High-throughput sequencing of 16S rRNA gene amplicons is a valuable tool for comparing microbial community structure among hundreds of samples for which researchers have primarily used the 454 Pyrosequencing platform. The Illumina platform produces far more reads than 454 (up to 1.5 billion reads per run, compared with 1 million reads per run on a 454 plate of comparable cost), but produces fewer base pairs (bp) per read (75–150 bp per read compared with 250–400 bp per read on 454). The shorter Illumina reads may reduce phylogenetic resolution, both in terms of picking operational taxonomic units (OTUs) and determining evolutionary distances between OTUs. The paired-end (PE) approach where each molecule is sequenced from both the 5′ and 3′ ends can double the number of bp per read for the Illumina platform. Some researchers have obtained overlapping PE Illumina reads covering the V3 (Bartram et al., 2011) or V6 (Zhou et al., 2011) regions of 16S rRNA genes, but many other useful primer regions (for example, V1–V2 or V4) are >200 bp in length in which case PE reads do not overlap. However, it is possible, as we present here, to use non-overlapping PE reads to pick OTUs and build a phylogenetic tree. We assessed the utility of using PE Illumina sequencing, compared with results from single-direction (SD) sequencing from the 5′ position of the V4 region of 16S rRNA genes. We compared alpha- and beta-diversity analyses (species richness and between-sample comparisons, respectively), using previously published non-overlapping PE Illumina sequence data from 16S rRNA gene surveys of 28 human microbiome and environmental samples (Caporaso et al., 2011).

Illumina sequences were quality filtered using the default pipeline in QIIME 1.2.1 (Caporaso et al., 2010b), including the default quality thresholds and a minimum read length of 75 bp. To avoid potential biases from quality filtering, only sequences passing the quality threshold in both 5′ and 3′ reads were kept for downstream processing (43% of the total reads were removed because one of the two directions did not pass the quality threshold). All reads were trimmed to 75 bp length (total of 150 bp per seq for PE). The PE and 5′ SD data were analyzed separately, using the default settings in QIIME, with the following additional steps for PE data: before OTU-picking with uclust (97% ID) (Edgar, 2010), we joined PE reads ‘inside-out’ such that the 3′ end of the 3′ read was to the left of the 5′ end of the 5′ read, required for uclust to perform accurate pairwise alignments. Based on a simulation using Greengenes sequences trimmed to the V4 region, we found that uclust assigned similar pairwise alignment distances to inside-out reads compared with the normal configuration (Supplementary Figure S1). OTU representative sequences (separate 3′ and 5′ reads for PE) were aligned with PyNAST (Caporaso et al., 2010a) against separate regions of the August 2007 Greengenes core (DeSantis et al., 2006), trimmed to positions 2250–2423 for 5′ reads and 3805–4069 for 3′ reads. Trimming reference sequences had advantages of rapid computation, more stringent discarding of non-16S reads, and has previously been shown to improve taxonomic classification (Werner et al., 2011b). We built a phylogenetic tree (FastTree) (Price et al., 2010) from aligned, filtered sequences (for PE, the separate 5′ and 3′ alignments were joined end-to-end to use all 150 bp for phylogeny). Separately for each method, we discarded OTUs with fewer than 10 total reads (2.6% of the 16.6 million high-quality reads). This was done to discard false diversity because of sequencing errors, by setting a 10 × threshold of evidence needed to support a true sequence. OTUs that failed to align >72 bp (0.007% of remaining reads) were also removed.

Log in or create a free account to read this content

Gain free access to this article, as well as selected content from this journal and more on nature.com

Access through your institution

or

Sign in or create an account
Continue with Google
Continue with ORCiD