HIV-1 genotypic resistance testing using the Vela automated next-generation sequencing platform

Abstract

Objectives

To evaluate the diagnostic performance of the Vela next-generation sequencing (NGS) system in conjunction with the Sentosa SQ HIV Genotyping Assay for genotyping HIV-1.

Methods

Plasma RNA was extracted and templates prepared with the Sentosa SX instrument before sequencing the HIV-1 polymerase on the Sentosa SQ301 Sequencer (PGM IonTorrent). The Vela NGS System was compared with direct sequencing and the 454 GS-FLX (Roche) and MiSeq (Illumina) systems for genotypic resistance testing on clinical samples.

Results

The Vela NGS system detected majority resistance mutations in subtype B and CRF02-AG samples at 500 copies/mL and minority variants with a sensitivity of 5% at 100 000 copies/mL. The Vela NGS system and direct sequencing identified resistance mutations with 97% concordance in 46 clinical samples. Vela identified 1/20 of the 1%–5% mutations identified by 454, 5/12 of the 5%–20% mutations and 60/61 of the >20% mutations. Vela identified 3/14 of the 1%–5% mutations identified by MiSeq, 0/2 of the 5%–20% mutations and 47/47 of the >20% mutations. The resistance mutation quantifications by Vela and 454 were concordant (bias: 2.31%), as were those by Vela and MiSeq (bias: 1.06%).

Conclusions

The Vela NGS system provides automated nucleic acid extraction, PCR reagent distribution, library preparation and bioinformatics analysis. The analytical performance was very good when compared with direct sequencing, but was less sensitive than two other NGS platforms for detecting minority variants.

Introduction

The great genetic variability of HIV enables it to adapt readily in order to evade the host immune system and/or ART. The virus in infected individuals forms a population composed of many genetically distinct, but related, variants that define the quasi-species. Patients infected with HIV-1 are screened for the presence of resistance mutations that decrease susceptibility to certain drugs before they are given antiretroviral drugs. Genotypic resistance testing is usually carried out by direct sequencing to identify variants accounting for ∼20% of the total virus population. However, minority variants accounting for <20% of the population may harbour resistance mutations and may emerge under drug selection pressure.^1,2 Minority variants that resist NNRTIs may impair the virological response to combination therapy with NNRTIs.^3–5 Deep sequencing techniques can detect minor variants.^6–8 These techniques have revealed that resistant variants are more prevalent than previously thought.⁹ The 454 GS-FLX (Roche) and Illumina sequencing platforms have been used to determine HIV-1 drug resistance by genotyping reverse transcriptase (RT), protease (PR) and integrase (IN).^10,11 Workflow on the next-generation sequencing (NGS) platforms was originally poorly automated and bioinformatics analysis required the development of an in-house pipeline for alignment, data cleaning and interpretation. Vela Diagnostics has developed a solution for HIV genotyping based on ultra-deep sequencing with the Ion Torrent PGM platform and a completely automated workflow.

We evaluated the performance of the Sentosa NGS system in conjunction with the Sentosa SQ HIV Genotyping Assay for sequencing and genotyping HIV samples. We compared results obtained with the Vela NGS sequencing with those obtained with the 454 GS-FLX and Illumina MiSeq platforms.

Materials and methods

Samples

HIV-1 samples of various subtypes harbouring resistance-associated mutations (RAMs) were used to validate the performance of the Vela NGS platform. Two samples of subtype B and CRF02-AG with majority resistance mutations were diluted with HIV-negative plasma to concentrations of 100 000 to 500 copies/mL and used to evaluate analytical sensitivity and specificity. We evaluated subtype determination by analysing 10 NIH reference strains of HIV-1 (culture supernatants, 10 000 copies/mL), 3 subtype B and 7 non-B subtypes. Sequencing methods were compared using 46 clinical samples (mean virus load: 6.15 log copies/mL; range 2.7–7 log copies/mL) tested by direct sequencing and with two NGS platforms, 454 GS-FLX (Roche) and MiSeq (Illumina).

Extraction, amplification and ultra-deep sequencing on the Vela platform

The workflow included the extraction of RNA from plasma samples and RT–PCR preparation using a Sentosa SX 101 instrument (Vela Diagnostics, Hamburg, Germany). RT–PCR was performed on a Veriti thermal cycler (Applied Biosystems, Foster City, CA, USA). Library preparation was completed on the Sentosa SX 101. Emulsion PCR and enrichment were done on dedicated instruments before sequencing. The RT, PR and IN genes were sequenced on a Sentosa SQ 301 Sequencer (Ion Torrent PGM) using dedicated reagent kits from Sentosa. The Vela software assembled two contigs per sample, one for the PR-RT and one for the IN, and provided lists of mutations.

HIV-1 genotyping by direct sequencing

HIV-1 RNA was extracted from 140 μL of plasma using QIAamp RNA extraction kits (Qiagen, Courtaboeuf, France), amplified and then sequenced using the ANRS protocol (http://www.hivfrenchresistance.org/ANRS-procedures.pdf). The resistance mutations list was the 2016 ANRS algorithm list (http://www.hivfrenchresistance.org/2016/Algo2016-hiv1.pdf). Mutations I15V in PR and A98S in RT were excluded from the analysis because the Vela software did not list them.

HIV-1 genotyping on the 454 GS-FLX Roche platform

Clinical samples were analysed on the Roche 454 GS-FLX platform as previously described.¹² Briefly, RNA was extracted, first-strand cDNA was generated with two gene-specific oligonucleotides and then used to produce six partly overlapping amplicons covering the HIV-1 pol gene (RT and PR codons 1–251) that were mixed in equimolar amounts. Clonal amplification was performed by emulsion PCR and sequenced on a Genome Sequencer FLX (Roche-454 Life Sciences). The sequences of the RT and PR regions were first processed using GS Amplicon Variant Analyzer (AVA) software (Roche Diagnostic, Meylan, France). The resulting .fna files were then analysed with DeepChek-HIV version 1.4 software (ABL, TherapyEdge, Luxembourg).

HIV-1 genotyping on the MiSeq Illumina platform

Clinical samples were analysed on the Illumina MiSeq platform. RNA was extracted and One-Step Qiagen kits (Qiagen, Courtaboeuf, France) used to generate a long (1.4 kb) PCR fragment encompassing the RT and PR regions. Then, nested PCR using the KAPA HiFi HotStart ReadyMix (Kapa Biosystem, Boston, MA, USA) was used to generate four overlapping amplicons from each patient sample. These four amplicons were purified, quantified using Quant-iT Picogreen dsDNA assay kits (Invitrogen) and pooled in similar quantities. Index tags were added to each pool of amplicons using the Illumina Nextera XT indexing procedure. Indexed MiSeq amplicons were purified using Agencourt AMPure XP magnetic capture beads and quantified using Quant-iT Picogreen dsDNA assay kits. Equimolar concentrations of amplicons from all patients were pooled in one library. The samples were prepared for sequencing using the MiSeq Reagent Kit version 2 (2 × 250 bp paired-end reads, 8.5 Gb output) and sequenced on an Illumina MiSeq at the Toulouse genomic platform (http://get-genotoul.fr). The raw MiSeq sequencing data were processed using the NG6 application (http://ng6.toulouse.inra.fr/) and an in-house pipeline so that only reads with 95% of nucleotides having a Q score >30 were conserved. FASTQ files were then analysed with DeepChek-HIV version 1.4 software.

Results

Analytical sensitivity

Samples, one of subtype B and one of CRF02-AG with majority resistance mutations, were diluted with HIV-negative plasma to concentrations of 100 000, 10 000, 2000, 1000 and 500 copies/mL and tested in triplicate (Table 1). The PR-RT and IN genes of both samples were successfully sequenced in all triplicates with mean coverages from 2278 to 6657 reads in the RT-PR gene and from 6264 to 15 160 reads in the IN gene (Table 1). The majority RAMs were correctly identified by Vela in both the subtype B (L63P, V77I, E138A, M184V) and CRF02-AG samples (G16E, K20I, M36I, L89M, V179I) in all triplicates between 100 000 and 500 copies/mL.

Table 1.

Analytical sensitivity of the Vela NGS system

	Virus load (copies/mL)
	HIV-1 subtype B					HIV-1 CRF02-AG
	100 000	10 000	2000	1000	500	100 000	10 000	2000	1000	500
PR-RT
mean coverage (reads)	5356	5768	2278	6588	6657	3939	6001	6394	4801	3694
triplicates detected	3	3	3	3	3	3	3	3	3	3
IN
mean coverage (reads)	12 219	11 197	7787	10 530	9341	15 160	9784	9461	6264	8100
triplicates detected	3	3	3	3	3	3	3	3	3	3
Unexpected mutations			PR 54T 6.38%a		RT 219R 13%a			IN 230N 3.97%a

Performance of Vela sequencing of two samples of subtype B and CRF02-AG with majority resistance mutations diluted with HIV-negative plasma.

a

Mutation observed in one of the triplicates.

Majority RAMs (RT: 184V 138A; PROT: 63P 77I) were identified in all triplicates of mixtures of mutated and WT strains containing 10% and 5% of the mutated strain at a concentration of 100 000 copies/mL. Thus, 5000 copies/mL of the mutated strain were identified in 100 000 copies/mL of HIV.

Analytical specificity

Specificity was evaluated using dilutions of the subtype B and CRF02-AG samples. The only mutations found in addition to the majority mutations detected at 100 000 copies/mL were: one mutation I54T (6.38%) in one triplicate of subtype B (dilution: 2000 copies/mL), one mutation K219R (13%) in one triplicate of subtype B (dilution: 500 copies/mL) and one mutation S230N (3.97%) in one triplicate of subtype CRF02-AG (dilution: 2000 copies/mL) (Table 1). Specificity was >99.9% taking into account the 276 mutations analysed by the Vela NGS system. The three false minority mutations were not present in the FASTQ sequences that were analysed independently of the Vela software.

Determination of HIV-1 subtype

We sequenced 10 NIH HIV-1 strains on the Vela NGS platform: 3 subtype B specimens and 7 non-B specimens (Table 2). The subtypes were correctly determined except for the subtype C, which was found to be CRF07 (recombinant of subtype B and C). Direct sequencing identified 27 mutations in RT-PR, 25 of which were correctly identified by Vela. Two samples had discordant sequencing results: one mixed population L89I/M was found to be a majority mutation (L89M 99.9%) by Vela and one mutation G16E was found as a minority mutation (6.4%) by Vela.

Table 2.

Vela sequencing of HIV-1 NIH reference strains

Subtype (NIH strain)	Vela subtype	Sanger PR mutations	Vela PR mutations (frequency)	Sanger RT mutations	Vela RT mutations (frequency)
B (BaL)	B	V77I	77I (99%)	—	—
B (JRFL)	B	—	—	K103R	103R (99%)
				V179I	179I (99%)
B (ADA)	B	—	—	M41L	41L (99%)
A	A1	L33F	33F (99.5%)	V179I	179I (99.7%)
		M36I	36I (99.5%)		103E (12.6%)
		L89M	89M (99.8%)
C	CRF07-BC	D60E	60E (99.8%)	—	—
		L63P	63P (99.8%)
		I93L	93L (99.8%)
D	D	—	—	—	—
F	F1	M36I/V	36I (47.3%) 36V (53.9%)	—	—
		L63wt/P	—
		L89I/M	89M (99.9%)
G	G	K20I	20I (98.8%)	I62wt/V	62V (42.83%)
		M36I	36I (99.5%)
		V82I	82I (99.9%)
		L89M	89M (99.9%)
CRF01-AE	CRF01	G16E	16E (6.4%)	V179I	179I (99.7%)
		K20R	20R (99.5%)		238R (99.9%)
		M36I	36I (99.8%)
		L89M	89M (99.1%)
CRF02-AG	CRF02	K20I	20I (99.3%)	—	—
		M36I	36I (99.6%)
		—	77I (7.7%)
		L89M	89M (99.8%)

Repeatability and reproducibility

A control specimen (10 000 copies/mL mixture of 80% WT JRCSF strain and 20% of a variant with M184V resistance mutation) was used to measure reproducibility. The control was tested in triplicate in a single run: the average frequency of the M184V mutation was 39.8% and the coefficient of variation (CV) was 13%. One control tested in three different runs gave an average M184V frequency of 62.2% and a between-runs CV of 33%.

Comparison of Vela NGS system and direct sequencing

We analysed 46 samples for RT and PR genotyping with both the Vela NGS and direct sequencing: 26 subtype B, 10 CRF02-AG and 7 other non-B subtypes (subtypes C, D, A1, CRF30 and CRF01-AE) according to phylogenetic analysis of Sanger sequences.

The resistance genotyping indicated that 39 patients harboured mutations. Sanger sequencing detected 103 PR-RT mutations (Table 3). These PR-RT mutations were identified by both sequencing methods except that the mutation L10V was identified in only 3/5 samples by Vela and one mutation T215D was not identified by Vela. Virus loads were 4.44 and 4.64 log copies/mL in the samples whose L10V was not detected by Vela, while virus loads were 6.33, 6.84 and 2.69 log copies/mL in the samples whose L10V was detected by Vela. The virus load of the sample harbouring the T215D was 3.19 log copies/mL. Overall, the Vela NGS system detected 100 of the 103 RAMs detected by Sanger sequencing (concordance 97%). The three RAMS not detected by the Vela platform were present in the FASTQ sequences that were reanalysed.

Table 3.

Concordance between Vela NGS system and Sanger direct sequencing for HIV-1 genotyping

	Mutation	Sanger (no. of mutations)	Vela (no. of mutations)
PR	10V	5	3
	16E	6	6
	20I	10	10
	36I	17	17
	62V	6	6
	63P	17	17
	69K	5	5
	71T	2	2
	71V	2	2
	77I	6	6
	89M	13	13
	93L	4	4
RT	41L	1	1
	90I	1	1
	103N	1	1
	138A	2	2
	179I	2	2
	179D	1	1
	215D	1	0
	215C	1	1

Comparison of Vela and other NGS platforms

We analysed 43 samples for RT and 21 samples for PR with both the Vela and 454 GS-FLX platforms: the 454 GS-FLX detected 92 mutations (1% threshold) (Table 4). The Vela NGS detected 1/20 of the 1%–5% mutations detected by the 454, 5/12 of the 5%–20% mutations and 60/61 of the >20% mutations. The 5%–20% mutations were detected in 8/12 samples when we reanalysed the FASTQ sequences generated by the Sentosa SQ 301 sequencer. The two platforms measured similar percentages of the 66 RAMs (Spearman correlation ρ = 0.6898; P < 0.0001). A Bland–Altman plot showed a mean difference of 2.3% between the Vela and 454 quantifications (Figure 1a).

Table 4.

Concordance between Vela and 454 GS-FLX platforms

Mutation frequency	Mutations detected by 454 GS-FLX	Mutations detected by Vela, n (%)a
1%–5%	20	1 (5)
5%–20%	12	5 (42)
>20%	61	60 (98)

a

Number of mutations detected by Vela among the mutations detected by 454 GS-FLX.

Figure 1.

Comparison of two NGS methods for quantifying mutated variants. (a) Bland–Altman diagram comparing the Vela NGS and the 454 GS-FLX platforms. (b) Bland–Altman diagram comparing the Vela NGS and the MiSeq platforms. Each point represents one of the mutated variants quantified by both platforms. The bias is the mean of the difference between the two methods. The 95% limit of agreement contains 95% of the possible differences.

We analysed 23 samples in the RT and PR genes using the Vela NGS and MiSeq Illumina platforms. MiSeq detected 63 RAMs (1% threshold) (Table 5). Vela NGS detected 3/14 of the 1%–5% mutations detected by the MiSeq platform, 0/2 of the 5%–20% mutations and 47/47 of the >20% mutations. The two platforms measured similar percentages of the 53 mutations (Spearman correlation ρ = 0.4060; P = 0.0026). A Bland–Altman plot showed a mean difference of 1.1% between the Vela and MiSeq quantifications (Figure 1b).

Table 5.

Concordance between Vela and MiSeq platforms

Mutation frequency	Mutations detected by MiSeq	Mutations detected by Vela, n (%)a
1%–5%	14	3 (21)
5%–20%	2	0 (0)
>20%	47	47 (100)

a

Number of mutations detected by Vela among the mutations detected by MiSeq.

The workflow and the run times of the three NGS assays and the Sanger sequencing are shown in Table 6. We estimated that the Vela took 4 h to handle 15 samples while the 454 GS-FLX required about 12 h, the Miseq 9 h and Sanger sequencing 7 h.

Table 6.

Comparison of the workflow of Sanger sequencing and three NGS platforms

	Sanger sequencing	454 GS-FLX	MiSeq Illumina	Sentosa Vela DX
Sample workflow
extraction	1 h	1 h	1 h	2 h 15 min
amplification	4 h	6 h	6 h	4 h 20 min
library preparation	—	13 h	12 h	9 h 45 min
sequencing	8 h	12 h	15 h	5 h
data analysis	3 h	1 h	1 h	4 h
Time for a run (15 samples)	16 h	33 h	35 h	25 h 20 min
Hands-on time	7 h	12 h 30 min	9 h 30 min	4 h

Discussion

We have evaluated the analytical performance of the Vela NGS platform, which provides a completely automated system for HIV genotyping from plasma samples, including identification and quantification of resistance mutations.

The Vela NGS system detected majority variants in subtype B and CRF02-AG samples at a virus load of 500 copies/mL. It also detected RAMs at 200 copies/mL (2/3 triplicates) and 100 copies/mL (1/3 triplicates). Thus, the system can be used when the HIV-1 viraemia is below that indicated in the package insert of the manufacturer (1000 copies/mL). Specificity was very good (>99.9%); false mutations were observed only at low virus loads and low frequencies.

The number of non-B isolates of HIV-1 was too small to evaluate the performance of subtype determination by the Vela NGS system, but the set of tested subtypes was successfully amplified and gave satisfactory resistance genotyping results. The Vela NGS system and Sanger direct sequencing agreed well (97%) for determining the genotypic resistance of HIV-1 in a set of 46 clinical samples. Similarly, the Vela NGS system and two other NGS platforms (454 GS-FLX and MiSeq) agreed in >98% of cases for identifying the mutations accounting for >20% of the virus population. The 454 GS-FLX and HiSeq 2000 NGS platforms have also been found to give very similar results.¹³

We evaluated the reproducibility of quantification with a control specimen tested in triplicate in the same run and then in separate runs, and by comparing the quantification obtained using another NGS platform. Within-run reproducibility was good (CV = 13%), while between-run reproducibility was lower (CV = 33%). A previous multicentre study determined the between-run reproducibility on the 454 GS-FLX platforms in different sites and found a CV of 40%.¹⁴ The quantification of the control specimen did not agree with the theoretical values in the mixture. This could be due to PCR bias in the artificial mixture of two HIV strains. On the contrary, the data obtained with the Vela NGS system and the 454 GS-FLX and MiSeq platforms for the quantification of RAMs in clinical samples were well correlated. The outputs of the three NGS platforms agreed well, with a mean difference of <3% between them, despite their different chemical processes.

We compared the new Vela NGS platform with the 454 GS-FLX and MiSeq platforms for the detection of minority variants. Vela identified 19% of the minority resistant variants that the 454 GS-FLX or MiSeq found, accounting for <20% of the quasi-species. As expected, the Vela NGS detected more minority variants when the frequency of mutation determined by the other NGS platform increased. Although the Vela NGS system was able to detect variants accounting for >20% of the quasi-species, this method had lower ability to detect minority variants than the other NGS platforms. However, NGS can provide better detection of majority mutations than Sanger sequencing, which may miss certain mutations above a 20%–30% threshold. The Vela NGS system provides automated RNA extraction, PCR reagent distribution, library preparation and bioinformatics analysis. This automation reduces considerably the handling time and the risk of human mistakes.

Bioinformatics analysis of the data from the Vela platform was completely automated with no options or settings that could be selected by the user. We reanalysed the FASTQ sequences from the Sentosa SQ 301 sequencer that harboured discordances with sequences obtained by the other assays. We found that errors were generated during automated bioinformatics analysis on a small proportion of samples. The result of the analyses could depend on the set of reference sequences used for alignment. The new version of the assay should offer improved bioinformatics analysis.

The Vela NGS system and HIV Genotyping Assay accurately identified HIV-1 resistance mutations. Nucleic acid extraction, PCR reagent distribution, library preparation and bioinformatics analysis are all automated. Vela sequencing identified the same RAMs as those found by direct sequencing. The Vela NGS seems to be less sensitive for detecting minority variants than the other NGS platforms. However, all three NGS platforms, Vela Sentosa, 454 GS-FLX and MiSeq, detected variants accounting for >20% of the quasi-species. This threshold is currently the relevant threshold for the analysis of HIV resistance in clinical practice in the absence of recommendations for the detection of minority variants.

Funding

Vela Diagnostics supplied the platform instruments and the reagents for this independent evaluation of the Sentosa SQ HIV Genotyping Assay. This study was supported by internal funding.

Transparency declarations

None to declare.