Microbial ecology along the buffalo digestive tract provides insights to their functions and interactions with the host

Buffalo is an important livestock in Asia. Like other ruminants, its digestive tract (DT) is the key to the quality and wellbeing of buffalo and heavily interact with microbes. Here, we present a comprehensive survey on the microbial ecology along buffalo’s DT, including eight sites in three sections (i.e., stomach, intestine and rectum). We collected 695 samples, performed metagenomic sequencing and obtained 4,960 high-quality metagenome-assembled genomes (MAGs), to which ~ 85% of the raw reads could be mapped. 90.7% of the MAGs are previously unidentied at species level. Overall, Firmicutes and Bacteroidetes are the most abundant phyla; their ratios showed an increasing trend along the DT, consistent with their functions in the DT sections. We identied known interactions between microbes and DT sites including the enrichment of ber-digesting and methane-producing microbes in the stomach. Strikingly, archaea were highly abundant in both stomach and intestine and showed positive-correlations with Fibrobacter, indicating their roles in methane-production and ber-degradation at both sections. We annotated 5,862,748 non-redundant proteins from the MAGs, many of which also showed different abundances and were related to site-specic functions. By comparing with the rumen microbiota of cattle, we found higher abundances of microbes in ber degradation and but lower in methane production. Our catalog of microbial genomes and encoded-proteins provides insight to their functions and interactions with distinct DT sites, and pave the way to microbial interventions for better buffalo quality.

We submitted the 695 samples to metagenomics next-generation sequencing (mNGS) using Illumina NovaSeq 6000 with read-length of 150. After removing vector and low-quality sequences, contaminations from host and food genomic sequences, we obtained in total 11Tb of clean data for further analyses (see Methods for details). On average, we obtained 41,842,231 pairs of clean reads and 6,244,074,222 bases for each sample.
To obtain high-quality metagenome-assembled genomes (MAGs), we adopted a customized bioinformatic analysis work ow (see Supplementary Fig. 1b for a graphical representation; Methods).
Brie y, the clean reads were assembled by metaSPAdes 58 and MEGAHIT 59 ; the resulting 109,471,448 contigs were grouped into 58,094 bins using metaBAT2 with default parameters. All bins were aggregated and dereplicated using dRep 60 (v.2.3.2); followed by CheckM 61 (v.1.0.18) for quality assessment. In the end, we obtained a non-redundant set of 4,960 bins (MAGs) with completeness ≥ 80% and contamination ≤ 10%. Rarefaction analysis indicated that the curves could be plateaued using samples from stomach, rectum or all combined, although not for intestines that had relatively fewer samples (Fig. 1b).
Among the resulting MAGs, 2,581(52.0%) were high-quality draft genomes as de ned by Bowers et al 64 with ≥ 90% completeness and ≤ 5% contamination, while 2,222(44.8%) met the score criterion de ned by Parks et al 62 (completeness -(5 × contamination) ≥ 50) (Fig. 1c, d). The sizes of the MAGs ranged from 402 kilobases (Kb) to 6.1 megabases (Mb), with a median of 2.1 Mb (Fig. 1f). The MAGs contained 1 to 679 contigs with a median of 102. The contig sizes of the MAGs ranged from 2.5Kb to 1.5Mb, with N50 values (50% of assembled bases in contigs larger than the N50 value) ranging from 5.4 kb to 1.4 Mb To check if our MAGs could improve the coverage of microbial genomes associated with buffalo's DT, we mapped all the metagenomic clean reads to the resulting MAGs and compared the overall mapping rates to MAGs of selected organisms (human 65 , chicken 54 , pig 66 and cattle 31 ) and reference microbial genomes from public databases (NCBI RefSeq genomes plus the Hungate collection genomes 63 , referred as to BFAP (bacterial, fungal, archaeal, and protozoan)). As shown in Fig. 1g, the buffalo raw reads showed the highest mapping rate to our MAGs with an average of 85%, followed by the cattle's Rumen Uncultured Genomes (RUGs) from Stewart et al 31 , BFAP and MAGs of human, pig and chicken.
Our stomach samples showed the lowest mapping rates to the reference genomes and MAGs as compared with samples from intestines and rectum (Fig. 1g), with the cattle dataset being an exception because it consisted of MAGs from rumen metagenomes 31 . These results are consistent with the fact that stomach metagenomes were less covered in public databases and most metagenomic researches focused on the gut. The overall mapping rates of the stomach samples to cattle and buffalo MAGs were close (Fig. 1g, orange boxes), suggesting their stomach metagenomes were similar; however, the mapping rates of buffalo intestine and rectum samples to cattle RUGs were signi cantly lower (less than 65%), suggesting gut metagenomes were under-represented in Stewart et al 's data 31 . Since the cattle and buffalo are closely related, we also mapped raw sequencing reads of cattle metagenomes obtained from 31 to our MAGs. On average 71% of the cattle reads could be mapped to our MAGs, as compared to 82.5% to cattle RUGs ( Supplementary Fig. 2).
Together, we obtained in total 4,960 high-quality MAGs that signi cantly improve the coverage of the microbes in buffalo's digestive tract, especially those in sections of intestine and rectum. Taxonomic Annotation Of Buffalo Mags We next assigned taxonomic classi cations to the MAGs using GTDB-TK 67 . According to GTDB-TK, genomes with < 99% average nucleotide identities (ANIs) belong to the same strains, while those with < 95% ANI belong to the same species. By these criteria, a total of 4,895 MAGs had < 99% ANI and 4,277 MAGs had < 95% ANI with GTDB-TK reference data, indicating potential new strains and species respectively. Of the 4,960 MAGs, almost all could be classi ed to known taxonomical classi cations at the higher levels such as kingdom, phylum, class and order; however, at more re ned levels especially the species level, only 460 (9.3%) of the MAGs could be classi ed as known species (Fig. 2a), indicating most the MAGs were novel (i.e., not present in the GTDB-TK database). Both bacteria and archaea showed similar classi cation results (Fig. 2b, c).

Taxonomic characteristics of MAGs in different sections along the digestive tract
To investigate the distributions of the 4,960 MAGs in different sections along the DT and their putative interactions with the host, we rst determined the coverages and relative abundances of all the MAGs in each sample. We mapped the clean reads of a sample to all MAGs and calculated the coverage of a MAG as the total aligned bases divided by the total bases of the MAG 68 , and relative abundance a MAG as the percentage of reads mapped to the MAG out of the total reads mapped to all MAGs (see Methods for details and Supplementary Table 3 for the results).
Using a coverage 1X as the cutoff of presence/absence, we found 3,032, 3,081 and 4,141 MAGs were present in at least one samples of the three sections, respectively (Fig. 3a). By this criterion, we found 1,692 (34% out 4,960) MAGs were present in all three sections, 1,910 MAGs (38.5%) in samples of two sections, while only small proportions of section-speci c MAGs (499 in stomach and 859 in rectum; Fig. 3a). We did not nd intestine-speci c MAGs by this criteria, likely due to the fact that it positioned between stomach and rectum (Figs. 1a and 3a). At the phylum level, ve phyla were stomach-speci c, among which three (UBP3, UBP6, and Eremiobacterota) are candidatus, indicating unknown phyla. Among the other two, Euryarchaeota consists of species related to the methane metabolism, while Synergistota consists of oral bacteria in human and was rst identi ed in the goat rumen, usually for the toxic compound degradation 69 . The main rectum-speci c phylum is Verrucomicrobiota, which was found in soil, water, and feces 70 .
We further analyzed the relative abundances of the MAGs along the DT and found signi cantly different alpha diversities among all sections, measured by both the Shannon and Simpson indexes (Fig. 3b). Among which, rectum had the highest diversities, followed by stomach and intestine (Fig. 3b), in part consistent with the numbers of MAGs identi ed in the three sections (Fig. 3a). We then applied Principal coordinates analysis (PCoA) to show Bray-Curtis distances among the samples, and found that the overall microbial pro les of intestine and rectum were similar (R = 0.048; P = 0.001, pair-wise nonparametric MANOVA test) while both signi cantly different from that of the stomach (R = 0.192, 0.270, P = 0.001, 0.001, pair-wise non-parametric MANOVA test; Fig. 3c). Interestingly, samples from jejunum of the intestine formed their own cluster and were far away from other DT sites (Fig. 3d), indicating its distinctive microbial structure; this is also consistent with the alpha diversity of jejunum that was clearly different from other parts of the DT (Fig. 3b).
Distinctive patterns of MAGs along the digestive tract coincide with their functions As shown in Fig. 3e, the overall microbial structures of stomach, intestine and rectum were different with each other (Fig. 3e), with Firmicutes and Bacteroidota being the two most abundant phyla. Firmicutes_all, including Firmicutes, Firmicutes_A, Firmicutes_B, and Firmicutes_C according to GTDB, on average accounted for 79%, 85%, and 90% of total microbial abundances in the three DT sections. Interestingly, Bacteroidota showed decreased abundances along the digestive tract (Fig. 4a), while Firmicutes_all showed the opposite (Fig. 4b). Consequently, Firmicutes to Bacteroidota ratio (F/B ratio, Fig. 4c) was lowest in stomach and highest in rectum. Previous results have linked increased F/B ratio with increased capacity for energy harvest from the diet 71,72 , consistent with the physiological roles of the three sections. In addition, the F/B ratio in the rumen has been linked to milk fat yield in cows [73][74][75] .
The decreasing abundances of Bacteroidota along DT was in part due to Prevotella, the main genus of Bacteroidota that is mostly abundant in stomach, especially in rumen, and is signi cantly lower in other sections and DT sites (Fig. 4h), accounting for 33.1% total microbial abundances in rumen. Prevotella species are associated with non-cellulose plant ber degradation, and are known the largest single bacterial group reported in the rumen of cattle and sheep under most dietary regimes 76 . Most of the species in this genus are unclassi ed, including the most abundant one ( Supplementary Fig. 4). In addition to the role in plant degradation, Prevotella species also played important role to prevent ruminal acidosis [76][77][78] .
Similarly, bacteria capable of digesting cellulose, the main component of the cell wall of plants, were also signi cantly abundant in the stomach 79,80 , and were less abundant in other DT sites (Fig. 4e, f, g). These included Fibrobacter_all (including Fibrobacter and Fibrobacter_A according to GTDB; Fig. 4e), Ruminococcus_all (including Ruminococcus_E, Ruminococcus_A, and Ruminococcus according to GTDB; Fig. 4f) and Butyrivibrio_all (including Butyrivibrio_A and Butyrivibrio according to GTDB; Fig. 4g). Among them, Fibrobacter_all was the most abundant taxon. Our results showed that the total abundances of Fibrobacter_all were signi cantly higher in all four stomach sites (i.e., rumen, reticulum, omasum and abomasum) than the other two sections; however, we found the highest abundances of Fibrobacter_all in omasum, indicated that the later may play important roles in cellulolytic digestion.
All the archaeal species we identi ed were methanogens; they were highly abundant in the stomach and intestine (Fig. 4d). These results contradicted our current understandings that only stomach especially rumen is the main organ of methane metabolism 81 , and highlight the importance of intestine in the methane metabolism. Strikingly, close examination revealed that the total abundances of these methanogens peaked in omasum (Fig. 4d), coincided with Fibrobacter_all (Fig. 4e); in fact, we found that the overall abundances of stomach and intestine had the highest correlations as compared with other categories (Fig. 4i). Together, these results highlighted the important roles of omasum in both methane metabolism and cellulose degradation, and possible functional link between the latter two 82 Functional characteristics of MAGs in different sections along the digestive tract We next explored the proteomic contents of buffalo metagenome and their putative functions. We predicted in total 9,470,238 proteins from the 4,960 MAGs; after clustering by CD-HIT 83 , we obtained a non-redundant proteome dataset of 5,862,748 proteins. We annotated these proteins by comparing the amino acid sequences with the eggNOG database 84 using eggNOG-mapper 85 and the CAZy 86 database using dbCAN2 87 . As a result, 4,787,680 proteins (81.7% out of total) could be annotated according to one or both methods; among which, 114,989 and 4,672,691 were annotated by CAZy and eggNOG, respectively.
We then calculated the relative abundances of all proteins in each sample and compared their distributions (see Methods). As shown in Fig. 5b-g, all the six above-mentioned CAZy families showed signi cant differences in terms of relative abundances (the sum abundance of all proteins is 100%) among the three DT sections. Interestingly, all CAZy protein families except Auxiliary Activity (AA) showed the highest abundances in stomach, followed by rectum and then intestine, supporting the central role of stomach in food digestion and processing ( Fig. 5b-g); rumen often had the highest abundances of these CAZy families among all DT sites (except for carbohydrate-binding module (CBM), omasum displayed the highest abundances), although other stomach sites also contained signi cantly higher CAZy proteins than other DT sections ( Fig. 5b-g). Again, we identi ed jejunum as an outlier which contained the lowest percentages of all six CAZy families ( Fig. 5b-g), likely due to its low microbial diversity (Fig. 3b).

Comparisons Of Rumen Microbiota Between Buffalo And Cattle
Rumen often is considered the most important stomach site in ruminants. As the rst section of DT, rumen is the largest compared with other stomach sites, governs the rst steps of feedstuffs degradation and is the main site for methane-production 26 . The microbiota in rumen played important roles in its function and were the rst to be studied by researchers 26,88 . We thus compared the taxonomic and functional pro les of buffalo's rumen microbiota with that of cattle which was recently made available by Stewart et al 31 . To make a fair comparison, we obtained in total 4,941 Rumen Uncultured Genomes (RUGs, similar to MAGs; see ref 31 ), 4,879,163 non-redundant proteins, and calculated their abundances in each sample using the same methods used in this study (Methods).
As shown in Fig. 6a, we found that buffalo and cattle differed signi cantly in the two most dominant phyla (Firmicutes_all and Bacteroidota) and consequently the F/B ratios (Fig. 6a). Bacteroidota species, especially those in its dominant genus Prevotella are capable of degrading non-cellulose plant bers. Therefore, the higher abundances of Bacteroidota and Prevotella in buffalo rumen than cattle, as well as the similar levels of Fibrobacter_all (responsible for cellulolytic plant ber digestion) between the two (Fig. 6a), suggested that buffalo was more adapted to coarse forage than cattle 81 . Conversely, we found signi cantly higher levels of Archaea, Butyrivibrio_all and Ruminococcus_all in cattle, all of which played important role in methanogenesis through biohydrogenation and glycolysis (Fig. 6b), suggesting that buffalo may produce signi cantly less methane than cattle.
We also compared the abundances of rumen protein families (i.e., CAZy) between buffalo and cattle. Surprisingly, all six families were signi cant abundant in buffalo (Fig. 6c). This result further suggested that buffalo had better capacity in carbohydrate metabolism. Besides, the comparison based on all CAZy proteins ( Supplementary Fig. 7) displayed signi cant abundance of GH and CBM in cattle, indicating that rumen of cattle had more proportion of enzymes that relative to the formation of glycosidic bonds.

Discussion
As an important livestock, buffalo provides humans with milk, meat, leather and draft power. Like other ruminants, its digestive tract (DT) is the key to the quality and wellbeing of buffalo and heavily interact with microbes. However, the lack of microbial reference genomes at different sites of the DT greatly hindered our understanding of the functional interactions between DT sites and their microbial ecology, and our ability to modulate buffalo's physiology and economically important phenotypes through DT microbiota. More importantly, a comprehensive pro ling on the methanogenic microbes that are presumably inhabit in rumen may provide us with insights on reducing emission of methane, an important source of greenhouse gases. Recent studies on ruminant microbiota have most focused on rumen, while the exploring for the digestive tracts was still missing. To ll in these gaps, we presented a comprehensive survey on the microbial ecology along buffalo's DT. We collected in total 695 samples from eight DT sites in three sections, namely stomach, intestine and rectum. To further increase the representativeness of our study, we took samples from six different locations (Guangxi, Henan, Anhui, Yunnan, Hainan, Hubei; Supplementary Tables 1, 4; Supplementary Fig. 1a), three breeds (river, swamp and hybrid; Supplementary Tables 1, 2), both sexes and two developmental stages (Supplementary   Tables 1, 5).
We performed metagenomic next generation sequencing (mNGS) on these samples and obtained 4,960 high-quality metagenome-assembled genomes (MAGs). These MAGs greatly improved the coverage of the raw sequencing reads from 64% of the public databases (the combination of reference genomes datasets including bacterial, fungal, archaeal and protozoan genomes from NCBI RefSeq, BAFP) to 85%. Taxonomic annotation revealed that all MAGs could be classi ed into known phyla; however, more than 90% of the MAGs are novel at species level. Thus, our dataset represents a great expansion of buffalo microbiomes.
Sampling at different sites of the DT allowed us to better understand the functional associations between the microbial ecology and the DT sites. For example, we found that Firmicutes and Bacteroidota, the two most dominant phyla, showed distinct abundance patterns along the DT: Firmicutes were increased along the digestive tract (Fig. 4a), while Bacteroidota showed the opposite (Fig. 4b). Consequently, Firmicutes to Bacteroidota ratio (F/B ratio, Fig. 4c) was lowest in stomach and highest in rectum. F/B ratio has been shown to be related to energy harvesting 71,72 ; its trend thus coincides the physiological transition from food digestion (stomach) to energy harvesting (intestine) along the DT. Our data also allowed us to validate known interactions between microbes and DT sites, including the enrichment of ber-digesting and methane-producing microbes in the stomach. Surprisingly, we found that Fibrobacter, a group of cellulolytic bacteria known to colonize mainly in rumen showed higher relative abundance in omasum; its distributions coincided with archaea, the main methane-producers whose abundances also peaked in omasum. Archaea were also highly abundant in both stomach and intestine and showed positive-correlations with Fibrobacter, indicating their roles in methane-production at both sections. Based on the abundance of the microbes at phylum and genus level, we identi ed several features that abundant in stomach, intestine, and rectum. These results highlighted the importance of having samples from all DT sites, especially those other than rumen.
We also evaluated the functional capacities of the microbial ecology of the MAGs at the DT sites by annotating the protein-coding genes from the MAGs and comparing them against CAZy and eggNOG databases. We found all of the six CAZy families showed signi cant differences among DT sections, suggesting their different roles associated with distinct sections of the DT. Our data showed poor sequence identity with public data, indicating previously unidenti ed protein sequences and thus novel functions encoded in the DT-associated MAGs.
Rumen is the most important section in the DT; its microbiota has been extensively explored recently in cattle. We thus took the opportunity to compared the rumen microbes between the two closely related model ruminants. Our results showed the signi cant differences between buffalo and cattle. For example, we found higher abundances of microbes with ber degradation capacity in buffalo's rumen than cattle. In addition, the relative abundance of methane-producing archaeal species in buffalo was signi cant lower than cattle, indicating the less production of methane and the more fully use of energy from feedstuff.
Together, our catalogs of microbial genomes and their encoded-proteins represented the largest effort so far to characterize the microbial ecology along all major sections of the DT; our study provided insight to the microbial functions and interactions with distinct DT sites, and valuable resources for the community who are interested in microbial interventions for better buffalo quality.

Sample collection
A total of 695 samples were collected for metagenome sequencing. To ensure the diversity of the samples, they were collected from three breeds (153, 430, 112 samples from river, swamp and hybrid buffaloes), six regions (Guangxi, Henan, Anhui, Yunnan, Hainan, Hubei from China), two sexes (females and males respectively) and two developmental stages (Adult buffalos and calfs; see Supplementary Data 1 for details). These samples included 296 content and 399 fecal samples; the content samples were taken from rumen, reticulum, omasum, abomasum, jejunum, cecum and colon, while the fecal samples taken from rectum (Supplementary Table 3). In this study, the above eight DT sites were divided into three sections, namely stomach (rumen, reticulum, omasum, and abomasum), intestine (jejunum, cecum, and colon) and rectum; see also Fig. 1. All samples were immediately frozen after collection in liquid nitrogen and stored at − 80°C until DNA extraction.

DNA Extraction, Library Construction And Metagenomics Sequencing
Three grams of each sample were taken for DNA extraction. DNA was extracted by a bead-beating method using a mini-bead beater (Biospec Products; Bartlesville, USA), followed by phenol-chloroform extraction. The solution was precipitated with ethanol, and the pellets were suspended in 50 µL of Tris-EDTA buffer. DNA was quanti ed using a NanoPhotometer® (IMPLEN, CA, USA) following staining using a Qubit® 2.0 Flurometer (Life Technologies, CA, USA). DNA samples were stored at − 80°C until further processing.
Library preparation was performed according to the TruSeq DNA Sample Preparation Guide (Illumina, 15026486 Rev. C) method and procedure using 500 ng DNA as template. Quali ed libraries were selected and subjected to the Illumina NovaSeq 6000 for pair-ended sequencing with read-length of 150 base pairs (PE150).

Quality Control And Removal Host-And Food-associated Genomes
Raw sequencing reads were rst trimmed by Trimmomatic 89  concordantly to references were removed as contamination. As a result, average 20.3% bases were removed. The remaining "clean reads" were used for further analysis.

Generation And Quality Assessment Of Metagenomics-assembled Genomes (mags)
MEGAHIT (v.1.2.8) and metaSPAdes (v.3.13.0) were used for single-sample assemblies. Before each run of metaSPAdes, the k-mer parameter was tested with a range of 21 to 141; then the k-mer with longest N50 and total reads length was chosen (if the result of N50 and total length were inconsistent, the k-mer with longest total length was chosen). For MEGAHIT, we used default parameters to assemble the reads. Co-assemblies were performed for each the three sections by combining all samples as input only using MEGAHIT because its less consumption of time and memory compared with metaSPAdes. However, As shown in Supplementary Fig. 8, metaSPAdes could generate contigs with longer N50 than MEGAHIT through comparison with N50, among other measurements for assembly qualities. Thus, we combined the assembly results from both tools to increase the coverage and quality of the resulting contigs.
MetaBAT2 95 (2.12.1) was used to group contigs into bins. First, BWA-MEM 96 (v.0.7.15) was used to map reads to the contigs (MEGAHIT) and scaffolds (metaSPAdes) to get the depths of contigs (MEGAHIT) or scaffolds (metaSPAdes) in each sample. The results were saved in SAM les. Second, Samtools 97 (v. 1.8) was used to convert SAM les to BAM format. Last, MetaBAT2 was used to calculate coverage from the resulting BAM les and output the results of bins. As a result, Single-sample binning produced a total of 58,041 bins, and while additional 53 bins were obtained from co-assembly binning. We referred these bins to as metagenome-assembled genomes (MAGs).
All bins were dereplicated using dRep 60 (v.2.3.2) with the opinion 'dereplicate_wf -p 16 -comp 80 -con 10str 100 -strW 0'. During this process, CheckM 61 (v.1.0.18) was rst used to access the quality of the resulting MAGs. After removing MAGs with completeness < 80% or contamination > 10%, the remaining high-quality MAGs were processed by two clustering steps to remove replicates with default parameters. The rst is a rapid primary algorithm (Mash, ANI = 0.9), and the second is a more sensitive algorithm (ANI = 0.99). After that, we removed MAGs which size larger than 10Mb. At the end, 4960 non-redundant MAGs were obtained.
To calculate the coverage of each MAG in each sample, clean reads of each sample were mapped to the 4,960 MAGs using BWA-MEM with default parameters. After converting the resulted SAM les to BAM format using Samtools 97 (v.1.8), BEDTools 68 (v.2.27.1) was used to calculate the coverage of MAGs, which de ned as the total bases mapped to a MAG in a sample divided by its length.

Comparisons with reference microbial genomes and MAGs associated with model organisms
To check if our MAGs could improve the coverage of microbial genomes associated with buffalo's DT, MAGs of model organisms including human 65 , chicken 54 , pig 66 and cattle 31 were downloaded from their respective sources. In addition, a BFAP dataset was created to include reference microbial genomes (bacterial, fungal, archaeal, and protozoan) from the NCBI RefSeq genome database, and the Hungate collection genomes 63 .
BWA-MEM was used to map the "clean reads" to the above datasets and our MAGs as references. A mapping rate was calculated for each sample as the percentage of clean reads mapped to each of the reference datasets.

Taxonomic Assignments Of Buffalo Mags
Taxonomic assignments of the 4,960 MAGs were performed using the GTDB-TK tool 67

Data Availability
The raw sequencing data were submitted to the NCBI SRA database under the accession ID PRJNA656389; the sequences and annotations of the 4,960 MAGs are available at ENA under the accession ID ERZ1741894. These data will be public after the manuscript is published. stomach (orange), intestine (purple), rectum (green) and all samples that could be mapped to the collected datasets. In addition to our MAGs (buffalo), chicken, pig and human MAGs were obtained from gut metagenomes [9,11,35]; BFAP was the combination of reference genomes datasets including bacterial, fungal, archaeal and protozoan genomes from NCBI RefSeq and the genomes from the Hungate collection 63; cattle MAGs were obtained from rumen metagenomes [37].     activities (AA), c) carbohydrate-binding module (CBM), d) carbohydrate esterase (CE), e) glycoside hydrolase (GH), f) glycosyl transferase (GT), and g) polysaccharide lyase (PL) in the three sections (upper part) and eight sites (lower part). Y-axis: relative abundances (i.e., sum of all proteins in a functional category). More details about the de nition of the relative abundance of a protein can be seen in methods Wilcoxon Rank Sum Test was used to perform pairwise comparisons between sections (the upper part); * P < 0.05, ** P < 0.01, *** P < 0.001, **** P <0.0001.

Figure 6
Comparisons of rumen microbiota between buffalo and cattle. a) Relative abundances of selected taxa between buffalo (red boxes) and cattle (blue boxes), except panel 3 of the rst row, which shows Firmicutes/Bacteroidota ratios. b) The schematic diagram of plant ber digestion and methane metabolism in ruminants. Highlighted are the key microbial genera during these processes that showed signi cant differences between buffalo and cattle; Red: signi cantly higher in buffalo rumen, blue: signi cantly higher in cattle rumen. c) Comparisons of protein families between buffalo (red) and cattle (blue). Y-axis shows the relative abundances of protein families; here the relative abundance of a protein family is de ned as the percentage of reads mapped to the code sequences of member proteins in a family out of all reads mapped to all coding sequences (the sum abundances of all proteins are 100%). GH, glycoside hydrolase; GT, glycosyl transferase; PL, polysaccharide lyase; CE, carbohydrate esterase; AA, auxiliary activities; CBM, carbohydrate-binding module. * P < 0.05, ** P < 0.01, *** P < 0.001, **** P <0.0001.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.