Pro- and synbiotic additives are associated with a reduction in relative abundance of Mycoplasma throughout the rainbow trout gut
During the experimental period the rainbow trout were fed control feed (CTRL), probiotic feed (PRO), and synbiotic feed (SYN), respectively (Table 1). Bacterial profiling of both the mid and distal intestinal content of 120 juvenile rainbow trout, using the V3-V4 region of the 16S rRNA gene, resulted in 382 amplicon sequence variants (ASVs). The five most abundant ASVs comprised 85.1% of the total number of microbial reads from rainbow trout in this trial, revealing a low intestinal microbiota diversity (mean effective ASV richness of 34.78 ± 15.8 Hill numbers). Diversity analysis, based on Hill numbers, revealed significantly higher diversity of the core microbiome in both the PRO and SYN groups compared to CTRL (Supplementary Fig. 1). Taxonomy assignment revealed that the five most abundant ASVs included genera of Mycoplasma, Pediococcus, Pseudomonas, Massilla, and endosymbiont8 (genus of Enterobacteriaceae). Bacterial profiling throughout the gut revealed significant changes in the microbial composition among different diet groups with fish from the CTRL group being dominated by Mycoplasma, compared to fish from the other two groups that were largely characterised by a higher relative abundance of Massilla and endosymbiont8, though Mycoplasma were still present with individual variation. (Fig. 1a). Further, the recovered Pediococcus ASV revealed an exact match with the administered probiotic strain of P. acidilactici MA18/5M.
Our analysis revealed a clear alteration of the microbiota as a result of feed type. A Principal Coordinates analysis (PCoA) revealed that 96.4% of the variance of the microbiota was explained by two principal components, and that the microbiota of CTRL clustered alone, whereas the microbiota of PRO and SYN clustered together (Fig. 1b). This pattern was repeated between both the mid and distal gut sections with no significant differences between the gut sections (Fig. 1b). Differential abundance analysis of the top 50 most abundant ASVs confirmed the significantly higher abundance of Mycoplasma in the CTRL group (Fig. 1c), indicating that feed additives may have a suppressing effect on the presence of Mycoplasma and Bifidobacterium (Fig. 1c). On the other hand, our data reveals an increase of the phylum Proteobacteria, including Pseudomonas, Enterobacteriaceae, Massilia. Also, an increase of the phylum Firmicutes, including Clostridiales, Weissella, Staphylococcus. The abundance of the probiotic Pediococcus ASV was significantly higher in the SYN group, compared to both the CTRL and the PRO groups, indicating that the usage of galacto oligosaccharides (GOS) as a supplemental prebiotic in the SYN group did increase the abundance of P. acidilactici MA18/5M (Fig. 1c).
Mycoplasma abundance is associated with microbial pathways of known relevance for salmonid metabolism
A random subset of individuals from each feeding group were selected to investigate inherent microbes in the rainbow trout intestinal content. Deep sequencing of each individual was required to get a decent coverage, since biomass of the microbes in intestinal samples was shown to be low from qPCR quantification of the V3-V4 region of the 16S rRNA gene (Supplementary Table S2.1). To cope with the high level of host DNA in the gut content, we generated more than 1.5 Tb of raw sequence data to obtain a metagenome from this low biomass microbiome. Raw reads were host filtered, assembled, binned, and a MAG database was curated, resulting in a metagenome of 5.01 Mb, consisting of no more than 5,574 genes from two MAGs and one bin (Fig. 2a). The metagenome for Candidatus Mycoplasma salmoninae mykiss (referred as Mycoplasma in this study) has previously been reported42, but here we present the whole intestinal metagenome data retrieved from six rainbow trout, including both the mid and distal gut sections. Mycoplasma had an identical match with our previously found Mycoplasma ASV from the 16S rRNA gene profiling. Further, a MAG of an unknown genus of Enterobacteriaceae corresponded to the presence of the endosymbiont8 ASV, which we hypothesise to be the corresponding MAG for the endosymbiont8 ASV. Our analysis also revealed a bin of an unknown Lactobacillus. Lastly, short read mapping of the metagenome revealed low levels of Pediococcus acidilactici MA18/5M genes present in rainbow trout from the PRO and SYN groups, indicating that the probiotic strain seems to be abundant at a low level in the intestinal content.
Our metagenomic analysis confirmed the bacterial composition found by 16S rRNA gene metabarcoding, where Mycoplasma was found to be highly dominant in CTRL and especially in the midgut, corresponding to 76.8–84.5 % of all microbial reads in the midgut and between 56.4–68.7 % of all microbial reads in the distal gut. This Mycoplasma dominance resulted in a Q2-Q3 mean coverage of 3,667-5,939 X in the midgut samples and 317–847 X in distal gut samples for CTRL, whereas the coverage of Mycoplasma in PRO and SYN was extremely low, except for one sample in SYN (Supplementary Table S3.1). Both Lactobacillus and Enterobacteriaceae were found at higher relative abundance in fish from the PRO and SYN groups, a reflection of a reduced Mycoplasma biomass (Fig. 2a). Interestingly, the coverage of Lactobacillus and Enterobacteriaceae were in general very low and ranged from 0.00 to 8.43 X Q2-Q3 mean coverage across all samples for Lactobacillus and 0.00 to 5.16 X Q2-Q3 mean coverage for Enterobacteriaceae, clearly indicating a low bacterial load even when Mycoplasma was reduced.
The functional potential of metagenomes also varied significantly among the feeding groups. Differential abundance analysis of the metagenome data revealed that 670 out of a total of 5,574 non-redundant genes were significantly more abundant in CTRL (adjusted p-value < 0.05) (Fig. 2b, Supplementary Table S3.2), including genes encoding for Arginine biosynthesis pathway, such as arcA, arcC, and otc (Fig. 2b), genes associated with the Cellobiose PTS system, referred to as cellulosome. Interestingly, we found Terpenoid Backbone Synthesis encoding genes, from the non-mevalonate (MEP) pathway, including ispE, ispF, ispG, and ispH to be enriched in the CTRL group (Fig. 2a-b). These MEP related genes were all present in the Mycoplasma MAG, clearly indicating that alterations of Mycoplasma abundance are the main driver of the observed metagenomic variation among feed groups. Surprisingly, we found that genes related to Terpenoid Backbone Synthesis had a dramatically higher log fold change than the rest of the Mycoplasma MAG related genes, which we hypothesise could reflect a higher gene copy number.
Diet and Mycoplasma abundance are associated with the intestinal metabolism of rainbow trout
To increase comprehensiveness of ionic properties and polarity in our metabolic analysis, we included UHPLC-MS/MS and IC HR-MS/MS data generation44,45 resulting in a total of 22,222 mass spectral features with associated tandem mass spectrometric data, which we here use as a proxy for metabolites. Out of the 22,222 metabolites, 12,706 metabolites were generated from UHPLC-MS/MS and 9,516 metabolites were generated from IC HR-MS/MS.
Using the molecular networks, we retrieved in silico annotated chemical classes for 7,190 metabolites (56.59%) of UHPLC-MS/MS. Out of the 12,706 metabolites, 741 metabolites were included in the study after filtering for false positives and zero elimination. Overall metabolic variations revealed a clear differentiation among the CTRL, PRO, and SYN (Fig. 3a). Specifically, we investigated metabolite classes putatively synthesised by enzymes encoded by genes found to be differentially abundant in the metagenomes (Fig. 2b). We found clear differentiations in compositions of metabolite subclasses, including amino acids, peptides, and analogues, terpenoids, bile Acids, alcohols, and derivatives (Fig. 3b-d). Composition of especially terpenoids did not only cluster samples based on feed type alone, but also revealed some clustering of samples with a high relative abundance of Mycoplasma irrespective of feeding types (Fig. 3c).
For UHPLC-MS/MS, 419 (3.29%) could be matched to known compounds in the GNPS library. For IC HR-MS/MS, 282 (2.96%) could be matched to known compounds in the mzCloud database, which in total resulted in 240 known compounds after deduplication of isoforms of compounds and filtering (Supplementary Table S3.4). Differential intensity analysis of the 240 known metabolites resulted in 25 differentially abundant metabolites, whereas 19 of these metabolites were more abundant in CTRL (Fig. 3e). These included pantothenic acid, indole-3-carboxylic acid, 5-methoxyindole, and 5-hydroxyindole-3-acetic acid, indicating a higher amount of vitamin B5 and degradation of tryptophan46 in the gut of rainbow trout from the CTRL group. These differences indicate alterations of important immune related metabolites among fish reared on the different feed types46,47. Furthermore, we found an increase of succinic semialdehyde in CTRL, indicating butyrate related short chain fatty acid (SCFA) metabolism occurring in the gut of rainbow trout48, which corresponds to previous findings that Mycoplasma dominates the microbiome of both wild and farmed Atlantic salmon36. SCFAs are known to be the end-products of dietary fibre fermentation by gut microbiota and have been suggested to be an essential nexus between microbiota and different host organ systems49. We found an increase of lauroyl-carnitine in PRO and SYN indicating fatty acid oxidation and thereby an increase in lipid metabolism. Furthermore, gluconic acid lactone were found increased in PRO and SYN, indicating induced sugar degradation, which we hypothesise is due to sugar formation by present Lactobacillus or P. acidilactici MA18/5M, which would make sense for SYN, where galacto oligosaccharides were added to the feed (Fig. 2a and Fig. 3e).
Furthermore, differential intensity analysis of metabolites with no spectral hits revealed a total of 168 metabolites from UHPLC-MS/MS with a significantly different abundance between CTRL and the two other groups after FDR correction for multiple tests (adjusted p-value < 0.05) (Supplementary Table S3.4). Furthermore, we found that 89 of the 168 metabolites were higher abundant in the CTRL group, whereas 79 of the metabolites were higher abundant in PRO or SYN. Metabolites higher abundant in the CTRL group included the metabolite classes: Prenol lipids, Steroids and Steroid Derivatives, Carboxylic Acids and Derivatives, and Benzenes and Substituted Derivatives (Supplementary Fig. 3, Supplementary Table S3.5). These metabolite classes clearly indicate a differentiation in steroid and terpenoid production in the intestinal environment. Especially prenol lipids, which include classes of terpenoids, were found to be highly affected by feed type and more abundant in the CTRL group thereby mimicking the differential abundance of Mycoplasma among feeding groups (Supplementary Fig. 3). Further investigation of differentially abundant metabolites observed across feeding types confirmed our previous finding of the CTRL group having a distinct metabolomic landscape compared to the PRO and SYN groups (Fig. 3a, Supplementary Fig. 4).
Deciphering unknown metabolites associated with gut microbiota
To investigate association between specific metabolites and presence of microbes, we computed the correlation between the relative abundance of the ASVs per sample to the concentrations of a subset of the metabolites. We restricted our analysis to the 26 samples, which included both 16S rRNA gene profiling and metabolomics. The 26 samples included 10 fish from the CTRL group, nine from the PRO group, and seven SYN samples. Filtering out rare ASVs resulted in a total of six ASVs (Fig. 1a), while zero inflation of metabolites validated a total of 569 metabolites for this association analysis. Association tests between metabolite intensities and the relative abundances of ASVs revealed four metabolites that are significantly associated with the ASV abundances after Bonferroni correction (Supplementary Table S3.6). We investigated the top 25 most significantly bacterial associated metabolites (BAMs) post Bonferroni correction, using an enhanced molecular network50–54 to infer these unknown metabolites in the intestinal metabolomic landscape of rainbow trout (Supplementary Table S3.7).
Network analysis of 350 metabolites, including the top 25 BAMs and their related molecular families, were used to decipher molecular structures of unknown metabolites (Fig. 4). GNPS successfully classified 92.5% of the metabolites, where 28.2% of the classifications were confirmed by SIRIUS + CSI:FingerID. Of the 25 BAMs and their related molecular families, we were able to classify 11 BAMs and their related molecular families. The molecular families included prenol lipids from terpenoid backbone synthesis, which were associated with intestinal bacteria. Furthermore, we found a molecular family of unknown lipids, with indications of water loss, indicating formation of steroids, suggesting formation of bacterial related steroids in the intestinal environment (Fig. 5). Interestingly, we found BAMs related to networks of benzenoids, including putative stilbenes and phenylpropanones, indicating production of antibacterial BAMs in the intestinal environment of rainbow trout, which could target bacterial cell walls55,56.
Furthermore, we found BAMs in molecular families of fatty acyls with a relatively low mass-charge, indicating conjugation of SCFAs. A molecular family of peptide structures, containing substructural motifs of SCFA related aminobutyrate, indicating degradation, biosynthesis or conjugation of an aminobutyrate-like peptide by bacteria in the intestinal environment. A network of putative peptides with BAMs revealed three shared substructural motifs between metabolites, including traces of histidine, alkylamine, and creatinine, indicating incorporation of ammonia derivatives into peptides by intestinal bacteria in rainbow trout (Fig. 4). These findings could confirm our metagenomic observation of arginine biosynthesis, which includes metabolism of ammonia rich peptides.
Feed additives and nutrient utilisation
Feed performance was evaluated based on bulk weights and counts of rainbow trout from five replicate tanks for each of the three feeding groups tested. Registration of total fed feed for each tank, as well as near infrared spectroscopically determined content of protein and lipid in each feed group was recorded (Supplementary Fig. 5a-e).
Our findings indicate that nutrient related phenotypes of juvenile rainbow trout can be affected by feed additives. Overall, our analyses revealed no significant differences in percent weight gain (WG), Feed Conversion Ratio (FCR), and the inverse of FCR, Feed Efficiency Ratio (FER), among feeding groups (Supplementary Fig. 5a-c). It should be noted that nutritional analysis of feed revealed a lower number of calories (MJ/Kg) and fat content (%) in PRO feed, and a lower amount of protein in the CTRL diet (Table 1). This may explain some of the observed differences in the performance data and complicates the further interpretations of feed efficiency indices (Supplementary Fig. 5).
However, the Lipid Efficiency Ratio (LER) suggest a significantly more effective conversion of feed lipids into biomass in the PRO group (F(2,12) = 9.84, p = 0.0029) (Supplementary Fig. 5d), indicating usage of Pediococcus acidilactici MA18/5M could improve efficiency of lipid utilisation in rainbow trout. Analysis of Protein Efficiency Ratio (PER) showed that the CTRL group had a significantly higher efficiency than the other feeding types (F(2,12) = 9.88, p = 0.0029). This could indicate that use of feed with pro- or synbiotic additives may decrease protein utilisation as a response to higher lipid efficiency ratio (Supplementary Fig. 5e). An accurate LER and PER determination would require isoenergetic/proteinic diets and analysis of whole-body fat and protein rather than bulk weight of the fish.