Rumen contents were obtained from four ruminal fistulated German Holstein cattle housed at the Institute for Animal Nutrition, Federal Research Institute for Animal Nutrition, Brunswick, Germany. Fistulation of the animals was approved by the Lower Saxony State Office for Costumer Protection and Food Safety, Oldenburg, Germany (33.9-42502-04-12/0876). All procedures involving animals were carried out in accordance with the German Animal Welfare Act.

Four RUSITEC experiments were carried out using two RUSITEC systems with in total 12 fermenters. Inoculation of the fermentation vessels was carried out as described previously (Eger et al., 2018), with the modification that for each experiment rumen contents of one of the four cow were used. Briefly, rumen ingesta were squeezed through two layers of medical gauze (Gazin^®, 80 cm × 5 cm, Lohmann & Rauscher International GmbH Co.KG, Rengsdorf, Germany) to separate the liquid and solid phase. Rumen contents were transported in preheated insulated containers to avoid heat losses. For each fermenter, a nylon bag (R712 Forage Bags in situ, ANKOM Technology, Gesellschaft für Analysetechnik HLS, Salzwedel, Germany) was filled with 70 g of solid rumen ingesta and placed in the inner fermentation vessel. A second nylon bag containing the daily feed was added. Feed bags contained 15 g of substrate. Components and chemical contents of the daily substrate are presented in Table 1. All components were shown to be free from glyphosate and aminomethylphosphonic acid (AMPA) before use (< 0.01 mg/kg, Wessling GmbH, Hannover). In each experiment, three fermentation vessels served as control, and one vessel each was assigned to one of the following nine treatments: 0.1 mg/l, 1.0 mg/l or 10 mg/l of pure glyphosate (40% N-(Phosphonomethyl) glycine, monoisopropylamine salt solution, Sigma-Aldrich, Darmstadt, Germany), 0.1 mg/l, 1.0 mg/l or 10 mg/l of glyphosate as Durano TF (052389-83/MOT, Bayer Crop Science, Langenfeld, Germany), or 0.1 mg/l, 1.0 mg/l or 10 mg/l of glyphosate as Roundup^® LB plus (024142-60/MOT, Nordland Agrar, Süderlügüm, Germany). The first two concentrations were chosen to meet the concentrations of 0.08 mg/d to 6.78 mg/d reported for in vivo uptake of glyphosate by von Soosten et al. (2016), the highest concentration exceeds in vivo uptake levels by 10-fold to provide a safety margin. The compounds were dissolved in methanol. The experiments consisted of an equilibration period of 6 days, a control day (day 7) and the experimental period from day 8 to day 16, during which, on each day either the respective glyphosate treatment was added to the fermentation vessels or the same amount of methanol (control vessels). A physiological buffer solution was infused into the fermenters to achieve a liquid turnover of once per day (Table 2). In order to measure microbial protein synthesis, ¹⁵N was included in the buffer solution (Table 2).

The pH and redox potential were measured daily throughout the experiment (Digital-pH-Meter 646, Knick GmbH & Co. KG, Berlin, Germany, electrodes: InLab^® Routine, InLab^® Redox Pro, Mettler-Toledo GmbH, Gießen, Germany). At control day and during the experimental period, fluid flow-through of the fermenters was recorded daily and samples for the analysis of NH₃-N and short-chain fatty acids (SCFA) were collected daily from the overflow. Short-chain fatty acid concentrations were measured by gas chromatography as described by Koch et al. (2006). The concentration of NH₃-N was determined by photometry as described by Riede et al. (2013). From day 10 to day 13, feed bags were collected for the analysis of nutrient degradation. To determine the degradation of organic matter (OM), crude fat (XL), crude protein (XP), crude fibre (XF), neutral detergent fibre (NDF) and acid detergent fibre (ADF) feed residues from the 4 days were pooled, analysed by Weender analysis (Chair of Animal Nutrition, Technical University of Munich, Munich, Germany) and compared to undigested substrate.

From day 10 to day 16, 20 ml of fermenter fluid was collected daily for the separation of liquid-associated microorganism (LAM) and frozen at −20°C. Solid-associated microorganism (SAM) was collected from the feed bags at days 14 and 15 by methyl cellulose incubation (Boguhn et al., 2013) and also frozen until further treatment. Liquid samples and the methyl cellulose samples were then thawed and subjected to differential centrifugation (Brandt and Rohr, 1981). First samples were centrifuged at 2,000 × g for 5 min. The supernatant was again frozen at −20°C and the pellet was discarded. The supernatant was thawed in a fridge, centrifuged at 2,000 × g and 5 min again and the supernatant was collected in a new tube. The samples were then centrifuged at 27,500 × g for 15 min. The supernatants of LAM samples, which represented the fermenter liquid, were stored for later ¹⁵N-analysis, while the pellets representing LAMs or SAMs were washed with 0.9% NaCl three times. The remaining pellets were weighted and frozen at −20°C. For the analysis, LAM and SAM pellets and fermenter liquid were freeze-dried and grinded with a glass stick. The amount of ¹⁵N was analysed using a mass spectrometer (TracerMAT, Thermo Fisher Scientific GmbH, Dreieich Germany) after incineration in an elementary analyser (EA1108, Fisons Instruments GmbH, Mainz-Kastel, Germany).

N assimilation in LAMs was calculated using the following formula:

Where ¹⁵N_input is input by the buffer solution; ¹⁵N_output is the content in the fermenter liquid; and ¹⁵N_plateau is the ¹⁵N percentage in the reference microbes.

The N assimilation in SAM was calculated using the following formula:

Where ¹⁵N_FR is the measured ¹⁵N content in the feed residues; ¹⁵N_nat is the natural content of 0.3663%; N_FR is the N content of the feed residues; and ¹⁵N_SAM is the measured ¹⁵N content in the reference microbes.

Microbial protein synthesis was calculated by multiplying N content by 6.25 (Boguhn et al., 2006). The efficiency of microbial protein synthesis was calculated by dividing totally synthesised protein by fermented organic matter.

Statistical analysis was performed using the Statistical Analysis System for Windows SAS, version 9.4 by using the SAS Enterprise Guide version 7.1 (SAS Institute Inc., Cary, NC, United States). A three-factorial analysis of variance was applied by using the procedure mixed to determine effects of ‘treatment’ (=glyphosate formulation), ‘concentration’ and ‘day’ for parameters measured daily. Treatment and concentration were set as independent factors. Statistical analysis applies for days 7 to 16; however, only days 7 and 16 (before application and end of application) are displayed in graphs and tables to improve readability. For parameters obtained once per treatment or from pooled samples, only effects of ‘treatment’ and ‘concentration’ were analysed by two-factorial analysis of variance. The levels of significance were set at ^*p < 0.05, ^**p < 0.01 and ^***p < 0.001. Data are presented as means ± SD.

For microbiome and metaproteome analysis, 10 ml of fermenter liquid was collected at days 10–16 and SAM was collected at days 14 and 15. Bacteria pellets were thawed and lysed, followed by protein and DNA extraction as previously described (Haange et al., 2020). Briefly, samples were resuspended in 1 ml lysis-buffer (50 mm Tris, 5 mm EDTA, 0.4% SDS, 50 mm NaCl and 1 mm PMSF, pH = 8). This was followed by lysis using a Fastprep 24 (MP Biomedicals, Santa Ana, CA, United States) set to three cycles, with each cycle consisting of 1 min disruption at a speed of 5.5 m/s and 1 min at rest. The resulting extract was heated in a thermomixer (Thermomixer comfort 5,355, Eppendorf, Hamburg, Germany) at 60°C with shaking at 1400 rpm for 15 min and spun at 10,000 × g at 4°C for 10 min. Supernatants, containing the DNA and protein content, were stored at −20°C.

For DNA purification, half of the supernatant was thawed and 260 μl 10 M ammonium acetate was added, mixed and incubated on ice for 5 min. Samples were then spun at 20,000 × g for 10 min at 4°C and the supernatants were kept. Isopropanol in equal volume was added to the supernatants, mixed thoroughly and incubated on ice for 30 min. The samples were then centrifuged at 20,000 × g for 15 min at 4°C to pellet the DNA. The pelleted DNA was washed with 10 μl pure ethanol, dried in a speed vac and finally dissolved overnight at 4°C in TE-Buffer (1 mm EDTA and 10 mm Tris, pH = 8). The dissolved DNA was purified and proteins removed using the QIAmp DNA Mini Kit (QIAGEN, United States) and following the manufacturer’s instructions. Finally, the DNA content recovered from each sample was quantified using NanoDrop (NanoDrop2000, Thermo Scientific, Rockford, IL, United States).

Quantification of glyphosate in the fermenter medium was done as previously described (Krause et al., 2020). Briefly, for the extraction of glyphosate, 1,000 μl of methanol:acetonitrile:water (2:3:1) was added to 100 μl of specimen. Samples were vortexed, sonicated and centrifuged at 14,000 × g. The supernatant was dried in a vacuum centrifuge (Concentrator Plus, Eppendorf AG Hamburg, Germany) and re-dissolved in 100 μl of Milli-Q water afterwards. A 10 μl of each sample of the resuspended extract was injected into a BEH Amide column (2.1 × 100 mm, 1.7 μm) supplied by Waters (Milford, United States). Chromatographic separation was performed with a gradient of solvent A (66% H₂O, 33% acetonitrile, 10 mm ammonium acetate and 0.04% ammonium hydroxide; pH 9) and solvent B (10% H₂O, 90% acetonitrile, 10 mm ammonium acetate and 0.04% ammonium hydroxide; pH 9). The LC was run at a constant flow rate of 0.4 ml/min. Initial equilibration for 2 min with 0% B, within 2.5 min gradient from 0 to 100% solvent B, hold 100% solvent B for 2 min, back to 0% solvent B for 3.4 min. Targeted multi-reaction monitoring (MRM) measurement of glyphosate was done on a QTRAP^® 5,500 (AB Sciex, Framingham, United States) in negative ionisation mode. Analyst^® Software (AB Sciex, Framingham, United States, version 1.6.2) was used for data acquisition and analysis.

Alpha-diversity and principal component analysis (PCA) were done in R (Ihaka and Gentleman, 1996) using the basic functions and the vegan package (Dixon, 2003). Statistics were done in R using in-house written scripts as previously described (Haange et al., 2020). Briefly, the statistical tests used were for complete sample data analysis Permutational Multivariate Analysis of Variance (PERMANOVA) using the adonis function in the vegan R package, and for single variables, Kruskal-Wallis group test followed by a post-hoc pairwise Dunn test. Where appropriate (number of tests >20), p-values were corrected for multi-testing by the Benjamini-Hochberg method (Benjamini and Hochberg, 1995). Heat maps were constructed with pheatmap R package (version 1.10)² and all other figures were constructed using the R package ggplots2 (Wickham, 2011).

The effects of different formulations and concentrations of glyphosate on rumen fermentation were assessed by measuring pH (Figure 1A), redox potential (Figure 1B), NH₃-N concentration (Figure 1C) and production of SCFA (Figure 1D). None of these parameters were significantly affected by the factors ‘glyphosate formulation’ or ‘concentration’. Slight variations over time were observed for all four parameters. The pH value slightly increased over time (p < 0.001) with a value of 6.58 ± 0.09 (mean ± SD) at day 7 and 6.66 ± 0.07 at day 16 (Figure 1A). Although this difference was significant, the change in pH value was very small and all values were within the physiological range. The redox potential also varied significantly over time (p < 0.001); however, all values were within physiological levels (Figure 1B). The concentration range of NH₃-N was between 11.4 ± 2.3 mmol/l at day 7 and 12.4 ± 1.9 mmol/l at day 16 (p < 0.001). The production of SCFA varied between 44.0 ± 4.5 mmol/d and 41.4 ± 4.3 mmol/d (effect of time p < 0.01). For all four mentioned parameters, no significant interactions were present between the factors. The molar proportions of the individual SCFAs were influenced neither by glyphosate formulation, nor by concentration, nor were interactions revealed. While the proportion of acetate remained unchanged at about 45–50% throughout the whole experiment, the proportions of propionate, butyrate, isobutyrate, valerate and isovalerate varied slightly over time (at least p < 0.05, means are presented in Supplementary Figure 1). The proportions of propionate ranged between 23.3 ± 3.1% and 23.6 ± 3.0%, butyrate accounted for 16.7 ± 1.1%–16.4 ± 1.5% of the SCFA mixture, while isobutyrate was below 1%. The proportion of valerate varied between 6.3 ± 0.9% and 6.6 ± 0.7% and the proportion of isovalerate between 4.5 ± 0.7% and 4.6 ± 0.8%. Although significant time-dependent alterations were observed for nearly all biochemical parameters, these were relatively small and most probable of low biological impact.

The degradation of organic matter did not differ based on treatment, glyphosate concentration or interaction of both (Table 3). A degradation of 51.4–55.9% was observed during the 48 h incubation period. The degradation of crude fat was more variable with 26.0–39.5% (Table 3). The degradation of crude protein reached between 70.5 and 72.0% and between 38.5 and 44.4% of crude fibre were degraded (Table 3). Within the fibre fraction, 41.9–48.1% of NDF and 32.7–38.4% of ADF were degraded (Table 3). None of these parameters was significantly altered by treatment, glyphosate concentration or an interaction of both (Table 3).

Rumen microorganisms produced about 1 g of protein per day in each fermenter. Most of this protein (781.8 ± 66.5 mg/d) was produced by the LAM (Figure 2A), while SAM produced only about 225.0 ± 41.7 mg/d (Figure 2B). For both, LAM and SAM, more than 80% of N for microbial protein synthesis was derived from the NH₃-pool (Figures 2C,D). The efficiency of microbial protein synthesis (Figure 2E) reached about 158.0 ± 35.5 g/kg degraded dry matter. None of the parameters for microbial protein synthesis was affected by treatment, glyphosate concentration or interaction of both.

To investigate any shifts in taxonomic structure of the microbiome associated with treatment with glyphosate or its formulations Durano and Roundup, 16S rRNA gene profiling was performed. The mean number of taxa based on amplicon sequencing variants (ASVs) was 508 ± 105 for LAM (Figure 3A) and 468 ± 70 for SAM (Figure 3B). There was no significant difference (LAM: p = 0.887, SAM: p = 0.874, Kruskal-Wallis test) in the number of taxa present for either the LAM or SAM when treated with glyphosate or a glyphosate formulation. A decrease in alpha-diversity, a combination of the number of taxa present in a sample and their distribution is often a sign of dysbiosis. Therefore, alpha-diversity based on Shannon-Effective was calculated (Figures 3C,D), but no significant change in Shannon-Effective between the control samples and those cultivated with different concentrations of glyphosate or a glyphosate formulation (LAM: p = 0.767, SAM: p = 0.651, Kruskal-Wallis test) was observed. To investigate if there was a global shift in taxonomic distribution between the control samples and the glyphosate and glyphosate formulation exposed microbial community’s, beta-diversity analysis based on principal component analysis was done. No clustering was observed between the control communities and those exposed to glyphosate at different concentrations of glyphosate. Neither for glyphosate, Durano nor Roundup exposure at the concentrations of 0.1 mg/l, 1 mg/l or 10 mg/l were significant differences in global community structure observed in the LAM (Figures 3E–G) or SAM (Figures 3H–J) communities.

On the taxonomic level of family, the most abundant families in the non-treated LAM communities were from the Firmicutes Ruminococcaceae (15.1 ± 4.3%, mean ± SD), Lachnospiraceae (12.9 ± 3.6%) and Veillonellaceae (6.8 ± 2.1%), and from the Bacteroidetes the families Prevotellaceae (15.2 ± 3.6%), Rikenellaceae (4.9 ± 1.1%) and the F082 (2.6 ± 0.8%). Other abundant families included Succinivibrionaceae (5.4 ± 6.6%), Spirochaetaceae (4.7 ± 2.2%), Bifidobacteriaceae (2 ± 2.2%) and the archaea family Methanobacteriaceae (2.9 ± 1.6%; Figure 4A). The SAM non-treated community exhibited a different distribution in bacterial families than the LAM. Here, the most abundant families were Lachnospiraceae (32.6 ± 3.2%), Prevotellaceae (15.6 ± 2.8%) and Veillonellaceae (10.8 ± 1.2%), Ruminococcaceae (7.5 ± 2.5%), Lactobacillaceae (6.6 ± 2.7%), Spirochaetaceae (6.5 ± 1.6%), Bifidobacteriaceae (4.8 ± 1.8%) and the archaeal family Methanobacteriaceae (2 ± 0.6%; Figure 4B). No significant changes in relative abundance of any bacterial family were observed in either LAM or SAM communities with treatment with glyphosate, Durano or Roundup.

To investigate if changes in taxa distribution could only be observed at higher taxonomic resolution, an analysis on genus level was conducted. In total, we identified 155 genera in LAM (Figure 4C) and 121 genera in SAM (Figure 4D) with a relative abundance of at least 0.01% in one sample. No significant change in abundance was observed for any single genus in either LAM or SAM treated with glyphosate or a glyphosate formulation.

Metaproteomics was performed to analyse if glyphosate or glyphosate formulation exposure affected the functions exhibited by the LAM or SAM communities. In the LAM communities, a total of 8,052 microbial protein groups were detected, whereas in the SAM communities, 5,992 microbial protein groups were detected. To determine global metaproteome wide significant changes associated with glyphosate exposure or its formulations, principle component analysis was performed with the relative abundance of protein groups. PERMANOVA was used to determined significance. No global significant differences between the untreated control LAM (Figures 5A–C) or the untreated SAM communities (Figures 5D–F) with those treated with glyphosate, Durano or Roundup were observed. To investigate possible specific functional changes, protein groups were annotated to Kyoto Encyclopedia of Genes and Genomes (KEGG)³ (Ogata et al., 1999) metabolic pathways and the relative abundances summed and compared between the treatment groups. In total, 134 KEGG metabolic pathways with a functional coverage of at least 10% were identified in LAM (Figure 5G), while 106 were seen in SAM (Figure 5H). Glyphosate, Durano or Roundup treatment at any of the three concentrations (0.1 mg/l, 1 mg/l and 10 mg/l) did not significantly enrich any KEGG functional pathway when compared to the untreated controls. The KEGG pathway map 00400 was investigated more closely since it included the shikimate pathways and the 5-enolpyruvylshikimate-3-phosphate (EPSP) synthase, which is the specific target for glyphosate. We identified 16 KEGG protein functions including the EPSP synthase in the metaproteomes of LAM and SAM (Figure 6A), but none of them were significantly altered depending on the abundance by glyphosate, Durano or Roundup treatment.

One potential reason for the lack of changes in the taxonomic or functional community structure of LAM or SAM observed after glyphosate, or glyphosate formulation treatment, could be a potential degradation of glyphosate in the microbial community. To control for this, we quantified glyphosate in the bioreactor supernatant by targeted multi-reaction monitoring mass spectrometry. As expected, in the untreated control samples, no glyphosate was detected (Figure 6B). The spiked in concentration in the bioreactor medium was 10 mg/l, 1 mg/l and 0.1 mg/l. In the glyphosate or glyphosate formulation treated bioreactors, we detected somewhat lower levels of glyphosate than expected. The concentrations in the bioreactors with pure glyphosate were 6.414 mg/l ± 0.438 mg/l, 0.717 mg/l ± 0.105 mg/l and 0.068 mg/l ± 0.009 mg/l. In the bioreactors with Roundup, the measured concentrations were 8.478 mg/l ± 0.753 mg/l, 0.791 mg/l ± 0.088 mg/l and 0.074 mg/l ± 0.011 mg/l. In the bioreactors containing Durano, glyphosate concentrations after the run were measured at 8.179 mg/l ± 1.152 mg/l, 0.791 mg/l ± 0.088 mg/l and 0.074 mg/l ± 0.011 mg/l. These determined concentrations values were consistent with no or little degradation of the glyphosate when taking the accuracy tolerance (78% relative recovery) for the glyphosate quantification method into account. This suggested that the majority of glyphosate was still present in the medium. To reveal potential correlations among glyphosate concentration in the fermenter, microbial abundances and functions, the measured glyphosate values were correlated with the relative abundance of the microbial genera detected by 16S rRNA gene profiling and the relative abundance of KEGG metabolic pathways in the metaproteomics study. No significant positive or negative correlations between the concentration of glyphosate and the abundance of genera or KEGG pathways in the LAM or SAM were identified (data not shown). In addition, the relative abundances of protein functions of the KEGG 00400 metabolic pathway map (aromatic amino acid biosynthesis) identified in LAM and SAM were not significantly positively or negatively associated with the glyphosate concentration (data not shown).