The test substances were glyphosate [N-(phosphonomethyl) glycine; Sigma–Aldrich, Merck KGaA, Darmstadt, Germany], the formulation Roundup^® LB Plus (purchased at a local retail store), and the metabolite AMPA [(aminomethyl)phosponic acid; Acros Organics BVBA, Geel, Belgium]. These compounds were dissolved in ultrapure water to obtain solutions of 56, 560, and 5,600 μg L⁻¹ for glyphosate and Roundup^® (containing 560 or 5,600 μg L⁻¹ gLyphosate, respectively) and 3,666 μg L⁻¹ for AMPA. To achieve the respective nominal concentrations, corresponding stock solutions were prepared with ultrapure water.

Brown trout (Salmo trutta f. fario) was purchased from a commercial trout farm in southern Germany (Forellenzucht Lohmühle, Alpirsbach-Ehlenbogen, Germany), which is listed as disease-free according to the EC Council Directive (2006). The fish were acclimated to laboratory conditions in a climate chamber in a 200 L tank with filtered (iron filter, activated charcoal filter, and particle filter) and aerated tap water 1 week prior to the exposure experiment. The fish were observed daily and fed with commercial trout feed (Inico Plus, Biomar, Brande, Denmark). All animal experiments were approved by the Animal Welfare Committee of the Regional Council of Tuebingen, Germany (approval number ZO 2/16 and ZO 02/21 G).

Each exposure experiment was conducted in a semi-static, three-block setup. Therefore, each treatment was set up in triplicates, and each block consisted of six aquaria tanks containing 15 L of the respective test medium. In each aquarium tank, 30 fish of 6 months old or 10 fish of 10 months old were exposed to water with no test medium as control, or to 56 μg L⁻¹, 560 μg L⁻¹, or 5,600 μg L⁻¹ of glyphosate, Roundup^® LB Plus containing 560 μg L⁻¹ or 5,600 μg L⁻¹ of glyphosate, respectively, or 3,666 μg L⁻¹ AMPA, which is equimolar to the highest concentration of glyphosate. Exposure experiments were conducted under a 12:12 light/dark cycle at 7°C, and the tanks were covered with black foil to protect the fish from direct light. Half of the respective test solutions were removed from each tank every 2–3 days and replaced with fresh test solutions to keep the concentration stable. Furthermore, water conditions (temperature, conductivity, pH, and oxygen content) were checked regularly. The fish were exposed for 3 weeks and then anesthetized and killed by an overdose of the anesthetic MS-222 (tricaine methanesulfonate; Pharmaq, Overhalla, Norway), followed by a cervical cut. Various samples were collected for different biomarkers including gill and gut samples for microbiome analysis.

For population analysis and the abundance of fish pathogens, 90 guts of 6-month-old fish and 90 guts of 10-month-old juvenile fish were utilized to extract bacterial DNA. For bacterial DNA extraction, the midgut part was used. Due to the small tissue size, no separation of mucus-associated biofilm and fish tissue was performed. Instead, the whole gut was dissolved by a 3-h incubation using lysing buffer and proteinase K provided by DNA Mini Kit by QIAGEN (Hilden, Germany). The DNA extraction and clean-up were performed by the DNA Mini Kit. The number of intestines analyzed individually ranged from 8 to 18 in biological triplicates and technical replicates, depending on the exposure and age of the fish.

Terminal Restriction Fragment Length Polymorphism (t-RFLP) was applied to assess population changes in the microbiome of 6-month and 10-month-old fish larvae. In principle, t-RFLP relies on the universal amplification of a large 16S rRNA fragment from Eubacteria. Using selected restriction enzymes, it is possible to generate labeled, terminal 16S rRNA fragments of different sizes depending on individual restriction sites of the present bacterial species. As a consequence, this fingerprint analysis complements and extends the results of amplicon sequences, aiming on possible shifts in the microbial population. The t-RFLP is not designed and qualified to detect specific species but specified terminal fragments of the 16S rRNA which can correlate with closely affiliated bacteria. To compensate for these uncertainties, the different fragments representing different species are called phylogenetic clusters.

Sample preparation for t-RFLP analysis was performed in technical duplicates. For the DNA-based t-RFLP analysis, a universal PCR was performed using a Fam-labeled primer system targeting the V1-V3 region of the 16S rRNA. The primer sequences utilized are presented in Table 1. The mastermix for each PCR reaction consisted of 19 μL nuclease-free water, 2.5 μL combination buffer, 0.5 μL dNTP (VWR, Darmstadt, Germany), 0.125 μL Hot Start Polymerase (VWR, Darmstadt, Germany), and 1 μl of each primer (20 nM) for a final reaction volume of 25 μL. The PCR profile included 15 min of initial denaturation at 95°C, 30 s of denaturation at 95°C, 30 s of annealing at 56°C, 60 s of cycle elongation at 72°C, and 10 min of final elongation at 72°C for 30 cycles. Afterward, the PCR product was restricted using the four-base cutting Fast Digest Enzyme HhaI (Thermo Fisher Scientific, Waltham, United States). In total, 10 U was used to restrict 0.2 μg of PCR product at 37°C for 3 h. After restriction, the technical duplicates were pooled again, and 2 μL of the restricted DNA was mixed with 15 μL of Hi-Di formamide; fragment analysis was performed at a SeqStudio device (Thermo Fisher Scientific, Waltham, United States) to detect the FAM-labeled terminal restriction fragment. The run utilized an injection time of 40 s at 1200 Volts and a run time of 1,440 s at 9000 Volts. Analysis range of the t-RFLP was set to 20–500 bp with a set bin width of 5 bp. All samples were analyzed within one run, and the normalization was performed over the sum of signals. The resulting fragmentation patterns were further analyzed using R and the vegan community ecology package (Oksanen et al., 2022). Calculations were performed using the respective peak areas.

For analysis of the virulence gene expression, a variety of different virulence factors of the observed fish pathogens were chosen (Table 2).

The reference bacteria Y. ruckeri (DSM 18506), E. faecalis (DSM 20478), A. salmonicida (DSM19634), and E. faecium (DSM 20477) were incubated overnight in full media at their respective temperature. After incubation overnight, 5 mL of the culture was pelleted, washed twice with PBS, and resuspended to an OD₆₀₀ of 0.1 with a new medium containing 0 μg L⁻¹, 1 μg L⁻¹, 100 μg L⁻¹, 560 μg L⁻¹, and 5,600 μg L⁻¹ glyphosate or AMPA and incubated. At the end of the exponential growth phase, the bacteria were pelleted again at 4°C, and the total RNA was extracted using the RNeasy Mini Kit (QIAGEN, Hilden), including additional DNase treatment, according to the manufacturer’s instructions. Reverse transcription was performed using random hexamers and the Superscript IV reverse transcriptase (Thermo Fisher Scientific, Waltham, United States), according to the manufacturer’s instructions. Quantification was performed by qPCR using the Maxima SYBR Green Master Mix (Thermo Fisher Scientific, Waltham, United States), utilizing the primers presented in Table 1. 16S rRNA expression was used as reference housekeeping gene for normalization (Huang et al., 2014; Bolhari et al., 2018). Analysis of all virulence factors was performed in biological triplicates.

The exposure of the reference bacteria Y. ruckeri (DSM18506) for the whole-transcriptome analyses was carried out in biological duplicates in the same way for the detection of virulence factors. After washing with PBS, the pelleted overnight culture was re-suspended in either full medium (tryptogenic soy broth) containing amino acids (VWR, Darmstadt, Germany) or mineral medium (M9 medium containing 6 g L⁻¹ Na₂HPO₄. 3 g L⁻¹ KH₂PO₄, 0.5 g L⁻¹ NaCl, 1 g L⁻¹ NH₄Cl, 0.1 mL 1 M CaCl_2, 2 mL 1 M MgSO₄, and 5 mL 40% glucose, adjusted to pH 7.4) without any amino acids. RNA isolation was performed using the RNeasy Mini Kit from QIAGEN (Hilden, Germany), including DNase digestion. Libraries were created using the ZymoResearch RiboFree Total RNA Library Kit (ZymoResearch, Freiburg, Germany) and subsequent electrophoretic quality control. An Illumina NextSeq 500/550 HighOutput Kit V2.5 with 400 million read pairs was used for sequencing. The quality control and the analysis of the sequencing were carried out using the following programs: FastQC (quality control), TrimGalore (removal of adapter sequences), SortMeRNA (removal of rRNA), STAR (alignment of sequences with gene databases), SAMtools (sorting and indexing of individual aligned sequences), FeatureCounts (counting the number of identical aligned sequences), and DESeq2 (normalization of count and expression analysis) (Kopylova et al., 2012; Liao et al., 2014; Love et al., 2014; Dobin and Gingeras, 2015; Krueger, 2015; de Sena Brandine and Smith, 2019).

Significance of the alpha diversity (mean Shannon coefficient) was evaluated starting from the Shannon coefficient of every sample using a two-sided Mann–Whitney test, since the Shapiro–Wilk test showed that not all samples were normally distributed. The significance level was chosen to be p < 0.05.

Experiments of virulence gene expression were performed in biological triplicates, each analyzed as technical duplicates. Significance testing was performed by a one-sided Mann–Whitney U test on a p = 0.05 level using the mean of the technical replicates.

To evaluate the significance of expression changes observed during transcriptome analysis, p-values were calculated, and Benjamini–Hochberg adjustments were performed considering a false positive rate of 10% acceptable, although no analyzed sample shows significant changes of the observed genes involved in the shikimate pathway compared with the control.

Obtained raw reads of the 16 s rRNA amplicon sequencing as well as the transcriptome analysis of Y. ruckeri can be accessed via the NCBI database under the BioProject: PRJNA1001608.

As the sample preparation of the gut microbiome of juvenile fish was challenging due to the small gut dimensions, previous screening and optimization of different sample preparation approaches were performed. It was not possible to extract prokaryotic DNA selectively during preparation without neglecting important compartments (i.e., mucosa-associated bacteria). The ratio of detected bacterial cell equivalents per ng of extracted DNA was an important parameter to be determined in order to assess sample quality. As the high concentration of eukaryotic DNA can negatively influence both DNA extraction and the following PCR-based molecular methods, the amount of eukaryotic DNA should be kept to a minimum. Neither the fish age nor the exposure to glyphosate, AMPA, or Roundup^® had a significant influence on the bacterial DNA yield of the extraction, suggesting that glyphosate, AMPA, and Roundup^® do not decrease the overall amount of bacteria associated with the gut. Nevertheless, the amount of DNA extracted from the older fish was, with an average of 3,729 ± 2,414 ng per gut sample, slightly higher than that of the 6-month-old fish (average 2,602 ± 904 ng DNA per gut sample). The high variance in total DNA yield is caused by the individual gut sizes and preparation variances. To visualize the amount of eukaryotic DNA, a graphical plot of the 16S rRNA copy number (as a reference for bacterial population) per ng DNA is presented in Supplementary Figure S1.

To get an overview of the microbial gut population of fish, amplicon sequencing based on 16S rRNA was performed, targeting the V1-V3 variable region. The bacterial community composition of 10-month-old fish is shown in Figure 1. Due to the DNA sample quality and amount needed for 16S rRNA amplicon sequencing, not all gut samples were suited and qualified for this analysis. Thus, the number of fish gut sample for each of the different tank treatments varied up to Δn = 4.

Each sample resulted in approximately 160,000 raw reads with 74,000 total read pairs with a mean read length of 278 bp. Depending on the sample, approximately 36,292 reads were matched. Detailed statistics for each sample are presented in Supplementary Tables S1, S2.

The control fish gut contained 17 different detectable genera. Exposure to glyphosate led to a reduction in nine different detectable bacterial genera for 56 μg L⁻¹ glyphosate and seven detectable genera for 560 μg L⁻¹ and 5,600 μg L⁻¹ glyphosate. Exposure to Roundup^® led to a reduction to six detectable genera and AMPA to eight detectable bacterial genera. Commonly eliminated genera through all treatments were Escherichia, Polymorphobacter, Paracoccus, and Lactococcus. Bacterial genera, which showed resistance against Roundup^® and only got eliminated by higher glyphosate concentrations, were Rhodobacter, Lactococcus, and Clostridium. The majority of the fish gut population consisted of Leuconostoc (61%) followed by Weissella (15%). Less abundant bacteria were Clostridium (4%), Streptococcus (3%), and Lactobacillus (3%). It can be observed at first glance that after exposure to glyphosate, the abundance of Leuconostoc spp. increased approximately 22%. As Leuconostoc spp. is known as a probiotic bacterium, its increase would contradict the potential negative effects of glyphosate on the health of hosts (Balcázar et al., 2009; Pérez et al., 2010). A major counterpart to the increase in Leuconostoc spp. was the decrease in Weissella spp. by 11%. The remaining 11% was compensated by the decrease or even elimination of low abundant species such as Clostridium, Streptococcus, and Pseudorhodobacter. Some bacteria such as Microtericola and Rhizobium were only detected after exposure to one of the test substances. Overall, these bacteria-specific responses to exposure to glyphosate are caused by a wide range of susceptibilities of different bacteria species to glyphosate (Nielsen et al., 2018). It should be noted that even an increase in a bacterial genus, which is identified as probiotic, is indeed a shift of microbial population, which can lead to dysbiosis and, therefore, can have a negative effect on the health of the host.

Although a glyphosate-induced intestinal population shift could be observed, the changes in the abundance of low-abundant species might influence the vitality of hosts in a positive and a negative way, since even low abundant species in the gut microbiome have a strong impact on the vitality of hosts (Sheehan et al., 2015; Yang et al., 2019). Therefore, even the weakly observed glyphosate-induced changes in the microbiome should not be neglected.

Fingerprint profile patterns, displaying each detected fragment obtained by the t-RFLP of both 6-month-old and 10-month-old fish, are presented in Figures 2, 3.

It became obvious by these analyses that the gut microbiome of 6-month-old fish was more diverse than the ones of 10-month-old fish. In the case of 6-month-old fish, 114 different phylogenetic clusters could be detected, while the t-RFLP of 10-month-old fish resulted in 91 different phylogenetic clusters. At both ages, dominant phylogenetic clusters, which are present in the majority of the analyzed samples and made up a majority of the population, could be observed. Additionally, some fragments were only present in some fish samples, indicating the presence of distinct bacteria in fish subpopulations. The presence of these bacteria was common for all types of fish within one tank, indicating a variance in the fish microbiome despite similar treatment. Especially for 10-month-old fish, the control fish contained some phylogenetic clusters (at the left side of the fragmentation pattern), which were absent in the majority of the gut of the exposed fish.

The decreased number of phylogenetic clusters of the microbiome of 10-month-old juvenile fish compared with 6-month-old fish leads to an assumption that the gut microbiome is altered between both age cohorts. Consequently, it was advised to analyze both age cohorts separately in order to address this diversity between the age cohorts.

The alpha diversity (Shannon coefficient) of the gut microbiome of both age cohorts is presented in Table 3. The Shannon coefficient shows the biodiversity within a single population, whereby a higher coefficient shows a higher diversity of the microbiome.

It was obvious that, irrespective of the exposure, young fish have a more diverse gut microbiome than older fish (Shannon coefficient of 1.93 vs. 1.82). The exposure of 6-month-old fish to glyphosate, AMPA, and Roundup^® did not show a significant effect on the microbiome diversity. Furthermore, the concentration of glyphosate did not play any role. However, in 10-month-old fish, AMPA and Roundup^® affected the gut microbiome diversity in the same way as glyphosate (1.65 for 5,600 μg L⁻¹ glyphosate formulated as Roundup^® and 1.61 for 3,666 μg L⁻¹ AMPA). The control group showed a Shannon coefficient of 1.93, while the exposure groups resulted in coefficients between 1.90 and 1.98.

The Shannon coefficient gives an overview of the diversity of a microbiome within itself. Therefore, it is not suitable to elaborate on the differences between the microbiome of two different groups (microbiome dissimilarity). For this purpose, Bray–Curtis dissimilarity analysis was utilized. It allows us to directly compare the microbiomes of the untreated control group with the ones exposed to different concentrations of glyphosate, AMPA, and Roundup^®. The resulting coefficient between 0 and 1 describes the species dissimilarity of both microbiomes, as 0 represents microbiomes with identical compositions, while a Bray–Curtis coefficient of 1 describes two microbiomes that have no common species. This allows a more direct comparison of two sets of population and gives an overview of composition changes as a result of exposure to glyphosate, AMPA, and Roundup^®. Bray–Curtis Index of both 6-month-old and 10-month-old fish is presented in Table 4 (6-month-old fish) and Table 5 (10-month-old fish).

The overall impact of glyphosate, AMPA, and Roundup^® on the microbiome composition was evident on a small scale. The range of Bray–Curtis coefficients (beta diversity) between the control group and treated fish varied between 0.09 and 0.22 for 6-month-old fish and 0.1 and 0.29 for 10-month-old fish.

The 6-month-old fish showed the most distinct effect for the lower glyphosate concentration of 56 μg L⁻¹ associated with a Bray–Curtis coefficient of 0.22. Increasing glyphosate concentration led to microbiome composition, which is again more similar to the microbiome of untreated fish. A similar hormesis effect on the microbiome of young fish could be observed in previous studies with the anti-diabetic compound metformin (Rogall et al., 2020). A concentration of 560 μg L⁻¹ and 5,600 μg L⁻¹ glyphosate-formulated as Roundup^® did not have an effect on the gut microbiome of 6-month-old fish larvae. The Bray–Curtis coefficient of the gut microbiome of 10-month-old fish increases with increasing glyphosate concentrations, starting with 0.22 for 56 μg L⁻¹ glyphosate up to 0.29 for 5,600 μg L⁻¹ glyphosate. Thus, glyphosate and AMPA caused comparable shift in the gut microbiome of 10-month-old fish (Bray–Curtis coefficient of 0.26 for glyphosate and 0.27 for AMPA).

Furthermore, the analyses showed that the differences between controls and treated fish were only slightly influenced by the formulation of the substance in high concentrations with a Bray–Curtis coefficient of 0.15 in 6-month-old fish and 0.26 in 10-month-old fish after exposure. However, a comparison of the Roundup©-treated fish with raw glyphosate-treated fish showed that their differences were in the same range (Bray–Curtis coefficient of 0.13 for 6-month-old fish and 0.16 for 10-month-old fish) as their comparison with the control fish, suggesting that the effect of glyphosate is not identical to the Roundup^® formulation containing the same amount of glyphosate.

On a final note, t-RFLP analysis verified the results of 16S rDNA amplicon sequencing while considering a wider range of fish gut microbiome samples. Population changes upon exposure to glyphosate, AMPA, and Roundup^® occur but are limited to a small fraction of the microbiome subpopulations. The extent of the shift does not only depend on the glyphosate concentration. The age of the fish and therein the initial composition of the microbiome and its initial diversity play an important and defining role.

It is hypothesized that even small changes in the microbiome can have a significant effect, especially if pathogens are involved (Sheehan et al., 2015). Each obtained gut microbiome sample was screened via qPCR for the selection of fish facultative pathogens, such as Y. ruckeri, E. faecalis, E. faecium, A. salmonicida, and F. psychrophilum. The frequencies in which fish microbiomes were colonized by each of these fish pathogens are presented in Figure 4. The deviation of the fish microbiome colonization rates between the different breeding tanks with identical conditions varied between 10 and 15%. It became apparent that younger fish of 6 months were more susceptible to colonization with fish pathogens than 10-month-old ones, which may also be a result of the more matured immune status and/or more stable microbiome of older fish. Y. ruckeri was analyzed with a percentage of colonization up to 70% being the most abundant fish pathogen. A. salmonicida, E. faecium, E. faecalis, and F. psychrophilum were less abundant as their percentage of colonization was up to 30% of all analyzed fish guts.

Exposure of 6-month-old fish to glyphosate resulted in a major rise in colonization with Y. ruckeri, starting from 10% for control fish to 70% for fish exposed to 560 μg L⁻¹ glyphosate. The percentage of fish colonization with A. salmonicida, E. faecium, and E. faecalis did also increase to 20% upon exposure to glyphosate. It became evident that an increase in the glyphosate concentration resulted in a higher percentage of colonization up to 560 μg L⁻¹ glyphosate. Rising the concentration further up to 5,600 μg L⁻¹ glyphosate no longer led to an additional increase. The colonization probability is 30% lower for 5,600 μg L⁻¹ glyphosate than for 560 μg L⁻¹ glyphosate. Exposure to Roundup^® also results in an increase in Y. ruckeri, E. faecium, and E. faecalis colonization, however, not in the same way as in glyphosate-exposed fish. This became obvious, especially for Y. ruckeri, exhibiting a colonization percentage rise of 40% upon exposure. The colonization percentages of A. salmonicida and F. psychrophilum in 6-month-old fish were not increased significantly by any exposure of the fish.

In contrast to this, in 10-month-old fish, the percentage of colonization with pathogens was less impacted by exposure to glyphosate, AMPA, and Roundup^®. Only Y. ruckeri could be frequently detected in all exposure groups. Whereas the colonization of fish guts from the control groups was at 29%, exposure to glyphosate or Roundup^® resulted in colonization between 5% and 45%. Thus, excluding the treatment with 560 μg L⁻¹ glyphosate (with a colonization rate of 45%), no significant change in colonization with Y. ruckeri could be observed. High concentrations of glyphosate and Roundup^® led to a slight increase in colonization with F. psychrophilum, which could neither be detected in the fish gut of control samples nor the lower test substance concentrations. The increased colonization rate of 10-month old fish exposed to 560 μg L⁻¹ glyphosate up to 45% colonization with Y. ruckeri goes in accordance with the colonization rates of 6-month old fish leading to a depiction of 560 μg L⁻¹ to be the most effective concentration instead of an increasing effect by increasing concentrations.

As this approach and the results obtained provide the first hint of the effect of glyphosate and Roundup^® on the colonization of fish pathogens, it has a major drawback just like most exposure experiments: the fish grow with time and they were exposed in artificial systems under controlled environments. As the trout breeding station, where the larvae were obtained is certified as pathogen-free, it is at least questionable if the lack of colonization is due to the missing effect of exposure or the overall absence of the pathogen in the environment. This especially accounts for pathogens with low colonization rates. Due to its relatively high abundance in this setup, Y. ruckeri can be observed as the only reliable indicator through all exposures and ages, resulting in the final conclusion that glyphosate, AMPA, and Roundup^® showed an effect on the colonization of fish guts, Although the scope of the effect is, among other factors, determined by the age of the fish and the glyphosate concentration. Here, the guts of younger fish with a more diverse microbiome are more affected by pathogens than those of older fish, verifying the results obtained by t-RFLP population analysis.