Development of a novel metric for evaluating diatom assemblages in rivers using DNA metabarcoding

Abstract

Fundamental differences in the nature of diatom assemblage composition data generated using light microscopy and molecular barcoding create problems when applying current paradigms and metrics developed for ecological assessment. We therefore describe the development of a new metric designed specifically for diatom rbcL barcode data gathered using high throughput sequencing (HTS). Although the structure of datasets collected using HTS is similar to that collected using light microscopy (LM), differences in the proportions of key species between the two methods mean that the use of metrics designed for LM on HTS data gives biased results. We therefore recalibrated the Trophic Diatom Index in order to produce a version that is sensitive to nutrient pressures in rivers but that can be used with HTS data. Correlation between the LM and HTS metrics is good (r = 0.86 on a cross-validated model); however, 30% of sites will change class when the current Water Framework Directive classification approach is applied. Although less than 15% of diatom taxa recorded from UK and Ireland are included in the rbcL barcode reference database, gaps in this database are not a major source of variation between the HTS and LM models. We argue that use of metrics calibrated using HTS data is a more realistic option than applying correction factors to enable HTS data to be used with existing indices. We also stress the importance of starting the process of integrating HTS into ecological assessments with relatively conservative approaches. This enables the data collected by HTS to be related to those generated by established approaches, both now and during long-term monitoring, making it possible for scientists, regulators and stakeholders to have an informed conversation about the benefits and challenges of HTS. Overall, the study demonstrates that it is possible to translate the legal requirements of an ecological assessment framework from LM to HTS, though differences in these two approaches mean that there is unlikely to be perfect agreement between their outputs.