The rumen microbiome inhibits methane formation through dietary choline supplementation

Enteric fermentation from ruminants is a primary source of anthropogenic methane emission. This study aims to add another approach for methane mitigation by manipulation of the rumen microbiome. Effects of choline supplementation on methane formation were quantified in vitro using the Rumen Simulation Technique. Supplementing 200 mM of choline chloride or choline bicarbonate reduced methane emissions by 97–100% after 15 days. Associated with the reduction of methane formation, metabolomics analysis revealed high post-treatment concentrations of ethanol, which likely served as a major hydrogen sink. Metagenome sequencing showed that the methanogen community was almost entirely lost, and choline-utilizing bacteria that can produce either lactate, ethanol or formate as hydrogen sinks were enriched. The taxa most strongly associated with methane mitigation were Megasphaera elsdenii and Denitrobacterium detoxificans, both capable of consuming lactate, which is an intermediate product and hydrogen sink. Accordingly, choline metabolism promoted the capability of bacteria to utilize alternative hydrogen sinks leading to a decline of hydrogen as a substrate for methane formation. However, fermentation of fibre and total organic matter could not be fully maintained with choline supplementation, while amino acid deamination and ethanolamine catabolism produced excessive ammonia, which would reduce feed efficiency and adversely affect live animal performance.

. The MMC rely on a simplified methanogenesis pathway that utilizes methylated compounds, such as methylamines to generate energy, requiring only one mole of H 2 per mole of CH 4 6 . Therefore, they were predicted to have a lower threshold for dissolved H 2 than other rumen methanogens 7 . As a consequence, they would be able to survive in situations of low H 2 concentration which would be unfavourable for ATP production in other methanogens. Theoretically, methylotrophic methanogenesis would out-compete hydrogenotrophic methanogenesis at a low H 2 concentration, as MMC would consume H 2 and reduce dissolved H 2 to a level that does not meet the thermodynamic conditions required to produce ATP by hydrogenotrophic methanogens 8 . Despite these thermodynamic and H 2 threshold advantages, and temporary H 2 limiting situations in certain periods of the feeding cycle, Methanobacteriales remains the dominant methanogen order in the rumen. Possible reasons for failure of MMC to dominate the rumen methanogen niche might be that MMC divert less energy towards ATP production from methanogenesis than other methanogens, or that their growth is limited by the insufficient availability of methylated substrates, such as mono-(MMA), di-(DMA) and trimethylamines (TMAs). The present study aimed to exploit this limitation and to use it to drive changes in the rumen microbiome towards a simplified rumen methanogen population with MMC as the dominant order by providing an abundant supply of either methylated substrate or their precursors as selection pressure. Such a simplified rumen methanogen population may be more vulnerable to CH 4 mitigation strategies that have previously failed due to the high adaptability of a diverse methanogen population.
A preliminary experiment carried out to test the effects of different methylated compounds demonstrated that, while DMA and TMA significantly enhanced the MMC population, MMC did not outcompete other methanogens (Supplementary information and Fig. S1). Unexpectedly, it was found that the methylamine precursor choline 9 , caused a strong inhibition of CH 4 production. Choline is a registered feed supplement and can be added to ruminant diets. It has been used before in a rumen-protected form to enhance milk yield, whereas choline undergoes extensive degradation within the rumen environment when provided in a non-protected form 10,11 . Consequently, a second aim of the present study was to investigate the underlying mechanism of CH 4 mitigation imposed by choline. Experiments were carried out using the in vitro system Rumen Simulation Technique (Rusitec) 12 . To establish a dose-response relationship using choline and differences in efficiency between choline compounds, a dose-response and a main experiment were carried out. In the main experiment, in depth metagenomics and metabolomics approaches were applied for the identification of the cholinemediated mechanisms affecting the ruminal microbiome associated with methanogenesis the use of alternative H 2 utilization pathways.

Results
Effect of methylated substrates on CH 4 production and MMC abundance (preliminary experiment). Supplementing of two of the three methylated substrates (MMA and TMA) and the TMA precursors betaine and ChCl numerically enhanced CH 4 production initially compared to control, whereas this was different with DMA (Fig. 1a). However, ChCl and the mixed treatment caused a continuous decline from day 5 to near complete termination of CH4 production on day 10. The MMA and TMA treatments numerically reduced the relative abundance of MMC to archaea (Fig. 1b). Both treatments also numerically reduced the relative abundance of archaea to bacteria. All other methylated substrates numerically enhanced the relative abundance of MMC to archaea when provided alone. Although MMC was increased to above 70% of archaeal population by DMA and TMA, this was not sufficient to remove all other methanogens within the 10 days of supplementation.
Response to choline dosage in CH 4 production and NH 3 concentration (dose-response experiment). A supplementation of ChCl at 6.5, 13, 26, 39, 52, and 100 mM together with 10 μM CoM stimulated the CH 4 production in a polynomial curve (Fig. 2). At 200 mM, the CH 4 production was reduced to a level below the detection limit of the gas chromatograph. There was a large increase in the NH 3 concentration of the incubation liquid when the ChCl dosage was increased.

Effect of choline on methanogenesis and ruminal fermentation (main experiment). A sup-
plementation of ChCl at 200 mM initially tended (P < 0.10) to increase CH 4 production compared to the control, but decreased it from day 6 onwards (Fig. 3a). On day 15, ChCl and ChHCO 3 reduced the CH 4 production to 2.1% and 3.5% of control respectively (P < 0.001). In three of the four replicates, there was no detectable CH 4 production from day 12 onwards in ChCl treatment (Supplementary Table S1), while ChHCO 3 was close but unable to reduce CH 4 production to 0 mmol/day by the end of the experiment. Both ChCl and ChHCO 3 increased the level of H 2 accumulated by 7.1-fold and 16.9-fold of control, respectively (Fig. 3b and Supplementary Table S2). A fruity smell of the incubation liquid could be detected with both ChCl and ChHCO 3 treatments, which may hint at production aromatic gas such as ethylene. The average pH of the incubation fluid differed (P < 0.001) between treatments from day 6 to day 15 ( Table 1). The choline treatments also increased (P < 0.05) total VFA concentration, which was associated with an increased (P < 0.05) acetate proportion, whereas proportions of propionate and valerate decreased (P < 0.05). A reduction (P < 0.05) in both in vitro ruminal organic matter digestibility and neutral-detergent fibre digestibility was also observed. The NH 3 concentration of the incubation liquid and the amount of N supplied and recovered in NH 3 increased (P < 0.05) to about 30-and 20-fold higher than values found in control. From the methylated compounds detected in the incubation liquid on d15, only choline and TMA, but not MMA and DMA were substantially elevated (P < 0.05) by the choline treatments (Table 2). More than 90% of the choline was depleted, with the residual choline being lower by 70% with ChHCO 3 than with ChCl. Ethanol was elevated by 192-fold (P < 0.05) and 153-fold (P < 0.05) of control in ChHCO 3 and ChCl treatment respectively. The compounds with the strongest negative correlation to CH 4 production were of TMA (r = − 0.99, P < 0.001), ethanol (r = − 0.95, P < 0.001) and NH 3  www.nature.com/scientificreports/ propionate concentration most strongly positively correlated with CH 4 production (r = 0.86, P < 0.001). The variables most strong positive correlation with H 2 production were concentrations of formate (r = 0.85, P < 0.001) and succinate (r = 0.84, P < 0.001).

Effect of choline and its chemical form on the rumen microbiome (main experiment).
The effects of two forms of choline supplementation on total bacteria and total protozoa counts were not significant (Table 1). This was different concerning the composition of the microbiome. The changes in α-diversity showed that, compared to the species richness of the rumen fluid used for inoculation, 15 days of treatment with ChCl and ChHCO 3 reduced species number to 21.2 ± 1.6% (mean ± standard error) and 13.4 ± 1.4% respectively, while under control condition 73.8 ± 8.7% of the species could be maintained (Fig. 4a). The Shannon evenness was also altered from 5.65 ± 0.13 (inoculum) to 3.28 ± 0.04 (ChCl), 2.44 ± 0.13 (ChHCO 3 ) and 4.50 ± 0.19 (control) on d15 (Fig. 4b). The β-diversity also illustrated a difference between treatment groups and control (Supplementary Fig. S1). The permutation analysis of variance indicated that microbiome composition differs between groups (P < 0.001). The relative prokaryotic abundance is illustrated in Fig. 5. At the phylum level (Fig. 4a), the ChCl treatment increased the relative abundance of Actinobacteria, while the ChHCO 3 treated microbiome was dominated by Firmicutes, along with an increased proportion of Proteobacteria. A marked reduction of Euryarchaeota was observed in both treatments indicating a decline of the methanogenic archaea. At the order level, the control group showed an increased relative abundance of Lactobacillales, Peptostreptococcales and Veillonellales. The ChCl treated microbiome had an even higher relative abundance of the same orders of microbes and of Coriobacteriales. By contrast, the ChHCO 3 -treated microbiome was dominated by Lactobacillales. The   Table S6 and S7). Both choline treatment groups were additionally pooled in order to identify conservation of differentially abundant species that may have contributed to CH 4 mitigation on d15. This comparison identified 227 less abundant species, including five clusters of Methanobrevibacter and two clusters of MMCs; eight species were identified to be more abundant (Supplementary Table S8). In addition, eight mOTUs clusters were identified by R s to be negatively associated with CH 4 production and 15 mOTUs clusters were positively associated with H 2 production (Table 3). Among the species associated with CH 4 mitigation, M. elsdenii (6.24% average relative abundance) possessed the strongest association, M. elsdenii also correlates with NH 3 (r = 0.801, P < 0.001). Enterococcus gallinarum (4.06% average relative abundance) was most strongly associated with H 2 production.  4 and NH 3 production. Choline was supplemented to artificial saliva at 0, 6.5, 13, 26, 39, 52, 100 and 200 mM for 15 days. The scatterplot displays the CH 4 production in orange and NH 3 concentrations in blue between day 11 and day 15, each dot representing measurement from a single day. The CH 4 production was normalized to that of the 0 mM group. The CH 4 production was fitted with a 3 rd order polynomial regression (y = 6 × 10 -07 x 3 − 0.0003 x 2 + 0.0267 x + 1.1659, R 2 = 0.908), the NH 3 concentrations was fitted with a 2nd order polynomial regression (y = 0.0078 x 2 + 0.5374 x + 9.1634, R 2 = 0.982).

Discussion
Choline has a marked effect on methanogenesis. The results of the preliminary experiment indicated that, although some methyl compounds did enrich MMC, this did not allow MMC to out-compete other methanogens to the point where they were the only methanogen remaining. In fact, the study showed that choline, after at first (day 1) promoting the rumen methanogen population and CH 4 formation at high supplementation level, eventually led to a near complete cessation of the methanogenic activity. Methanogens were negatively affected already from day 2 onwards, as shown by the decline of CH 4 production. In order to ensure that the influence on CH 4 was due to choline itself, two different chemical forms of choline -ChCl and ChHCO 3 -were supplemented. The maximal level of CH 4 reduction achieved at 200 mM was nearly the same with ChCl (98%) and ChHCO 3 (97%). The present study henceforth sought to answer how these phenomena can be explained using results from ruminal fermentation and rumen microbiome composition. According to the results of the hNMR analysis, more than 90% of the choline was utilized ( Table 2) and likely converted to TMA and acetaldehyde by choline TMA-lyase 13 . The substantial increase found in TMA concentration in the incubation with both forms of choline along with CH 4 mitigation might therefore be an indicator of inhibition of methanogenesis. The supplementation of 200 mM TMA did not reduce CH 4 production. Therefore, it is likely that the acetaldehyde as end product of choline metabolism may play an active role leading to the CH 4 mitigation, and choline may act as a selection pressure to encourage the bacteria that can take advantage of it. The group of bacteria possessing choline TMAlyase would be the first to benefit from acetaldehyde. The choline TMA lyase and its activating enzyme have been identified in the differentially abundant mOTUs clusters, including members of the Anaerovoraceae and Olsenella umbonata enriched by ChCl supplementation and Enterococcus avium, Alkaliphilus spp., Proteus mirabilis/vulgaris and unknown Lachnotalea spp. enriched by ChHCO 3 . Accordingly, the two forms of choline stimulated entirely different species capable of degrading choline 13 . E. avium, Proteus mirabilis, unknown Alkaliphilus and unknown Lachnotalea could be the species metabolizing choline in the ChHCO 3 group, and Olsenella umbonata and unknown Anaerovoraceae those metabolizing choline in the ChCl group (Fig. 6). All but Lachnotalea possess eut gene clusters associated with choline utilization via microcompartment 13,14 . Inside microcompartments choline can be metabolized to acetaldehyde and ammonia, the acetaldehyde may subsequently be converted to ethanol and acetate 15 . All of the aforementioned species can produce acetate, but only a subset can produce propionate and butyrate (Table S11). The reductive acetogen Alkaliphilus may have contributed to the elevated acetate concentration in the treatment groups. The two Enterococcus species dominant with the ChHCO 3 treatment are also predicted to be able to metabolize ethanolamine. The organisms that can cleave choline to TMA and acetaldehyde are presented in Fig. 6 along with the average abundance of each species in each condition. Acetaldehyde enters the central carbon metabolism and thus can lead to the production of lactate, succinate, ethanol and formate, which are alternative H 2 -sinks, and influence the downstream microbial crosstalk. Figure 6 illustrates simplified pathways of ruminal degradation of choline and of certain structural carbohydrates (xylan, cellulose, pectin) as well as of the production of ethanol, formate, succinate and VFA along with the predicted capability of species of high abundance and species of interest.
Choline affects ruminal NH 3 formation, which can potentially inhibit CH 4 formation. Theoretically even the very high level of 200 mM choline chloride could be used in livestock nutrition in the European Union, because the responsible organization, the European Food Safety Authority 16 , has not set a limit for this particular supplement in feed in their regulations. However, the NH 3 concentration in the incubation liquid found at 200 mM choline supplementation by far exceeded the critical level for NH 3 toxicity of > 110 mM 17 . The increase in NH 3 is likely to come from choline metabolism, but the total amount of NH 3 -N produced exceeds that added as choline-N, so NH 3 is likely also produced from sources other than choline.
The high level of NH 3 produced by choline treatment may also contribute to the lowering of CH 4 production. At the physiological ruminal pH of 6.5 or lower, almost all NH 3 exists in the form of the NH 4 + ion 18 . NH 3 can pass through the cell membrane and requires cellular H + to form NH 4 + . In methanogens this may divert H + away from methanogenesis 19 . When methanogen cultures were inhibited with 400 mM NH 4 Cl, the cytoplasmic NH 3 concentration ranged from 100 mM to above 200 mM 19 . The ammonia concentration in the choline supplemented Rusitec incubation liquid was well above 400 mM in the current study, which suggests that NH 3 may be acting to cause CH 4 inhibition. Signs of toxicity to the ruminant were observed when ruminal NH 3 exceed 110 mM 20 . The high level of NH 3 registered in the present experiment therefore suggests this treatment should never be carried out in live animals. www.nature.com/scientificreports/ In a previous study 21 , there was increased production of N containing microbial compounds, likely due to the improvement in efficiency of synthesis of such compounds when methanogenesis was inhibited, despite decreased OM digestibility. Also, some species identified in the present study are able to ferment amino acids and produce NH 3 from them, while some are able to reduce nitrate to NH 3 , or catabolize ethanolamine to produce NH 3 22 . A particularly important role in explaining the NH 3 excess in the present study could be attributed to the presence of M. elsdenii (ref_mOTU_v25_01516), its high abundance correlates positively with NH 3 concentration (r = 0.801, P < 0.001). This species has been observed to produce NH 3 nearly as fast as obligatory amino acid fermenting bacteria 23 . In the present study, the N retained by NH 3 exceeded that of the N input from the feeds provided with the nylon bags, which suggest microbial N fixation may have occurred from the N 2 gas used to keep Rusitec anaerobic 24 . This phenomenon has been described previously 25 . It is possible that the inhibition of methanogenesis may increase nitrogenase activity, as was found under rice paddy conditions 26 . However, www.nature.com/scientificreports/ nitrogenase requires 16 mol of ATP to fix one mole of N 2 27 , which makes it inferior in a competitive environment such as the rumen. It is possible to power the nitrogenase by a proton membrane potential via a FixABC membrane complex 28 , but this operon was only predicted to be present in Proteus mirabilis which does not harbour a nitrogenase reductase complex. Instead, nitrogenase reductase has been identified in Prevotella bryantii and Lachnospira multipara 29 , and predicted in M. elsdenii, i.e., members of the Lachnospiraceae, Lachnotalea spp. and Anaerovoracaceae spp. Therefore, it is unlikely the N fixation could be channelled by FixABC and the plausibility of N fixation with high choline supplementation requires further study.
Ethanol was among the metabolites most strongly associated with reduced CH 4 production. Its high concentration in the incubation liquid suggests that this compound is an important alternative H 2 -sink 30 , which could have been a consequence from microbiome adaptation to the absence of methanogenesis. Unlike other alternative H 2 -sinks such as succinate and lactate that are readily converted to propionate by bacteria, ethanol might have been primarily utilized by the methanogens 31,32 . Therefore, the lack of methanogens likely led to the accumulation of ethanol.
Alternative  33 . This Table 3. Metagenomic-based Operational Taxonomic Units (mOTUs) associated with CH 4 mitigation, H 2 production and ethanol concentration as identified by Spearman's Rank correlation coefficient (Rs) (data from main experiment). Spearman R s : Spearman rank correlation coefficient. BH padj: Benjamin Hochberg false discovery rate adjusted p-value. Inoculum was excluded from this analysis (n = 24) as no gas production measurements were available. In case of ethanol concentration, this analysis only includes inoculum and day 15 samples with corresponding hNMR metabolite data (n = 16). Only mOTU clusters that meet the required P < 0.05 and Benjamini-Hochberg padj < 0.05 cutoff are presented. a Unique mOTU ID within mOTUs database (https:// motu-tool. org/).  38 . Among these organisms, M. eldsenii and D. detoxificans were negatively correlated with CH 4 production in the present experiment. Furthermore, D. detoxificans gains energy by oxidizing nitrogenous compounds such as nitroethane, 2-nitroalcohol and 3-nitro-1-propionate, using trimethylamine from choline metabolism as electron acceptor 38 . These nitrogenous compounds can be accumulated in forages, in particular legumes such as the alfalfa used in the basal diet in the present study and be made available within the rumen 39 . The N compounds mentioned are known to act as methanogen inhibitors 40 , and thus may have contributed to the CH 4 inhibition observed in the present study.
Although lactate and succinate were not detected in high concentrations in the incubation liquid of the current study, the high abundance of lactate producing lactic acid bacteria (LAB), which led the lactate consuming M. elsdenii to prosper. M. elsdenii is negatively correlated to CH 4 production, which suggest strongly that lactate were a prominent H 2 -sink 41 in the present study as well. Both lactate and succinate are intermediate metabolites that can be readily converted to propionate 42 and the production of propionate could compete with CH 4 production 43 . However, a significant decrease of propionate was observed in the choline treatments. In previous batch cultures, inhibition of CH 4 was accompanied by metabolic H redirected from acetate to propionate; however, continuous cultures like Rusitec behave differently: when CH 4 was mitigated by > 50%, generally no overall metabolic H redirection to propionate or butyrate had been observed 30,44 . It was speculated 29 that the H was diverted to other H 2 sinks or microbial cell mass. This means that lactate may not have been primarily converted to propionate in the present study. Lactate can also be oxidized to pyruvate by an NAD-independent lactate dehydrogenase 45 connected to electron bifurcation from electron-transferring flavoprotein Etf, butyryl-CoA dehydrogenase Bcd and ferredoxin/flavodoxin-NAD + reductase Rnf complex 46,47 . All of these enzymes are present in M. elsdenii, which may then use the pyruvate to increase microbial cell mass 41 . The increased abundance of LAB and radical decline of Bacteroidetes in the ChHCO 3 treatment is similar to what has been observed during rumen acidosis, where the pH sensitive Gram negative Bacteroidetes perish 48 . However, instead of acidosis, the www.nature.com/scientificreports/ pH increased in the ChHCO 3 treatment due to the buffering capacity of bicarbonate, which suggests that there is more than pH that caused the Bacteroidetes to perish, perhaps an inhibition by LAB 49 . The production of CH 4 from H 2 by methanogens prevents H 2 accumulation and thereby avoids inhibition of fermentation of nutrients via negative feedback loops, especially of fibre where most H 2 is produced. Therefore, the inhibition of CH 4 production by choline metabolism was expected to have a negative effect on ruminal nutrient degradation as observed earlier in continuous culture experiments 30 . Hydrolysis of plant structural carbohydrates xylan, cellulose and pectin releases hexoses, which are metabolized via the glycolytic pathways to produces pyruvate, a branching point to acetate, propionate, butyrate, lactate, formate or ethanol production 44 . The individual steps in the glycolytic pathway are not affected by the increased H 2 partial pressure, but the regeneration of NAD + required for glycolysis is negatively impacted 50 , and organisms may be driven to use alternative H 2 -incorporating reactions, such as succinate, lactate and ethanol production which directly regenerates NAD + . In this way, the microbiome is able to adapt, and the surviving microbiome likely harbours alternative H 2 -incorporating pathways such as lactate or succinate-mediated propionate production 32,47 . This allows fermentation to continue but at a reduced capacity 51 .
Assuming the H 2 concentration between liquid phase and gas phase is in equilibrium according to Henry's law, both ChCl and ChHCO 3 in fact increased H 2 partial pressure (7.1-fold and 16.9-fold H 2 accumulation compared to control, respectively), which governs the Gibbs free energy (∆G) of VFA production 43 . Furthermore, the ∆G must be greater than the minimum amount of energy required for ATP production for the reaction to be viable in bacteria. Therefore, the surviving microbiome after choline treatment is likely capable of decoupling energy production from H 2 partial pressure by various means, including usage of alternative H 2 -sinks.
The richness of the microbiome in the control group after 15 days of operation indicates that Rusitec is a good simulation system for the rumen prokaryotes. The microbiome revealed a decline of Euryarchaeota, i.e., the methanogens, in the choline treatment groups, which corresponds to the reduced CH 4 formation. Some methanogens have syntrophic interaction with specific H 2 producers via adhesins 52 . The reduced alpha diversity suggests that syntrophic interaction may have been broken, which could contribute to the reduced CH 4 production. All of the most differentially abundant taxa found in the present study are either able to utilize H 2 and produce metabolites such as lactate and ethanol, or they are able to make use of the alternate H 2 -sink metabolites produced by other bacteria. It is unknown whether the use of alternative H 2 sinks is the result of CH 4 reduction, or a contributor to CH 4 reduction.

Conclusion
As a model, treatment with choline, especially in the form of choline chloride, has demonstrated a new way to inhibit methanogenesis and to reduce CH 4 to below the detection limit for in vitro continuous culture systems. This treatment could be used to study how the energy, otherwise lost through CH 4 production could be redirected, and how rumen fermentation takes place in the absence of methanogenesis. This treatment reduces digestibility and massively enhances ruminal ammonia concentration, and is thus not suitable to be carried out in live animals for animal health and welfare reasons.

Methods
Experimental design. All methods were performed in accordance with the relevant guidelines and regulations.
The preliminary experiment was designed to investigate which methyl-compound enhanced the MMC community most strongly. Different methylated substrates, namely MMA, DMA and TMA, as well as the TMA precursors betaine and choline (provided as chloride (ChCl)), all from Sigma-Aldrich (St. Louis, MO, USA), were supplemented at 200 mM (equivalent to 750 g/kg basal diet supplied at 15 g dry matter/day) to the artificial saliva flowing into the Rusitec. In addition, an aliquot of an equimolar mixture of all five supplements was included at 100 mM each. The duration of the experiment was 10 days.
In the dosage response experiment, the dosage of ChCl required to achieve CH 4 reduction was investigated during 15 days in a dose-response design within one Rusitec run. The ChCl was supplemented via artificial saliva at 0 (control), 6.5, 13, 26, 39, 52, 100, and 200 mM, by continuous flow into the Rusitec.
In the main experiment, three treatments were investigated in four 15-day Rusitec runs in a complete, randomized design. Treatment 1, control: 0 mM ChCl, Treatment 2: 200 mM ChCl, Treatment 3: 200 mM choline bicarbonate (ChHCO 3 ) were supplemented with the artificial saliva in four replicates each. In all experiments, artificial saliva was supplemented with 10 µM coenzyme M (CoM; Sigma-Aldrich), as MMC relies on external supply of CoM to grow 6 and a deficiency thereof might have effects on methanogenesis not related to the supply with methylated substrate.
Origin of the rumen fluid. The starting rumen fluid was collected from two available lactating rumencannulated Brown Swiss cows. Cow 1 was fed ryegrass hay ad libitum and concentrate (1 kg/day), cow 2 was fed hay from a biodiverse meadow ad libitum. The rumen fluid was always collected at 07:00 a.m. just prior to refilling hay troughs and offering concentrate (cow 1 only). Procedures imposed on the rumen-fluid donor animals in the present study were approved by the Committee on Animal Experimentation (Ethics Committee) of the Cantonal Veterinary Offices of Zurich (Licence no. ZH 38/14; cow 1) and Berne (Licence no. VB BE 20/17; cow 2).The rumen fluid from each cow was separately used as starting inoculum in two of the four runs each in the main experiment to be able to offer two biological replicates each. Rumen fluid from cow 1 was used for both preliminary experiments. The rumen fluid was kept warm in a thermos flask during transport and inoculation took place within 2 h after rumen fluid harvest. The rumen fluid was strained through four layers of medical gauze (pore size 1 mm) prior to transfer into the Rusitec vessels.  53 . The artificial saliva 12 had a composition ensuring a continuous supply of substrates required for optimal fermentation. The artificial saliva was sterilized by passing a 0.2 μm filter and stored in 10 L Nalgene autoclavable carboy (Thermo Fisher Scientific, Waltham, MA, USA). All Tygon tubes connecting the artificial saliva to the fermenter were also sterilized prior to the experiment. The overflown incubation liquid was collected in flasks to measure flow rate and immediately frozen at − 20 °C to terminate fermentation. To simulate the rumen washout effect, the average artificial saliva flow rate was 403 mL/day, equivalent to a dilution rate of 40.3%/day.

Sample collection.
Incubation liquid samples were taken daily 3 h prior to feed bag exchange to assess pH, NH 3 concentration and VFA content. A portion of the incubation liquid was centrifuged for 5 min at 4000g, the supernatant was stored at − 20 °C for later high performance liquid chromatography (HPLC) and proton nuclear magnetic resonance (hNMR) analysis. After 48 h of incubation, respectively, the feed bags were processed for subsequent nutrient analysis as described by Soliva et al. 53 , detergent fibre fractions were assessed by Fibertherm system FT 12 (Gerhardt GmbH & Co. KG, Koenigswinter, Germany) as described by Terranova et al. 54 . The fermentation gases were collected during 24 h periods in gas-tight aluminium bags (TECOBAG 8 L, PETP/AL/ PE-12/12/75 quality; Tesserau Container, Bürstadt, Germany). A portion of the initial rumen fluid inoculum and the subsequent incubation liquid were snap frozen in liquid nitrogen and stored at − 80 °C for microbiome assessment via metagenomics.

Incubation liquid and fermentation gas analysis. Protozoal and bacterial counts in the incubation
liquid were obtained daily with Neubauer haemocytometers (0.1 and 0.02 mm depth, respectively; Blau-Brand, Wertheim, Germany) following the manufacturer's recommendation. The pH and NH 3 concentration were measured by corresponding electrodes (Unitrode easyClean Pt1000 and NH 3 -selective gas membrane electrode) connected to a pH meter (model 713; Methrom, Herisau, Switzerland). The concentration of VFA was analysed using HPLC (System Hitachi Lachrom; Merck, Tokyo, Japan) following the procedure of Ehrlich et al. 55 . Various metabolites were identified and quantified by NMR (Table 2). Samples were processed by filtration via Nanosep 3 k Omega (Pall, Port Washington, NY, USA), with 3 kDa cut-off to remove protein molecules. An amount of 440 μL of sample was mixed with 100 μL of NaHPO 4 buffer (1 M, pH 7), and 60 μL of 5 mM sodium trimethylsilylpropionate-d 4 (TSP) (Armar AG, Döttingen, Switzerland) in deuterated water (D 2 O) was used as internal standard. All NMR experiments were performed at 25 °C on a 600 MHz Bruker Avance III HD spectrometer equipped with a Prodigy triple-resonance probe with z-gradient. Quantitative 1 H spectra were recorded using a 1D-NOESY sequence (τ mix = 10 ms) with presaturation of the water resonance during the relaxation delay. The relaxation delay was 7.5 s and the CW presaturation field strength was set to 30 Hz. The acquisition time was 5 s. The spectral width was 22 ppm centred on the water signal at 4.7 ppm. After 8 dummy scans, 512 scans with 131,072 total data points were accumulated for each spectrum. All spectra were processed with MestReNova14 (Mestrelab Research S. L.). Prior to Fourier transformation the time domain was extended to twice its size by zero-filling and multiplied with an exponential function (LB = 0.15 Hz). The baseline of the resulting spectra was corrected with a polynomial of 3rd order. Metabolites were quantified by comparing their integrals to the integral of the internal standard. The integration method was set to "sum". For metabolites with more than one DNA sequencing and data processing. The microbiome was assessed by metagenomic sequencing.
A total of 1089 metagenome assembled genomes (MAGs) were reconstructed. The relative abundance of the prokaryotes was quantified by mOTUs2 profiler 59 using species level clusters of metagenomic-based Operational Taxonomic Units (mOTUs). The validity of the sequencing pipeline was validated by the Zymo Microbial Community DNA Standard (Supplementary Table S10). The extracted DNA was analysed using the sample library prepared by Illumina Truseq Nano, and sequenced on Novaseq SP 300 cycles Flowcell by Illumina Novaseq 6000 (Illumina, San Diego, CA, USA). Twenty-eight samples were sequenced in the present study generating between 26 and 85 M 150-base-pair paired-end reads per sample. All raw sequences can be accessed through the NCBI at BioProject PRJEB43305, and the code for all the analyses were detailed in submitted publication by Paoli et al. 60 .
The sequencing reads from all metagenomes were quality filtered using BBMap 61 (v.38.71. Available from: https:// sourc eforge. net/ proje cts/ bbmap/). We first removed adapters from the reads, and then removed reads that mapped to quality control sequences (PhiX genome). We discarded low quality reads by applying the parameters trimq = 14, maq = 20, maxns = 0 and minlength = 45. Reads were then merged using bbmerge.sh with a minimum overlap of 16 bases. The merging step results into merged and unmerged reads that are both used from hereon for all analysis steps. Assembly was performed using metaSPAdes 62 (v3.14) in metagenomic mode. The resulting scaffolded contigs (hereafter scaffolds) were filtered by length (≥ 1000 bp). MAG reconstruction was performed by mapping sequences from all samples against all filtered scaffolds using bwa 63 (v0.7.17-r1188) with the -a flag and alignments were filtered to be at least 45 bases in length, with an identity ≥ 97% and covering ≥ 80% of the query sequence. Alignment files were processed using the jgi_summarize_bam_contig_depth script to create abundance profiles that were used as input for MetaBAT2 64 (v2.12.1). Quality of resulting bins were estimated using checkM 65 (v1.0.13). A total of 1189 bins with a completion ≥ 50% and a contamination < 10% were reported as MAGs and used for downstream analysis. Marker genes from the 1189 MAGs from this study, 410 genomes from the Hungate collection 66 and 4941 publicly available rumen MAGs 67 were extracted using fetchMGs (v1.2, available at http:// motu-tool. org/ fetch MG. html) and 6197 MAGs with ≥ 6 marker genes were used to extend the mOTUs 59 (v2.5) database. 1154 MAGs were added to existing mOTUs and 5043 MAGs created 2311 new mOTUs (Illustration of workflow given in Supplementary Fig. S2). Next, the 28 Rusitec samples were profiled taxonomically using the mOTUsv2 tool in combination with the extended database using default parameters. Gene calling of the 28 Rusitec assemblies and the 6540 MAGs were called using Prodigal 68 (v2.6.3) with the parameters -c -q -m -p meta and -c -q -m -p single respectively. Genes were subsequently clustered at 95% identity, keeping the longest sequence as representative using CD-HIT 69 (v4.8.1) with the parameters-c 0.95 -M 0 -G 0 -aS 0.9 -g 1 -r 0 -d 0. Representative gene sequences were aligned against the KEGG database 70 (release 2020-02-10) using DIAMOND 71 (v0.9.30) and filtered to have a minimum query and subject coverage of 70% and requiring a bitScore of at least 50% of the maximum expected bitScore (reference against itself). The MAGs affiliated under mOTUs of interest were collectively analysed as pangenome by OrthoMCL 72 (v2.0) via Kbase 73 . mOTUs cluster capability prediction. Prediction of capability was based on the presence of predicted genes listed in Table S11.
Statistical evaluation. The statistics program R studio 74 was used for all evaluations other than multiple comparisons, which was carried out in GraphPad Prism 8.0.0 (GraphPad Software, San Diego, California USA, www. graph pad. com). In preliminary experiment 2, the CH 4 production was normalized to that of the control (0 ChCl group). Data from the main experiment were subjected to analysis of variance with choline treatment as fixed effect and Rusitec fermenter as experimental unit. Tukey's method was applied to perform multiple comparisons among treatment means. The mOTUs results were analysed using the vegan package 75 of R studio, and the Richness and Shannon evenness index was calculated to assess α-diversity. The Mann-Whitney-Wilcoxon Test was performed to establish significant differences in population distribution. Constrained principal coordinate analysis based on Bray Curtis dissimilarity was used to assess β-diversity and dimension reduction. Permutation analysis of variance was used to determine the significance of difference between groups. Differential abundance analyses were carried out by DESeq2 76 . The mOTUs clusters satisfying the statistical cutoff of P < 0.05 (Wald-test), the Benjamini-Hochberg false discovery rate adjusted P value (padj) of < 0.05 and a log2 fold change of ≥ 2 were considered differentially abundant. Spearman's Rank correlation coefficient (R s ) was used for associations of mOTUs with CH 4 mitigation and H 2 production. Relation between metabolites and CH 4 mitigation were established as Pearson Correlation coefficients (r). Cutoffs of P < 0.05 and padj < 0.05 were applied to the R s of each mOTUs.