A general approach to identify low-frequency variants within influenza samples collected during routine surveillance

Abstract

Influenza viruses exhibit considerable diversity between hosts. Additionally, different quasispecies can be found within the same host. High-throughput sequencing technologies can be used to sequence a patient-derived virus population at sufficient depths to identify low-frequency variants (LFV) present in a quasispecies, but many challenges remain for reliable LFV detection because of experimental errors introduced during sample preparation and sequencing. High genomic copy numbers and extensive sequencing depths are required to differentiate false positive from real LFV, especially at low allelic frequencies (AFs). This study proposes a general approach for identifying LFV in patient-derived samples obtained during routine surveillance. Firstly, validated thresholds were determined for LFV detection, whilst balancing both the cost and feasibility of reliable LFV detection in clinical samples. Using a genetically well-defined population of influenza A viruses, thresholds of at least 104 genomes per microlitre and AF of ≥5 % were established as detection limits. Secondly, a subset of 59 retained influenza A (H3N2) samples from the 2016–2017 Belgian influenza season was composed. Thirdly, as a proof of concept for the added value of LFV for routine influenza monitoring, potential associations between patient data and whole genome sequencing data were investigated. A significant association was found between a high prevalence of LFV and disease severity. This study provides a general methodology for influenza LFV detection, which can also be adopted by other national influenza reference centres and for other viruses such as SARS-CoV-2. Additionally, this study suggests that the current relevance of LFV for routine influenza surveillance programmes might be undervalued.

Recommendations for Uniform Variant Calling of SARS-CoV-2 Genome Sequence across Bioinformatic Workflows

“…In support of this suggestion, prior work on respiratory syncytial virus sequencing found that, below allele frequencies of 25%, variant calling across a range of coverages, variant callers, and a range of simulated error profiles resulted in a precision of less than 50%, approaching 0% at allele frequencies below 5% [ 30 ]. Additionally, sequencing analysis on influenza samples has found that minor alleles cannot be confidently called at frequencies below 5% [ 31 ], and that without correction to the methods, minor variants were overestimated by as much as 10-fold [ 32 ]. These studies also identified several methods to improve confidence in variant calls, including generating technical replicates for each sample and adjusting allele-frequency thresholds based on Polymerase Chain Reaction (PCR) cycle threshold values, both of which underscore the need for rich metadata to support thorough and accurate use of public sequence data and the need to consider consensus sequences carefully in the absence of such metadata.…”