AbstractMetabarcoding has the potential to become a rapid, sensitive, and effective approach for identifying species in complex environmental samples. Accurate molecular identification of species depends on the ability to generate operational taxonomic units (OTUs) that correspond to biological species. Due to the sometimes enormous estimates of biodiversity using this method, there is a great need to test the efficacy of data analysis methods used to derive OTUs. Here, we evaluate the performance of various methods for clustering length variable 18S amplicons from complex samples into OTUs using a mock community and a natural community of zooplankton species. We compare analytic procedures consisting of a combination of (1) stringent and relaxed data filtering, (2) singleton sequences included and removed, (3) three commonly used clustering algorithms (mothur, UCLUST, and UPARSE), and (4) three methods of treating alignment gaps when calculating sequence divergence. Depending on the combination of methods used, the number of OTUs varied by nearly two orders of magnitude for the mock community (60–5068 OTUs) and three orders of magnitude for the natural community (22–22191 OTUs). The use of relaxed filtering and the inclusion of singletons greatly inflated OTU numbers without increasing the ability to recover species. Our results also suggest that the method used to treat gaps when calculating sequence divergence can have a great impact on the number of OTUs. Our findings are particularly relevant to studies that cover taxonomically diverse species and employ markers such as rRNA genes in which length variation is extensive.
