Two samples of seaweed naturally occurring in Iceland were studied for their effect on TMR formulated for cattle in a Rusitec system (Czerkawski and Breckenridge, 1977). The TMR was composed of 20% corn grain, 20% soybean meal, 40% corn silage, and 20% grass silage and served as the control treatment. One seaweed was Ascophyllum nodosum (AN), harvested in August 2018, and the other was Fucus vesiculosus (FV), harvested in June 2019. Both seaweeds were used at 2.5% (AN2.5 and FV2.5) and 5% (AN5 and FV5) inclusion in the TMR on a dry matter (DM) basis in exchange for TMR. The seaweeds and all ingredients of the TMR were dried and ground to pass a 1 mm screen. The analyzed nutrient composition, bromoform, and total phenolic content (TPC) of the seaweeds and TMR and the calculated nutrient composition of the experimental treatments are shown in Table 1.

Rumen content was collected from three rumen-fistulated non-lactating Jersey cows before the morning feeding. Animals had free access to water and a diet composed of 33% corn silage, 33% grass silage, 23% grass hay, 10% barley straw, and 1% mineral mixture (on a DM basis). During the daytime, the cows were kept on pasture. Two liters of rumen fluid were taken from each cow into prewarmed thermos flasks, comprised of 1 L pumped from the liquid phase and 1 L squeezed out from the solid phase. Additionally, 200 g of squeezed solid phase from each cow was transferred into prewarmed, isolated containers. Rumen fluid was strained through two layers of cheesecloth and mixed with equal parts from the donor animals. Subsequently, rumen fluid was mixed with a buffer solution (1:1), flushed with CO₂, and stirred at 39°C until fermenters of the Rusitec were inoculated. The buffer solution was prepared according to the suggested composition of artificial saliva by McDougall (1948) with the addition of ¹⁵N enriched NH₄Cl (0.0378 g/L; 104 mg ¹⁵N/g N) used for the calculation of microbial protein synthesis.

Each run lasted 14 days (days 0–13), with 7 days of adaptation, followed by 7 days of sampling. Ten fermenters were arranged side by side, with five fermenters sharing one circulation thermostat (Lauda ECO E 4 S, Lauda-Königshofen, Germany) and one buffer pump (Ismatec IPC ISM 931, Wertheim, Germany) (Figure 1). Each of the five treatments was replicated in two fermenters for two runs each, resulting in four replications per treatment. Fermenters were randomized in each run. The circulation thermostat was used as a block design, and each treatment was used once in each block. Due to technical arrangements, an additional empty fermenter was connected to each circulation thermostat. The glass fermenters were kept at a constant temperature of 39°C with a heating jacket. The buffer pump continuously infused buffer solution into each fermenter at a daily rate of 75% of the fermenter’s capacity (950 ml) to stimulate salivation. Each fermenter contained a feed container with continuous vertical movement ensured by a lift motor to simulate rumen motility (10–12 strokes/min; Figure 2). The fluid effluent was separated from the fermentation gas in a glass cylinder and then collected in 1 L glass bottles. The glass cylinder and effluent bottles were placed in a 4°C tempered water bath. Gaseous effluent was passed through gas counters (BlueVCount, BlueSens gas sensor GmbH, Herten, Germany) to measure gas production via gas-tight tubes and was subsequently collected quantitatively in gas-tight five-layered plastic-aluminum bags (Dr.-Ing. Ritter Apparatebau GmbH & Co., KG, Bochum, Germany). The methane concentration of total gas production was measured using an infrared CH₄ analyzer from the gas collected in the plastic-aluminum bags (PRONOVA Analysentechnik GmbH & Co., KG, Berlin, Germany).

Feed and solid rumen content were weighed into nylon bags (120 mm × 70 mm) with a 100 μm pore size and closed with cable ties. To start the system, one feed bag containing 15 g of the respective treatment and one feed bag containing 60 g of solid rumen content were put into the container of each fermenter. After 24 h, the bag with rumen content was replaced by another feed bag. Subsequently, each feed bag was removed after 48 h, rinsed with 50 (adaptation period) or 100 ml (sampling period) of buffer solution, squeezed moderately, and replaced by a new bag containing the specific treatment. The resulting liquid was returned to the respective fermenter.

The total gas production and CH₄ concentration of total gas were measured in 24 h intervals during the sampling period. The temperature, pH, and redox potential of the fermenter fluid were measured daily in the fermenter before the feed bags were changed (SenTix ORP, WTW, Weilheim in Oberbayern, Germany; BlueLine 14 pH IDS, SI Analytics, Mainz, Germany). From days 7 to 13, a sample from the fermenter liquid phase (FL; 30 ml/d) was taken daily and pooled by the fermenter to obtain the fraction of liquid-associated microbes (LAM) by centrifugation according to Boguhn et al. (2006), with minor modifications. In brief, the suspension was centrifuged twice at 2,000 × g for 5 min at 4°C. Then, the supernatant was centrifuged three times at 15,000 × g for 15 min at 4°C. Afterward, samples were freeze-dried and pulverized with a ball mill (MM 400; Retsch GmbH, Haan, Germany). On day 13, solid-associated microbes (SAM) were separated from the feed residues (FR), as described by Ranilla and Carro (2003). Ammonia nitrogen (NH₃-N), ¹⁵N abundance, and volatile fatty acids (VFA) were analyzed in the effluent (E), which was weighed and sampled daily (70 ml/d), pooled by the fermenter, and stored at −20°C until it was centrifuged for 15 min at 24,000 × g. The analysis of VFA was performed by vacuum distillation and gas chromatography (Hewlett-Packard 6890; Agilent, Waldbronn, Germany) measurement, as described by Wischer et al. (2013). NH₃-N was analyzed using Kjeldahl steam distillation with phosphate buffer (Vapodest 50, Gerhardt, Königswinter, Germany). The daily production of VFA and NH₃-N was calculated by relating the analyzed concentrations to the daily measured amount of E. The removed feed bags were dried for 24 h at 65°C, weighed, and pooled by the fermenter to determine nutrient degradation from days 7 to 12. Feed and FR were analyzed according to official methods in Germany (Verband Deutscher Landwirtschaftlicher Untersuchungs- und Forschungsanstalten, 1976) for DM (method 3.1), crude protein (CP; method 4.1.1), neutral detergent fiber on an ash-free basis (aNDFom; method 6.5.1), and acid detergent fiber on an ash-free basis (ADFom; method 6.5.2). Seaweeds and TMR were additionally analyzed for crude ash (method 8.1) and ether extract (method 5.1.1). Degradation of nutrients was calculated as the difference between the input and output of each fermenter in relation to input and expressed as a percentage.

Bromoform was analyzed in seaweeds according to method 8260B (EPA, 1996) by ALS Global (Miami, FL, United States). For the analysis of TPC, dried seaweeds were extracted in Milli-Q water (1:7 sample:water) for 1 h at room temperature (400 rpm). The TPC was determined on the sample supernatant according to the method by Singleton and Rossi (1965) adapted to the microplate format. For the analysis of heavy metals and minerals in the seaweeds, an ultraWAVE acid digestion system (Milestone Inc., Italy) and an Agilent 7900 quadrupole inductively coupled plasma mass spectrometer (Agilent Technologies, Singapore) were used according to NMKL (2007). Analyzed heavy metals and minerals are shown in Supplementary Table 1. Feed, FR, freeze-dried particle-free E, LAM, SAM, and NH₄Cl were analyzed for ¹⁵N abundance using an elemental analyzer (EA, 1108; Carlo Erba Instruments, Biberach, Germany) combined with an isotope mass spectrometer (MS Finnigan MAT; ThermoQuest Italia S.p.A., Milan, Italy). SAM was additionally analyzed for total N.

Microbiota composition was analyzed in rumen solid phase (RSP) and rumen fluid (RF) from the liquid phase of each cow in both experimental runs (n = 6), and the mixture of rumen fluid from all cows and buffer solution, treated as RF replicates (n = 2) on day 0. On days 7 and 13, 1 ml of FL and E were taken from each fermenter for the analysis of microbiota composition (n = 40). Feed residues from both feed bags remaining for 24 and 48 h in the fermenter were sampled on day 13 (n = 40). Additionally, the liquid after SAM treatment of both bags was sampled on day 13 (n = 20). Samples were stored at −20°C immediately after collection.

DNA was extracted with the commercial DNA extraction kit FastDNA™ Spin Kit for soil (MP Biomedicals, LLC, Solon, OH, United States), following the manufacturer’s instructions. DNA was quantified with a NanoDrop 2000 spectrophotometer (Thermo Fisher Scientific, Waltham, MA, United States) and stored at −20°C. Two sequencing libraries were prepared as previously described to assess bacterial (V1–V2 region; Kaewtapee et al., 2017) and archaeal communities (Arch349 and Arch806 primers; Lee et al., 2012; Deusch et al., 2017). In brief, targeted regions of the 16S rRNA gene were amplified by a first polymerase chain reaction (PCR) using forward primers with 6-nt barcodes and a 2-nt linker attached and both primers were complementary to the Illumina adapters. Then, 1 μl of the resulting product was used as a template for a second PCR, this time using a reverse primer containing a specific sequence to multiplex and index primers. Obtained amplicons were checked by agarose gel electrophoresis and normalized using the SequalPrep Normalization Kit (Invitrogen Inc., Carlsbad, CA, United States). Sequencing was performed on the 250 bp paired-end Illumina NovaSeq 6000 platform.

The estimated microbial protein synthesis (EMPS) was calculated in accordance with Boguhn et al. (2006) and Hildebrand et al. (2011). In brief, the daily outflow of microbial N from a fermenter (N_m; mg/d) was calculated as the sum of N originating from the LAM in the effluent (N_LAM–E; mg/d) and SAM (N_SAM; mg/d) fractions:

Therefore, N_LAM–E was calculated with the sum of the daily input of ¹⁵N via buffer solution and feed (¹⁵N_in; μg/d), the sum of the daily output of ¹⁵N via FR and NH₃-N (¹⁵N_out; μg/d), and the analyzed abundance of ¹⁵N in N_LAM–E (¹⁵N_LAM–E; μg/mg N):

The N_SAM was calculated using the amount of N in FR (N_FR; mg/day), the analyzed abundances of ¹⁵N in the FR N (¹⁵N_FR; μg/mg N), the feed N (¹⁵N_feed; μg/mg N), SAM N (¹⁵N_SAM, μg/mg N), and the assumed natural abundance of ¹⁵N in unlabeled N_SAM (3.66303 μg/mg):

Microbial CP (CP_m; mg/d) was calculated as N_m (Eq. 1) multiplied by 6.25:

In the next step, EMPS [g/kg degraded organic matter (OM)] was determined using the CP_m (mg/d), the degraded organic matter (OM_deg; mg/d), and the amount of OM originating from SAM in the FR (OM_SAM; mg/d):

Where OM_SAM was calculated as described by Boguhn et al. (2006) with the amount of N_SAM (Eq. 3), the analyzed N concentration in SAM (N_%SAM; %), the concentration of ash in SAM [12%, Boguhn et al. (2006)], and the proportion of DM in the isolated SAM fraction [0.93, Boguhn et al. (2006)]:

Statistical analysis of gas data, nutrient degradation, NH₃-N, VFA, and EMPS was done with a one-way ANOVA in SAS 9.4 using the mixed procedure. The treatment was the fixed effect, and the run, circulation thermostat, fermenter, and day were used as random effects. When treatment differences were identified, multiple t-tests (Fisher’s LSD test) were used to distinguish between treatments. The residuals were checked graphically for the normal distribution and homogeneity of variance.

Fastq files were demultiplexed with Sabre¹ and processed by Qiime2 (v.2021.2/8; Bolyen et al., 2019). Primers were removed by the q2-cutadapt plugin (Martin, 2011). Reads were quality filtered, error corrected, dereplicated, and merged by the q2-dada2 plugin (Callahan et al., 2016). Taxonomy assignment of produced amplicon sequence variants (ASVs) was performed using VSEARCH-based consensus (Rognes et al., 2016) and pre-fitted sklearn-based classifiers (Pedregosa et al., 2011) against the Silva SSU-rRNA database (v.138.1, 16S 99%; Quast et al., 2013), formatted by RESCRIPt (Robeson et al., 2021). Reads from organelles and unassigned sequences were removed from the analysis. A phylogenetic tree was constructed with the q2-phylogeny plugin, implementing MAFFT 7.3 (Katoh and Standley, 2013) and FastTree 2.1 (Price et al., 2010). For diversity assessment, ASV tables were rarefied to 3580 (archaea) and 5856 (bacteria) sampling depths. Alpha diversity was estimated by Faith’s phylogenetic diversity (Faith, 1992) and Shannon’s entropy (Shannon, 1948) indices. For beta diversity, Bray–Curtis (Bray and Curtis, 1957) and Jaccard (Jaccard, 1912) distances were employed. Ordination of the beta-diversity distances was implemented with a principal-coordinate analysis (PCoA; Halko et al., 2011). Differences in diversity metrics were tested with the Kruskal–Wallis H-test (Kruskal and Wallis, 1952; alpha diversity and relative abundances), followed by Dunn’s pairwise test (Dunn, 1964) and a PERMANOVA test (Anderson, 2001; beta diversity) with 999 permutations. The obtained P-values were corrected for multiple comparisons using the Benjamini–Hochberg procedure (Benjamini and Hochberg, 1995).

For correlation analysis, ASV tables were filtered to remove genera with less than 1% relative abundances. Absolute abundances of the remaining genera were correlated using “Sparse Cooccurrence Network Investigation for Compositional Data” or the SCNIC tool (Shaffer et al., 2020). Genera were correlated using the SparCC method (Friedman and Alm, 2012) and with numerical metadata using the Spearman correlation (Dodge, 2010). All P-values were considered to be significant if ≤0.05 (except for q2-ALDEx2 output) and correlation coefficients (r) if their absolute values were ≥0.3.

The q2-ALDEx2 differential abundance plugin was used to test the effect of seaweed supplementation on ASVs raw counts (relative abundance ≥1%) with a significance threshold of Wilcoxon test 0.1 (Fernandes et al., 2013, 2014). Sequences are available at the European Nucleotide Archive (ENA) under accession number PRJEB50942 and bioinformatics².

The total gas production was decreased by both seaweeds compared to TMR alone, with a greater effect for the 5% seaweed inclusion level (P < 0.001, Table 2). No significant difference between AN and FV was detected for total gas production. For the CH₄ concentration of total gas and the methane production per g of degraded OM, only the 5% inclusion level of the two seaweeds showed a difference compared to TMR alone (P < 0.001 and 0.017, respectively). The reduction was bigger for AN5 (8.9 and 16.9% reduction for CH₄ concentration and CH₄/g degraded OM) compared to FV5 (3.6 and 11.2% reduction for CH₄ concentration and CH₄/g degraded OM).

Seaweed supplementation also decreased the degradation of all analyzed nutrients in the feed bags (P < 0.001, Table 3). Only for ADFom, AN2.5 showed no significant difference to TMR alone. In all other cases, nutrient degradation was significantly lower with seaweed supplementation. The 5% inclusion level caused at least a numerically greater reduction than the 2.5% inclusion. The highest decrease compared to TMR alone was observed for CP, with a minimum reduction of 8.9% percentage points. The two seaweed products only differed at the 5% inclusion level for DM and OM degradation, where FV showed a greater decrease than AN. The EMPS was decreased by both seaweeds, with a greater effect of the 5% inclusion level (P < 0.001, Table 3). No significant difference was detected between the inclusion of AN and FV.

The temperature (P = 0.576; 38.6°C) and redox potential measured in FL (−241 mV; P = 0.062) did not differ among the treatments. The pH in the fermenter was different between the TMR alone (pH = 6.82) and the seaweed supplemented treatments (all pH = 6.85; P = 0.004). At both supplementation levels, seaweed decreased analyzed NH₃-N and VFA production compared to TMR alone, except for propionate production in FV2.5 (Table 4; P ≤ 0.001). For isovalerate and valerate, the 5% inclusion level of both seaweeds showed a less pronounced reduction than the 2.5% inclusion level. For butyrate, this was only the case for seaweed AN and acetate only for seaweed FV. For NH₃-N and the other VFA, the decrease was equal to or greater for the 5% inclusion level than the 2.5%.

After sequencing and demultiplexing, a total of 4,843,225 and 5,851,973 read pairs were obtained for archaeal and bacterial datasets, respectively. Quality filtering, denoising, merging of paired reads, and chimera removal resulted in 2,648,201 archaeal and 4,237,733 bacterial ASVs. Archaeal Faith’s PD and Shannon indices were approximately equal in RF and RSP at the start of the experiment (Figures 3A,C and Supplementary Table 2) and failed to reject the null hypothesis when tested against each other (Faith’s PD P = 0.881, Shannon P = 0.531). Archaeal phylogenetic diversity was greater in the Rusitec samples compared to the inoculum; meanwhile, Shannon entropy fluctuated in the system among sampling days. Compared to RF, Faith’s PD was greater in E and FL at day 7 (P = 0.001 and 0.044) and in E at day 13 (P = 0.026). E phylogenetic diversity at day 13 decreased compared to day 7 (P = 0.048). Shannon entropy at day 7 was greater in E than RF (P = 0.006). The entropy of E and FL sample types was reduced from days 7 to 13 (P = 0.001). At day 13, both metrics demonstrated no differences between FR and SAM samples (all P > 0.05) but were greater in FR samples from bags with a 48-h incubation period compared to 24 h (Faith’s PD P = 0.028, Shannon P = 0.001). In the bacterial dataset, the alpha diversity of samples at day 0 (RF and RSP) was also at the same level (all P > 0.05) (Figures 3B,D and Supplementary Table 2). Unlike archaea, bacterial communities demonstrated no changes in E and FL samples compared by sampling days and the initial RF at day 0 (all P > 0.05). Meanwhile, the Shannon index decreased in E and FL at day 7 (P = 0.029 and 0.017) and in E at day 13 (P = 0.019). Both tested alpha diversity metrics of FR and SAM revealed no differences between these sample types (all P > 0.05) but were lower for both of them than in all samples from days 0 to 7 and E and FL at day 13 (all P < 0.05). When testing FR samples based on incubation time, no differences were observed for phylogenetic diversity, but Shannon entropy was lower in 48-h samples (P = 0.026).

In archaea, seaweed additives affected the Shannon entropy of E at days 7 and 13 and FL at day 13 (Supplementary Figure 1). At both days 7 and 13, diets AN5, FV2.5, and FV5 resulted in lower entropy than the TMR alone (all P < 0.05). In FL, the Shannon index of FV2.5 and FV5 samples was also lower than that of TMR (all P < 0.05). Among bacteria, no significant changes between treatments were observed according to both alpha diversity metrics (all P > 0.05).

Concerning beta diversity, PCoA analysis (Figure 4) was performed on the ASVs, using Jaccard and Bray–Curtis distances. Archaeal and bacterial samples clustered into three big groups (RF and RSP from day 0 as one group, E and FL samples from days 7 to 13 as another one, and FR and SAM from day 13 as the third), with clearer clusters in the archaeal dataset. PERMANOVA analysis revealed that microbial composition was affected by the sample type (all P < 0.05), sampling day (all P < 0.05), and between FR samples incubated for 24 and 48 h (all P < 0.01), except for archaeal Jaccard distances.

Seaweed inclusion affected archaeal beta diversity according to both Jaccard and Bray–Curtis distances of E samples at days 7 and 13 and FL at day 13 (all P < 0.05). Additionally, Bray–Curtis distances were also affected in FR and SAM samples (all P < 0.05). Pairwise tests with correction for multiple comparisons found differences in Bray–Curtis distances of FR samples between the FV5 diet and TMR alone and AN2.5 (all P = 0.025). No effect of seaweed on bacterial beta diversity was detected (all P > 0.05).

Most of the archaeal sequences (≈ 95% of the reads) were assigned to Methanomicrobium, Methanobrevibacter, unclassified to genus level members of the Methanomethylophilaceae family, Candidatus Methanomethylophilus, and Methanimicrococcus (Figure 5A). Methanobrevibacter was the most abundant genus in RSP and RF samples, with a gradual shift toward Methanomicrobium dominance in E and FL. Considering all samples, regardless of the treatment by sample type and day (Supplementary Table 3), relative abundances of Methanomicrobium were barely detectable in the samples that served as inoculum RF and RSP but increased up to the dominance in samples from the Rusitec (all P < 0.05). On the contrary, Methanobrevibacter ratios mostly decreased in the Rusitec samples (all P < 0.05, except for E at day 7 compared to RF: P = 0.065). Moreover, Methanomicrobium relative abundances in E and FL were also greater at day 13 than at day 7 (P = 0.01 and 0.001); meanwhile, by the same comparison, the presence of Methanobrevibacter at day 13 was lower for E (P = 0.03). Except for SAM, the proportions of unclassified Methanomethylophilaceae in RF were lower than in Rusitec sample types (all P < 0.05). Methanomicrobium in FL at day 13 (P = 0.023), unidentified Methanomethylophilaceae in E at day 7 (P = 0.029), Methanimicrococcus in E at day 7 (P = 0.023), and FL at day 13 (P = 0.008) had higher relative abundances in FV5 samples than TMR alone (Supplementary Table 4). At day 13, Methanimicrococcus was also more abundant in FV2.5 samples from SAM (P = 0.043).

On average, among bacteria, Lactobacillus showed the highest abundance across all samples, followed by Pseudoscardovia, unclassified Bacteroidales, marked by Silva database as F082 (further F082), Limosilactobacillus, and Prevotella (Figure 5B). The high abundance of Lactobacillus was mostly observed in FR and SAM samples. In E and FL samples, F082 dominated the community. Prevotella accounted for the highest abundance in the inoculum (day 0) but was less abundant in E, FL, FR, and SAM (all P < 0.05) compared to RF, except for E at day 7. Compared with sample types and sampling days (Supplementary Table 3), relative abundances of Lactobacillus, Pseudoscardovia, and F082 were about the same in E and FL at both days 7 and 13 and increased their proportions compared to cow samples (all P < 0.05, except for Pseudoscardovia at day 13) after incubation in the Rusitec. Notably, Lactobacillus and Pseudoscardovia were poorly represented in RF and RSP. Treatment AN5 decreased relative abundances of F082 in E samples on day 7 (P = 0.025) and FR on day 13 (P = 0.005) compared to TMR alone. Meanwhile, Ruminobacter proportions in E (P = 0.044) and SAM samples (P = 0.05) increased at day 13. FV5 also increased relative abundances of Ruminobacter in FL samples at day 13 (P = 0.022) (Supplementary Table 4).

In E and FL sample types, 10 archaeal and 10 bacterial ASVs were differentially abundant in seaweed-supplemented diets compared to TMR alone (Figure 6). In E, archaeal ASV with the first four characters in corresponding id digest 7f9f (further indicated in round brackets), assigned to Candidatus Methanomethylophilus, was more abundant in treatments AN5 (P = 0.054) and FV5 (P = 0.049). Methanobrevibacter wolinii ASVs (80e7, b43f) were less abundant in FV5-supplemented treatments (P = 0.013, 0.01); meanwhile, Methanomicrococcus (48e8) abundances increased (P = 0.002). Unclassified Methanomethylophilaceae ASVs (d2f4, 8b4b, 9af5, 898c, and 37c0) were mostly associated with seaweed-supplemented samples, with one ASV (c795) more abundant in TMR alone and decreased when the FV2.5 (Wilcoxon P = 0.008) was added. In FL, Methanomicrococcus (48e8) (P = 0.005) and Methanomethylophilaceae (9af5, d2f4) (P = 0.067, 0.069) were more abundant in FV5 diet. Methanomethylophilaceae (d2f4) also increased when FV2.5 seaweed was added (P = 0.052).

Among bacteria of E and FL sample types, ASVs (7718, ea75), assigned to the Bacteroidales F082 group, Pseudoscardovia (9a4e), Atopobium (e1d3), Prevotellaceae (8969), and Succiniclasticum (1814) were mostly associated with TMR alone. However, Prevotellaceae (8969) (P = 0.083) and Pseudoscardovia (9a4e) (P = 0.085) were more abundant in FL samples of AN5. Prevotella (30d0) was more abundant in AN2.5, AN5, and FV5 samples (P = 0.002, 0.001, and 0.011 in FL and 0.002, 0.003, and 0.005 in E). Rikenellaceae RC9 gut group (81ff, 01fe) also increased in E and FL sample types when AN seaweed was added (P in the range of 0.002–0.003). Finally, in E unclassified Spirochaetaceae (8023) was more abundant in AN5 (P = 0.004) and FV5 (P = 0.04) supplemented diets.

A correlation analysis was performed to investigate potential relationships between microbiota genera, total gas, CH₄ concentration, and VFA production (Supplementary Tables 5–7). At the genus level, correlations of microbial abundances with total gas production, CH₄ concentration, and VFA were performed for E, FL, and FR sample types (Supplementary Figure 2), with only significant correlations plotted (absolute value of Spearman r ≥ 0.3 and P ≤ 0.05). The largest number of genera that negatively correlated with the above-mentioned traits was detected in E samples; meanwhile, in FL and FR, positive correlations prevailed.

A significant reduction of CH₄ concentration in the total gas was found to be caused by the supplementation of both seaweed species at the 5% inclusion level. However, the effect was not as high as previously reported for A. taxiformis (Kinley et al., 2016; Machado et al., 2018; Roque et al., 2019). Additionally, the lower CH₄ concentration in total gas was accompanied by a reduced rate of overall fermentation. The microbiota produced not only less methane but also fewer metabolites relevant to animals. This was indicated by the decreased total gas production, VFA, NH₃-N, and nutrient degradation. Reduced fermentation and nutrient degradation would generally mean a diminished amount of nutrients and energy available for the host animal. However, in the case of CP, a reduced degradation could benefit the animal because the ruminal undegradable protein could be used as a bypass protein like it was already shown for other seaweed species (Tayyab et al., 2016). The reduced CP degradation may have resulted from the formation of complexes between the phlorotannins contained in both seaweeds and other proteins (Belanche et al., 2016) and resulted in a lower EMPS, but more research is needed to confirm this.

The CH₄ reduction effect of A. taxiformis is attributed to the presence of bromoform (Machado et al., 2016), which is known to be a carcinogen (National Toxicology Program, 1989). The suggested modes of action in the brown seaweeds AN and FV used in the present study were likely related to phlorotannins as the bromoform content [8.0 (AN) and <0.8 μg/kg DM (FV)] was negligibly lower compared to the one analyzed by Machado et al. (2016) for A. taxiformis (1,723,000 μg/kg DM). The TPC, an indicator of phlorotannin content, was very similar in both seaweeds (7.9 vs. 7.4 phloroglucinol equivalent g/100 g sample). The greater CH₄ reduction with the supplementation of AN compared to FV could be explained by either the slightly greater TPC or the greater bromoform content in this sample, even if it is still very low. However, both species showed the same CH₄ concentration of total gas at the low seaweed inclusion level. Both bromoform and terrestrial tannins influence methanogens (Cieslak et al., 2014; Machado et al., 2018). Additionally, it has been shown that the composition of archaeal methanogens is associated with CH₄ production (Tapio et al., 2017). Consistent with this finding, our study showed that the decline in methane concentration of total gas in seaweed supplemented samples was accompanied by the separation of archaeal and bacterial ASVs in a dose-dependent manner, indirectly supporting the importance of methanogen composition. Interestingly, AN5 seaweed supplementation resulted in more differentiation of microbial ASVs abundances and a stronger reduction of CH₄ concentration when compared to TMR than FV5.

The treatment showing the highest CH₄ reduction in this study, AN5, also contained the highest iodine concentration (Supplementary Table 1). The iodine content of AN was more than 10 times greater than that of FV. Therefore, the application of practical feeding should be done with caution. Iodine toxicity in dairy cows is reported by the National Research Council (U.S.) (2001) for a concentration of 5 mg iodine/kg dietary DM. The European Food Safety Authority recommends a maximum iodine content of 2 mg/kg of complete feed (EFSA, 2013). In the current experiment, the iodine concentration was 35 mg/kg DM for treatment AN2.5 and 2.75 mg/kg DM for treatment FV2.5. Additionally, it was shown that the iodine concentration in milk follows a dose-response relationship with iodine intake (Antaya et al., 2015; Newton et al., 2021). Humans, especially children, are even more sensitive to iodine poisoning than ruminants (Zimmermann et al., 2005). Therefore, the seaweed inclusion levels used in the present study do not apply to practical ruminant feeding, except for FV2.5. Seaweed supplementation at an exceptional iodine inclusion level would therefore not significantly reduce methane compared to TMR.

In our study, Prevotella, unclassified members of Prevotellaceae and Clostridia negatively correlated with CH₄ concentration (Supplementary Table 6) and archaeal methanogens (Supplementary Table 7). Prevotella spp. and other members of the Prevotellaceae family are known for their ability to consume H₂ and produce propionate (Strobel, 1992; Mitsumori et al., 2012; Denman et al., 2015), which in turn is negatively associated with CH₄ formation and able to act as an alternative hydrogen sink (Ungerfeld, 2015). Inverse associations of Prevotella spp. with CH₄ concentration were previously reported (Mitsumori et al., 2012; Aguilar-Marin et al., 2020). Positive correlations with CH₄ concentration were observed for Fibrobacter, Lactobacillus, Pseudoramibacter, Olsenella, Shuttleworthia, and unclassified Oscillospiraceae (NK4A214). Fibrobacter, a fibrolytic bacteria that produce formate, can also be utilized by methanogens as a substrate for CH₄ production (Rychlik and May, 2000). Positive correlations of the Fibrobacter genus with CH₄ formation were previously reported (Wang et al., 2016). The addition of some Lactobacillus and Limosilactobacillus spp. (L. mucosae) in an in vitro rumen fermentation technique led to the rise of CH₄ production (Soriano et al., 2014). Pseudoramibacter can utilize carbohydrates as an energy source (Deusch et al., 2017), producing VFA during fermentation, including butyrate, acetate, formate, and hydrogen (Palakawong Na Ayudthaya et al., 2018), and increasing methane production by methanogens. Olsenella contains genes encoding choline trimethylamine lyase (Kelly et al., 2019) and is involved in CH₄ production (Broad and Dawson, 1976; Neill et al., 1978). Shuttleworthia is positively correlated with CH₄ formation in dairy heifers (Cunha et al., 2019) and produces acetate and butyrate through glucose fermentation (Downes et al., 2002). Oscillospiraceae members were linked with acetate production (Tanca et al., 2017) and, in our study, correlated with acetate, the A:P ratio, and CH₄ concentration. In general, those correlations are consistent with previous studies. It was shown that the addition of AN to an in vitro fermentation system led to a decrease in the growth of Fibrobacter succinogenes (positively correlated with CH₄ concentration) and an increase in Prevotella bryantii (negatively correlated with CH₄ concentration) (Wang et al., 2009). As it was reviewed by Abbott et al. (2020), it is possible that seaweeds modulate the growth of cellulolytic rumen bacteria by altering polysaccharide availability (Lee et al., 2019).

Although we allowed rumen microbiota to adapt for 7 days in the Rusitec prior to sampling, archaeal community distribution demonstrated a shift in the dominant genera at the end of the adaptation period and revealed a variation between the samplings at days 7 and 13. Partially, it can be explained by lower metabolite digestibility in the Rusitec than in vivo conditions and lack of or low protozoa counts (Martínez et al., 2010; Hristov et al., 2012). Relative abundances of Methanobrevibacter, which is the most abundant methanogen genus in cow samples, continuously decreased in the Rusitec. This tendency was somehow more profound in seaweed-treated samples, although no statistically significant differences with TMR were detected. In the Rusitec samples, dominance gradually shifted from Methanobrevibacter to Methanomicrobium by day 13. Although it has been shown that the Rusitec provides a stable condition for the methanogens (Lengowski et al., 2016), additional studies with longer in vitro incubation periods may be performed to investigate Methanobrevibacter/Methanomicrobium ratio dynamics in such systems since abundances of Methanobrevibacter relative to other archaea may affect CH₄ production (Cunha et al., 2019). When the total gas production and methane concentration are additionally statistically analyzed not only by treatment but also by sampling day and the interaction between treatment and day, it becomes evident that both traits increased during the experimental period (P = 0.005 and <0.001, respectively; Supplementary Table 8). This coincides with changes in the proportion of the most abundant methanogens in the system (Methanobrevibacter abundances decreased, while Methanomicrobium increased) between days 7 and 13 and indicates that adaptation of the microbes to the system was not finalized after day 7. Previous studies have shown a decrease in the protozoa population, also known as a methane producer, in Rusitec fermenters over time (Ziemer et al., 2000; Martínez et al., 2010; Morgavi et al., 2012). Protozoa were not analyzed in the current study, but this potential shift could also have led to an altered archaeal community distribution.

The North Atlantic seaweed species used herein can modify the microbiota in the rumen toward reduced methane production. However, methane reduction was not great, and the seaweed inclusion in the ration caused reduced in vitro fermentation overall. Therefore, the tested seaweed species are not a viable option to be used as feed for ruminants to mitigate methane production.