Selection of appropriate metagenome taxonomic classifiers for ancient microbiome research

Abstract

AbstractMetagenomics enables the study of complex microbial communities from myriad sources, including the remains of oral and gut microbiota preserved in archaeological dental calculus and paleofeces, respectively. While accurate taxonomic assignment is essential to this process, DNA damage, characteristic to ancient samples (e.g. reduction in fragment size), may reduce the accuracy of read taxonomic assignment. Using a set of in silico-generated metagenomic datasets we investigated how the addition of ancient DNA (aDNA) damage patterns influences microbial taxonomic assignment by five widely-used profilers: QIIME/UCLUST, MetaPhlAn2, MIDAS, CLARK-S, and MALT (BLAST-X-mode). In silico-generated datasets were designed to mimic dental plaque, consisting of 40, 100, and 200 microbial species/strains, both with and without simulated aDNA damage patterns. Following taxonomic assignment, the profiles were evaluated for species presence/absence, relative abundance, alpha-diversity, beta-diversity, and specific taxonomic assignment biases. Unifrac metrics indicated that both MIDAS and MetaPhlAn2 provided the most accurate community structure reconstruction. QIIME/UCLUST, CLARK-S, and MALT had the highest number of inaccurate taxonomic assignments; however, filtering out species present at <0.1% abundance greatly increased the accuracy of CLARK-S and MALT. All programs except CLARK-S failed to detect some species from the input file that were in their databases. Ancient DNA damage resulted in minimal differences in species detection and relative abundance between simulated ancient and modern datasets for most programs. In conclusion, taxonomic profiling biases are program-specific rather than damage-dependent, and the choice of taxonomic classification program to use should be tailored to the research question.ImportanceAncient biomolecules from oral and gut microbiome samples have been shown to preserve in the archaeological record. Studying ancient microbiome communities using metagenomic techniques offer a unique opportunity to reconstruct the evolutionary trajectories of microbial communities through time. DNA accumulates specific damage over time, which could potentially affect taxonomic classification and our ability to reconstruct community assemblages accurately. It is therefore necessary to assess whether ancient DNA (aDNA) damage patterns affect metagenomic taxonomic profiling. Here, we assessed biases in community structure, diversity, species detection, and relative abundance estimates by five popular metagenomic taxonomic classification programs using in silico-generated datasets with aDNA damage. Age-related damage patterns had minimal impact on the taxonomic profiles produced by each program, and biases were intrinsic to each program. Therefore, an appropriate classification program should be chosen that minimizes the biases related to the questions being addressed.