association requires further investigation to understand its role in trout physiology. The differential abundance of Lactobacillus and Bacillus, their correlations with metabolic gene expression and nal weight, taken together, could be indicative of the prebiotic-microbiome-host axis. Nevertheless, studies on the recognition and uptake mechanisms of SCFAs in the intestine and their transport and assimilation in different metabolic organs need further investigation. The application of meta-omic approaches in future studies could provide insights into such interactions and open new opportunities for microbiome-mediated nutritional optimization in aquaculture.

Diet and experimental setup
A total plant-based diet either with (experimental) or without (control) different prebiotics was prepared. The formulation and proximate composition of the diet is given in table 1 and supplementary table 1, respectively. The diets were prepared to meet the nutritional requirements of the rainbow trout [61] and manufactured and tested at the INRAE experimental sh farm in Landes, France. The diets were isoproteic (~50% crude protein), isolipidic (~20% crude fat) and isoenergetic (~24 KJ -g dry matter). The experimental diet was supplemented with either fructo-oligosaccharides (FOS), inulin or mannanoligosaccharides (MOS), each at 2 different doses, 1g -100g and 2g -100g feed. These doses were selected based on previous studies on various carnivorous teleosts [13]. The basal diet contained only the mixture of plant ingredients and vegetable oils supplemented by free amino acids. Thirty-six juvenile rainbow trout (~0.13g -L ) in equal numbers of males and females were randomly distributed in each of the 130 L berglass tanks. There were seven diet groups in total and 3 tanks were assigned to each of the diet groups. The sh were kept throughout the experimental period under standard rearing conditions with water oxygen levels 9 mg -L , temperature 17 °C and pH 7.5, water ow rate 0.3L -s , with daylight of 12h and 12h of darkness. The sh were manually fed twice a day (with an interval of 8h) either with the control diet or one of the experimental diets for 12 weeks. The tanks were checked for mortalities (if any) daily.
The sh were weighed in bulk every three weeks to evaluate the growth parameters.

Sampling
After the feeding experiment, sh were randomly sampled from each tank 24 h after the last feeding. The randomly selected sh were rst anaesthetised with benzocaine (30 mg-L) before being euthanized with a benzocaine overdose of 100 mg-L, followed by blood collection for plasma isolation. After the blood collection, sh were aseptically dissected and the digestive tract was separated. Liver, adipose and muscle were dissected and immediately frozen with liquid nitrogen and stored at -80 °C for longterm storage. The posterior intestine was separated from the rest of the digestive tract and was cut open longitudinally using sterile instruments. The posterior intestine was chosen because most polysaccharide-utilizing bacteria are found in the distal intestine of other animals. The contents and mucus were sampled together to analyse the total microbial pro le of the intestine and a part of the tissue was also collected for the RNA extraction. Intestinal contents and distal intestinal tissue samples for SCFA and electron microscopy, respectively, were also ash frozen using liquid nitrogen and then stored at -80 °C.
Diet and whole-body proximate composition

Measurement of the plasma biochemical parameters
Blood samples were collected using the heparinised syringe and tubes. Samples were centrifuged at 3000 g for 10 min to isolate plasma and stored at -20°C until use. Commercial kits designed to be used along with the microplate reader were used to measure the plasma glucose (Glucose RTU, bioMérieux, Marcy l'Etoile, France), triglycerides (PAP 150, bioMérieux), cholesterol (Cholesterol RTU, bioMérieux) and free fatty acid (NEFA C kit, Wako Chemicals, Neuss, Germany). Total free amino acid was quanti ed according to the method of Moore [62], with glycine as standard.

Hepatic fat measurement
Hepatic fat was measured according to the protocol described previously [63]. The lyophilised liver samples (100mg) were mixed and homogenised with Folch reagent to extract the lipids. The lipids were extracted three times using this protocol and sodium chloride was added to the recovered supernatant after centrifugation for phase separation. The bottom layer containing the lipids was harvested and transferred to a glass tube, and the solvents were evaporated using nitrogen gas at 40°C. After drying, the lipids were weighed.

Hepatic glycogen measurement
Liver glycogen was measured according to the protocol described by Good et al [64]. Brie y, lyophilised liver (100 mg) was homogenised with 1M HCL (VWR, France). After this step, the samples were divided into 2 aliquots and one part was neutralised with 5M KOH (VWR), centrifuged and the supernatant was used to measure free glucose with a commercial kit (Glucose RTU, bioMérieux). The second aliquot was boiled with 5M KOH (VWR) for 2.5h at 100°C before neutralization. After centrifugation, total glucose (free glucose + glucose released by hydrolysis of glycogen) was measured in the supernatant. The glycogen content was calculated by subtracting the free glucose from the total glucose. products were pooled for each sample and run on a gel to con rm ampli cation of a ~550 bp product. A positive and a negative control sample were also included in the run. After con rmation of PCR ampli cation, the PCR products were shipped to La Plateforme Génome Transcriptome de Bordeaux (PGTB, Bordeaux, France) for the following steps. Index PCR was performed to add the unique dual indexes to the sequences in a sample-speci c manner. For this purpose, the Nextera XT index kit was used according to the manufacturer's protocol (Illumina, France). Thermocycling conditions were the same as in step 1, except that PCR was performed for only 8 cycles. After PCR clean-up, libraries were quanti ed using the KAPA library quanti cation kit for Illumina platforms (Roche, France) according to the manufacturer's instructions. Libraries were pooled at an equimolar concentration (4nM) and sequenced on a MiSeq platform using a 250 bp Paired End Sequencing Kit v2 (Illumina, France).

Data analysis
The initial data analysis was performed using the UPARSE pipeline [67]. First, the forward and reverse reads of each sample were fused. Also, the primer binding sites were removed from the assembled sequences. Then, the sequences were quality ltered using the strategy of ltering by the maximum expected error rate [68]. Sequences with an error rate of > 0.01 per sequence were removed from the dataset. After quality ltering, sequences from different samples were labelled with the sample name and combined in to a single le. This collection of sequences was dereplicated and the sequences that were observed only once in the dataset (singletons) were removed, as these singletons were most likely sequencing artefacts [69]. The sequences were further clustered into OTU (Operational Taxonomic Unit) based on a sequence similarity of 97%. The raw reads were then mapped back to the OTUs to obtain the abundance of each OTU in different samples. The taxonomy was assigned to the OTUs using SINTAX [70]. Taxonomies with bootstrap con dence value less than 0.8 were removed from the dataset. Furthermore, OTUs belonging to the phylum Chloro exi were removed from the analysis as it is likely that these OTUs originated from the plant components of the diet. A phylogenetic tree in Newick format was constructed based on the OTU sequences . The OTU table, taxonomy table and phylogenetic tree were imported into the phyloseq package [71] in R (version 3.6.3) along with the sample metadata. Alpha diversity measures, observed OTUs and Shannon index [72] were calculated. Beta diversity was calculated using the Bray-Curtis distance with the phyloseq package. All of these data analysis steps were also applied to the Firmicutes amplicon dataset. Gene expression analysis RNA from liver, intestine and muscle was extracted using TRIzol reagent method (Invitrogen, Carlsbad, CA) according to the manufacturer's protocol. RNA from adipose tissue was extracted using the RNeasy kit for fatty tissues (Qiagen, France). One µg RNA was converted to cDNA using the enzyme superscript III reverse transcriptase and random hexamers (Invitrogen, France) in the case of liver and muscle, while the quantinova reverse transcription kit (Qiagen, France) was used for adipose tissue and intestine. After the reverse transcription, cDNA was diluted 75-fold(liver) and 20-fold for muscle, adipose tissue and intestine before its use in RT-qPCR.
Reactions were performed on a 384 well plate. The total volume of reaction was 6 µL, consisting of 3 µL stable reference genes across samples. These calculations were performed using geNorm [75], which is implemented in the R [76] package SLqPCR [77]. The reference genes eef1a and rna18s were used to calculate the normalization factor in liver, while a combination of actb and eef1a was used in muscle, actb and eef1a in adipose tissue, and actb and rna18s in intestine. Normalized relative quantities were analysed for statistical signi cance.

Statistical analysis
Zootechnical parameters, including feed e ciency (FE) and speci c growth rate (SGR), are calculated per tank (n=3). HSI, VSI and nal growth data are collected for each sampled individual (n=9). Samples from the same nine sh were also used for gene expression analysis and plasma parameters. For microbiome analysis, three sh were sampled in addition to the 9 already mentioned. Three additional sh per group were sampled for SCFA analysis and for electron microscopy.

Competing interests
The authors declare no con ict of interest.  Bacterial alpha diversity represented in terms of observed OTUs (a) and Shannon index (b). Alpha diversity between diet groups was compared using one-way ANOVA and p<0.05 was considered signi cant. Beta diversity is presented as Bray Curtis distance between samples and depicted using NMDS (c). Beta diversity was compared using pairwise PERMANOVA and p<0.05 was considered signi cant, n=12.