The Planktonic Core Microbiome and Core Functions in the Cattle Rumen by Next Generation Sequencing

The cow rumen harbors a great variety of diverse microbes, which form a complex, organized community. Understanding the behavior of this multifarious network is crucial in improving ruminant nutrient use efficiency. The aim of this study was to expand our knowledge by examining 10 Holstein dairy cow rumen fluid fraction whole metagenome and transcriptome datasets. DNA and mRNA sequence data, generated by Ion Torrent, was subjected to quality control and filtering before analysis for core elements. The taxonomic core microbiome consisted of 48 genera belonging to Bacteria (47) and Archaea (1). The genus Prevotella predominated the planktonic core community. Core functional groups were identified using co-occurrence analysis and resulted in 587 genes, from which 62 could be assigned to metabolic functions. Although this was a minimal functional core, it revealed key enzymes participating in various metabolic processes. A diverse and rich collection of enzymes involved in carbohydrate metabolism and other functions were identified. Transcripts coding for enzymes active in methanogenesis made up 1% of the core functions. The genera associated with the core enzyme functions were also identified. Linking genera to functions showed that the main metabolic pathways are primarily provided by Bacteria and several genera may serve as a “back-up” team for the central functions. The key actors in most essential metabolic routes belong to the genus Prevotella. Confirming earlier studies, the genus Methanobrevibacter carries out the overwhelming majority of rumen methanogenesis and therefore methane emission mitigation seems conceivable via targeting the hydrogenotrophic methanogenesis.


INTRODUCTION
Ruminants are one of the most successful groups of herbivorous mammals on the planet, having evolved the forestomach, the rumen, where feed is degraded before it enters the true stomach and the rest of the digestive system. The rumen provides an environment for a diverse consortium of anaerobic microbes. These microbes produce enzymes which are needed to break down complex molecules, primarily plant polysaccharides (Dai et al., 2015). The decomposition of cellulose-rich fibers involves intimate symbiotic relationships in the microbiota (Moran, 2005). The hydrolysis of lignocellulose often limits the kinetics and efficiency of metabolite production (Chapleur et al., 2014;Güllert et al., 2016). During rumination, sophisticated mechanical pretreatment of the lignocellulosic substrate and the delicate oxygen gradient along the route of feed play important roles in making the degradation of recalcitrant substances more efficient (Bayané and Guiot, 2011). The rumen microbiome is a well-studied, although not thoroughly understood microbial ecosystem, which is able to utilize the plant material in an organized food chain (Krause et al., 2003;Chapleur et al., 2014). In the first step, Protozoa, Fungi, and Bacteria carry out the hydrolysis of polymers, such as cellulose and other complex carbohydrates, proteins and lipids to low molecular weight compounds (Lynd et al., 2002;Jindou et al., 2008;Güllert et al., 2016). Subsequently, fermentative bacteria convert these metabolites to short chain fatty acids (SCFAs) like acetate, propionate and butyrate, CO 2 , H 2 , alcohols and other compounds. Methanogenic Archaea produce methane in the final step of the microbial food-chain (Le Van et al., 1998;Hegarty and Gerdes, 1999;Lopez et al., 1999;Janssen and Kirs, 2008;Hook et al., 2010). These communities co-evolve with their host according to their ability to convert some of the SCFAs, methylamines, CO 2 and H 2 to CH 4 , CO 2 and water (Jami and Mizrahi, 2012;Meale et al., 2012;Poulsen et al., 2013;Jewell et al., 2015). The host benefits most if the process generates SCFAs (Weimer, 2015), which are absorbed through the rumen wall and used in the biosynthesis of sugars, lipids, and amino acids for the animal. Nonetheless, the thermodynamic and redox balances function properly only if a certain amount of chemical energy is released in the form of biogas, i.e., a mixture of CH 4 and CO 2 (Ungerfeld, 2014). The synthesis of acetate and butyrate involves the formation of reduced co-factors that require re-oxidization by methanogens (Ungerfeld, 2015). Conversely, propionate fermentation, which is the main precursor of glucose biosynthesis in ruminants, competes with methanogenesis for H 2 (Janssen, 2010). The release of CH 4 is an energetic loss for the animal and a major factor in global climate change (Martin et al., 2010;Opio et al., 2013).
Our knowledge of rumen microbiology has been accumulated by employing culture-based techniques. Studies on isolated cultures have provided a basic understanding of the biochemistry of the isolated strains (Bryant and Burkey, 1953;Hungate, 1960Hungate, , 1966Hobson, 1969). Unfortunately, most of the microbial strains in complex habitats cannot be cultivated in pure cultures; therefore, these approaches are of little help when the goal is the elucidation of the relationships between community members. The advent of nucleic acid-based molecular technologies opened up a new culture-independent perspective in microbial ecology. The majority of microbial identification studies rely on 16S ribosomal RNA (rRNA) gene sequencing. The 16S rRNA gene exists across bacterial and archaeal taxa and contains both highly variable and conserved regions (Woese and Fox, 1977;Skillman et al., 2006;Case et al., 2007;Wu et al., 2012), making it suitable for phylogenetic analyses (Case et al., 2007;Rajendhran and Gunasekaran, 2011). Nevertheless, 16S rDNA sequencing has some drawbacks as it suffers from biased choice of primers, PCR errors, and does not give functional information (Oulas et al., 2015). The meta-omic approaches give a much broader genomic and functional profile generated directly from environmental samples (Venter et al., 2001(Venter et al., , 2004Tyson et al., 2004). "Next generation sequencing" (NGS) has emerged, which employs various chemical reactions for the rapid determination of DNA sequences (MacLean et al., 2009;Metzker, 2010;Goodwin et al., 2016), produces huge datasets from relatively short sequence fragments and uses sophisticated bioinformatics to analyze the results (Huson et al., 2007(Huson et al., , 2011Raes et al., 2007). Initially, the most widely used NGS method in the identification of rumen communities and its functional mechanisms was based on 454-pyrosequencing (Mccann et al., 2014;Weimer, 2015;Güllert et al., 2016;Kala et al., 2017). Nowadays, other NGS approaches like Illumina TM , and Ion Torrent TM have also been employed to investigate these communities (Singh et al., 2014;Mao et al., 2015;Pitta et al., 2016;Comtet-Marre et al., 2017).
At low resolution, the ruminal community appears quite stable (Russell and Rychlik, 2008). Studies have pointed out that diet is the major factor, which has more of an influence than animal species on the rumen microbiome (Jami and Mizrahi, 2012;Henderson et al., 2015;Kala et al., 2017). Nevertheless, there is still limited information available about the core rumen microbiome, i.e., microbes that are shared between individual animals, and about the common functional elements comprising the metabolic networks (Shi et al., 2014;Henderson et al., 2015;Li and Guan, 2017). The aim of this study is to contribute to our understanding of these aspects by examining 10 individual Holstein dairy cow whole metagenome and transcriptome datasets.

Animals
Ten multiparous cows in the second to fourth lactation period were sampled. The dairy cows were of the Holstein breed, selected from the herd maintained at the Pilot Farm, Faculty of Agriculture, University of Szeged, Hódmezövásárhely, Hungary. This study was carried out in accordance with the recommendations of Hungarian Law of Animal Protection and Charity XXVIII/1998 and Government Decree 40/2013 (II.14.) about animal protection. The protocol was approved by the Ethical Committee of the Faculty of Agriculture, University of Szeged. The best practice veterinary care has been followed and informed consent has been granted by the University of Szeged. The animals were fed with a mixed forage-concentrate diet, using a forage wagon, and water ad libitum. Supplementary Table S1 shows the composition of the diet.
The stomach tube method was used to obtain rumen samples. The plastic collecting tube was 25 mm in diameter and was operated by manual aspiration of rumen fluid. The first 300 mL was discarded to avoid possible contamination with saliva. Sterile plastic containers were filled with 300 mL of rumen contents, flushed briefly with CO 2 gas, snap-frozen in liquid nitrogen and transported to the laboratory within 1 h. The samples were stored at −20 • C. After thawing, the material was filtered through a sterile mesh to remove suspended solids and used for further processing.

DNA and RNA Isolation for Meta-Omic Studies
For total community DNA isolation, 2 mL of rumen liquid samples were used. DNA extractions were carried out using a slightly modified version of the Zymo Research Fecal DNA kit (D6010, Zymo Research, Irvine, CA, United States). The lysis mixture contained 100 µL CTAB (cetyltrimethylammonium bromide) to improve the efficiency (Wirth et al., 2015a,b). After lysis (bead beating), the Zymo Research kit protocol was followed.
For total RNA isolation, 1 mL of rumen liquid samples were taken. The RNA extractions were carried out with the Zymo Research Soil/Fecal RNA kit (R2040, Zymo Research, Irvine, CA, United States). After lysis (bead beating), the Zymo Research kit protocol was followed. The DNA contamination was removed by Thermo Scientific Rapidout TM DNA removal kit (K2981, Thermo Fisher Scientific, Waltham, MA, United States).
The quantity of DNA and RNA was determined in a NanoDrop ND-1000 spectrophotometer (NanoDrop Technologies, Wilmington, DE, United States) and a Qubit 2.0 Fluorometer (Life Technologies, Carlsbad, CA, United States). DNA purity was tested by agarose gel electrophoresis and on an Agilent 2200 TapeStation instrument (Agilent Technologies, Santa Clara, CA, United States). The quality of the RNA preparation was checked by agarose gel electrophoresis (data not shown).

Next Generation Sequencing and Data Handling
In sample preparation for total metagenome and metatranscriptome sequencing, the recommendations of the Ion Torrent PGM TM sequencing platform were followed (Life Technologies, United States) using Ion Torrent PGM TM 316 chips. Before metatranscriptome sequencing, rRNA was depleted from metaRNA by using the Gram+/Gram-depletion kit in 60:40 ratio (RiboMinus A15020 Life Technologies, United States).
The sequences which were produced by Ion Torrent PGM were denoised, normalized and sequencing artifacts were removed by MG-RAST software pack (DRISEE, Dinamic Trim) (Cox et al., 2010;Meyer et al., 2011;Keegan et al., 2012). The filtered data were downloaded and were further analyzed by Diamond applying default LCA (Lowest Common Ancestor) algorithm (Buchfink et al., 2014). Diamond was set as follows: Blast Mode: BlastX, Min Score: 50, Max Expected: 0,01, Top Percent: 10, Min Support: 1. The data were further analyzed by Megan6 (Huson et al., 2007(Huson et al., , 2011(Huson et al., , 2016. The NCBI Taxonomy database was used for taxonomic alignment, which is a standard nomenclature and classification repository for the International Sequence Database Collaboration (INSDC) comprising the GenBank, ENA (EMBL) and DDBJ databases (Federhen, 2015). For functional annotations, the InterPro database was applied. This database integrates several pieces of information about protein families, domains and functional sites (Finn et al., 2017). For carbohydrate active enzyme identification, the CAZy database was used 1 (Lombard et al., 2014). Computing core microbiome and functions Megan6 co-occurrence plot function was applied with the following parameter sets: Threshold (%): 0.01, Min Prevalence (%): 0 Max Prevalence (%): 100, and Probability (%): 100. For functional statistic calculations, the R software was used 2 . The metatranscriptomic sequences were used to link functions to taxa by exporting the data from Megan and further analysis by the R software. The detailed sequence fragment parameters are summarized in Supplementary Table S2. The raw data were deposited in NCBI SRA database under the submission SRP148947.

Metagenomic Profiles
Sequencing genomic rumen DNA by Ion Torrent PGM produced an average of 305,884 reads with an average length of 168 bp. In average, 286,786 reads passed the quality control performed by the MG-RAST software package. These were subjected to quality control and were further analyzed by Diamond and Megan6. An average of 106,809 reads was obtained as known taxonomy units based on NCBI Taxonomy. Taking into account that 10 rumen samples were involved in the analysis, the overall read number exceeded 1 million (Supplementary Table S2). The most widespread phyla among Prokaryota are Bacteroidetes, Firmicutes, Proteobacteria, Euryarchaeota, and among Eukaryota are Ciliophora, Ascomycota, and Neocallimastigomycota. The taxonomic distribution at class level showed that the most abundant classes within Bacteria were Bacteroidia followed by Clostridia, Bacilli, Spirochaetia and Gammaproteobacteria (Figure 1). Within Archaea, the class Methanobacteria prevails. Litostomatea, Saccharomycetes and Neocallimastigomycetes are the most abundant classes among Protozoa and Fungi. Rarefaction analysis was performed at genus level. The curves reach its asymptotic nature, indicating that the sequencing depth at this scale is adequate (Supplementary Figure S1).

Core Microbiome of the Rumen
Hundreds of microbial genera were identified after the double filtering approach, corroborating that the rumen microbiome is indeed a very diverse community. Co-occurrence analysis was used both on DNA and RNA data to select the planktonic core microbiome, i.e., the group of microbes found in all 10 rumen fluid samples, which were metabolically active. The core included 48 genera (Figure 2), which comprised about 18 and 23% of all genera, based on DNA and RNA sequences (Supplementary Figure S2).
The class Bacteroidia of the phylum Bacteroidetes contributed the highest number of core DNA sequences, although the species diversity of this group was relatively low. Prevotella prevailed over other Bacteroidia genera. The genus Bacteroides was among the top ten abundant genera in this class. The phylum Bacteroidetes contained unknown and uncultured species, like bacterium P3, P201, F082, F083 or the uncultured class Prevotellaceae (Stevenson and Weimer, 2007) (Figure 2). Alistipes, Parabacteroides, Paraprevotella, and Porphyromonas were less frequent in the class Bacteroidia based on DNA data. Nevertheless, it should be noted that from these genera Parabacteroides, Paraprevotella, and from the uncultured and unclassified species bacterium P3, P201 and F082 showed high metabolic activity in RNA data.
The class Clostridia (phylum Firmicutes) produced fewer sequences in the core but was the most diverse higher taxon in the rumen fluid. Considerable amount of DNA sequence reads pointed at representatives of the genera Ruminococcus, Eubacterium, Clostridium, Butyrivibrio, Oribacterium, and Sarcina. Other genera such as Blautia, Lachnoclostridium, Faecalibacterium, Roseburia, Mogibacterium, Ruminoclostridium, Oscillibacter, and Dorea, were detected at a lower abundance. The representatives of genera Butyrivibrio, Oribacterium, and Lachnoclostridium had lower apparent metabolic activity than their DNA data suggested. Unclassified sequences were also found, which belonged in the families Ruminococcaceae, Lachnospiraceae and the phylum Clostridiales. It is noteworthy that more than 13 genera of the core 48 taxa belonged in the class Clostridia (Figure 2).
Also belonging in the phylum Firmicutes, the class Erysipelotrichi contributed to the planktonic core rumen microbiome with unclassified Erysipelotrichaceae, although this class was represented in relative small numbers. The genera Lactobacillus, Streptococcus, Bacillus, and Paenibacillus, in the class Bacilli, and the genus Mycoplasma (class Mollicutes) were barely detectable in the DNA based core microbiome. Few sequences of the genera Succiniclasticum and Selenomonas, cataloged in class Negativicutes, from which Selenomonas had a few related functions in the functional core compared to the DNA based hits (see Metatranscriptomic Profiles).
The class/phylum Spirochaetes was represented by the highly abundant genus Treponema.
The genus Fibrobacter, belonging in the phylum Fibrobacteres, class Fibrobacteria, and genus Bifidobacterium in the phylum/class Actinobacteria were also identified within the core genera. Surprisingly, significantly less RNA sequences mapped to the genus Fibrobacter than in the DNA data. Among the minor components of the functionally active core microbiome were other members of Actinobacteria, such as the genera Olsenella, and Slackia.
From the phylum Proteobacteria, class Gammaproteobacteria, the genera Acinetobacter and Ruminobacter were identified. The genus Chlamydia (class/phylum Chlamydiae) was observed in diminishing quantity in DNA reads, and it had even smaller number of RNA hits.
About 1% of all the identified sequences belonged in the phylum Euryarchaeota, class Methanobacteria. Only the genus Methanobrevibacter was verified from this taxon among the core microbes. It is noteworthy that, a comparable number of RNA and DNA sequences were mapped to the genus Methanobrevibacter.
The protozoa content of the rumen was about two orders of magnitude lower in cell number than the bacterial cell number (∼0.002%). The genus Entodinium (protozoan phylum Ciliophora, class Litostomatea) found its place in the DNA based planktonic core microbiome.

Metabolic Pathways
Ion Torrent PGM RNAseq produced an average of 680,886 reads with an average length of 185 bp. In average, 661,310 reads passed the MG-RAST software quality control (Supplementary Table S2). From the mRNA sequences, in average 296,618 contained predicted protein coding regions with known functions, based on the InterPro database.
Most of the reads were related to metabolic or electronproton exchange activity. Enzymes involved in biosynthetic, small molecule metabolism, carbohydrate metabolism and redox activities were among the most frequent. Housekeeping functions like translation, transport, RNA, DNA and amino acid metabolism were less persistent. The multifaceted picture was confirmed when putative protein sequences derived from the mRNA data set were projected onto KEGG metabolic pathways providing an integrated picture of global cell Frontiers in Microbiology | www.frontiersin.org

CAZymes
A great number of cellulases, hemicellulases, and oligosaccharide-degrading enzymes were identified when the filtered reads were probed against the CAZy database. The share of putative CAZymes was 4% in the total RNA database (Figure 3). It is noteworthy that DNA-based metagenome data gave similar results, with 6% of DNA reads coding for carbohydrate-degrading genes (Figure 3). Moreover, the order of identified enzyme families showed similar tendency. Differences habitually appeared in the relative numbers of corresponding genes; the DNA sequences gave more matches compared to the RNA data. This may be explained by the difference in the number of coded and expressed genes. It is conceivable that not all of the genes in the genome are expressed at all times. Nevertheless, similar trends in the two sets of data clearly indicate that metagenomic DNA data, obtained by NGS, can provide useful information on the main metabolic functions. Among the CAZymes, five glycoside hydrolase (GH) families were prevalent among cellulases in our rumen samples, i.e., GH3, GH5, GH8, GH9, GH44, and GH48. The GH3, GH5, and GH9 families were the most copious among cellulases (Figure 3). The GH3, G5 and GH8 families are widely distributed in bacteria and fungi and have a variety of hydrolyzing activities. The GH9 and GH44 families mainly contain endoglucanases, while GH48 is a reducing end-acting cellobiohydrolase. Eight GH families (GH3, GH8, GH10, GH11, GH16, GH28, GH30, GH43, GH51, GH53, GH67, and GH78) are primarily hemicellulose degraders, and have both endo-acting and debranching enzymes. The identified frequent oligosaccharide-degrading GH families are as follows: GH1, GH2, GH13, GH27, GH29, GH31, GH35, GH36, GH38, GH42, and GH77. GH1, GH2, GH13, and GH77. Carbohydrate-binding modules (CBMs) are specialized domains in plant cell wall polysaccharide-degrading enzymes. They promote the intimate attachment of catalytic domains to the substrates and enhance the activity of the enzymes 3 . Nine CBM families, CBM2, CBM3, CBM6, CBM13, CBM20, CBM26, CBM32, CBM34, and CBM48, were identified in both the DNA and mRNA databases. CBM2, CBM3, and CBM6 are specific for cellulose, CBM20, CBM26, and CBM34 have a preference for starch and CBM13, CBM32, and CBM48 families have oligo-and monosaccharide-binding functions. Supplementary Table S3 contains the entire list of the identified CAZymes.

Core Metabolic Functions
Functional analysis of whole transcriptome data pinpointed thousands of active gene transcripts, confirming the metagenomic data from a distinct view, i.e., the rumen microbiome is a highly active metabolic system that is cumulatively equipped with copious numbers of biochemical functions. Co-occurrence analysis was performed to filter out the most common functions present in our samples. Filtering of the mRNA sequence database yielded at least 587 genes, which may form the core functions indicated by their interactions and presence in all data sets (Supplementary Figure S7 and Supplementary Table S4). According to the InterPro database, 62 genes from this pool apparently code for enzymes involved in metabolic pathways (Figure 4). This is a small number for the enzymes to comply with the requirements of basic biochemical life-sustaining fermentative pathways, which is likely due to the stringent filtering conditions used in the co-occurrence analysis. The remaining transcripts are housekeeping genes required for the maintenance of basic cellular functions. The core metabolic processes provided satisfactory coverage of the entire microbiological food chain, from carbohydrate metabolism to methanogenesis. From the 62 core metabolic function coding genes, 18 belonged to carbohydrate processing, which comprised 26% of all core transcripts. Most of them encoded glycoside hydrolase families.
A diverse group of redox enzymes such as pyruvate-flavodoxin oxidoreductase (PFOR), glyceraldehyde/erythrose phosphate dehydrogenase family, the alpha subunit of indolepyruvate ferredoxin oxidoreductase, malic oxidoreductase, and Llactate dehydrogenase indicated highly vigorous redox processes, which made up 28% of all core transcripts. Glutamate dehydrogenase, glutamate synthase and glutamate-5-semialdehyde dehydrogenase within this group signaled active ammonia assimilation (Yuan et al., 2009) by the rumen community. HydE, responsible for the maturation of [FeFe], hydrogenase suggests the presence of hydrogen metabolism in each rumen sample (Figure 4 and Supplementary Figure S6).
A distinct group of the represented core functions take part in the synthesis of precursor metabolites and energy supply including ATP dependent 6-phosphofructokinase, fructose-1,6-bisphosphate aldolase, phosphoglycerate kinase, enolase, phosphoglucose isomerase, adenylate kinase and pyruvate kinase were found in high transcript numbers. Numerous additional enzymes were identified in the core functional group, which are also implicated in various biosynthetic pathways (Figure 4).
Overall, 7% of the total transcripts represented enzymes playing a part in organic acid metabolic processes, i.e., acetate/propionate kinase, methylmalonyl-CoA, acetyl-CoA carboxylase carboxyl transferase, succinate CoA transferase and butyrate kinase (Figure 4 and Supplementary Figure S4).
Although they represent only 1% of the total core transcripts, Archaea enzymes performing methanogenesis play important functions in maintaining the proper redox balance in the rumen. Methyl-coenzyme M reductase alpha, beta, and gamma subunits and 5,10-methylenetetrahydromethanopterin reductase were detected in all of the rumen samples investigated (Figure 4 and Supplementary Figure S6).

Linking Core Functions to Taxa
In most metagenomic studies, the composition of the microbial community is determined and changes in relative abundances are recorded as the results of changes in various environmental conditions or feeding regimes. More recently, important experiments have complemented this information aiming at functional aspects by metatranscriptome, proteome and metabolome investigations concentrating on the "who does what?" question. The metabolic functions are grouped according to the InterPro classification 4 .
In all of the functional groups associated with Bacteria, the genus Prevotella (phylum Bacteroidetes) prevailed (Figure 5), due to its vast predominance in the rumen planktonic microbiota and the versatile metabolic pathways possessed by this genus. Accordingly, the share of Prevotella in carbohydrate (∼63%), small molecule (∼50%), organic acid (∼53%) metabolisms, oxidation-reduction processes (∼44%), and generation of precursor metabolites (∼42%) was high, indicating that the majority of the metabolic "work" taking place in the rumen fluid is carried out by representatives of this taxon (Figure 5). Seven additional genera complemented the picture participating in the various metabolic pathways with varying activity. These 4 https://www.ebi.ac.uk/interpro/ were as follows: Bacteroides, Ruminococcus, Bifidobacterium, Clostridium, Lactobacillus, Paraprevotella and Eubacterium. It is interesting to note that the diversity ranking at the genus level is different from the abundance distribution based on the metagenomic data (Figure 2). The divergences between DNA and RNA based taxonomy data suggest that abundance and functional importance are not necessarily coupled.
The genus Bacteroides extensively contributed to carbohydrate, small molecule and organic acid metabolisms and played a significant role in other Bacteria-linked metabolic processes as well. The genus Ruminococcus was primarily active in the generation of precursor metabolites and energy although these species also participated in other bacterial tasks. Clostridium was more active in handling oxidation-reduction processes than Ruminococcus, and lagged behind Ruminococcus in carbohydrate metabolic process performance. In spite of their low relative abundance in the community, the genus Lactobacillus displayed remarkable contribution to tasks related to the generation of precursor metabolites and energy. Bifidobacterium showed considerable involvement in carbohydrate metabolism-related pathways and a noticeable one in organic acid metabolism.
Paraprevotella and Eubacterium were detected in the carbohydrate and small molecule metabolic processes. In the functional group "oxidation-reduction processes, " an uncultured bacterium was observed as the second dominant genus.
The genera Kandleria, Parabacteroides, Porphyromonas, Selenomonas, Alistipes, Catenibacterium, Barnesiella, and Mitsuokella were found less frequently among the various metabolic functions, perhaps due to their low abundance in the planktonic rumen microbiota. Lachnospiraceae and Treponema were the apparent exceptions as they were present at a relatively high number in the core rumen microbiome and proportionally low related activity.

The Core Microbiome
The following discussion is restricted to genera present in the core microbiome and have functions associated to them (Supplementary Table S4). This information will be related to the results of previous rumen studies. In this study, our fundamental assumption has been that the taxa occurring together in the various rumen microbiota have shared functional connections and are therefore in some sort of metabolic relationship. This may not be the case in every individual interacting partner pair, but it is a reasonable postulation to find FIGURE 5 | Coupling functional groups to taxonomic genera determined from the metatranscriptomic dataset.
interrelationships between taxonomic and functional analyses in the core community.
Within the domain Bacteria, the majority of the identified sequences were annotated to the genus Prevotella (class Bacteroidia) (Figure 2 and Supplementary Figure S2). Previous studies, using qPCR and/or 16S rDNA sequencing, also identified this genus as the most abundant in the rumen bacterial population (Stevenson and Weimer, 2007;Jami and Mizrahi, 2012). Further confirmation came from the analysis of the Suslike (Starch utilization system) polysaccharide utilization loci (Naas et al., 2014;Rosewarne et al., 2014;Güllert et al., 2016). Changing dietary conditions did not alter Prevotella abundance (Stevenson and Weimer, 2007;Bekele et al., 2010;Purushe et al., 2010;Pitta et al., 2014;Lyons et al., 2017), although an agerelated shift from Prevotella to Succinivibrio was noted recently (Liu et al., 2017). This is likely due to their ability to serve the wellbeing of their host in various ways, exploiting their high degree of genetic diversity (Avgustin et al., 1994;Ramšak et al., 2000) and remarkable metabolic versatility (Bryant et al., 1957;Wen et al., 1996;Matsui et al., 2000). In addition, Prevotella is a highly diverse taxon comprising various functional niches in different systems (Purushe et al., 2010).
The second most widespread group in the class Bacteroidia was the genus Bacteroides. This genus, like Prevotella, can utilize polysaccharides as an energy source. Apparently, some rumen Bacteroides can also decompose cellulose (Naas et al., 2014). Smaller genera in the class Bacteroidia include Parabacteroides, Paraprevotella and Porphyromonas, in addition to the uncultured Prevotellaceae family. These are usually detected in cow manure and their relative abundances change during the transition from developing and mature rumen (Dowd et al., 2008;Wu et al., 2012).
The classes Bacteroidia (phylum Bacteroidetes) and Clostridia (phylum Firmicutes) together represented ∼47-30% of the rumen core microbiome. In a recent study, the ratio of the two taxa in the whole rumen community was close to 1:1 (Güllert et al., 2016). Interestingly, in the biogas producing anaerobic microbial community metabolizing similar lignocellulosic substrates, Clostridia surpass greatly Bacteroidia (Güllert et al., 2016;Bozan et al., 2017). Apparently, the two classes differ in their colonization strategies. Only a few genera of the class Bacteriodia were present in the rumen, although in large relative abundances, whereas a remarkable diversity of genera characterized the class Clostridia.
Among Clostridia, the thoroughly studied genus Ruminococcus was frequently found. Their interactions with other species may make this genus more important than their abundance implies (Fondevila and Dehority, 1996;McSweeney et al., 1999;Koike and Kobayashi, 2009;Christopherson et al., 2014;Dai et al., 2015). Members of the family unclassified Lachnospiraceae and the genera Eubacterium and Blautia are among the taxa in the core of the cow rumen and present also in the kangaroo forestomach (Henderson et al., 2010;Godwin et al., 2014). These bacteria acquired the Wood-Ljungdahl pathway, which is alternatively called the reductive acetyl-CoA or the reductive acetogenesis pathway (Hattori, 2008;Ragsdale and Pierce, 2009;Xu et al., 2009;Gagen et al., 2015;Kelly et al., 2016) (Supplementary Figure S4).
The genus Clostridium possess high cellulolytic activity and actively produces H 2 (Lin et al., 2007). The partial pressure of H 2 determines the rate of methanogenesis and the assortment of short chain fatty acids generated in the rumen (Hegarty and Gerdes, 1999;Janssen, 2010). Methanogens outcompete the microbes involved in reductive acetogenesis because acetogens are less efficient at obtaining energy from the oxidation of H 2 (Le Van et al., 1998;Siriwongrungson et al., 2007;Ungerfeld, 2015). The genus Butyrivibrio is present in a large selection of ruminants and has been characterized as an oligosaccharidedegrading bacterial taxon (Dai et al., 2015;Henderson et al., 2015).
The genus Oribacterium was observed as a particle associated bacterium in the rumen and the genus Mogibacterium was identified as a predominant member of the cattle gastrointestinal tract, although little is known about their functions (Mao et al., 2015;Schären et al., 2017). The genera Faecalibacterium, Roseburia, Ruminiclostridium, Oscillibacter, and Dorea were represented in low abundance. These diverse genera are active in cellulose and hemicellulose decomposition and ferment various sugars to short chain fatty acids (Duncan et al., 2006;Flint et al., 2008;Koike and Kobayashi, 2009;Thoetkiattikul et al., 2013;Ravachol et al., 2015).
The genera Sharpea, Lactobacillus, Streptococcus, Bacillus, and Paenibacillus represented the core members of the class Bacilli (phylum Firmicutes). Perhaps not surprisingly, the lactic acid bacteria Streptococcus and Lactobacillus tolerated aciduric challenge well in the cow rumen (Petri et al., 2013). Representatives of the genera Bacillus and Lactobacillus produce organic acids, some of which may endow probiotic effects (Qiao et al., 2010;Seo et al., 2013;Ushakova et al., 2013). The genus Paenibacillus has been noted for its hemicellulose-degrading activity (Kala et al., 2017), while the genus Sharpea tends to form and metabolize lactate an thereby increased its relative abundance in low methane yield sheep rumen (Kamke et al., 2016).
Within the core rumen planktonic microbiome, the genera Succiniclasticum and Selenomonas, were also present, belonging to the class Negativicutes (phylum Firmicutes), although in low abundance. The genus Selenomonas is lactate-utilizing bacteria (Hackmann and Firkins, 2015). Succiniclasticum members apparently differed in quantity between hay diet-and high grain diet-fed goats and have been described as starch degraders (Kim et al., 2011;Huo et al., 2014).
Three identified taxons in the core microbiome belonged in the class Actinobacteria (phylum Actinobacteria), these were the genera Bifidobacterium, Olsensella, and Slackia. Certain members of these genera are considered probiotic in both ruminants and humans as they metabolize oligosaccharides and release lactic acid, which helps to control the normal microflora (Picard et al., 2005;Nagai et al., 2010;Kamke et al., 2016).
The genus Treponema belong to the class/phyla Spirochaeta. Microbes in this genus are important fatty acid producers in the rumen (Yamada and Yukphan, 2008;Henderson et al., 2010;Li et al., 2012). The genus Fibrobacter, member of the class Fibrobacteria (phylum Fibrobacteres), contains well-known cellulose degraders (Dai et al., 2015). Our transcriptomic data indicate that the activity of this genus is lower than suggested by the DNA based hits. Similar phenomenon could be observed in case of the genera Sharpea, Treponema, Butyrivibrio, bacterium P3, bacterium P201 and Lachnoclostridium. This may imply that, not always the predominant genera play essential role and some microbes present in small numbers may have significant activity in the microbial ecosystems.
In the rumen core microbiome the single genus Methanobrevibacter was found as the most prominent Archaea taxon. They are hydrogenotrophic methanogens and reduce CO 2 to CH 4 when H 2 is present as an electron donor (Deppenmeier et al., 1996). The high rate of short chain fatty acid passage through the rumen wall via active uptake may explain the predominance of hydrogenotrophic Archaea, as the short chain fatty acid (SCFA) uptake is faster than the growth rate of acetate-utilizing methanogens (Dijkstra, 1994).
The total Protozoa and Fungi cell number in the rumen is two orders of magnitude lower than that of the Bacteria. In our data among Protozoa, the single genus Entodinium (class Litostomateae) positioned itself in the DNA based core microbiome, which reflects the uneven distribution of Protozoa in the individual rumen microbiota. The genus Entodinium is very efficient in starch fermentation and is a frequently detected protozoan in the rumen. (Jouany and Ushida, 1999;Zhang et al., 2017). Surprisingly, none of the Fungi were present in the DNA and RNA based core microbiome, suggesting that the fungal community is extremely varied across the individual rumen communities.

Core Functions and Their Associated Microbes
Numerous studies recognized that the functional and phylogenetic distribution of microbes in the rumen comprise an integrated system; therefore, both aspects should be investigated together to better understand the complex metabolic processes taking place in this ecosystem (Deusch et al., 2015;Bielak et al., 2016;Mayorga et al., 2016;Shabat et al., 2016;Wang et al., 2016;Shen et al., 2017). In the present study, core metabolic functions and their accompanying microbes were determined from double filtered metatranscriptomic data by pairing the functional groups with the most probable taxonomic units. This resulted in a comprehensive coverage of the microbial food chains from complex carbohydrates to end products, i.e., SCFA and CH 4 , to be utilized or released by the host animal. In the following section, we therefore attempt to reconstruct a metabolic map of the physiological events taking place in the cattle rumen (Figure 6) and discuss the identified microbes associated with these biochemical reactions, extending the findings of similar previous studies (Stevens and Hume, 1998;Russell and Rychlik, 2008;Ungerfeld, 2014Ungerfeld, , 2015Jiang et al., 2016;Deusch et al., 2017).
Taking the transcriptomic and taxonomic data together, the genera Prevotella, Bacteroides and Ruminococcus seem to predominate in the deconstruction of the lignocellulose rich feed.
Although the genera Bifidobacterium and Bacteroides contained the FTHFS gene, the enzyme does not exclusively serve the Wood-Ljungdahl pathway in these microbes; it functions rather as a methyltransferase in purine and glycine degradation and in the metabolism of some sulfate-reducing bacteria (Xu et al., 2009;Henderson et al., 2010). The genera Ruminococcus, Clostridium (∼1%), Blautia (∼1%) and order Clostridiales (∼0.5%) include known homoacetogens but they are represented in low transcript frequency (Supplementary Table S3).
Along the pathway of heterotrophic acetate production, acetyl-CoA becomes phosphorylated to acetyl-phosphate (Schuchmann and Müller, 2014). The acetate/propionate kinase enzyme (IPR004372) can carry out this reaction. Acetate kinase catalyzes the acetyl-phosphate to acetate reaction with concomitant ATP generation (Hasona et al., 2004). The corresponding transcript was widespread in the genus Prevotella (∼69%) followed by Bacteroides (∼7%), Clostridium (∼3%), Ruminococcus (∼3%) and Lactobacillus (∼3%). Acetate kinase can be involved both in the reductive acetogenesis and heterotrophic acetate production (Schuchmann and Müller, 2014); thus, the genera attached to this reaction are probably mixed. Butyrate is produced by butyrate kinase (IPR011245) from butyryl-CoA (Louis et al., 2004). The corresponding members of the rumen fluid microbial community were Prevotella (∼52%) and Bacteroides (∼14%), respectively. Our data indicated that several gene products contributed to the regulation of acetate and butyrate production, i.e., the alpha subunit of acetyl-CoA carboxylase (IPR001095), the beta subunit of acetyl-CoA carboxylase carboxyl transferase (IPR000438) and the acetyl-CoA biotin carboxyl carrier (IPR001249) (green lines in Figure 6). These enzymes carry out the carboxylation of acetyl-CoA to malonyl-CoA, thus they tightly regulate the SCFA synthesis (Rock and Jackowski, 2002). The taxa linked to this regulation were the genera Prevotella and Ruminococcus (Supplementary Table S3).

CONCLUSION
The metagenome and metatranscriptome of the rumen fluid fractions of ten lactating cows showed a fairly consistent picture, which corroborated and extended our knowledge from similar previous studies. Ten separate rumen samples were analyzed, about one third of the average 305,844 Ion Torrent DNA reads could be associated with taxonomic information in the NCBI Taxonomy database. In the RNAseq experiments an average of 680,886 reads were obtained and approximately half of them predicted proteins of known functions. Each set of data was subjected to quality control and filtering before co-occurrence analysis, which was based on the assumption that those microbes which occurred together were likely to form physiological relationships.
A core microbiome, i.e., a collection of taxa present in all ten sequenced samples, was established. The taxonomic core microbiome consisted of 48 genera belonging in Bacteria and 1 Archaeon. Based on their relative abundances, about a dozen genera formed the majority of the microbiota; the genus Prevotella exceedingly predominated the community. Fungi were not placed in the core microbiome, which might suggest that either fungi formed a diverse community or they were present at a very low number in our samples and their DNA sequence reads could not reach the threshold.
Thousands of mRNA sequences pointed to active proteins, implicating a complex and diverse metabolism taking place in the rumen. From the 587 functions, a core functional group was distinguished based on co-occurrence analysis, but only 62 of them could be assigned to metabolic functions. This is clearly a low number to cover all possible metabolic pathways and is ascribed to very stringent filtering conditions and to the selection of common elements in 10 separate rumen fluid samples. Nevertheless, even this minimal functional core revealed key enzymes participating in various metabolic processes. As expected, a diverse and rich collection of enzymes was involved in carbohydrate metabolic processes, but other functional groups were also sufficiently represented. Transcripts coding for enzymes active in methanogenesis made up 1% of the core functions.
The genera associated with the core metabolic functions were identified. The main conclusion drawn from this investigation was that for all metabolic functions, performed primarily by Bacteria, several genera could provide the necessary activity. In other words, there always seems to be a "back-up" microbial team to substitute the predominant bacteria to accomplish any given metabolic function. Nevertheless, the key actors in most metabolic functions belong to the genus Prevotella. The Prevotella predominance is partly due to their massive abundance and partly to the metabolic versatility of this taxon. Contrary to the extensive potential to share the contribution to the tasks among bacterial taxa, methanogenesis seemed to have limited possibilities, i.e., only the hydrogenotrophic pathway existed with a limited possible role for "unidentified" methanogens. The potential bypass biochemical routes make the diversion of metabolism toward advantageous pathways by managing the rumen community composition a considerable challenge.
Nonetheless, methane emission mitigation seems conceivable via targeting the hydrogenotrophic genus Methanobrevibacter.

AUTHOR CONTRIBUTIONS
KK conceived the study, participated in its design and evaluation. GK and JH collected the rumen samples and took part in the assessment of the data. RW, GM, and BK performed the sequencing experiments and statistically evaluated the metagenomic data sets. KK, RW, ZB, GR, and ÁS composed the manuscript. All the authors agreed in publishing the final version of this paper.

ACKNOWLEDGMENTS
The support and advice of Professor János Minárovits, Dean Kinga Turzó (Faculty of Dentistry, University of Szeged) are gratefully acknowledged.
FIGURE S2 | Co-occurrence analysis of the metagenome (DNA) samples. Enlargement of the picture shows the taxon belonging to the colored dot in the network. Taxa identified in the core of all ten rumen samples were included in the calculation.