Quality-score refinement of SSU rRNA gene pyrosequencing differs across gene region for environmental samples

Abstract

Due to potential sequencing errors in pyrosequencing data, species richness and diversity indices of microbial systems can be miscalculated. The "traditional" sequence refinement method is not sufficient to account for overestimations (e.g., length, primer errors, ambiguous nucleotides). Recent in silico and single-organism studies have revealed the importance of sequence quality scores in the estimation of ecological indices; however, this is the first study to compare quality-score stringencies across four regions of the SSU rRNA gene sequence (V1V2, V3, V4, and V6) with actual environmental samples compared directly to corresponding clone libraries produced from the same primer sets. The nucleic acid sequences determined via pyrosequencing were subjected to varying quality-score cutoffs that ranged from 25 to 32, and at each quality-score cutoff, either 10 or 15 % of the nucleotides were allowed to be below the cutoff. When species richness estimates were compared for the tested samples, the cutoff values of Q27(15%), Q30(10%), and Q32(15%) for V1V2, V4, and V6, respectively, estimated similar values as obtained with clone libraries and Sanger sequencing. The most stringent Q tested (Q32(10%)) was not enough to account for species richness inflation of the V3 region pyrosequence data. Results indicated that quality-score assessment greatly improved estimates of ecological indices for environmental samples (species richness and α-diversity) and that the effect of quality-score filtering was region-dependent.

Figure 1

Species richness estimates for the V1V2 (a), V3 (b), V4 (c), and V6 (d) SSU rRNA gene regions. Full and enlarged rarefaction curves are displayed for each region of the SSU rRNA gene. Operational taxonomic units (OTUs) are clustered at 97 % similarity. The wide gray line in the enlarged rarefaction curves represents 95 % confidence intervals for the clone library species richness predictions

Figure 2

Chao1 diversity estimates for the same samples at different Q filtering compared to the respective clonal library. Error bars denote 95 % confidence intervals

Figure 3

Phylogenetic comparison pre- and post-quality filtering. The phylum (class for Proteobacteria) distribution was compared for each region of the SSU rRNA gene at the Q suggested by the rarefaction curves in Fig. 1 (Q27_15% for V1V2 (a), Q32_10% for V3 (b), Q30_10% for V4 (c), and Q32_15% for V6 (d)). The coordinates for each taxon correspond to the abundance by fraction of unfiltered sequences (x-axis) and fraction of filtered high-quality sequences (y-axis). The scale differs across graphs to maximize point separation. Taxa along the line of y = x did not show a shift in percent abundance during filtering while those left and above the line represent phylogenetic groups that shifted to higher abundance post-filtering, and those right and below the line had a lower abundance post-filtering. Linear regression analysis to the line y = x yielded R ² values that indicate how well each region fits the assumption that the sequences removed were not phylogenetically biased

Figure 4

Shannon's evenness for each respective library with increasing stringency of Q filtering

Figure 5

Fraction of total sequences removed from clusters which have been trimmed but not quality filtered during Q analysis. Sequences were clustered pre-quality checking, and the cluster in which sequences were removed during quality checking was monitored for the V1V2 (a), V3 (b), V4 (c), and V6 (d) SSU rRNA gene regions. The Q parameter was Q27_15% for V1V2, Q32_10% for V3, Q30_10% for V4, and Q32_15% for V6. The majority of sequences were removed from the largest and smallest clusters