RIM-DB: a taxonomic framework for community structure analysis of methanogenic archaea from the rumen and other intestinal environments

1Ruminants such as sheep and cattle are among the most numerous farmed animals, and enteric methane formation in their rumens contributes substantially to global greenhouse gas emissions, as well as being a conversion of part of the feed energy into a form unavailable to the ruminant animals. Methane is produced by methanogenic archaea, a phylogenetically diverse group of microorganisms. Depending on the species and substrate availability, methanogens can grow hydrogenotrophically (using hydrogen or formate), aceticlastically (using acetate) and/or methylotrophically (using methanol or other simple methyl-compounds). Rumen methanogens consume hydrogen, formate, and methyl-compounds that are among the products formed during the degradation and fermentation of ingested feed through the combined activities of bacteria, fungi, and protozoa. Aceticlastic methanogenesis does not appear to be a significant source of methane in the rumen and has only been observed in exceptional cases. Methane cannot be used by the ruminant and it is lost to the atmosphere, mainly through eructation. To mitigate emissions of methane from ruminants into the atmosphere, interventions are being developed to reduce the number or activity of methanogenic organisms in the rumen. Interventions like targeted vaccines and inhibitors are based on genome sequences of key methanogens. In addition, tools based on DNA markers are being used to monitor the effects of experimental interventions or to uncover differences in microbial community structures in animals with different productivity traits, such as differences in feed conversion efficiency. Therefore, the accurate identification and classification of rumen methanogens is an important step, whether to identify target methanogens, for subsequent community analysis during the development and testing of mitigation technologies, or to uncover details of rumen community differences in animals with different production characteristics.

2Next generation sequencing technologies have revolutionised cultivation-independent methods of characterising microbial communities, and it has become feasible to sequence amplicons at unprecedented depth and from large numbers of samples. Having overcome the hurdle of generating large enough amounts of sequencing data to cover a large part of the diversity of the microorganisms present in a sample, the focus now needs to shift to another crucial aspect of community structure characterisation, namely the correct taxonomic assignment of sequencing reads. This step relies on the availability of high-quality reference sequences for the relevant environment. The rumen is known to contain methanogens from at least four different orders: Methanobacteriales, Methanomicrobiales, Methanosarcinales and Methanomassiliicoccales. The availability of reference 16S rRNA gene sequences for each order varies considerably and is particularly low for Methanomassiliicoccales, which have recently been proposed as a new order of methanogens (for which the synonymous name ‘Methanoplasmatales’ has also been proposed). Only six almost full-length sequences (>1,200 bp) from isolates or enrichment cultures are currently publicly available for Methanomassiliicoccales, making it difficult to obtain a comprehensive overview of this order and to define taxonomic groups. This limitation can partially be overcome by including almost full-length PCR-amplified sequences from cultivation-independent studies, especially older clone library-based investigations. The downside of this approach is that these sequences may potentially be of lower quality. Cultivation-independent studies rely on sequenced PCR products and may include artefacts, such as chimeras. The removal of such low quality sequence data from reference databases is therefore necessary to produce high quality and trustworthy taxonomic assignments and to reliably define taxonomic groups that do not yet have cultured representatives.

3The aim of the work described here was to develop a taxonomic framework that improves the quality and accuracy of taxonomic assignment of rumen and intestinal methanogens. We therefore developed a specialised taxonomic framework, a R umen and I ntestinal M ethanogen D ata b ase (RIM-DB). RIM-DB includes 16S rRNA gene sequences from ruminal methanogens, and, to make it more widely useful, from various other intestinal environments where methanogens are known to be important hydrogen consumers, such as the human and termite intestinal tract. The database contains long (>1,200 bp) 16S rRNA gene sequences that have been checked for potential chimeras. We believe that the use of specialised and/or refined taxonomic frameworks allows more detailed and accurate taxonomic assignments than other publicly available databases, such as Greengenes or SILVA, which are far more comprehensive and contain highly diverse sets of 16S rRNA sequences from many different environments, but which, due to their size, are not curated to the same extent. We tested the newly developed RIM-DB by analysing the composition of the methanogen microbiota in the rumens of sheep and cattle in New Zealand and compared the results to those obtained using SILVA and Greengenes taxonomies.

4Archaeal 16S rRNA gene sequences were exported from SILVA version 111 to develop a database that represents the seven orders of methanogens and the order Archaeoglobales. Initially, all sequences that fell into one of the five established orders (Methanobacteriales, Methanococcales, Methanomicrobiales, Methanopyrales, and Methanosarcinales) and into the two recently-proposed orders, Methanocellales and Methanomassiliicoccales, were selected. Sequences from the thermophilic genera of the order Methanobacteriales were removed from this selection because they are not known to be major colonisers of the rumen or the intestinal tracts of animals or humans. Sequences of thermophilic methanogens belonging to other orders and the non-methanogenic order Archaeoglobales were retained in the dataset so that it included representatives of all methanogenic and closely-related orders. Twenty-eight sequences of methanogens not represented in the SILVA 111 database were added to the sequence database (Table S1). Another twenty sequences were made publicly available as part of this study and were also added to the database (Table S1). These sequences include three novel cultured representatives of the order Methanomassiliicoccales. All RIM-DB sequences were subjected to rigorous chimera checks using a BLAST-based chimera checking protocol and the UCHIME pipeline, which are both part of the QIIME package, version 1.5. Both methods were run in reference mode, using 16S rRNA gene sequences from methanogen type strains and the Greengenes database (release GG_05_13) as reference sets, respectively. Sequences identified as chimeric by the parallel_identify_chimeric_seqs.py script, using the options –d 4 and –n 2, were removed from the database. Questionable sequences that were identified by UCHIME were manually inspected by BLAST analysis of 5′ and 3′ ∼400 bp-long fragments and discarded when considered to be chimeric. Duplicate sequences were eliminated unless they were derived from isolates or stable cultures. Following these steps, 2,379 sequences remained for analysis, and these are provided as a qiime-compatible database file (File S1).

5Sequences were aligned using the SINA aligner using default settings, then imported into ARB and the alignment, where necessary, manually curated. Aligned sequences were exported in phylip format to construct phylogenetic trees using all available base positions. Maximum likelihood phylogenetic trees based on aligned archaeal 16S rRNA gene sequences were generated using RAxML version 7.0.3. Unless stated otherwise, the parameters “-m GTRGAMMA -# 500 -f a -x 2 -p 2” were used. Taxonomic strings were generated for each of the database sequences according to the naming scheme used in the Greengenes taxonomy, consisting of seven different taxonomic levels: k_kingdom; p_phylum; c_class; o_order; f_family; g_genus; s_species. Two different qiime-compatible taxonomy files (File S2 and File S3) are provided, both intended to be used in conjunction with the reference sequence database (File S1). The taxonomy file in File S2 contains an eighth field, “i_isolate”, which has been introduced to designate which sequences originate from isolates or cultures. This file contains the taxonomic strings used in the ARB database (File S4). In the taxonomy file in File S3, most sequences were manually binned into species-level clades to define an association of the majority of sequences to known species. This file is intended for taxonomic assignment of sequences to designated taxa, for example when analysing large datasets of archaeal 16S rRNA gene sequences originating from the rumen or other gastrointestinal systems. Phylogenetic trees shown in the paper are also included in the ARB-database that is provided (File S4).

6Sequence coverage at each alignment position was calculated to determine if RIM-DB was suitable for taxonomic assignments of amplicon reads obtained from different regions of the archaeal 16S rRNA gene. Variability of base composition for sequences in RIM-DB was analysed in R version 3.0 using the seqinR package, and results were plotted using the ggplot2 library. Shannon index values for each alignment position were calculated using the frequencies of bases A, T (U and T were used synonymously), G, C and blank positions. Variable regions for the analysis were selected as described previously.

7Sequences of poorly-resolved clades were binned into defined groups to improve the accuracy and detail of taxonomic assignments as follows. All groupings of sequences was done in ARB. The groupings were primarily based on sequence identity by identifying sequences within a chosen identity cut-off (mostly 95%–97%) for a selected sequence and strong bootstrap support (≥70%) for any defined group.

8Benchmarking was performed on isolate sequences and on an amplicon test dataset. Twenty-four sequences of rumen and intestinal methanogens were selected for benchmarking with isolate sequences (see Table S2). The selected sequences were either exported from SILVA or are published as part of this study (see Table S2 for isolates). Analysis was performed on long length (>1,000 bp) sequences and on sequences of the V6–V8 variable regions of the 16S rRNA gene. Taxonomic assignment of sequences was carried out using the parallel_assign_taxonomy_blast.py script in QIIME, version 1.5. The three different reference databases used for taxonomic assignments of sequences were RIM-DB (File S1 and File S3), SILVA (release 111, ) and Greengenes (release GG_13_05, ). QIIME-compatible SILVA and Greengenes databases were downloaded from http://qiime.wordpress.com. Specific options/files used for taxonomic assignments with SILVA were: –id_to_taxonomy Silva_111_taxa_map_full.txt and –blast_db Silva_111_full_unique.fasta; and with Greengenes: –id_to_taxonomy gg_13_5_taxonomy and -blast_db gg_13_5.fasta. Abundance tables were generated and only OTUs with a mean minimum relative abundance of 1% across all samples were retained.

9A test set of amplicon sequence data was generated by combining the following sequence datasets (for accession numbers see Table S4). These datasets contain partial 16S rRNA gene sequences covering nucleotide positions 935–1,385 (Escherichia coli 16S rRNA nucleotide numbering). Sequence data were processed using the QIIME package, version 1.5. Reads were quality filtered and assigned to the corresponding sample by barcodes using the QIIME split_library.py script. Only reads with average quality scores >25 were included in the analysis. The resulting fna-files from all experiments were concatenated and denoised using combined flowgram-files, using the denoise_wrapper.py script with default settings. The output was subjected to the inflate_denoiser_output.py script (default settings). Denoised sequence reads were chimera-checked with the QIIME script parallel_identify_chimeric_seqs.py, using the parameters –d 4 and –n 2, and using RIM-DB as the reference database. The chimeric sequences that were identified were removed from the dataset using the QIIME filter_fasta.py script. Subsequently, the denoised and chimera-checked dataset was processed with the QIIME pipeline. Sequences were clustered into operational taxonomic units (OTUs) used the default clustering method UCLUST with a sequence similarity cut-off of 99% (pick_otus.py option: -s 0.99). Abundance tables were generated and only OTUs with a mean minimum relative abundance of 1% across all samples were retained. Taxonomic assignment of representative sequences was carried out as described for the isolate sequences.

10Taxonomic frameworks are essential tools for describing the composition of microbial community structures. This study had two main aims. We wanted to develop a taxonomic framework for rumen methanogens and methanogens residing in other intestinal habitats based on long (>1,200 bp) 16S rRNA gene sequences that resolves taxonomic groups that are currently unresolved by taxonomies such as Greengenes or SILVA. We achieved this by better defining taxonomic groups (corresponding to the genus and species levels) within the genera Methanobrevibacter and Methanosphaera and the order Methanomassiliicoccales to develop RIM-DB. We then wanted to test the utility of this framework to provide more detailed taxonomic assignments for shorter sequence data (covering the V6–V8 region) obtained from rumen microbial communities.

11RIM-DB contains a subset of currently available archaeal 16S rRNA gene sequences and is primarily designed for taxonomic assignment of methanogenic archaea from the rumen and other intestinal environments. A maximum likelihood-based phylogenetic analysis of 2,379 high-quality sequences indicated the existence of seven clearly-defined orders of methanogenic archaea (Fig. 1). Previous studies have shown that ruminants from different geographic locations around the world are dominated by members of only two of the seven orders, the Methanobacteriales and the Methanomassiliicoccales. Within the Methanobacteriales, the genera Methanobrevibacter and Methanosphaera have primarily been detected in the rumen; and both genera contain species that have also been found in a variety of other intestinal environments. Based on sequence similarity and bootstrap support, we combined database sequences into groups and developed a naming scheme for clades within the genera Methanobrevibacter and Methanosphaera and within the order Methanomassiliicoccales (Fig. S1 and Fig. S2).

12The high degree of 16S rRNA gene sequence similarity between some Methanobrevibacter spp. is reflected by the grouping of sequences from a number of species into one clade that cannot be separated further, which prevents a more detailed taxonomic assignment of some sequencing reads. Within the genus Methanobrevibacter, Mbb. ruminantium and Mbb. gottschalkii were each defined as clades that include some other formally described species, specifically Mbb. ollyeae, Mbb. millerae, and Mbb. thaueri. The latter three species have been defined based on DNA hybridisation and other phenotypic characteristics, but sequence divergence between the 16S rRNA gene sequences of these strains is in some cases <2%. In RIM-DB, the type strains of Mbb. millerae, Mbb. olleyae and Mbb. thaueri have been assigned their given species names, but due to the high sequence similarity, Mbb. thaueri and Mbb. millerae are placed in the Mbb. gottschalkii clade, while Mbb. ollyeae is placed in the Mbb. ruminantium clade as previously suggested by. The low degree of sequence difference means that, based on 16S rRNA gene sequences, these Methanobrevibacter spp. (five named species plus further potential unrecognised species) cannot be further broken down below the level of these two clades that are typified by the species Mbb. ruminantium and Mbb. gottschalkii. Outside these two clades, sequences from other named Methanobrevibacter spp. and good-quality sequences can be defined to the species level, in some cases forming coherent groups that probably represent new species of Methanobrevibacter (Fig. S1). Among these are the already recognised Mbb. smithii, Mbb. woesii, and Mbb. wolinii, and potentially new species from rabbits, rats, and termites.

13The genus Methanosphaera comprises methylotrophic methanogens within the order Methanobacteriales, and some of the characterised members are known to be restricted to methanol and hydrogen as substrates for methane formation. Msp. stadtmanae and Msp. cuniculi are currently the only formally-described species of this genus. The previously unpublished isolate sequences of Methanosphaera sp. ISO3-F5 (J Jeyamalar, GE Naylor, G Henderson, RS Ronimus, PH Janssen, 2010, unpublished data) and Methanosphaera sp. A4 (CC Kim, GE Naylor, R Ronimus, G Henderson, PH Janssen, 2012, unpublished data) indicate that at least two additional species may exist. These two isolates have less than 98% sequence identity to other described Methanosphaera species, but await formal description to confirm their species status. A fifth species level group was defined based on sequence divergence to other isolates and strong bootstrap support for the group (≥70%; Methanosphaera sp. Group5) (Fig. S1). No isolate sequences are available for this group to date. Based on the 16S rRNA gene, this group is most similar to Methanosphaera sp. ISO3-F5 (97.1% sequence similarity).

14Using RIM-DB, 17 of 195 Methanomassiliicoccales sequences were assigned to the genus Methanomassiliicoccus, while the remaining 178 sequences were distributed into 12 newly-defined groups (Fig. S2). Most of these groups are strongly supported by high bootstrap values. Group 7 is the only group that contains only a single sequence in RIM-DB, which represents the enrichment culture Methanogranum caenicola Kjm51a, and no other (>1,200 bp) sequences appear to be currently available. Minimum sequence identities between some groups are higher than 95%, indicating that not all groups may represent potential new genera. In addition, sequence identities of less than 88% between the described Methanomassiliicoccus species and the previously unpublished cultures ISO4-G1, ISO4-G11, ISO4-H5 indicate a wide sequence divergence of species/genera within this order. It could also indicate that the recently proposed family Methanomassiliicoccaceae is not the only family in this order. The confirmation and proposal of a second family awaits formal description of some of its representative members. In the meantime, all Methanomassiliicoccales sequences are placed in the family Methanomassiliicoccaceae in the current version of this database. It has to be emphasised that these groupings (and group names) do not represent official nomenclature and that they await confirmation by successful isolation and description of representative isolates. They can, however, be used to provide a better definition of subgroups within the order.

15Where applicable, sequences have also been distributed into groups of gastrointestinal methanogens from the genera Methanobacterium, Methanimicrococcus and Methanosarcina. The low abundance of these groups in our test dataset, and the lack of publicly-available datasets from intestinal samples that include significant numbers of sequences from these organism groups, currently prevent sufficient testing. However, their inclusion in the database will allow future identification of samples where these genera might be abundant, which could then provide more sequence data and justify further effort in delineating sub-groups.

16Currently available next generation sequencing methods target relatively short segments of the 16S rRNA gene (of up to approximately 500 bp in length). RIM-DB contains primarily sequences longer than 1,200 bp that cover large parts of the ∼1,540 bp long gene. Shorter sequences were only retained in the database if they originated from isolates or enrichment cultures. An analysis of the nucleotide position coverage by the database revealed that the coverage is even throughout the majority of the 16S rRNA gene (Fig. 2). This indicates that RIM-DB could also be suitable for the analysis of amplicon data that have been generated for variable regions other than the V6–V8 region, which was the target for amplicon sequencing of the samples included in the test dataset.

17To compare the detail and accuracy of taxonomic assignment of sequences using RIM-DB in comparison to using Greengenes (release gg_13_5) or SILVA (release 111) databases, we analysed the taxonomic assignment of 24 different isolates and of a test dataset of amplicons.

18The 24 selected isolate sequences of rumen and intestinal methanogens spanned the orders Methanobacteriales, Methanomicrobiales, Methanosarcinales and Methanomassiliicoccales. Two versions of this sequence set were tested: the complete sequences and an extracted set of their V6–V8 regions. A comparison at the genus level revealed a high accuracy of taxonomic assignment, and little difference between assignments with the three different databases (Figs. 3A and 3B). At the species level, however, taxonomic assignments differed considerably. RIM-DB and SILVA performed well at species level assignment of these sequences, and assigned most long length and V6–V8 region sequences correctly (Figs. 3A and 3B). Resolution of taxonomic assignment was limited for highly similar sequences, such as Methanobrevibacter boviskoreani and Methanobrevibacter sp. AbM4, which share more than 99% sequence identity. Some sequences, e.g., of Methanobrevibacter olleyae, were assigned correctly by SILVA but not by RIM-DB, because Mbb. olleyae sequences and closely related Mbb. ruminantium sequences have been combined to a species level “Mbb. ruminantium clade” in RIM-DB. The “i_isolate” field in RIM-DB (File S2) did allow identification of all 24 isolates correctly when analysing long length sequences, and 19 isolates when analysing V6–V8 region sequences. It needs to be emphasized that sequencing errors and limited sequence divergence may make it difficult to classify sequences from environmental sequence data beyond the species level. Because of this, we developed a more conservative taxonomy in which some difficult-to-delineate sequences are grouped into clades. We therefore recommend that the taxonomic assignment of large datasets of short archaeal 16S rRNA gene sequences originating from the rumen or other gastrointestinal systems be performed using the more conservative taxonomy (File S3) in conjunction with the sequence database (File S1). Taxonomic assignments at the isolate level and at high similarity thresholds (>99%) are more useful for the analyses of samples of known composition, such as mock-communities, and could use the more detailed taxonomy file (File S4). Taxonomic assignment using the Greengenes database resulted in one of 24 sequences being correctly assigned at the species level (both for long length 16S rRNA sequences and for sequences spanning the V6–V8 region). Twenty-two sequences were not assigned at the species level, and one sequence was assigned erroneously (Figs. 3A and 3B, for full taxonomic assignments see Table S2).

19This approach of testing and comparing taxonomic assignment with different databases is biased towards isolates present in the databases. For example, only few sequences of described Methanomassiliicoccales species are available. We therefore also analysed 520,563 sequencing reads of a test dataset of amplicons of V6–V8 regions of archaeal 16S rRNA gene sequences (Fig. 3C). All datasets were generated from sheep and cattle on different feeds, from different geographic locations within New Zealand, and of varying age (see Table S4 for details). The sequences in the datasets were combined, denoised, chimera-checked, clustered into 322 OTUs based on a clustering cut-off of 99%, and taxonomically assigned using the three different databases. The total relative abundances for each of the three most abundant organism groups (Methanobrevibacter, Methanosphaera, Methanomassiliicoccales) were similar, and, in general, differed by less than five percent between assignments with different taxonomies (see Table S3 for details). However, the detail of taxonomic assignment varied, depending upon which database was used. RIM-DB was able to distinguish between distinct Methanobrevibacter clades and assigned them as such (i.e., Mbb. gottschalkii clade, Mbb. ruminantium clade), whereas SILVA and Greengenes did not provide specific taxonomic assignments beyond the genus Methanobrevibacter genus level (Fig. 3C). SILVA did assign the Methanobrevibacter-associated reads to two groups, but there were not given any clear taxonomic designations to indicate what they might be. Greengenes assigned all Methanobrevibacter-associated reads to one undifferentiated Methanobrevibacter group (Fig. 3). Obtaining more detailed taxonomic assignments is important for characterising methanogen communities, because Methanobrevibacter spp. are among the most dominant methanogens in intestinal tracts of many mammals and termites, and types of species may vary depending on host or host-tissue/organ. In addition, the abundance of different clades of Methanobrevibacter spp. may be correlated with the abundance of certain bacterial groups. Strains of the Mbb. gottschalkii and Mbb. ruminantium clades have been described as dominant species in the rumen, while Mbb. curvatus, Mbb. filiformis, and Mbb. cuticularis were isolated from termite guts and Mbb. smithii is from human faeces. Other Methanobrevibacter strains, such as the less dominant rumen methanogen Methanobrevibacter sp. AbM4, a member of Mbb. wolinii, have been found to be involved in differences in feed efficiencies in ruminants. These examples underline the importance of detecting intra-genus differences for the analysis of amplicon data.

20Methanosphaera sequences in the test sequence reads with RIM-DB were primarily assigned to Methanosphaera sp. ISO3-F5 (13% ± 7.7% [mean ± standard deviation]), a potential species-level group containing an isolate from a sheep rumen. Using RIM-DB, a small percentage (1.2% ± 1.7%) of reads of the test data set were assigned to the Methanosphaera sp. Group5 within the genus that does not contain any isolates, indicating that some species of Methanosphaera in the rumen are as yet uncultured. Taxonomic assignment with SILVA and Greengenes were restricted to the genus-level assignment and did not distinguish between different species of Methanosphaera in the test dataset (see Table S3 for full assignments).

21Accurate taxonomic assignments of Methanomassiliicoccales sequences are difficult, because this order has just recently been proposed and Methanomassiliicoccus is currently the only validly described genus of this diverse order. While assignment with SILVA and Greengenes suggested only two and one taxonomic group(s) respectively, taxonomic assignment with RIM-DB revealed the presence of five different potential species-level groups within Methanomassiliicoccales in the test dataset (Fig. 3). Not unexpectedly, most of the assigned OTUs could be assigned to groups 10 and 12, which tend to contain sequences from the rumen. already showed that Methanomassiliicoccales 16S rRNA and mcrA gene sequences could potentially be distributed into environment specific clades. We have extended these groupings by distributing additional reference sequences into clades, some of which appear to be dominated by sequences associated with a specific habitat. For example, group 12 of Methanomassiliicoccales contains primarily sequences from rumen (Fig. S2). Future efforts in community profiling of various environments using RIM-DB will verify whether these groupings are indeed habitat-specific. Both phylogeny and taxonomy of the Methanomassiliicoccales are at an early stage, and further efforts of isolating and characterising new species are required to fully understand the diversity of this interesting order of methanogens.

22Taxonomic frameworks require constant updating to accommodate changes in nomenclature and to include new sequence information. RIM-DB incorporates recent advances, such as the newly proposed order Methanomassiliicoccales, for which little taxonomic information is available. RIM-DB can provide new and detailed insights into the composition of the rumen archaeal microbiota, and our testing shows that this specialised taxonomy could be useful to complement analyses made using larger and more general databases such as Greengenes and SILVA. Some specialised taxonomies already exist (for examples see; ), but developing and refining additional specialised taxonomies should be considered, as these could also contribute to the refinement of general and more comprehensive databases.

23RIM-DB is intended for taxonomic assignment of next generation amplicon sequence data of rumen methanogens, but it also includes reference 16S rRNA gene sequences from other intestinal habitats, such as human faeces and the termite gut. There is currently a lack of suitable publicly-available amplicon datasets, which prevents sufficient testing of RIM-DB on sequence data from habitats other than the rumen, but RIM-DB may also be a valuable resource for the analysis of methanogenic community structures from other intestinal environments, including the human gut. Like other taxonomic frameworks, RIM-DB will require updating to include novel sequences of rumen and intestinal methanogens and to address changes in nomenclature. The most recent release of RIM-DB (including fasta-, taxonomy- and ARB-file) is available for download from www.globalrumencensus.org.nz.