Taxonomic annotation of 16S rRNA sequences of pig intestinal samples using MG-RAST and QIIME2 generated different microbiota compositions

Environmental microbiome studies rely on fast and accurate bioinformatics tools to characterize the taxonomic composition of samples based on the 16S rRNA gene. MetaGenome Rapid Annotation using Subsystem Technology (MG-RAST) and Quantitative Insights Into Microbial Ecology 2 (QIIME2) are two of the most popular tools available to perform this task. Their underlying algorithms differ in many aspects, and therefore the comparison of the pipelines provides insights into their best use and interpretation of the outcomes. Both of these bioinformatics tools are based on several specialized algorithms pipelined together, but whereas MG-RAST is a user-friendly webserver that clusters rRNA sequences based on their similarity to create Operational Taxonomic Units (OTU), QIIME2 employs DADA2 in the construction of Amplicon Sequence Variants (ASV) by applying an error model that considers the abundance of each sequence and its similarity to other sequences. Taxonomic compositions obtained from the analyses of amplicon sequences of DNA from swine intestinal gut and faecal microbiota samples using MG-RAST and QIIME2 were compared at domain-, phylum-, family- and genus-levels in terms of richness, relative abundance and diversity. We found significant differences between the microbiota profiles obtained from each pipeline. At domain level, bacteria were relatively more abundant using QIIME2 than MG-RAST; at phylum level, seven taxa were identified exclusively by QIIME2; at family level, samples processed in QIIME2 showed higher evenness and richness (assessed by Shannon and Simpson indices). The genus-level compositions obtained from each pipeline were used in partial least squares-discriminant analyses (PLS-DA) to discriminate between sample collection sites (caecum, colon and faeces). The results showed that different genera were found to be significant for the models, based on the Variable Importance in Projection, e.g. when using sequencing data processed by MG-RAST, the three most important genera were Acetitomaculum, Ruminococcus and Methanosphaera, whereas when data was processed using QIIME2, these were Candidatus Methanomethylophilus, Sphaerochaeta and Anaerorhabdus. Furthermore, the application of differential filtering procedures before the PLS-DA revealed higher accuracy when using non-restricted datasets obtained from MG-RAST, whereas datasets obtained from QIIME2 resulted in more accurate discrimination of sample collection sites after removing genera with low relative abundances (<1%) from the datasets. Our results highlight the differences in taxonomic compositions of samples obtained from the two separate pipelines, while underlining the impact on downstream analyses, such as biomarkers identification.