Animals, diets and rumen content collection
All animal procedures were performed in accordance with the Institution of Animal Care and Use Committee (IACUC) at University of California, Davis under protocol number 19263. Rumen content was collected from two rumen fistulated cows, one Jersey and one Holstein, housed at the UC Davis Dairy Unit. Animals were fed a dry cow total mixed ration (50% wheat hay, 25% alfalfa hay/manger cleanings, 21.4% almond hulls, and 3.6% mineral pellet (Table 1). Three liters of rumen fluid and 60 g of rumen solids were collected 90 min after morning feeding. Rumen content was collected via transphonation using a perforated PVC pipe, 500 mL syringe, and Tygon tubing (Saint-Gobain North America, PA, USA). Fluid was strained through a colander and 4 layers of cheesecloth into two 4 L pre-warmed, vacuum insulated containers and transported to the laboratory.
In-vitro feed and feed additive composition and collection
Due to its wide utilization in the dairy industry for cows during lactation, super basic ration (SBR) was used as feed in the in-vitro experiment. SBR was composed of 70% alfalfa pellets, 15% rolled corn, and 15% dried distillers’ grains (Table 3). Individual components were dried at 55 °C for 72 h, ground through a 2 mm Wiley Mill (Thomas Scientific, Swedesboro, NJ) and manually mixed. Asparagopsis taxiformis used as feed additive was provided in kind from the Commonwealth Scientific and Industrial Research Organization (CSIRO) Australia. The macroalgae was in its filamentous gametophyte phase when collected near Humpy Island, Keppel Bay, QLD (23o13'01"S, 150o54'01"E) by MACRO (Center for Macroalgal Resources and Biotechnology) of James Cook University (JCU) in Townsville, QLD. The collected biomass was frozen and stored at − 15 °C then shipped to Forager Food Co. in Red Hills, Tasmania, AUS, where it was freeze dried and milled (2–3 mm) to ensure a uniform product. Chemical composition of SBR and of A. taxiformis were analyzed at Cumberland Analytical Services (Waynesboro, PA).
Engineered (in-vitro) rumen system
An advanced semi-continuous fermentation system, with six 1 L vessels with peristaltic agitation, based on the rumen simulation technique (RUSITEC) developed by Czerkawski and Breckenridge  was used to simulate the rumen in the laboratory.
Equilibration (Day 0): Temperature, pH and conductivity of the rumen fluid and solids were recorded using a mobile probe (Extech Instruments, Nashua, NH). Rumen fluid, 3 L, from each cow were combined with 2 L of artificial saliva buffer  homogenized and then split into two 3 L aliquots. Rumen solids, 15 g, from each animal were sealed in Ankom concentration bags (Ankom, Macedon, NY) and added to each equilibration vessel (30 g of rumen solids per vessel total). Three concentrate bags containing 10 g of SBR each were added to each vessel. One of the equilibration vessels was amended with 5% (w/w) of A. taxiformis 24 h prior to the start of the experiment (Fig. 1). Content of the equilibration vessel without A. taxiformis was used to inoculate control vessels of the in-vitro system, whereas content of the equilibration vessel with A. taxiformis was used to inoculate the treatment vessels (Fig. 1). SBR was ground in a 2 mm Wiley Mill before being added to each concentrate bag to increase substrate availability and therefore producing similar particle sizes that which the mastication function in-vivo provides to the animal. The two vessels were then placed in a 39 °C water bath and stirred with a magnetic stir bar for a 24 h equilibration period.
Fermentation (Days 1–4): After 24 h of equilibration, temperature, pH, and conductivity of the rumen fluid were recorded to determine stability of the vessels and their content. Each of the 6 in-vitro rumen vessels were randomly designated as either treatment or control vessel and filled with 750 mL of the corresponding fluid from the equilibration vessels. Location of the vessels within the in-vitro platform were randomly allocated.
Each vessel received one concentrate bag of SBR from its respective equilibration vessel and one new concentrate bag. Control concentrate bags contained 10 g SBR. Treatment concentrate bags contained 10 g SBR plus 5% (OM) A. taxiformis. To simulate rumen retention time, each of the feedbags were incubated in the allocated fermentation vessel for 48 h. Temperature, pH, and conductivity were measured every 24 h prior to exchanging one of the concentrate bags (feeding). After each feeding, all vessels were flushed with N2 to maintain anaerobic conditions within the reactors. Individual reactor vessels of the artificial rumen system were connected to a reservoir containing artificial saliva buffer. A peristaltic pump delivered 0.39 mL/min of buffer to each vessel throughout the course of the experiment. Gas bags (Restek, USA) and overflow vessel were used to continuously collect generated gas and effluent fluid. Effluent vessels were chilled with ice to mitigate residual microbial activity. An outline of the experimental set-up and the preparation of the treatment and control vessels is provided in Fig. 1.
Sample collection and analysis
Liquid and gas sample collections took place at 3 time points every 24 h for 4 days. Time point intervals were 4, 12, and 24 h post-feeding each day. Fluid samples were collected in 1.5 mL tubes, flash frozen in liquid nitrogen, and stored at − 20 °C until processed. Gas bags were collected at each time series interval for analysis of total gas production, CO2 and CH4 concentrations. Gas volume was measured with a milligas flow meter (Ritter, Germany) by manual expulsion of the collection bag.
Volatile fatty acid and greenhouse gas analysis
To determine VFA profiles, Gas Chromatography-Flame Ionization detection (GC-FID) was used. Fermentation fluid was prepared for VFA analysis by mixing with 1/5th volume 25% metaphosphoric acid, and centrifugation. Supernatant was filtered through a 0.22 μm filter and stored in amber autosampler vials at 4 °C until analysis. The GC conditions were as follows: analytical column RESTEK Rxi® – 5 ms (30 m × 0.25 mm I.D. × 0.25 μm) film thickness; the oven temperature was set to 80 °C for 0.50 min, and followed by a 20 °C/min ramp rate until 200 °C, holding the final temperature for 2 min; carrier gas was high purity helium at a flow rate of 2.0 mL/min, and the FID was held at 250 °C. A 1 μL sample was injected through Split/Splitless Injectors (SSL), with an injector base temperature set at 250 °C. Split flow and split ratio were programmed at 200 and 100 mL/min respectively. To develop calibration curves, certified reference standards (RESTEK, Bellefonte, PA) were used. All analyses were performed using a Thermo TriPlus Autosampler and Thermo Trace GC Ultra (Thermo Electron Corporation, Rodano Milan, Italy).
Methane and CO2 were measured using an SRI Gas Chromatograph (8610C, SRI, Torrance, CA) fitted with a 3’× 1/8″ stainless steel Haysep D column and a flame ionization detector with methanizer (FID-met). The oven temperature was held at 90 °C for 5 min. Carrier gas was high purity hydrogen at a flow rate of 30 ml/min. The FID was held at 300 °C. A 1 mL sample was injected directly onto the column. Calibration curves were developed with an Airgas certified CH4 and CO2 standard (Airgas, USA).
DNA extraction was performed using the FastDNA SPIN Kit for Soil (MP Biomedicals, Solon, OH) with ~ 500 mg of sample according to the manufacturer’s protocol. DNA was subsequently purified with a Monarch® PCR & DNA Cleanup Kit (New England Biolabs, Ipswich, MA) following the manufacturer’s instructions. Extracted DNA was stored at − 20 °C until subsequent PCR amplification and amplicon sequencing.
PCR amplification, library preparation, and sequencing
The V4-V5 hypervariable region of the 16S rRNA gene was sequenced on Illumina’s MiSeq platform using the 515yF (3′-GTG YCA GCM GCC GCG GTA A-5′) and 926pfR (3’-CCG YCA ATT YMT TTR AGT TT-5′) primer pair (Research and Testing, Lubock Texas; [44, 45] For sequencing, forward and reverse sequencing oligonucleotides were designed to contain a unique 8 nt barcode (N), a primer pad (underlined), a linker sequence (italicized), and the Illumina adaptor sequences (bold).
Forward primer: AATGATACGGCGACCACCGAGATCTACAC-NNNNNNNN- TATGGTAATT-GT-GTGYCAGCMGCCGCGGTAA;
Reverse primer: CAAGCAGAAGACGGCATACGAGAT-NNNNNNNN-AGTCAGTCAG- GG-CCGYCAATTYMTTTRAGTTT.
Barcode combinations for each sample are provided in Additional file 1: Table S4. Each PCR reaction contained 1 Unit Kapa2G Robust Hot Start Polymerase (Kapa Biosystems, Boston, MA), 1.5 mM MgCl2, 10 pmol of each primer, and 1 μL of DNA. The PCR was performed using the following conditions: 95 °C for 2 min, followed by 30 cycles at 95 °C for 10 s, 55 °C for 15 s, 72 °C for 15 s and a final extension step at 72 °C for 3 min. Amplicons were quantified using a Qubit instrument with the Qubit High Sensitivity DNA kit (Invitrogen, Carlsbad, CA). Individual amplicon libraries were pooled, cleaned with Ampure XP beads (Beckman Coulter, Brea, CA), and sequenced using a 300 bp paired-end method on an Illumina MiSeq at RTL Genomics in Lubbock Texas. Raw sequence reads were submitted to NCBI’s Sequence Read Archive under the SRA ID: SRP152555.
Sequencing resulted in a total of 1,251,439 raw reads, which were analyzed using mothur v1.39.5  using the MiSeq SOP accessed on 3/10/2018 . Using the make.contigs command, raw sequences were combined into contigs, which were filtered using screen.seqs to remove sequences that were > 420 bp or contained ambiguous base calls to reduce PCR and sequencing error. Duplicate sequences were merged with unique.seqs, and the resulting unique sequences were aligned to the V4-V5 region of the SILVA SEED alignment reference v123  using align.seqs. Sequences were removed if they contained homopolymers longer than 8 bp or did not align to the correct region in the SILVA SEED alignment reference using screen.seqs. To further denoise the data, sequences were pre-clustered within each sample allowing a maximum of 3 base pair differences between sequences using pre.cluster. Finally, chimeric sequences were removed using VSEARCH .
Quality filtered sequences were grouped into OTUs based on 97% sequence identity and classified using the Bayesian classifier and the Greengenes database (August 2013 release of gg_13_8_99)  with classify.seqs. Sequences that classified as mitochondria, chloroplasts, eukaryotic, or of unknown origin were removed using remove.lineage. Samples were rarefied to 6467 sequences per sample, the smallest number of sequences across all collected samples. Singleton abundances were calculated with filter.shared. Chao1 diversity , Good’s coverage , Shannon , and inverse Simpson indices were calculated using summary.single to quantify coverage and α-diversity.
To estimate the microbial diversity within each group, first, rarefaction analyses were performed (Additional file 1: Figure S1) and species richness and diversity indices were calculated (Additional file 1: Table S2.). Variance of the microbial community between and among the different vessels were quantified using a θYC distance matrix .
To investigate slow-acting effects of seaweed addition on microbiome communities, we computed Bray-Curtis dissimilarity (β-diversity)  between pairs of samples, both within vessels at different time points, and between vessels at identical time points. We also considered Jaccard dissimilarity which only reflects community composition and not relative abundance, but found similar results and so only report the results for Bray-Curtis dissimilarity. We independently computed β-diversity at the genus, family, order, class, and phylum level to assess whether the observed patterns were dependent on taxonomic resolution. For regression statistics, we computed 95% confidence intervals using non-parametric bootstrap resampling, and significance values using permutation tests. Both of the latter approaches gave qualitatively similar results. All analyses were performed using custom-written Java, SQL, and Bash code available at https://github.com/jladau.
Analysis of molecular variance (AMOVA)  was used to identify significant differences in community structure between treatment and control vessels using a θYC distance matrix for the amova command in Mothur. The complete results of these statistical tests between each time interval combination is included in the supplementary data.
Gas, VFA, and Euryarchaeota abundance data were analyzed using the linear mixed-effects model (lme) procedure using the R statistical software (version 3.1.1) [56, 57]. The statistical model included treatment, day, time point, treatment×day×time point interactions, treatment×day interactions, treatment×time point interactions, day×time point interactions and the covariate term, with the error term assumed to be normally distributed with mean = 0 and constant variance. Orthogonal contrasts were used to evaluate treatments vs. control, linear, and quadratic effects of treatments. Significant differences among treatments were declared at p ≤ 0.05. Differences at 0.05 < p ≤ 0.10 were considered as trend towards significance.