The effects of parameter choice on defining molecular operational taxonomic units and resulting ecological analyses of metabarcoding data

Abstract

Abstract:The combination of DNA barcoding and high-throughput (next-generation) sequencing (metabarcoding) provides many promises but also serious challenges. Generating a reliable comparable estimate of biodiversity remains a central challenge to the application of the technology. Many approaches have been used to turn millions of sequences into distinct taxonomic units. However, the extent to which these methods impact the outcome of simple ecological analyses is not well understood. Here we performed a simple analysis of dietary overlap by skinks and shrews on Ile Aux Aigrettes, Mauritius. We used a combination of filtering thresholds and clustering algorithms on a COI metabarcoding dataset and demonstrate that all bioinformatics parameters will have interacting effects on molecular operational taxonomic unit (MOTU) recovery rates. These effects generated estimates covering two orders of magnitude. However, the effect on a simple ecological analysis was not large and, despite the wide variation in estimates of niche overlap, the same ecological conclusion was drawn in most cases. We advise that a conservative clustering programme coupled with larger sequence divergences to define a cluster, the removal of singletons, rigorous length filtering, and stringent match criteria for Molecular Identifier tags are preferable to avoid MOTU inflation and that the same parameters be used in all comparative analyses.

Spatiotemporal and demographic variation in the diet of New Zealand lesser short‐tailed bats (Mystacina tuberculata)

“…Second, we used MOTU and a clustering threshold of 94%. This is relatively low compared to the suspected reality of species identified (see Discussion in Clare et al., 2016), but is recommended to reduce MOTU inflation (Clare et al., 2016; Flynn, Brown, Chain, MacIsaac, & Cristescu, 2015). We have used the empirical recommendations of Clare et al.…”

“…We have used the empirical recommendations of Clare et al. (2016) to be conservative, and note that MOTU should not be equated to “species” (see Floyd et al., 2002) but as a comparable taxonomic entity for ecological and statistical interpretations. All things considered, our results likely give a reliable insight into the insectivorous habits of M. tuberculata , although we must note that bats were sampled during the summer and winter months and our results represent only a snapshot of the bats’ dietary habits.…”

“…Sequences were clustered into molecular operational taxonomic units (MOTU; Floyd, Abebe, Papert, & Blaxter, 2002), and a representative sequence of each MOTU was picked for analysis with the QIIME pick otu and uclust methods (http://qiime.sourceforge.net; Caporaso et al., 2010). MOTU were clustered using a similarity threshold of 94% to minimize spurious OTU generation (see Clare, Chain, Littlefair, Cristescu, & Deiner, 2016 for the appropriateness of MOTU cluster levels for diet analysis). We identified MOTU to order level using BLAST analyses and a reference database of >600,000 DNA barcodes extracted from GenBank with a wider taxonomic profile (including potential contaminants bacteria, fungi, mammals).…”

Barcoding the food chain: from Sanger to high-throughput sequencing

“…All studies must make extensive use of both field and laboratory controls to monitor contamination. The inclusion of mock communities alongside mixtures of unknown samples should be an important part of ensuring that laboratory and bioinformatics methods development are fit for purpose, for example, resulting in the construction of reliable estimates of Molecular Operational Taxonomic Unit (MOTU) diversity (Clare et al 2016). Finally, amplicon sequencing is problematic, due to taxonomic bias and the potential amplification of contaminants, but PCR-free sequencing is on the horizon (Liu et al 2016).…”

“…In addition to molecular laboratory techniques, metabarcoding datasets also require the development of bioinformatics pipelines. Programmes and data cleaning steps vary between research groups; the different clustering methods and data cleaning steps can considerably change the estimate of MOTU diversity (Brandon-Mong et al 2015;Clare et al 2016). Analysis of how bioinformatics pipelines affect ecological conclusions will be valuable, but as yet they are rare (although see Clare et al 2016;Roslin and Majaneva 2016).…”