Error-correcting barcoded primers for pyrosequencing hundreds of samples in multiplex

Abstract

We constructed error-correcting DNA barcodes that allow one run of a massively parallel pyrosequencer to process up to 1,544 samples simultaneously. Using these barcodes we processed bacterial 16S rRNA gene sequences representing microbial communities in 286 environmental samples, corrected 92% of sample assignment errors, and thus characterized nearly as many 16S rRNA genes as have been sequenced to date by Sanger sequencing.

Figure 1

(a) A Hamming code to transmit one bit of information (k=1, n=3). Consider a hypersphere centered at 000 (blue): any single-bit error (010, 001, and 100) falls within a radius of 1 and thus can be corrected. Likewise with the hypersphere centered at 111 (red). (b) Regions of a codeword of length 16 (or longer) checked by parity bits at positions 0, 1, 2, and 4: bits that are checked by each position are marked with 1. (c) Example of decoding a “received” codeword containing the binary value of 3 (0011) (n=7, k=4): the first case contains no errors; the second contains a single-bit error at position 6 that is detected and corrected.

Figure 2

UniFrac clustering of samples from cystic fibrosis lung (red), Guerrero Negro microbial mat (green), air (gray), and North American rivers (blue) demonstrates the essentially perfect clustering by community when using UniFrac on samples obtained by pyrosequencing. Of 61 replicate samples, all but one pair clustered.