Evaluation of different analysis pipelines for the detection of HIV-1 minority resistant variants.

Marine Perrier,Christophe Rodriguez,Vincent Calvez,Mélanie Bertine,Eve Todesco,Alexandre Storto,Benoit Visseaux,Nathalie Désiré,Anne-Geneviève Marcelin,Quentin Le Hingrat,Charlotte Charpentier,Diane Descamps

doi:10.1371/journal.pone.0198334

Abstract

ObjectiveReliable detection of HIV minority resistant variants (MRVs) requires bioinformatics analysis with specific algorithms to obtain good quality alignments. The aim of this study was to analyze ultra-deep sequencing (UDS) data using different analysis pipelines.MethodsHIV-1 protease, reverse transcriptase (RT) and integrase sequences from antiretroviral-naïve patients were obtained using GS-Junior® (Roche) and MiSeq® (Illumina) platforms. MRVs were defined as variants harbouring resistance-mutation present at a frequency of 1%–20%. Reads were analyzed using different alignment algorithms: Amplicon Variant Analyzer®, Geneious® compared to SmartGene® NGS HIV-1 module.Results101 protease and 51 RT MRVs identified in 139 protease and 124 RT sequences generated with a GS-Junior® platform were analyzed using AVA® and SmartGene® software. The correlation coefficients for the MRVs were R2 = 0.974 for protease and R2 = 0.972 for RT. Discordances (n = 13 in protease and n = 15 in RT) mainly concerned low-level MRVs (i.e., with frequencies of 1%–2%, n = 18/28) and they were located in homopolymeric regions (n = 10/15). Geneious® and SmartGene® software were used to analyze 143 protease, 45 RT and 26 integrase MRVs identified in 172 protease, 69 RT, and 72 integrase sequences generated with a MiSeq® platform. The correlation coefficients for the MRVs were R2 = 0.987 for protease, R2 = 0.995 for RT and R2 = 0.993 for integrase. Discordances (n = 9 in protease, n = 3 in RT, and n = 3 in integrase) mainly concerned low-level MRVs (n = 13/15).ConclusionWe found an excellent correlation between the various UDS analysis pipelines that we tested. However, our results indicate that specific attention should be paid to low-level MRVs, for which the use of two different analysis pipelines and visual inspection of sequences alignments might be beneficial. Thus, our results argue for use of a 2% threshold for MRV detection, rather than the 1% threshold, to minimize misalignments and time-consuming sight reading steps essential to ensure accurate results for MRV frequencies below 2%.

Highlights

Human Immunodeficiency Virus (HIV) sequencing is used to detect resistance-associated mutations (RAMs) at the time of virological failure to assess acquired drug resistance as well as at the time of diagnosis to assess transmitted drug resistance [1]
101 protease and 51 reverse transcriptase (RT) minority resistant variants (MRVs) identified in 139 protease and 124 RT sequences generated with a GS-Junior® platform were analyzed using AVA® and SmartGene® software
Geneious® and SmartGene® software were used to analyze 143 protease, 45 RT and 26 integrase MRVs identified in 172 protease, 69 RT, and 72 integrase sequences generated with a MiSeq® platform

Summary

Introduction

Human Immunodeficiency Virus (HIV) sequencing is used to detect resistance-associated mutations (RAMs) at the time of virological failure to assess acquired drug resistance as well as at the time of diagnosis to assess transmitted drug resistance [1]. Developed ultradeep sequencing (UDS) technologies produce a massive data volume and they are able to detect minority viral variants that harbor RAMs down to a frequency of 1% [2,4,5]. Most of the studies to date using UDS technologies define a frequency of 1% as the threshold for detection of minority resistant variants (MRVs) [6,9,11,12]. A good alignment quality generated by bioinformatics analysis software for UDS reads is mandatory for reliable detection and/or quantification of MRVs, especially when they occur at low frequencies. Several software alignment algorithms are available to analyze UDS data These are, not necessarily designed for HIV MRV analysis, and the analysis of HIV MRVs can present problems for alignment algorithms due to the existence of viral quasispecies and the presence of several homopolymeric regions along the viral genome. This study analyzed the results of UDS data generated with GS Junior or MiSeq platforms from HIV-1 clinical samples using three different bioinformatics analysis pipelines: Amplicon Variant Analyzer (AVA1), SmartGene NGS HIV-1 CE-labeled (SmartGene, Zug, Switzerland) and Geneious

Methods

Results

Discussion

Conclusion