Evaluating bias of illumina-based bacterial 16S rRNA gene profiles

Abstract

Massively parallel sequencing of 16S rRNA genes enables the comparison of terrestrial, aquatic, and host-associated microbial communities with sufficient sequencing depth for robust assessments of both alpha and beta diversity. Establishing standardized protocols for the analysis of microbial communities is dependent on increasing the reproducibility of PCR-based molecular surveys by minimizing sources of methodological bias. In this study, we tested the effects of template concentration, pooling of PCR amplicons, and sample preparation/interlane sequencing on the reproducibility associated with paired-end Illumina sequencing of bacterial 16S rRNA genes. Using DNA extracts from soil and fecal samples as templates, we sequenced pooled amplicons and individual reactions for both high (5- to 10-ng) and low (0.1-ng) template concentrations. In addition, all experimental manipulations were repeated on two separate days and sequenced on two different Illumina MiSeq lanes. Although within-sample sequence profiles were highly consistent, template concentration had a significant impact on sample profile variability for most samples. Pooling of multiple PCR amplicons, sample preparation, and interlane variability did not influence sample sequence data significantly. This systematic analysis underlines the importance of optimizing template concentration in order to minimize variability in microbial-community surveys and indicates that the practice of pooling multiple PCR amplicons prior to sequencing contributes proportionally less to reducing bias in 16S rRNA gene surveys with next-generation sequencing.

FIG 1

Graphical representation of the experimental design, including soil samples (6TD and 10AS), stool samples (S1, S3, and SX), and treatments. The dark squares represent amplicons prepared on the first day of experimental manipulation (i.e., intended for the first MiSeq lane, except for sample SX), whereas the light squares represent amplicons prepared on a second experimental day (i.e., intended for the second MiSeq lane).

FIG 2

NMDS based on a Bray-Curtis dissimilarity matrix of all sample replicates from Illumina data for S1, S3, 6TD, and 10AS.

FIG 3

NMDS ordination of replicates for each sample using Bray-Curtis dissimilarity matrices. Open circles represent high-template-concentration, not-pooled, and lane 1 replicates; shaded circles represent low-template-concentration, pooled, and lane 2 replicates.

FIG 4

Distance to the centroid of a given sample group when using ordination distances for replicates of each sample. Within PERMDISP, ANOVA was implemented to test the statistical significance of sample distances to the corresponding centroids. The significance of differences is denoted by asterisks (*, P < 0.05; **, P < 0.01; ***, P < 0.001).