In Vivo Competitions between Fibrobacter succinogenes, Ruminococcus flavefaciens, and Ruminoccus albus in a Gnotobiotic Sheep Model Revealed by Multi-Omic Analyses

Ruminant animals, including cattle and sheep, depend on their rumen microbiota to digest plant biomass and convert it into absorbable energy. Considering that the extent of meat and milk production depends on the efficiency of the microbiota to deconstruct plant cell walls, the functionality of predominant rumen cellulolytic bacteria, Fibrobacter succinogenes, Ruminococcus albus, and Ruminococcus flavefaciens, has been extensively studied in vitro to obtain a better knowledge of how they operate to hydrolyze polysaccharides and ultimately find ways to enhance animal production.

C ellulose typically represents ;40% of plant biomass (1) but is not endogenously degradable by mammalian species, including ruminants, due to a lack of an endogenously encoded ability to hydrolyze this important carbohydrate resource. Mammals are instead dependent on members of their gastrointestinal tract microbiota for providing this capability (2). Ruminant animals, including cattle and sheep, harbor a core rumen microbiome (3), including a limited number of taxa known to specialize in the deconstruction of cellulose (4,5). The three best described and cultivable cellulolytic bacteria, Fibrobacter succinogenes, Ruminococcus albus, and Ruminococcus flavefaciens, each employ unique methods of fulfilling this function. R. flavefaciens facilitates cellulose deconstruction by virtue of a cellulosome (6); R. albus instead appears to utilize noncellulosomal, but coalescing membrane-attached enzymes (7), while the organization of the cellulolytic system of F. succinogenes remains enigmatic, with outer membrane vesicles (OMVs) possibly involved (8,9). Each of these species is frequently detected among a variety of adult ruminants throughout the world and often cooccur in the rumen (3,4). Previous in vitro studies have provided evidence of competition among these cellulolytic species for adhesion to plant biomass and for growth on cellulose substrates (10)(11)(12). However, little is known about their symbiotic relationships in vivo or the antagonistic or synergistic impact of their cooccurrence on fibrolysis. We sought to determine how the three species interact molecularly in vivo within a previously described and tractable gnotobiotic sheep model (13,14) and to evaluate how variations in their composition affect ruminal fibrolysis and metabolism alongside a methanogenic hydrogenotroph.

RESULTS
Fibrobacter succinogenes is outcompeted following inoculation of ruminococci. Fibrobacter succinogenes S85 (week 3), Methanobrevibacter sp. strain 87.7 (week 7), Ruminococcus albus 8, and R. flavefaciens FD1 (week 28) were sequentially inoculated into two gnotobiotic lambs that were raised aseptically from birth to 37 weeks (Fig. 1). Microbial composition dynamics were monitored from weeks 22 to 33 using quantitative PCR (qPCR) (Fig. 2) and most probable number (MPN) enumeration of total cultivable and total cultivable cellulolytic bacteria (see Fig. S1 in the supplemental material). Using qPCR, the total number of bacteria observed throughout the sampling period showed little variation at 10 10.3 6 0.16 16S rDNA copies/g of rumen content (Fig. 2). These numbers did not vary between the two animals (P = 0.5) and were consistent with those measured by MPN (10 10.2 6 0.4 cells/ml) ( Fig. S1A and S1B). qPCR data indicated that F. succinogenes was initially dominant, being enumerated at 10 9.4 6 0.25 16S rDNA copies/g prior to inoculation of the two Ruminococcus strains. Following inoculation with the two Ruminococcus strains, F. succinogenes was gradually outcompeted in both animals. By week 33, R. albus and R. flavefaciens were enumerated at 10 8.8 6 0.2 and 10 5.9 6 0.4 16S rDNA copies/g, respectively, while F. succinogenes was enumerated at 10 4.7 6 0.8 16S rDNA copies/g (Fig. 2). Methanobrevibacter sp. 87.7 was enumerated at 10 5.8 6 0.9 16S rDNA copies/g, and total cultivable methanogens were enumerated by MPN at 10 6.2 6 1.01 cells/ml throughout the sampling period and exhibited noticeable fluctuation over the first 3 weeks following inoculation of the Ruminococcus spp. (Fig. S1C). Consistent with both qPCR and MPN data, most metagenomic and metatranscriptomic reads aligned to the F. succinogenes S85 reference genome at week 27 with only low to modest alignment to the Methanobrevibacter sp. 87.7 and the two Ruminococcus genomes (Fig. S2A). F. succinogenes was also well represented in metagenomic, but not metatranscriptomic, data from weeks 29 and 31 and obtained almost no coverage by week 33. This transition, was also seen in the increasing alignment of these data to the R. albus and R. flavefaciens genomes between weeks 29 and 33 (Fig. S2A).
Gene expression. Carbohydrate-active enzymes (CAZymes) and accessory systems putatively involved in bacterial adhesion or polysaccharide utilization represented 11%, 21%, and 25% of the total transcriptomes of F. succinogenes, R. albus, and R. flavefaciens, respectively (Fig. 3), while Methanobrevibacter sp. 87.7 expressed mostly genes involved in methanogenesis and redox systems (on average 5.5% of total gene expression) (Fig. 3). The expression of CAZymes by each cellulolytic species reflected each bacterium's relative abundance in each sample ( Fig. S2A and S2B). Although the inoculated cellulolytic strains each encoded approximately 100 CAZyme genes (see http:// www.cazy.org/), only a small subset of those genes were expressed at .1% of their total CAZy gene expression ( Fig. 4; see Table S1 in the supplemental material). The 15 most highly expressed CAZyme genes from each of the three cellulolytic strains across all time points were genes targeting cellulose (glycoside hydrolase 5 [GH5], GH8, GH9, GH48, and GH51), hemicelluloses (GH10, GH11, GH26, GH43, and carbohydrate esterase 4 [CE4]) and pectins (polysaccharide lyase 11 [PL11] and CE12). Notably, each of R. albus' and R. flavefaciens' 15 most expressed CAZyme transcripts encoded proteins harboring either a family 37 carbohydrate-binding module or a dockerin, respectively (Fig. 4). A more diverse carbohydrate transport and metabolism gene expression profile was measured when the ecosystem harbored high numbers of F. succinogenes S85 (period 1) compared to that observed following the introduction of the ruminococci (period 2) (Fig. 5). R. albus and R. flavefaciens were observed to most abundantly FIG 2 Dynamics of total bacteria and F. succinogenes, R. albus, and R. flavefaciens. The numbers of 16S copies/gram of rumen material evaluated by quantitative PCR in rumen contents pre-and postinoculation of the ruminococci (at week 28) are shown.

FIG 1
Overview of animal trial. The lambs were initially fed sterilized cow milk (until weaning at week 14) and then sterilized lucerne hay pellets. They were inoculated with F. succinogenes S85 (week 3), Methanobrevibacter (Mbb.) sp. 87.7 (week 7), and R. albus 8 and R. flavefaciens FD1 (week 28). Periods 1 and 2 correspond to pre-and postinoculation of the ruminococci, respectively. Sampling of rumen contents during both periods are indicated by black triangles. In sacco incubations of lucerne and wheat straw cell walls are indicated by green and yellow rectangles, respectively. express GH9 and GH48 family mRNAs and their inoculation resulted in a dramatic increase in the level of these transcripts that corresponded to decreases in GH2 and GH43 mRNAs (Fig. 5). In addition to CAZymes, F. succinogenes exhibited high expression of genes predicted to be involved in the production of extracellular structures, including outer membrane vesicles (OMVs) (4% of total expression). R. albus 8 expressed type IV pilus genes (14% of total gene expression), and R. flavefaciens FD1 expressed both cellulosomal and type IV pilus genes (12 and 4% of total expression, respectively) (Table S1). Fibro-slime domain containing genes mostly exhibited only low to moderate expression by F. succinogenes (0.35% of total gene expression), the exception being FSU_RS11870 (FSU_2502) that was expressed at ;1,000 reads per kilobase per million (RPKM) initially prior to the introduction of the ruminococci (Table S1).
Adhesion to plant biomass. Using qPCR, we further sought to determine any preferential binding of the three inoculated cellulolytic organisms to plant biomass. The distribution of the three species adherent to cannula-inserted in sacco samples of either lucerne (alfalfa) or wheat straw reflected their measured abundances in the rumen content with no obvious preference of each cellulolytic species to one or other substrate (Fig. S3). Consistently, scanning electron microscopy (SEM) analyses of both substrates revealed high densities of adherent rod morphotypes consistent with F. succinogenes' cell morphology (Fig. S4) prior to the inoculation of the two cellulolytic ruminococci. After inoculation of R. albus and R. flavefaciens, adherent cocci were observed consistent with the Ruminococcus cell morphology (Fig. S4). F. succinogenes-like morphotypes remained evident in these later samples although they were never observed alongside Ruminococcus-like morphotypes, suggesting spatial partitioning.  Table S1 in the supplemental material, in addition to the calculations of the percentages of each group of expressed genes within each genome. These percentages were used to generate the four histograms presented here.  Table S1. CAZyme transcripts of F. succinogenes S85, R. albus 8, and R. flavefaciens FD-1 were sorted (Continued on next page) In Vivo Competitions among Rumen Cellulolytic Bacteria ® Fibrolytic activity. Zymogram profiles revealed carboxymethyl cellulase (CMCase) and xylanase enzyme profiles to be consistent with those of a pure culture of F. succinogenes initially at week 27 but to be more similar to pure R. albus and R. flavefaciens culture profiles at week 33 (Fig. 6). Total rumen CMCase and xylanase specific activity increased 2.6-fold during this same period (Table 1). However, increases in these fibrolytic activities were not reflected by in sacco experiments, where no significant increase was seen in the rates of dry matter (DM), neutral detergent fiber (NDF), or acid detergent fiber (ADF) disappearance of either lucerne or wheat straw (P . 0.05) ( Table 2).
Rumen metabolomics. The rumen metabolome (Table S2) varied by animal and sample day ( Fig. 7A and B). Profiles obtained before (week [wk] 27 [initial]) and immediately following the addition of the ruminococci (wk 30 [transitory]) revealed five discriminant metabolites that included a decrease in the free sugar's glucopyranose and galactofuranose and the cellulose-derived keto acid levulinic acid (Fig. 7C). As the ruminococci outcompeted F. succinogenes (wk 30 versus wk 33 [final]), 11 discriminant metabolites were observed, including decreases in ferulic acid and increases in several phenolic metabolites (Fig. 7D). Comparisons of initial and final time points revealed nine discriminant metabolites (Fig. 7E), including an increase in 3-(3-hydroxyphenyl) propionic acid and a decrease in oxalic acid. Targeted metabolomic analyses of fermentation end products revealed that acetate and butyrate increased in their relative abundance following establishment of the ruminococci at the expense of propionate (Fig. S5).

DISCUSSION
The functional ecology of the major cellulolytic bacteria of the rumen was studied for the first time in vivo within a tractable animal model using complementary molecular, microbiological, enzymatic, and nutritional techniques. The animal model harbored in its rumen an immature microbiota that was limited to those microbes acquired during the first 17 h of contact with the mother (see Text S1 in the supplemental material). The lack of cultivable cellulolytic and methanogenic communities and measurable activities in our animal model allowed us to control the cellulolytic and methanogenic populations in a sequential manner by inoculating specific strains harboring these functions and then observe competition among three major rumen cellulolytic species. Multiple analyses revealed that R. albus became numerically, transcriptionally, and enzymatically dominant within 4 weeks of inoculation. This observation agrees with several in vitro coculture experiments that have also shown that R. albus was the most competitive in initial adhesion to, and growth, on cellulose (10,12,15). The high adherence of R. albus has been described in vitro to be the result of cell-surface-located type IV pili (16)(17)(18), CBM37-containing glycoside hydrolases (19,20), and a glycocalyx (21). We observed high expression of type IV pilus genes by both R. albus and R. flavefaciens supporting an important role for these extracellular appendages in adhesion to cellulose in vivo (16)(17)(18)22). Our data also support the importance of CBM37-harboring CAZymes by R. albus and cellulosome by R. flavefaciens for fiber breakdown in vivo with GH9 and GH48 family enzymes appearing to be preferred for ruminococcal cellulolytic function (7,19,20,22).
R. albus has also been shown to produce in vitro a bacteriocin active against R. flavefaciens but not against F. succinogenes (23,24). The gene encoding this bacteriocin in R. albus 8 has been described (25), and we found that it is very poorly transcribed in our animal model whatever the period (data not shown). Hence, bacteriocin production by R. albus cannot explain the fact that R. flavefaciens did not establish at high levels. R. flavefaciens was, however, able to numerically outcompete F. succinogenes. FIG 4 Legend (Continued) in Table S1 from the highest to lowest expression counts (expressed in RPKM) using the sum of counts of each transcript in the two animals (421 and 455) at the three sampling times. For each gene, the encoded protein identifier (ID) is given along with its modular composition (dbCAN hits) and complementary annotation provided on the CAZy site (http://www.cazy.org/) or determined by manual annotation using BLASTP on NCBI (https://blast.ncbi.nlm.nih.gov/Blast.cgi).
Yeoman et al.  (26,27), most studies performed with conventional sheep have instead found F. succinogenes to outnumber both Ruminococcus spp. (28)(29)(30)(31)(32). The difference between in vitro studies, our animal model, and these studies on conventional sheep may be attributable to differences in the broader ecology, where a newborn immature microbiota could be more ecologically favorable to the development of R. albus and R. flavefaciens, while a conventional adult microbiota may better favor F. succinogenes through unmeasured ecological interactions. This assumption agrees with studies of newborn calves and lambs in which cellulolytic ruminococci colonized the rumen earlier than F. succinogenes (33,34). Diet may also influence the composition of both the cellulolytic and broader community (32,(35)(36)(37).
F. succinogenes expressed a more diverse CAZyme repertoire that included GH8, GH51, GH5, and GH9 family glycoside hydrolases. These genes are also highly expressed in in vitro continuous cultures of F. succinogenes S85 (38) and include a functionally characterized GH51 endoglucanase 2 (Eg2, CelF, and EgF) and the GH9 endoglucanase 1 (Cel2 and End1). Our results suggest that GH8 CAZymes may also be important to the fibrolytic function of F. succinogenes on lucerne. Additionally, four F. succinogenes genes (FSU_2078, FSU_2396, FSU_2397, and FSU_2502) have previously been proposed from in vitro proteomics studies to encode a four-protein complex within outer membrane vesicles (8). These vesicles are hypothesized to serve as vehicles of CAZymes that target plant biomass by means of the 180-kDa cellulose-binding protein encoded by 1 of 10 Fibro-slime-domain-containing gene, FSU_2502. Our metatranscriptomic data confirmed the high expression of these four genes in vivo and also suggest that OMV formation may involve additional genes, including a cluster of eight highly expressed genes (FSU_RS11370 = FSU_2396 to FSU_RS11405 = FSU_2404). As transcriptional changes do not necessarily reflect the translation and activity of the associated enzymes, we accompanied the metatranscriptomic analyses with functional analyses using in sacco incubations, zymograms, and metabolomics.
DM disappearance varied over time, corresponding to differences in the relative  abundances of each cellulolytic species, but also between animals and by substrate and did not reflect measured CMCase and xylanase activities, which were 2.6-fold higher when ruminococci became dominant. This finding suggests that factors in addition to total fibrolytic enzyme activities must influence fibrolytic efficiency. The resulting changes in fermentation end products, including a reduction in propionate and increases in acetate and butyrate, as the two Ruminococcus strains outcompeting F. succinogenes may impact the resulting productivity of the animal. Indeed, the amounts and profiles of short-chain fatty acids (SCFAs) formed in the rumen have consequences for the efficiency of energy utilization, production of methane, risks of ruminal acidosis, and composition of animal products (39,40). F. succinogenes is an acetate and succinate producer, and succinate is a precursor of propionate (41). Our metatranscriptome data confirmed that two genes (FSU_RS14505 and FSU_RS14510) encoding fumarate reductase/succinate dehydrogenase involved in the succinate pathway were highly transcribed by F. succinogenes in vivo, which was not observed with the two ruminococci (data not shown). Hence, the disappearance of this species could in part explain the decrease in propionate. R. albus and R. flavefaciens produce acetate, formate, and hydrogen. All three products could account for the increase in acetate in rumen fluids knowing that formate and H 2 can be utilized by acetogenic bacteria to produce acetate. We previously showed that acetogenic bacteria were active in similar animal models as long as methanogens did not colonize the ecosystem at high levels (14). The increase in butyrate may indicate unmeasured ecological interactions that favor growth of butyrate-producing bacteria as ruminococci become dominant.   In addition to SCFAs, several phenolic metabolites, including 3-(3-hydroxyphenyl) propionic acid were found to change in association with the abundance of the two ruminococci. These metabolites are catabolites of plant polyphenols, including hydroxycinnamic acids (42). It is possible that the ruminococci enhance the production of these molecules by hydrolyzing esterified hydroxycinnamic acids from lucerne hemicelluloses. While phenolic metabolites exhibit antimicrobial properties, R. albus and R. flavefaciens may be resistant to these effects, given findings that the dehydroxylated derivative of 3-(3-hydroxyphenyl)propionic acid, 3-phenylpropanoic acid has instead been shown to enhance their growth and fibrolysis in vitro (43)(44)(45)(46).
In conclusion, this study describes for the first time in vivo competition between the three predominant cellulolytic rumen bacteria. It shows that global metabolic modifications in the rumen can be engendered by a simple disequilibrium in the cellulolytic community. It also provides in vivo evidence of the expression by the three cellulolytic bacterial species of different enzymatic and structural systems previously thoroughly studied in vitro that allow them to ensure their adhesion and hydrolytic function and survive in a very complex and competitive microbial digestive ecosystem.

MATERIALS AND METHODS
Ethics statement. The experimental protocol was validated by the local ethics committee (Comité d'Ethique pour l'Experimentation Animale Auvergne) before beginning the trial, and the protocol was registered under the number CE18-08. Procedures were in accordance with the guidelines for animal research of the French Ministry of Agriculture and all other applicable national and European guidelines and regulations for experimentation with animals.
Animal model. The gnotobiotic sheep model has been described previously in detail (14,47). Briefly, two INRAE line 401 lambs were born naturally and removed from their dams 17 h after birth. Lamb 421 (female) and lamb 455 (male) were then placed in a sterile isolator and reared under gnotobiotic conditions in the animal experimental facility of UMR MEDIS (Theix, Saint-Genes-Champanelle, France). Lambs were fed ultrahigh-temperature (UHT)-sterilized cow (Bos taurus) milk until they were 14 weeks old. At 2 weeks, they were also provided access to pelleted rations created from dehydrated lucerne (alfalfa) hay (7-mm diameter; SAFE, Augy, France). Lucerne pellets were sterilized by gamma irradiation (4 Mrad; Ionisos, Dagneux, France) prior to feeding. After 14 weeks, the animals were fed once daily at 8:00 a.m. with the lucerne pellets only. Sheep 421 and 455 were fitted within the sterile isolator with a permanent rumen plastisol cannula (diameter, 2.5 cm) at 17 and 18 weeks, respectively. Withdrawal of rumen contents and in sacco incubations began after the lambs had recovered from surgery at 22 weeks of age. The experiment ran for 12 weeks, and at the end of the experiment, sheep 421 and sheep 455 were both 34 weeks old and weighed 32 and 42 kg, respectively.
Cellulolytic inoculation. Lambs were monitored for the first 2 weeks after birth for the cultivable presence of cellulolytic and methanogenic activities, as well as for cultivable protozoa and fungi as previously described (14). None of these metabolisms or microbial groups were detected during this period. At 3 weeks of age, lambs were inoculated on 3 consecutive days with 20 ml (10 9 cells ml 21 ) of a 2-day-old pure culture of Fibrobacter succinogenes S85 (ATCC 19169) fed on filter paper strips. The lambs were subsequently inoculated at 7 weeks of age with a pure culture of the ruminal methanogen Methanobrevibacter sp. 87.7 to enable hydrogenotrophy and ensure fibrolytic activity would not be inhibited by hydrogen buildup in this animal model (14,47,48). Finally, at 28 weeks of age, they were inoculated via cannula on 3 consecutive days with 160 ml (10 8 cells ml 21 ) of 2-day-old pure cultures of both Ruminococcus flavefaciens FD1 (sourced from the Department of Animal Science, University of Illinois, Champaign-Urbana, IL, USA) and Ruminococcus albus 8 (sourced from USDA-National Center for Agricultural Utilization Research [NCAUR], Peoria, IL, USA) each supplied on filter paper strips. An overview of the animal trial and sampling procedure is presented in Fig. 1.
Rumen sampling. Rumen contents were withdrawn through the cannula on 2 consecutive days (each Tuesday and Wednesday) of weeks 22, 24, and 27 (F. succinogenes S85 was the only cellulolytic strain inoculated at these sample points) and on weeks 29, 31 and 33 (after R. albus 8 and R. flavefaciens FD1 had been inoculated) (Fig. 1). Sampling was performed before morning feeding (time zero [T0]) and 3 h after feeding (T3). T0 samples are reported with comparisons to T3 samples described in Text S1 in the supplemental material. The rumen content (50 ml) withdrawn on the first day of sampling was immediately used to determine ruminal pH and for culture-based microbial enumerations.
The rumen content (250 ml) withdrawn on the second day of sampling was treated as follows: a 50- In Vivo Competitions among Rumen Cellulolytic Bacteria ® ml aliquot was immediately mixed on ice with 100 ml RNAlater (Applied Biosystems, Courtaboeuf, France), aliquoted to four fractions of equal volumes, and stored at 280°C until nucleic acid extraction. A second 50-ml aliquot was kept on ice and treated as described below for polysaccharidase enzymatic analyses. A third 50-ml aliquot was frozen at 220°C (as sample saving) to determine the dry matter content of the rumen material. The remaining aliquot was centrifuged at 10,000 Â g for 15 min at 4°C, and 1 ml of supernatant was taken and stored at 280°C for metabolomic analyses. Enumeration of microbial communities. Total cultivable bacteria, total cultivable cellulolytic bacteria, and total cultivable methanogens were enumerated by using the most probable number (MPN) method as described previously (14).
In sacco digestibility assays. To evaluate plant substrate degradation in our model, we used the lucerne (alfalfa) pellets, as provided in the animal diet, and cell walls prepared as described previously (49) from ground and sieved (160 to 400 mm) wheat straw. Cell wall residues were prepared using an Soxhlet extractor by refluxing in 1:2 ethanol-toluene and then 95°C ethanol until the extracts became colorless (49). Nylon bags (R510 in situ bags [100 Â 50 mm; porosity of 50 mm]; Ankom Technology, NY, USA) were filled with either 1 g dry matter (DM) of the lucerne pellets or 0.5 g DM of wheat straw cell walls. In sacco incubations of lucerne bags were carried out for both sheep beginning on weeks 23 (post-fibrobacter inoculation but pre-ruminococci inoculation) and 30 (post-ruminococci inoculation), while incubations of wheat cell wall bags began on weeks 25 (also post-fibrobacter inoculation but preruminococci inoculation) and 32 (post-ruminococci inoculation; Fig. 1). During 8 consecutive days, two bags were introduced into the rumen before morning feeding and removed after 24 h of incubation. One bag was then immediately washed in RNase-free phosphate-buffered saline (PBS) until the buffer was clear. It was then placed in a sterile tube containing a mixture of 20 ml RNAlater (Applied Biosystems, Courtaboeuf, France) and 10 ml RNase-free water and frozen at 280°C for subsequent molecular analyses. The other bag was washed with tap water until the water was visibly clear and then frozen at 220°C for subsequent fiber composition analysis. Extra bags were also incubated to provide materials for electron microscopy observations of adhering bacteria, as described below.
Dry matter disappearance was determined by measuring changes in the weight of nylon bags between insertion and after removal and drying at 60°C for at least 72 h. Contents of the bags corresponding to the same plant substrate, same week of incubation, and same animal were pooled together for analyses. Residues were analyzed for neutral detergent fiber (NDF) and acid detergent fiber (ADF) contents using the Ankom fiber analyzer (Ankom Technology, NY, USA). Unincubated (control) bags were treated identically and used to calculate the base lines of DM, NDF, and ADF disappearances (DMdis, NDFdis, and ADFdis, respectively).
Enzyme activity assays and zymograms. Microbial enzymes from rumen contents were recovered as follows: a freshly collected rumen sample (50 ml) was centrifuged at 10,000 Â g for 15 min at 4°C, and the pellet was washed several times with an anaerobic mineral solution (50) until supernatant was clear. The washed pellet containing rumen microbes and lucerne residues was then frozen at 280°C until analysis. Thawed pellets (1 g) from weeks 27 and 33 (end of each period) were mixed with 5 ml phosphate buffer (50 mM, pH 7) containing the Complete protease inhibitor cocktail (Roche, Boulogne-Billancourt, France). Cell lysis was performed by sonication (four pulses of 30 s at 20 KHz and 60 W, separated by 1min pause on ice). The resulting mixture was then centrifuged at 8,000 Â g for 30 min at 4°C, and the supernatant was assayed for polysaccharidase specific activity and zymogram analysis as previously described (21), using carboxymethyl cellulose (medium viscosity) and oatspelt xylan (Sigma-Aldrich, Saint-Quentin-Fallavier, France) as the substrates.
Electron microcopy. The lucerne and wheat straw cell walls from the in sacco incubations were fixed with 3% glutaraldehyde in PBS for 1h at 4°C. They were then treated and examined with a Philips SEM 505 scanning electron microscope as described previously (21).
Nucleic acid extraction. Rumen contents and sacco residues, both in suspension in RNAlater (Applied Biosystems, Courtaboeuf, France) were thawed on ice and centrifuged at 15,000 Â g for 15 min at 4°C, and the resulting supernatant was discarded. Pellets from sacco residues corresponding to the same plant substrate, same week of incubation, and same animal were pooled to obtain sufficient quantities of material for nucleic acid extraction. DNA was extracted in triplicate from 0.25 g of pellet using an MP Biomedical DNA extraction and purification kit (Fast DNA Spin kit and Gene Clean Turbo; MP Biomedicals, Illkirch, France) following the manufacturer's recommendations. RNA was extracted in quadruplicate from 0.70 g of rumen content pellet using a Nucleospin RNAII kit (Macherey-Nagel, Hoerdt, France) following the supplier's instructions with the following exceptions. (i) During the cell lysis step, approximately 160-mg zirconia beads of 0.1-mm diameter (Bio Spec Products, Bartlesville, OK, USA) were added to the sample in addition to RA1 buffer and b-mercaptoethanol, for bead-beating (twice for 30 s each time at maximum power of 6.5) using the FastPrep instrument (MP Biomedicals, Illkirch, France). (ii) During the washing step, buffer RA2 was replaced by buffer FW1 as recommended by the supplier. Eluted RNA was submitted to an additional RNase-free DNase treatment (provided in the RNAII kit). RNA integrity was assessed using an Agilent 2100 bioanalyzer as previously described (51), and all samples were found to have a RNA integrity number (RIN) of .8. Nucleic acid concentrations were estimated by absorbance at 260 nm (NanoDrop ND-1000). They were kept at 280°C until utilization.
Quantitative real-time PCR. Quantitative PCR (qPCR) targeting total bacteria, methanogenic archaea, and each of the cellulolytic species-R. albus, R. flavefaciens, and F. succinogenes-was performed using SYBR green chemistry with a Mastercycler ep realplex 2S (Eppendorf, Le Pecq, France) in 96-well plates (Twin.tec plate 96 skirted; Eppendorf). Results were analyzed using realplex version 2.0 software (Eppendorf). The qPCR conditions (i.e., primers, reaction mixture, program, standard curves) have been Yeoman et al. ® described previously (29), except that we used the IQ SYBR Green Supermix (Bio-Rad, Marnes-la-Coquette, France). qPCRs were performed in triplicate from three separate extracts of DNA per sample (biological replicates). Results were expressed as rrs copies of microbial target per gram (dry matter) of sample (centrifuged rumen content or sacco residue).
Genomic reference data. Representative genomes from each of the cellulolytic species have previously been reported for R. flavefaciens 17 (52), R. albus 8 (53), and Fibrobacter succinogenes S85 (54). Methanobrevibacter sp. 87.7 was sequenced at Montana State University via a single run of an Illumina MiSeq. DNA was extracted using the DNeasy PowerLyzer microbial kit (Qiagen, Germantown, MD, USA), and genomic data were assembled using SPAdes v. 3.6.2 and gave 500-fold coverage. Genomic data were annotated using NCBI's Prokaryotic Genome Annotation Pipeline (https://www.ncbi.nlm.nih.gov/ genome/annotation_prok/) and are deposited in GenBank under NZ_MRCT01000000 and BioProject PRJNA339853. The four genomes used as reference in this study are presented in Table 3.
Metagenomic and metatranscriptomic analyses. Rumen samples withdrawn before feeding (time zero [T0]) at weeks 27, 31, and 33 for animal 421 and weeks 27, 29, and 33 for animal 455 (Fig. 1) were subjected to metagenomic and metatranscriptomic analyses. Triplicate DNA and quadruple RNA extracts were each pooled by animal and time point. RNA extracts were transcribed to cDNA by first strand cDNA synthesis, and both DNA and cDNA were independently sequenced with one lane each of pairedend 2 Â 125 nucleotide (nt) HiSeq 2500 (Illumina) sequencing.
Sequence data were quality and adapter trimmed using Trimmomatic v 0.33 (55). Metatranscriptome data were then further preprocessed using SortMeRNA v 2.1b (56) to separate sequences into rRNA and non-rRNA fractions. Full metagenomic data and the rRNA and non-rRNA subset of metatranscriptomic data from the SortMeRNA analysis were aligned to all four reference genomes as a single index file, using the Burrows-Wheeler Aligner v 0.7.15 (57). Coverage data for all alignments (used for downstream visualization) were generated using deepTools v. 2.5.0.1 (58) and visualized using the Integrative Genomics Viewer v2.3.92 (59). Basic expression normalization to reads per kilobase per million (RPKM) was performed using the edgeR v3.18.1 package in R v3.4.0 (60). Heat-maps were generated using the R package, ggplots on expression data scaled to the standard deviation from the mean. De novo CAZy-specific functional analysis was performed using HMMER v 3.1b2 (61) and the Release 5.0 protein motifs obtained from dbCAN (62). Read counts for all coding sequence annotation from each genome were generated using featureCounts v1.9.2 (63). All metagenomic and metatranscriptomic data were aligned using DIAMOND v. 0.8.37 (64) against the NCBI nonredundant protein database (July 2017). Taxonomic and functional information assignment as well as initial analyses and summary figures for taxonomic and functional classification were generated using MEGAN v6.8.13 (65). Metagenomic and metatranscriptomic data are deposited in GenBank under BioProject PRJNA646650.
Metabolomic analyses. Using complementary samples to those utilized for metagenomic and metatranscriptomic analyses, 1 ml of the rumen supernatant (collected as described above) was dried and derivatized as previously described (66). A 5-ml aliquot of an internal standard (C 31 fatty acid) was added to each sample prior to derivatization. Samples were injected with a split ratio of 7:1 into a gas chromatography (GC)-mass spectrometry (MS) system, consisting of an Agilent 7890A (Agilent Inc., Palo Alto, CA, USA) gas chromatograph, an Agilent 5975C mass selective detector, and an Agilent 7683B autosampler. Gas chromatography was performed on 60-m HP-5MS columns with 0.25-mm inner diameter and 0.25-mm film thickness (Agilent Inc., Palo Alto, CA, USA) and with an injection temperature of 2,500°C , the interface set at 2,500°C, and the ion source adjusted to 2,300°C. The helium carrier gas was set at a constant flow rate of 1.5 ml min 21 . The temperature program of 5-min isothermal heating at 700°C, followed by an oven temperature increase of 50°C min 21 to 3,100°C and a final 20 min at 3,100°C. The mass spectrometer was operated in positive electron impact mode (EI) at 69.9 eV ionization energy in the m/z 30 to 800 scan range. The spectra of all chromatogram peaks were then compared with electron impact mass spectrum libraries NIST08 (National Institute of Standards and Technology [NIST], MD, USA) and WILEY08 (Palisade Corporation, NY, USA) and a custom library of the University of Illinois Roy J. Carver Metabolomics Center. To allow direct comparisons between samples, all data were normalized to the internal standard in each chromatogram. The chromatograms and mass spectra were evaluated using the MSD ChemStation (Agilent, Palo Alto, CA, USA) and AMDIS (NIST, Gaithersburg, MD, USA). The retention time and mass spectra were implemented within the AMDIS method formats.
Analysis of volatile fatty acid (VFA) was performed without derivatization on rumen fluids using gas chromatography as previously described (67). Fermentation end products were assessed by comparing prefeeding (0-h) and 3-h postfeeding profiles throughout the animal experiment.
Statistical analyses. A Student's t test was used to compare DM disappearance, as well as polysaccharidase specific activities in animals between period 1 (weeks 22 to 27) with F. succinogenes only and period 2 (weeks 28 to 33) after inoculation of cellulolytic ruminococci. Differences were considered significant at P , 0.05.  33]) and an orthogonal PLS-DA for the two animals (421 and 455). The raw matrix with metabolite concentrations was log 10 transformed before analyzing the data with the ropls R package (68,69). To identify the metabolites responsible for discrimination between two time points (initial versus transitory, transitory versus final, and initial versus final), a pairwise comparative analysis using a t test was used to test the statistical significance. Only the variables (metabolites) with a variable importance projection (VIP) of .1.0 in the PLS-DA analysis were considered relevant for time discrimination. These tests were performed using the MetaboDiff R package (70) and were normalized using the variance stabilization method (71) before the differential analyses.
Data availability. The data sets supporting the results of this article are included within the article and its additional supplemental material. The genome sequence of Methanobrevibacter sp. 87.7 was deposited in GenBank under NZ_MRCT01000000 and BioProject PRJNA339853. The metagenome and metatranscriptome data are available in GEO data sets at NCBI (http://www.ncbi.nlm.nih.gov/) under BioProject PRJNA646650.

SUPPLEMENTAL MATERIAL
Supplemental material is available online only. TEXT S1, DOCX file, 0.04 MB.

ACKNOWLEDGMENTS
We thank Christel Bera-Maillet for advice regarding the experimental protocol for obtaining gnotobiotically raised sheep and B. Gaillard-Martinie who was in charge of the electron microscopy work. We also thank Gérard Andant and Gérard Vert for technical assistance, especially for their input in the maintenance, sampling, and analyses performed on these animals maintained for several months in sterile isolators. This work was funded by MEDIS laboratory (INRAE France) with support from the Roy J. Carver Biotechnology Center, the Montana Agricultural Experiment Station (project MONB00113), the National Institute of General Medical Sciences of the National Institutes of Health (NIH-NIGMS; award number P20GM103474) and is a contributing project to Multistate Research Projects, W4177, Enhancing the Competitiveness of U.S. Beef (MONB00195).
We declare that we have no competing interests. P.M. and B.A.W. conceived the study. C.J.Y., P.L., and P.M. performed the experiments. C. J.Y., C.J.F., P.R., E.F., and P.M. analyzed the data. C.J.Y. and P.M. wrote the manuscript with the help of all authors. All authors read and approved the final manuscript.