Distinguishing highly similar gene isoforms with a clustering-based bioinformatics analysis of PacBio single-molecule long reads

Background: Gene isoforms are commonly found in both prokaryotes and eukaryotes. Since each isoform may perform a specific function in response to changing environmental conditions, studying the dynamics of gene isoforms is important in understanding biological processes and disease conditions. However, genome-wide identification of gene isoforms is technically challenging due to the high degree of sequence identity among isoforms. Traditional targeted sequencing approach, involving Sanger sequencing of plasmid-cloned PCR products, has low throughput and is very tedious and time-consuming. Next-generation sequencing technologies such as Illumina and 454 achieve high throughput but their short read lengths are a critical barrier to accurate assembly of highly similar gene isoforms, and may result in ambiguities and false joining during sequence assembly. More recently, the third generation sequencer represented by the PacBio platform offers sufficient throughput and long reads covering the full length of typical genes, thus providing a potential to reliably profile gene isoforms. However, the PacBio long reads are error-prone and cannot be effectively analyzed by traditional assembly programs.

Results: We present a clustering-based analysis pipeline integrated with PacBio sequencing data for profiling highly similar gene isoforms. This approach was first evaluated in comparison to de novo assembly of 454 reads using a benchmark admixture containing 10 known, cloned msg genes encoding the major surface glycoprotein of Pneumocystis jirovecii. All 10 msg isoforms were successfully reconstructed with the expected length (~1.5 kb) and correct sequence by the new approach, while 454 reads could not be correctly assembled using various assembly programs. When using an additional benchmark admixture containing 22 known P. jirovecii msg isoforms, this approach accurately reconstructed all but 4 these isoforms in their full-length (~3 kb); these 4 isoforms were present in low concentrations in the admixture. Finally, when applied to the original clinical sample from which the 22 known msg isoforms were cloned, this approach successfully identified not only all known isoforms accurately (~3 kb each) but also 48 novel isoforms.

Conclusions: PacBio sequencing integrated with the clustering-based analysis pipeline achieves high-throughput and high-resolution discrimination of highly similar sequences, and can serve as a new approach for genome-wide characterization of gene isoforms and other highly repetitive sequences.