Distinct approaches for the detection and removal of chimeric 16S rRNA sequences can significantly affect the outcome of between-site comparisons

Abstract

Comparative analyses of 16S rRNA clone libraries represent a standard tool in microbial ecology. Chimeric sequences are generally removed prior to such comparisons. A literature survey re- vealed a general pattern: (1) most commonly a single chimera identification approach (CIA) has been used; (2) putative chimeras have routinely been discarded without manual examination; (3) chimera fil- tered datasets have been submitted to repositories. To explore the effects of various CIAs on the study of microbial β-diversity relationships using complete primary data, 4 bacterial and 4 archaeal clone li- braries were generated from a submarine spring and analyzed together with 3 bacterial and 3 archaeal published primary datasets. The primary datasets were compared with their 8 different CIA filtered datasets using Chimera_check, CCODE, Pintail, Chimera Slayer and Bellerophon, the last with 4 dif- ferent settings. When CIA filtered datasets were pooled according to the CIA used, no significant dif- ferences between them could be observed, although there was not complete congruency between the different CIAs. When CIA filtered datasets of the same clone library were compared, generally no sig- nificant differences could be observed. In contrast, when CIA filtered datasets of different clone libraries were compared, the statistical significance of the relationships shifted from significant to in- significant or vice-versa in many cases depending on the CIA used. This precludes a correct identifica- tion of β-diversity. To solve this problem, we treated all CIA filtered datasets and primary data of a single clone library as CIA replicates in non-parametric MANOVA. This enabled unambiguous delin- eation of environmental samples by taking into account all CIA introduced data modifications.