High carbohydrate to protein ratio promotes changes in intestinal microbiota and host metabolism in rainbow trout (Oncorhynchus mykiss) fed plant-based diet

To ensure the sustainability of aquaculture, it is necessary to change the “menu” of carnivorous fish such as rainbow trout from a fish-based diet to one with plant-based ingredients. However, there is a major problem with the growth performance decrease of fish fed with a 100% plant-based diet due to the reduction in feed intake and feed efficiency. For the first time, we incorporated high levels of digestible carbohydrates (high-starch diet) in a 100% plant-based diet during a 12-week feeding trial in order to improve protein utilization for growth (protein sparing effect) and reduce nitrogen waste. We measured the changes in the intestinal microbiota, Short-Chain Fatty Acid (SCFA) levels and metabolic responses in liver. Dietary carbohydrates had a strong effect on alpha and beta diversity and abundance of 12 genera, including Ralstonia and Bacillus in digesta associated microbiota whereas mucosa associated microbiota was less affected. The change in microbial diversity might be linked to the change observed in SCFA production. High levels of Mycoplasma were observed in the intestinal mucosa. Overall, hepatic gene expression was significantly altered by the CHO/protein ratio. Up-regulation of genes involved in glucose metabolism (gcka, gckb, g6pcb2a), down-regulation of genes involved in lipid metabolism (hadh, acox3, srebp2a, and cyp51a) were associated with higher enzymatic activities (such as glucokinase or pyruvate kinase) and higher glycogen levels in the liver, suggesting adequate adaptation to diet. Interestingly, strong correlations were observed between abundances of certain bacterial OTUs and gene expression in the liver. The inclusion of digestible carbohydrates in combination with a 100% plant-based diet, could be a promising way to improve and reduce the use of plant proteins in rainbow trout. In addition, the relationship between intestinal microbiota and host metabolism needs further investigation to better understand fish nutrition.


Abstract 23
To ensure the sustainability of aquaculture, it is necessary to change the "menu" of carnivorous fish 24 such as rainbow trout from a fish-based diet to one with plant-based ingredients. However, there is a 25 major problem with the growth performance decrease of fish fed with a 100% plant-based diet due to 26 the reduction in feed intake and feed efficiency. For the first time, we incorporated high levels of 27 digestible carbohydrates (high-starch diet) in a 100% plant-based diet during a 12-week feeding trial 28 in order to improve protein utilization for growth (protein sparing effect) and reduce nitrogen waste. 29 We measured the changes in the intestinal microbiota, Short-Chain Fatty Acid (SCFA) levels and 30 metabolic responses in liver. Dietary carbohydrates had a strong effect on alpha and beta diversity and 31 abundance of 12 genera, including Ralstonia and Bacillus in digesta associated microbiota whereas 32 mucosa associated microbiota was less affected. The change in microbial diversity might be linked to 33 the change observed in SCFA production. High levels of Mycoplasma were observed in the intestinal 34 mucosa. Overall, hepatic gene expression was significantly altered by the CHO/protein ratio. Up-35 regulation of genes involved in glucose metabolism (gcka, gckb, g6pcb2a), down-regulation of genes 36 involved in lipid metabolism (hadh, acox3, srebp2a, and cyp51a) were associated with higher 37

46
Sustainable and efficient seafood production is urgently needed to meet the growth in world's 47 7 Gribheimer, Germany). Starch content was determined using the Megazyme© (Bray, Ireland) total 146 starch assay procedure. 147 148 Blood was sampled for plasma collection from the caudal vein and then were directly centrifuged at 149 12,000g at 4°C for 5 min, and stored in heparinized tubes at -20°C until use. Commercial kits were used 150 to determine the level of several plasma metabolites: glucose (Sobioda, Montbonnot-Saint-Martin, 151

Measurement of the plasma biochemical parameters
France), lactate (kit Randox, Crumlin, United Kingdom), triglycerides (Sobioda), and cholesterol 152 (Sobioda). These kits were adapted to 96-well plates formats according to the manufacturer's 153 instructions. 154 2.6. Hepatic metabolites measurement 155 Liver glycogen was measured using a protocol previously described (Good et al., 1933). For glycogen 156 determination, 250mg of liver were homogenized with 1M HCL and divided in two aliquots. One part 157 was neutralized with 5M KOH, centrifuged and the supernatant was used to measure the free glucose 158 with a commercial kit (Sobioda). The second group of aliquots was hydrolyzed during a boiling step 159 (2h30 at 100°C) and then neutralized with 5M KOH. After centrifugation, the total glucose (free glucose 160 and glucose released by hydrolysis of glycogen) was measured in the supernatant. For hepatic 161 cholesterol determination 100mg of liver was homogenized in 5% Igepal (sigma Aldrich, Saint- Louis,162 MO, USA). Samples are placed twice at 90°C during 5min and then centrifuged 2 min at 5,000g. 163 Supernatants were recovered and the cholesterol concentration measured in 96-well plates thanks to 164 a commercial kit (Sobioda). 165 same than in step 1, except that PCR was performed for only 8 cycles. After PCR cleaning, libraries were 193 quantified using the KAPA library quantification kit for Illumina platforms (Roche). Libraries were 194 pooled at an equimolar concentration (4nM) and sequenced on a MiSeq platform using a 250 bp Paired 195 End Sequencing Kit v2 (Illumina). 196 2.7.3. Data analysis 197 The initial data analysis was performed using the FROGS pipeline according to standard procedures 198 (Escudié et al., 2017). First, the forward and reverse reads of each sample are merged. Only amplicons 199 with size between 380 bp and 500 bp corresponding to the size of the V3-V4 region of the 16S rRNA 200 gene, without ambiguous bases and with the two primers were kept. The adapter sequences were 201 removed and the sequences with not expected lengths or ambiguous bases (N) were deleted. After 202 this step 7800687 sequences have been kept, which represents 79.42 % of initial input sequences. The 203 clustering swarm algorithm was used to group together amplicons with a maximum of one nucleotide 204 difference between two amplicons (Mahé et al., 2014), 476371 clusters were created. PCR-generated 205 chimeras, typically created when an aborted amplicon acts as a primer, are removed. Clusters present 206 in less than 4 samples, and having a minimum abundance of 0.005% are removed. After this step, 261 207 clusters have remained with 6716271 sequences. The PhiX database was used to removed 208 contaminants such as chloroplasmic or mitochondrial sequences (Mukherjee et al., 2015). Then, 209 taxonomic affiliations are carried out for each OTUs (Operational Taxonomic Unit), using the silva138.1 210 pintail100 16S reference database ("https://www.arb-silva.de/documentation/release-138/," n.d.  (Bradford, 1976), using a protein assay kit (Bio Rad, 278 München, Germany) with bovine serum albumin as a standard. 279 2.11. Statistical analysis 280 Zootechnical parameters, including initial and final body weight, specific growth rate, feed intake, feed 281 efficiency, and protein retention efficiency were calculated per tank (n=3). The hepatosomatic index 282 (liver weight*100/fish weight) were obtained during the final sampling of the trial experimentation 283 (n=12). All data were presented as mean ± SEM. All statistical analysis were performed with R software 284 (version 4.0.3). Data were tested for normally distribution using the Shapiro-Wilk test and 285 homogeneity of variance using the Bartlett's test. All Data were analyzed using the one-way ANOVA. 286 These statistical tests were followed by Tukey's HSD as post hos test when the normal distribution is 287 respected, otherwise a Wilcoxon non-parametric test is carried out. Results with a P-value ˂ 0.05 were 288 considered statistically significant. In the figures: *, P < 0.05; **, P < 0.01; ***, P < 0.001. The microbial 289 composition was analyzed using the phyloseq package. For alpha and beta diversity, data samples were 290 rarified. Beta-diversity was analyzed with the Bray-Curtis distance using permutational multivariate 291 analysis of variance (PERMANOVA). The mixOmics package was used to perform a Partial Least Square 292 Discriminant Analysis (PLS-DA) to determine the most discriminant OTUs. The rCCA (regularized 293 canonical correlation analysis) function of the same package was used to understand the correlations 294 between the bacterial OTUs and different host parameters. The core microbiota visualization was 295 made using the phyloseq package on R and the Venn diagram using a website. 296  (Table 2). Plasma parameters such as triglycerides, glucose, lactate and cholesterol were 301 also measured. Statistical analysis did not reveal any change between the two experimental groups 302 (Table 3). Regarding feed utilization, protein efficiency ratio increased (p = 2.41e-4) in trout fed with 303 high starch diet (HS), whereas no significant differences in daily feed intake and feed efficiency were 304 measured (table 2). Whole-body composition (dry matter, gross energy, crude protein and lipid) of fish 305 for the two experimental groups were evaluated (Supplementary table 3). The two diets did not lead 306 to any significant differences of whole-body composition of fish. Moreover, plasma metabolites 307 (glucose, lactate, triglycerides, cholesterol) were also not affected by the diets (Table 3). Finally, 308 several significant changes were observed for hepatic parameters. The use of the high-starch diet 309 increased significantly the average liver weight (hepatosomatic index). The level of glycogen stored in 310 the liver was also significantly higher (p = 1.894e-8) in the group of fish fed with the high-starch diet. 311

Results
Inversely, a decrease of the hepatic cholesterol was observed when trout were fed with the high-starch 312 diet. 313

Microbiota diversities and composition 315
Sequence data were rarefied to 17000 sequences per sample. Alpha diversity measures including the 316 indexes Observed OTUs, Chao1, Shannon, Simpson, and InvSimpson were calculated in the digesta 317 ( Figure 2A) and in mucosa ( Figure 2B). A significant decrease (ANOVA, p < 0.05) of all the alpha 318 diversities indices only in the digesta associated microbiota was observed for the high-starch group. 319 Most of the alpha diversity indices did not differ between digesta and mucosa samples except for the 320 Shannon index which is significantly (p = 0.01) higher in digesta (data not shown). Beta diversities were 321  abundant phyla regardless the sample origin. Independently of the diet, we observed significant 334 difference between mucosa associated microbiota and digesta associated microbiota. In digesta, 335 Proteobacteria was the dominant phylum (76.21 ± 11.22 %) ( Figure 4B) which is not the case in the 336 mucosa associated microbiota where Firmicutes dominates (66.58 ± 14.10 %) ( Figure 4A). In both 337 mucosa and digesta associated microbiota, Proteobacteria were dominated by the genus Ralstonia 338 belonging to the Burkholderiaceae family, whereas the genus Mycoplasma (Mycoplasmaceae family) mucosa, no change in the relative abundance was observed at phylum, family, and genus levels ( Figure  341 4A, 4D, 4G), except for three genera significantly lower in high-starch i.e Bacillus, Peptoniphilus, and 342 Clostridium sensu stricto 1 (Table 4). Only in digesta associated microbiota, these three phyla were 343 significantly affected by the diet, resulting in an increase in the relative abundance of the 344 Proteobacteria (81.84 ± 5.25 % in HS and 70.58 ± 12.91 % in LS, p = 0.010) and the Actinobacteria

Gene expression in liver 384
We evaluated the expression of different genes involved in gluconeogenesis, glycolysis, lipogenesis 385 and fatty acid oxidation in the liver, the center of the intermediary metabolism. ANOVA revealed that 386 there was not a significant effect of CHO on the expression of glut2a and glut2b, two genes involved 387 in glucose transport (Table 5). The mRNA level of genes implicated in the first (gcka, gckb), the third 388 (pfkIa, pfkIb) and last (pk) glycolysis steps was measured. The expression of genes involved in the first 389 steps of glycolysis (gcka: p<0.001 ***; gckb: p<0.01 **) increased whereas the expression of the gene 390 implicated in the last-step (pk: p<0.001 ***) decreased in the high-starch group. In the same way, the pck1 (p<0.05 *) and fbp1b1 (p<0.05 *) mRNA gene expression involved in gluconeogenesis were 392 significantly lower in the high-starch group, while in this group the mRNA level of g6pcb2a (p<0.05 *) 393 was increased. High-starch diet did not affect the expression of genes involved in lipogenesis. By 394 contrast, the hadh (p<0.01 **) and acox3 (p<0.05 *) expression implicated in beta oxidation of lipids 395 decreased in trout fed with the high-starch diet. Regarding cholesterol biosynthesis gene, the 396 expression of the srebp2a, hmgcs, and cyp51a genes decreased significantly in high-starch diet group. 397

Hepatic enzymatic activities 398
To validate the gene expression at functional level, the specific activities of glucokinase, pyruvate 399 kinase, glucose-6-phosphatase, and fatty acid synthase were measured in the liver. The enzymes 400 involved in the first (glucokinase) and in the last (PK) steps of glycolysis were both significantly 401 increased in trout fed with high starch diet ( Figure 7A, B). Indeed, the average activity of glucokinase 402 was higher in the high-starch group (+ 0.023 mU/mg of protein, p<0.001 ***). Glucose-6-phosphate, 403 implicated in gluconeogenesis presented a significant (p<0.05 *) higher specific activity in high-starch 404 group, 0.36 ± 0.098 mU/mg of protein, than low-starch group, 0.26 ± 0.11 mU/mg of protein ( Figure  405 7C). Then, for lipogenesis the key enzyme, Fatty Acid Synthase (FAS), did not show significant 406 difference between the two groups ( Figure 7D). 407 3.6. Correlations between the OTUs and the hepatic gene expression, zootechnical, liver, plasma 408 parameters and enzymatic activities 409 The correlations between discriminatory bacterial genera and hepatic gene expression was firstly 410 evaluated using regularized canonical correlation analysis (rCCA) ( Figure 8A) and Fibrella were positively correlated with fbp1b1, and pck1 genes evolved in gluconeogenesis, as 413 well as pk involved in glycolysis. These genera were all negatively correlated with pck2 414 (gluconeogenesis) and g6pcb2a (glycolysis). Sphingomonas, Undibacterium, Acinetobacter, Veillonella, gluconeogenesis), and negatively with pk and pck1. Ralstonia was negatively correlated with all of 417 these gene expressions. Secondly, OTUs abundances were correlated with zootechnical, liver, and 418 plasmatic parameters as well as enzymatic activities ( Figure 8B). The first 9 OTUs i.e. Lactococcus, 419 other OTUs belonging to R. pickettii were significantly in lower proportion showing we need to go 504 further the species level by using shotgun metagenomics to fully understand the role of these bacteria. 505 Interestingly, in our study, Ralstonia pickettii was dominant in the digesta microbiota, regardless of the 506 diet, and has been found in several environment such as soil, rivers or lakes. In mammals, Ralstonia 507 pickettii has been linked to the development of obesity and type 2 diabetes (Udayappan et al., 2017). 508 In this, the increase of Ralstonia was not associated to any negative effect of fish physiology. OTUs Bacillus amyloliquefaciens, could alleviate the metabolic phenotypes caused by a high-carbohydrate 528 diet by enriching the acetate-producing bacteria in Nile tilapia intestines (Xu et al., 2022). Furthermore, 529 a study in zebrafish has revealed that Cetobacterium improves glucose homeostasis, mediated by a 530 potential effect of acetate (A. . In our study, a significant increase of the valerate 531 concentration was observed in the microbiota of trout fed with the high-starch diet whereas acetate, 532 butyrate and propionate showed the same trend. Regarding SCFA levels, their increase observed in the 533 high-starch group could at least partially explain differences in gene expression in the liver (Morrison 534 and Preston, 2016). In addition, our data suggested a high production of lactate probably linked to 535 lactate-producing bacteria, Weissella and Limosilactobacillus, in the high-starch group microbiota. 536

The high-starch diet did not affect growth performance and glucose homeostasis but have 537 expected effect on glucose and lipid metabolism in liver 538
Overall, the incorporation of 20% of dietary starch to a 100% plan-based diet has resulted in strong 539 change in the digesta associated microbiota of the midgut. Indeed, we observed a decrease of without affecting the growth performance. Interestingly, while rainbow trout use high levels of protein 542 for growth (Cleveland and Radler, 2019;Seiliez et al., 2008), decreasing the proportion of plant protein 543 in the high-starch diet did not affect the trout final weight and even increase the protein efficiency 544 ratio suggesting that increasing the CHO/protein ratio could prevent protein catabolism for energy 545 needs as shown previously in fish fed with marine resources (Kamalam et al., 2017). For the first time, 546 we showed that it is possible to incorporate high levels of digestible carbohydrates (20%) without any 547 negative effects on zootechnical parameters and whole-body composition even in fish fed with a 100% 548 plant-based diet. While trout are often described as poor users of glucose caused by a persistent post-549 prandial hyperglycemia when fed a diet containing more than 20% of carbohydrates (Polakof et al., 550 2012), this metabolic disorder has not been observed with the high-starch diet in our study. Other 551 studies have also shown low blood glucose levels in mature brood stock trout fed with high levels of 552 CHO in their diets (Callet et al., 2020). suggesting that CHO can be efficiently metabolized and/or stored 553 as glycogen in the liver at least at later stage of development. Indeed, in rainbow trout fed with the 554 high-starch diet, the glycogen level is higher in the liver resulting in a higher hepatosomatic index. 555 Regarding the metabolism of glucose through glycolysis, our results reveal that the hepatic glucokinase 556 mRNA gene expression (gcka and gckb) and their enzymatic activities were significantly higher in fish 557 with the high-starch diet. The increase of glucokinase activity and glycogen level in trout fed with 558 carbohydrates suggests that the rainbow trout can adapt at a metabolic level to the carbohydrate 559 intake in fish fed with a 100% plant-based diet, as previously observed in fish fed with marine resources 560 (Capilla et al., 2003;Pereira et al., 1995). We observed a significant lower expression of the pk gene 561 (coding for the pyruvate kinase enzyme) involved in the last step of the glycolysis which is different to 562 what is found in mammals (Yamada and Noguchi, 1999). However, this result is consistent in rainbow 563 trout where it has been shown that the expression of the pk gene was poorly controlled with high 564 Lactobacillus and Bacillus to fully understand the role of these bacteria. 604

605
The present work allows us to show clear differences between the digesta associated microbiota and 606 mucosa associated microbiota. Phyla from Proteobacteria and Firmicutes are dominant in both 607 contents and mucosa associated microbiota. CHO/protein ratio strongly modify the bacterial 608 community and diversity especially in digesta-associated microbiota, associated with differences in 609 concentration of SCFA. Nevertheless, we cannot discard that some of these effects can also be related   InvSimpson, in digesta (A) and mucosa (B) according to the experimental diversity. Alpha diversity 993 between diet groups was compared using one-way ANOVA and p<0.05 was considered significant. Beta 994 diversity is presented by a nMDS representation (Bray-Curtis distance, Weighted-Unifra analysis) in 995 digesta (C) and in mucosa (D). Beta diversity was compared using pairwise PERMANOVA and p<0.05 996 was considered significant and indicated with asterisk. n=12 997 998    Data are presented as mean ± SD (n=12 fish). statistical differences were calculated by one-ways 1082 ANOVA (P<0.05). NS not significant.