This study was conducted using a RUSITEC system containing six fermentation vessels with samples taken over a 15-day period (7 days of adaptation followed by 8 days where samples were taken). The experiment was a completely randomized design with two treatments and three replicates per treatment. The fermentation substrate consisted of equal portions of concentrate and ryegrass that was included either as FM preparation (FM treatment) or FDG to 2 mm forage (FDG treatment). Fermentation parameters (total gas, CH₄, and volatile fatty acid [VFA] production) were analyzed on a daily basis over the experimental period and the archaeal and bacterial microbiota (liquid-associated microbes [LAM] and solid-associated microbes [SAM]) were characterized on days 0, 5, 10, and 15.

Plant material of ryegrass (Lolium perenne) was collected on 15th January 2015 at the May Farm research site of the University of Sydney, Camden Campus, NSW, Australia (34°04′ S; 150°81 69′ E). The climate is warm-temperate with a mean annual minimum and maximum temperature of 10.7 and 23.3°C, respectively. The annual average rainfall is 738 mm (1900–2010). Multiple samples were randomly selected and harvested at grazing height ≥ 5 cm above ground level to mimic grazing by cattle. Plant material was processed immediately upon return to the laboratory (within 45 min of collection) and divided into two equal portions. One half of the plant material (FM sample) was prepared according to the methodology described by Chaves et al. (2006) as follows: material was frozen at -20°C and while still frozen, cut into 2–3 cm lengths using scissors and minced using a meat mincer (Rovtek MG-22SS, Commercial Meat Mincer, 800 W, 250 kg/h, blade speed 190 rotation/h, Sydney, NSW, Australia) fitted with a screen plate with 12 mm holes. The samples were stored at -20°C until the day of incubation. Preparation of FDG samples involved freeze-drying the ryegrass, grinding and then passing the ground ryegrass through a 2 mm screen. Samples were kept at room temperature until the day of incubation.

Ryegrass and concentrate were analyzed for dry matter (DM) (method 967.03), ash (method 942), and ether extract (EE) content by extraction with diethyl ether using an Ankom XT10 Extraction System (Ankom^® Technol. Corp., Fairport, NY, USA; method 920.39), following Association of Official Analytical Chemists [AOAC] (2006) methods. The concentration of neutral detergent fiber (NDF) was determined using procedures detailed by Van Soest et al. (1991) and modified for an Ankom 200/220 Fiber Analyzer (Ankom Technol. Corp., Fairport, NY, USA) using sodium sulfite and heat stable α-amylase. Crude Protein (CP) was analyzed by combustion [method 990.03 (Association of Official Analytical Chemists [AOAC], 2006)] using FP628 Food/Protein Analyzer (LECO, St Joseph, MI, USA) following the manufacturer’s guidelines. The feed chemical composition is presented in Supplementary Table S1.

The rumen inoculum was collected on the initial day of the experiment from one ruminally fistulated (Soliva et al., 2011, 2015) Holstein dairy cow (9 years old, 750 kg). The donor animal was housed at The University of Sydney Corstorphine Dairy farm and was cared for in accordance with the guidelines of The University of Sydney Animal Ethics Committee (Project number 2015/835). The cow was fed the same diet (Supplementary Table S1) that was used as a substrate in the RUSITEC fermenters and the inoculum was collected 2 h after the morning feeding. The rumen digesta was filtered through four layers of cheesecloth to separate the solid and liquid portion to be used as an initial inoculum. The rumen inoculum was transported immediately to the laboratory using insulated containers.

Approximately 5 g (DM basis) of each of the treatments was weighed into pre-weighed nylon bags (70 mm × 140 mm; pore size = 150 μm) and approximately 5 g (DM basis) of concentrate feed was weighed into a separate nylon bag (70 mm × 100 mm; pore size = 150 μm). At the start of the experiment, 4.7 L of rumen fluid was evenly distributed into six fermentation vessels, and equal amounts (i.e., ∼50 g) of rumen solids were weighed into nylon bags.

The RUSITEC apparatus was equipped with six 800 mL fermentation vessels. Each vessel had an inlet for the infusion of buffer and an effluent output port. At the start of the experiment each fermentation vessel was filled with 780 mL of rumen fluid. The nylon bags (containing about 30 g of wet weight of rumen solids and the experimental diets) were placed inside each fermenter according to the randomized treatments. Fermenters were then submerged in a 39°C water bath and infused with McDougall’s buffer [3.69 g/L Na₂HPO₄ (anhydrous), NaHCO₃ 9.8 g/L, NaCl 0.47 g/L, KCl 0.57 g/L, MgCl₂6H₂O 0.061 g/L, CaCl₂2H₂O 0.0336 g/L] at a dilution rate of 30 mL/min.

After 24 h of incubation, the rumen solids bag was replaced with new nylon bags containing the experimental treatments, and from day 2 onward, the nylon bags were replaced each day, meaning that four bags were present at any given time (Narvaez et al., 2013; Avila-Stagno et al., 2014).

Total gas produced was collected on a daily basis in gas-tight bags (Plastigas, Linde AG, Munchen, Germany) connected to the effluent flasks. The total volume of gas was determined by water displacement. The gas bag was connected to a flask filled with water and the gas was evacuated by applying manual pressure to the flask. The evacuated water was then collected in a graduated cylinder and the volume collected was measured as daily gas production and expressed in mL/d.

From day 8 until the end of the experiment, 15 mL from each gas bag was removed with a syringe and transferred to an exetainer tube (Labco Ltd, Lampeter, UK) for analysis of methane concentration by gas chromatography (Bruker 450 GC, Bruker Technologies, Australia, with two packed columns and a Compass CDS data acquisition software (Bruker Technologies, Australia). Argon and helium were used as carrier gases at a flow rate of 30 mL/min. The oven, injector and detector temperatures were 50, 70, and 180°C, respectively.

Dry matter disappearance was determined using the residue remaining in each nylon bag after 48 h of fermentation. Nylon bags were removed, washed with cold water (tap water/distilled water) until the water was clear (Narvaez et al., 2013; Li et al., 2014) and then dried at 100°C for 24 h. The residue weight was recorded and used for the calculation of DM disappearance.

Fermentation parameters, such as pH and volume outflow, were determined on a daily basis during the 15-day experimental period (i.e., 7 days of adaptation and 8 days of collection). Volume was measured daily at the time of feed bag exchange with a graduated cylinder. Starting at day 8, individual samples from each fermentation vessel were taken for quantification of VFA. Samples were preserved with metaphosphoric acid (25% w/v; 1:5 dilution) and stored at -20°C until analysis. Capillary gas chromatography was performed to analyze VFA (C:2 to C:6) according to the Biochemistry Laboratory Method NTR-7, Issue No. 5, 05/05/2015 (University of Western Australia). An Agilent 6892 Series GC with Agilent 7696 sample preparation station and HP 6890 injector with HP Chemstation software were used. The capillary column was a HP-FFAP, 30 m × 0.53 mm × 1.0 μm (HP Part. No. 199095F-123). Oven temperature was 240°C and hydrogen was used as carrier with a total flow of 55.7 mL/min at 3.5 psi. The HP Chemstation system calculated the concentration for the VFA in mmol/L. Concentration of volatile fatty acids production was calculated per day by multiplying concentration and volume of outflow.

In order to investigate the effect of forage preparation on the rumen microbial community, samples were taken at the start of the experiment (day 0) from the original rumen fluid and solids, and on days 5, 10, and 15 for rumen fluid from the RUSITEC fermenters. To characterize LAM, a 10 mL sample from each fermenter was collected into a sterilized 25 mL falcon tube, immediately frozen in liquid nitrogen and kept at -80°C until DNA extraction. To assess SAM, on the last day of the experiment (i.e., day 15), the digesta in the nylon bags was frozen in liquid nitrogen and then subjected to freeze drying before DNA extraction.

Total DNA was extracted using a QIAamp Fast DNA stool mini kit (Qiagen), according to manufacturer’s instructions and a lysis temperature of 80°C. The DNA yield and purity was assessed using a NanoDrop 2000 Spectrophotometer (Thermo Scientific) and extracted DNA was stored at -20°C. The 16S rRNA gene libraries were generated using a two-step PCR protocol. The first PCR step amplified the V4 region of the 16S rRNA gene using the universal bacterial and archaeal primers 515-F (5′-GTGCCAGCMGCCGCGGTAA-3′) and 806-R (5′-GGACTACVSGGGTATCTAAT-3′) (Caporaso et al., 2011). The second PCR step was used to add a unique 10-bp barcode at the 5′ end of the amplicon as well to add Illumina adapters. All PCR amplification and sequencing steps were carried out at Genome Quebec (Montreal, QC, Canada). Briefly, the 16S rRNA gene amplicons were quantified using a Quant-iT PicoGreen dsDNA assay kit (Invitrogen, Burlington, ON, Canada), pooled in equimolar ratios, and then purified with AMPure XP beads (Beckman Coulter, Mississauga, ON, Canada). The 16S rRNA gene amplicons were then sequenced using an Illumina MiSeq (2 × 250) and the MiSeq Reagent Kit v2 (500 cycles; Illumina, San Diego, CA, USA) according to manufacturer’s instructions.

The 16S rRNA gene sequences were processed and analyzed within the QIIME software package v. 1.9.1 (Caporaso et al., 2010b). Paired-end reads were joined using fastq-join with a minimum overlap of 35 bp and a maximum percent difference of 15 (Aronesty, 2013). Joined sequences were quality filtered with sequences being truncated following three consecutive base calls of a Phred score of less than 25. Sequences were retained only when 75% or more of the original sequence remained after truncation. Chimeric sequences were removed using the UCHIME algorithm (Edgar et al., 2011) implemented in USEARCH v. 6.1544 (Edgar, 2010). Sequences were then clustered into operational taxonomic units (OTUs) at 97% similarity using an open reference OTU picking method and the SILVA database v. 111 (Pruesse et al., 2007). Sequences that did not match OTUs in the SILVA database were clustered into OTUs using the de novo approach and USEARCH. The UCLUST consensus taxonomy assigner (Edgar, 2010) was used to assign taxonomy to OTUs using the SILVA database, with a minimum similarity of 0.8 and max accepts of 3. Representative sequences for the OTUs were aligned using PyNast (Caporaso et al., 2010a) and a phylogenetic tree was created using FastTree (Price et al., 2010). OTUs containing fewer than 10 sequences were excluded from further analysis as well as those OTUs classified as chloroplasts.

Each sample was randomly subsampled to 22,500 sequences per sample to account for uneven sequencing depth. The bacterial and archaeal diversity in each sample was calculated within QIIME using the Shannon index (Shannon, 1948) and phylogenetic diversity (PD whole tree) (Faith, 1992). Unweighted UniFrac distances (Lozupone and Knight, 2005) were used to assess the archaeal and bacterial community structure (beta-diversity) of each forage preparation type, sampling time, and sample type (solid vs. liquid). The subsequent distance matrices were visualized as principal coordinate analysis (PCoA) plots using Emperor (Vazquez-Baeza et al., 2013).

All 16S rRNA gene sequences were submitted to the NCBI Sequence Read Archive (SRA) under bio-project accession PRJNA304765¹.

Fermentation and alpha-diversity data were analyzed using the MIXED procedure of SAS (SAS, Inc., 2015; SAS Online Doc 9.1.3). The model included the fixed effects of forage preparation, day and forage preparation × day interaction. Therefore, the individual fermenter was used as the experimental unit for statistical analysis. The minimum values of Akaike’s information criterion were used to select the covariance structure. Significance was declared at P ≤ 0.05 and a trend was discussed when 0.05 < P ≤ 0.10.

Unweighted UniFrac distances were compared using ANOSIM (analysis of similarities) with 999 permutations. OTUs that were differentially abundant by sample type, forage preparation type and sampling time were identified using the G-test of independence with a false discovery rate (FDR) < 0.05. Linear discriminant analysis effect size (LEfSe) was used to determine which phyla and genera were enriched at each sampling time, forage preparation type, and sample type. LEfSe uses the Kruskal–Wallis test to identify different (P < 0.05) genera among sample groups and uses linear discriminant analysis (LDA) to estimate the effect size of each of these (Segata et al., 2011). A LDA score of 3.5 and an overall relative abundance of greater than 0.01% was used as the threshold for identifying differentially abundant genera.

The type of forage substrate preparation did not affect total gas, CH₄ (%, mg/g DM, and mg/g DDM) or digestibility of substrates (P ≥ 0.13; Table 1). Total VFA, propionate, BCVFA (branched-chain volatile fatty acids), acetate to propionate ratio and pH were also not affected by type of preparation, but there was a tendency of increase in daily CH₄ (mg/d) production in FM compared to FDG (Table 2). An interaction between forage preparation and sampling time was observed for acetate, valerate and caproate (Figure 1; P ≤ 0.05). Butyrate production was greater (Table 1; P = 0.01) in FM compared to FDG treatments.

A total of 858,881 sequences with an average length of 262 bp among all samples remained following quality filtering and removal of primer sequences. These sequences were clustered into 3,349 OTUs. Overall, the most relatively abundant phyla were: Bacteroidetes (60.1%), Firmicutes (24.3%), Spirochaetes (3.8%), Euryarchaeota (3.7%), Planctomycetes (2.6%), and Proteobacteria (1.6%) (Supplementary Figure S1). At the genus-level, Prevotella was the most relatively abundant (44.8%), while Dialister (8.4%), Succiniclasticum (3.5%), Lactobacillus (2.6%), and Treponema (2.1%) were the only other genera with an overall relative abundance > 2.0% (Figure 2).

In regards to methanogens, only the solid fraction from the original rumen sample had a relatively high abundance of methanogens (classes Methanobacteria and Methanomicrobia; 22.9%). Nearly all of these methanogenic sequences were identified as Methanobrevibacter (22.6%). Excluding the original rumen samples, the relative abundance of the methanogens was 1.6% among all samples. Apart from Methanobrevibacter, the three other methanogenic genera identified among all samples were Methanosphaera, Methanomicrobium, and Methanimicrococcus. Overall, there were 28 OTUs that were classified into one of the four genera above.

Alpha-diversity (within-sample diversity) measures are presented in Table 3 for each sample preparation type. Archaeal and bacterial diversity and richness were not affected by forage preparation type (P > 0.05) in LAM samples, however, lower phylogenetic diversity was observed in LAM samples at day 5 compared with days 10 and 15 (68.84 vs. 74.8 and 74.2, respectively; P < 0.05). The archaeal and bacterial diversity and richness in SAM samples were also not affected by the forage preparation (P > 0.05; Table 3).

The effect of microbial sample type (LAM vs. SAM), forage preparation, and sampling time, on the structure of the archaeal and bacterial microbiota was analyzed using unweighted UniFrac distances (Figures 3–5). Visual inspection of the PCoA plots and ANOSIM of unweighted UniFrac distances demonstrated that sample type (LAM vs. SAM) was the most important factor in determining the structure of the microbiota (Figure 3, R-value = 0.982; P = 0.001). Similar to the within-sample diversity analysis, sampling time also had a significant effect on the structure of the microbiota in LAM samples. In particular, samples from day 5 clustered separately from those of days 10 and 15 (Figure 4, R-value = 0.423; P = 0.005). Samples also clustered together based on forage preparation type (Figure 5, R-value = 0.261; P = 0.002).

Differentially abundant OTUs were identified using the G-test of independence. Excluding the original rumen samples, there were 244 OTUs that were differentially abundant (FDR < 0.05) between sample type (LAM vs. SAM) and 118 differentially abundant OTUs between the two forage preparations (both SAM and LAM samples included). Phyla enriched in one of the two sample types (LAM vs. SAM) were identified using LEfSe (Figure 6A; LDA [log₁₀] score > 3.5). There was a total of six phyla enriched in the LAM samples and two in the SAM samples, with Firmicutes notably more relatively abundant in the SAM samples. At the genus-level, Succiniclasticum, Bacteroides, and Treponema were among the genera that were more relatively abundant in the LAM samples, while Lactobacillus and Roseburia were enriched in the SAM samples (Figure 6B).

Only the Proteobacteria phylum was more relatively abundant in the FDG samples and Bacteroidetes was the only phylum enriched in the FM samples (LDA [log₁₀] score > 4.0; P < 0.05). In the FDG samples, the genera Campylobacter and Lactobacillus were more relatively abundant while Butyrivibrio, Prevotella, and Roseburia were more relatively abundant in the FM samples (LDA [log₁₀] score > 3.5; P < 0.05). Interestingly, Campylobacter was almost completely absent from the FM samples, as 9 of the 11 samples analyzed had no Campylobacter sequences, while all but two of the 11 FDG samples had greater than 0.1%. Among sampling times in the LAM samples, Firmicutes was also enriched at day 10 and Planctomycetes, Synergistetes, and Verrucomicrobia more relatively abundant at day 15. No phyla were enriched at day 5 compared to the other two sampling times. Similarly, among the genera with an overall relatively abundance greater than 0.01%, only Dialister was enriched at day 10 and Acidaminococcus and Megasphaera in day 15 samples (LDA [log₁₀] score > 4.0; P < 0.05).