Abstract
Like most RNA viruses, influenza viruses generate defective viral genomes (DVGs) with large internal deletions during replication. There is accumulating evidence supporting a biological relevance of such DVGs. However, further understanding of the molecular mechanisms that underlie the production and biological activity of DVGs is conditioned upon the sensitivity and accuracy of detection methods, that is, next-generation sequencing (NGS) technologies and related bioinformatics algorithms. Although many algorithms were developed, their sensitivity and reproducibility were mostly assessed on simulated data. Here, we introduce DG-seq, a time-efficient pipeline for DVG detection and quantification, and a set of biological controls to assess the performance of not only our bioinformatics algorithm but also the upstream NGS steps. Using these tools, we provide the first rigorous comparison of the two commonly used sample processing methods for RNA-seq, with or without a PCR preamplification step. Our data show that preamplification confers a limited advantage in terms of sensitivity and introduces size- but also sequence-dependent biases in DVG quantification, thereby providing a strong rationale to favor preamplification-free methods. We further examine the features of DVGs produced by wild-type and transcription-defective (PA-K635A or PA-R638A) influenza viruses, and show an increased diversity and frequency of DVGs produced by the PA mutants compared to the wild-type virus. Finally, we demonstrate a significant enrichment in DVGs showing direct, A/T-rich sequence repeats at the deletion breakpoint sites. Our findings provide novel insights into the mechanisms of influenza virus DVG production.