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

Abstract

18 Background: The low cost of 16S rRNA gene sequencing facilitates population-scale 19 molecular epidemiological studies. Existing computational algorithms can parse 16S 20rRNA gene sequences to high-resolution Amplicon Sequence Variants (ASVs), which 21 represent ecologically coherent entities. Assigning species-level taxonomy to these ASVs 22is the critical remaining barrier to drawing ecologically/clinically relevant inferences from 23 and comparing data across 16S rRNA gene-based microbiota studies. 24Results: To overcome this barrier, we developed a broadly applicable method for 25 constructing a phylogeny-based, high-resolution, habitat-specific training set. When used 26 with the naïve Bayesian Ribosomal Database Project (RDP) Classifier, this training set 27 achieved species/supraspecies-level taxonomic assignment to 16S rRNA gene-derived 28ASVs. The key steps for generating such a training set are 1) constructing an accurate 29 and comprehensive phylogenetic-based, habitat-specific database; 2) compiling multiple 30 16S rRNA gene sequences to represent the natural sequence variability of each taxon in 31 the database; 3) trimming the training set to match the sequenced regions if necessary; 32 and 4) placing species sharing closely related sequences into a supraspecies taxonomic 33 level to maintain subgenus resolution. As proof of principle, we developed a V1-V3 region 34 training set for the bacterial microbiota of the human aerodigestive tract using our 35 expanded Human Oral Microbiome Database (eHOMD). In addition, we overcame 36 technical limitations to successfully use Illumina sequences for the 16S rRNA gene V1-37 V3 region, the most informative segment for classifying bacteria native to the human 38 aerodigestive tract. We also generated a full-length eHOMD 16S rRNA gene training set, 39 which we used in conjunction with an independent PacBio Single Molecule, Real-Time 40 (SMRT)-sequenced sinonasal dataset to validate the representation of species in our 41 training set. The latter also established the effectiveness of a full-length training set for 42 assigning taxonomy of long-read 16S rRNA gene datasets. 43Conclusion: Here, we present a systematic approach for constructing a phylogeny-44 based, high-resolution, habitat-specific training set that permits species/supraspecies-45 level taxonomic assignment to short-and long-read 16S rRNA gene-derived ASVs. This 46 advancement enhances the ecological and/or clinical relevance of 16S rRNA gene-based 47 microbiota studies. 48In microbiota studies of most ecosystems and/or habitats, achieving ecologically and/or 53 clinically relevant results requires species-level identification of constituents. For 54 example, species-level identification is often critically important for host-associated 55 microbial communities because these often include commensal and pathogenic species 56 of the same genus, e.g., [1, 2]. Also, some microbial genera include species that are site 57 specialists and inhabit distinct niches of a given environment [3]. Although hig...

Impact of 16S rRNA gene redundancy and primer pair selection on the quantification and classification of oral microbiota in next-generation sequencing

“…However, this limited approach does not reflect the complexity of the oral microbiota where around 700 species have been identified (12,13). On the other hand, the identification, at least at the species level, is highly desirable in 16S rRNA sequencingbased studies of the oral microbiota (14). This is because it has been demonstrated how different species from the same genus are associated with different oral conditions (15)(16)(17).…”

OTUs clustering should be avoided for defining oral microbiome

“…Schloss et al (Schloss 2021) found that, with a 97% similarity threshold and applying the OptiClust algorithm, 31.7%, 34.3% and 34.8% of the OTUs assessed had 16S rRNA amplicons from distinct species in the V3-V4, V4, and V4-V5 regions, respectively (Schloss 2021). However, these investigations did not focus on taxa inhabiting a specific environment, despite the importance of conducting 16S rRNA genebased research using habitat-specific databases (Escapa et al 2020). Consequently, we used primer pairs targeting several gene regions (Regueira-Iglesias et al 2021.a) to determine the number of different oral-bacterial and oral-archaeal species with an ASI≥97%, as well as the potential clusters that might contain distinct species.…”

“…This would, however, result in both an overabundance of the single species representing the cluster and an underestimation of the diversity of the community, with other species within the OTU overlooked. Consequently, despite possible difficulties, it would be better to use the lowest possible level of resolution, i.e., the variant level (Callahan et al 2017), and databases specifically designed for taxonomic identifications of taxa at this level (Escapa et al 2020).…”

RESCRIPt: Reproducible sequence taxonomy reference database management for the masses

“…Sequence reference databases are often subsetted by investigators to focus on particular clades of interest or to perform additional curation of public datasets. Some researchers have generated environment-specific databases, founded in the belief that such databases increase taxonomic classification accuracy by removing sequences that are genetically related but ecologically distinct from species found in a specific environment [52,[55][56][57][58][59], although this can elevate the risk of false-positive errors [76]. RESCRIPt contains several methods to support and evaluate such filtering decisions, which then become embedded in provenance to facilitate transparent and reproducible use of these databases.…”

“…Additionally, issues with amplicon length and sequence heterogeneity can limit the ability to identify species, especially from short marker-gene sequences or metagenome fragments [51]. Hence, many researchers choose to perform additional curation to focus on type strains [52], quality filtering [14,53], or construct environmentspecific databases that are constrained to contain species found within a given environment [52,[54][55][56][57][58][59]. Database customization is also often performed to add new accessions that are absent in some database releases to increase database coverage [50], or to incorporate outgroups [14].…”