Accuracy of protist diversity assessments: morphology compared to cloning and direct pyrosequencing of 18S rRNA genes and ITS regions using the conspicuous tintinnid ciliates as a case study

Deep-sequencing technologies are becoming nearly routine to describe
microbial community composition in environmental samples. 18S rDNA pyrosequencing
has revealed a vast diversity of infrequent sequences, leading to the proposition of
the existence of an extremely diverse microbial \"rare biosphere\". While rare
microbes no doubt exist, critical views suggest that many rare sequences may
actually be artifacts. However, information about how diversity revealed by
molecular methods relates to that revealed by classical morphology approaches is
practically non-existent. To address this issue, we used different approaches to
assess the diversity of tintinnid ciliates, a species-rich group in which species
can be easily distinguished morphologically. We studied two Mediterranean marine
samples with different patterns of tintinnid diversity. We estimated tintinnid
diversity in these samples employing morphological observations as well as both
classical cloning and sequencing and pyrosequencing of two different markers, the
18S rDNA and the ITS regions, applying a variety of computational approaches
currently used to analyze pyrosequence reads. We found that both molecular
approaches were efficient in detecting the tintinnid species observed by microscopy
and revealed similar phylogenetic structures of the tintinnid community at the
species level. However, depending on the method used to analyze the pyrosequencing
results, we observed discrepancies with the morphology-based assessments up to
several orders of magnitude. In several cases, the inferred number of operational
taxonomic units (OTUs) largely exceeded the total number of tintinnid cells in the
samples. Such inflation of the OTU numbers corresponded to \"rare biosphere\" taxa,
composed largely of artefacts. Our results suggest that a careful and rigorous
analysis of pyrosequencing datasets, including data denoising and sequence
clustering with well-adjusted parameters, is necessary to accurately describe
microbial biodiversity using this molecular approach.