The effect of genetic markers and reference databases on analyses of fungal communities were estimated using fungal large subunit (LSU) and internal transcribed spacer (ITS) amplicon datasets in consecutive years of rhizosphere samples from three candidate biofuel crops, corn (Zea mays), switchgrass (Panicum virgatum), and miscanthus (Miscanthus × giganteus). These two marker genes were selected to contrast possible differences in biological conclusions. In addition, two ITS schemes based on two ITS reference databases were used to assess differences due to reference database composition. A taxonomy-supervised method was invoked using the Ribosomal Database Project naive Bayesian classifier that accesses all three databases. The UNITE classification scheme had the highest number of classified taxa in the raw classification result; however, it also had the highest proportion of unknown taxa (sequences that were classified to “unclassified,” “unidentified,” incertae sedis or, in the case of Warcup, to matches containing two unique names). After removal of these unknown taxa, LSU had the highest classification rate followed by Warcup and UNITE. As expected, the communities resolved using the two ITS databases, based on the same sequences, were relatively more similar than those from the lower-coverage LSU classification scheme. The choice of marker gene or even the same reads with different classification databases revealed different community patterns due to database coverage, e.g., the relative abundance of the most abundant groups changed or were only detected in one or two of the classification schemes, such as for Mortierella, Fusarium, and Phoma. No marked difference in fungal beta-diversity was identified among the three methods. Differentiation between the three biofuel crops and between the drought and normal rainfall years was apparent, regardless of method. Though classification rates, taxonomic conflicts, and coverage differences within the high-abundance fungal groups varied according to classification scheme, there was no overall impact on beta diversity among the three methods.
Read full abstract