Improved taxonomic assignment of rumen bacterial 16S rRNA sequences using a revised SILVA taxonomic framework

Abstract

The taxonomy and associated nomenclature of many taxa of rumen bacteria are poorly defined within databases of 16S rRNA genes. This lack of resolution results in inadequate definition of microbial community structures, with large parts of the community designated as incertae sedis, unclassified, or uncultured within families, orders, or even classes. We have begun resolving these poorly-defined groups of rumen bacteria, based on our desire to name these for use in microbial community profiling. We used the previously-reported global rumen census (GRC) dataset consisting of >4.5 million partial bacterial 16S rRNA gene sequences amplified from 684 rumen samples and representing a wide range of animal hosts and diets. Representative sequences from the 8,985 largest operational units (groups of sequence sharing >97% sequence similarity, and covering 97.8% of all sequences in the GRC dataset) were used to identify 241 pre-defined clusters (mainly at genus or family level) of abundant rumen bacteria in the ARB SILVA 119 framework. A total of 99 of these clusters (containing 63.8% of all GRC sequences) had no unique or had inadequate taxonomic identifiers, and each was given a unique nomenclature. We assessed this improved framework by comparing taxonomic assignments of bacterial 16S rRNA gene sequence data in the GRC dataset with those made using the original SILVA 119 framework, and three other frameworks. The two SILVA frameworks performed best at assigning sequences to genus-level taxa. The SILVA 119 framework allowed 55.4% of the sequence data to be assigned to 751 uniquely identifiable genus-level groups. The improved framework increased this to 87.1% of all sequences being assigned to one of 871 uniquely identifiable genus-level groups. The new designations were included in the SILVA 123 release (https://www.arb-silva.de/documentation/release-123/) and will be perpetuated in future releases.

A restriction enzyme reduced representation sequencing approach for low-cost, high-throughput metagenome profiling

“…16S rRNA gene sequencing, WGS and the reference-based RE-RRS approach all require a reference database to assign taxonomic information to the sequences. 16S rRNA gene reference databases tend to be more comprehensive because only a single gene needs to be sequenced, enabling both culturable and unculturable microbes to be captured (10). WGS and the reference-based RE-RRS approach both need reference databases containing genome assemblies.…”

“…Historically, there have been two approaches used for sequencing metagenome samples: targeted sequencing and whole genome shotgun (WGS) sequencing. Targeted sequencing amplifies specified phylogenetically informative genes from a sample, such as the 16S rRNA gene (16S) of microbes, which typically distinguishes taxonomic groups well due to large, comprehensive databases of 16S rRNA sequences that include both culturable and unculturable organisms (9, 10). This approach usually relies on having long sequence reads (11), only captures phylogenetic variation at one gene, and is subject to PCR primer bias due to mismatches in the flanking regions where the primers bind (12).…”

A restriction enzyme reduced representation sequencing approach for low-cost, high-throughput metagenome profiling

“…Historically, there have been two approaches used for sequencing metagenome samples: targeted sequencing and metagenome shotgun sequencing. Targeted sequencing amplifies specified phylogenetically informative genes from a sample, such as the 16S rRNA gene (16S) of microbes, which typically distinguishes taxonomic groups well due to large, comprehensive databases of 16S rRNA sequences that include both culturable and uncultured organisms [10,11]. This approach usually relies on having long sequence reads [12], only captures phylogenetic variation at one gene, and is subject to PCR primer bias due to mismatches in the flanking regions where the primers bind [13].…”

Construction of habitat-specific training sets to achieve species-level assignment in 16S rRNA gene datasets

“…For habitats that have yet to be deeply interrogated, the access to this breadth outweighs the risk of misclassification due to annotation error. However, once a habitat is sufficiently explored, a habitat-specific database enables accurate fine-level phylogenetic resolution for taxonomic assignment to ASVs [20-30]. Existing habitat-specific databases are constructed with different methods and can be used to assign taxonomy via different approaches.…”

“…Existing habitat-specific databases are constructed with different methods and can be used to assign taxonomy via different approaches. Examples of this include the following: 1) stand-alone habitat-specific databases consisting of curated collections of close-to-full-length 16S rRNA gene sequences compiled both from other repositories and by generating new sequences from the habitat of interest, e.g., eHOMD for the human aerodigestive tract [20, 29], HITdb for the human gut [23] and RIM-DP for rumen [22]; 2) custom addition of compiled sequences from a specific habitat of interest to augment a broad general database, e.g., HBDB for honey bee [21], DictDB for termite and cockroach gut [27], SILVA19Rum for rumen [30] and MiDAS for activated sludge [24, 26]; 3) both a general and a habitat-specific database combined in the same pipeline, e.g., a general database followed by a most common ancestors approach with a custom species-level phylogeny of selected human-associated genera with pathogenic members [31] and FreshTrain with the TaxAss workflow for freshwater [28]. Many of these databases are used to train classifiers for taxonomy assignment.…”