Prospective Evaluation of the Vela Diagnostics Next-Generation Sequencing Platform for HIV-1 Genotypic Resistance Testing

Abstract

Genotypic antiretroviral drug resistance testing is a critical component of the global efforts to control the HIV-1 epidemic. This study investigates the semiautomated, next-generation sequencing (NGS)-based Vela Diagnostics Sentosa SQ HIV-1 Genotyping Assay in a prospective cohort of HIV-1–infected patients. Two-hundred sixty-nine samples were successfully sequenced by both NGS and Sanger sequencing. Among the 261 protease/reverse transcriptase (PR/RT) sequences, a mean of 0.37 drug resistance mutations were identified by both Sanger and NGS, 0.08 by NGS alone, and 0.03 by Sanger alone. Among the 50 integrase sequences, a mean of 0.3 drug resistance mutations were detected by both Sanger and NGS, and 0.08 by NGS alone. NGS estimated higher levels of drug resistance to one or more antiretroviral drugs for 6.5% of PR/RT sequences and 4.0% of integrase sequences, whereas Sanger estimated higher levels of drug resistance for 3.8% of PR/RT sequences. Although the samples successfully sequenced by the Sentosa SQ HIV Genotyping Assay demonstrated similar predicted resistance compared with Sanger, 44% of Sentosa runs failed quality control requiring 17 additional runs. This semi-automated NGS-based assay may aid in HIV-1 genotypic drug resistance testing, though numerous quality control issues were observed when this platform was used in a clinical laboratory setting. With additional refinement, the Sentosa SQ HIV-1 Genotyping Assay may contribute to the global efforts to control HIV-1.

The United Nations' 2030 Agenda for Sustainable Development and the World Health Organization's Global Health Sector Strategy on HIV, 2016 to 2021, has committed to end the AIDS epidemic by 2030 (World Health Organization, https://www.who.int/hiv/strategy2016-2021/ghss-hiv/en, last accessed February 6, 2019). To monitor progress toward achieving this goal, the World Health Organization and the Joint United Nations Programme on HIV/AIDS have set the target of 90% of individuals living with HIV on antiretroviral therapy (ART) to achieve virologic suppression by 2020 (World Health Organization, https://www.who.int/hiv/pub/drugresistance/tackling-hiv-drug-resistance/en, last accessed February 6, 2019). This target rises to 95% by 2030. As of 2017, an estimated 36.9 million individuals were living with HIV, 21.7 million were accessing ART, and approximately 81% (17.6 million) of those on ART had achieved virologic suppression (UNAIDS, http://www.unaids.org/en/resources/fact-sheet, accessed April 30, 2019). However, there remain more than 4 million individuals worldwide who have failed to achieve virologic suppression while on ART. This may be due to transmitted drug resistance, acquired drug resistance, poor adherence to ART, or a combination of these factors.

In high-resource settings, antiretroviral drug resistance testing is recommended at diagnosis, at the initiation of therapy, and at virologic failure to optimize ART. In low-resource settings, resistance testing is not routinely available for clinical decision-making, and virologic failure is managed via adherence counseling and then empirical transition to second-line ART regimens. However, resistance testing for low-resource settings is important for epidemiological purposes to identify rates of pretreatment drug resistance and to inform local and international HIV policy, particularly on the selection of ART regimens.

Antiretroviral drug resistance testing is performed via sequence analysis of the HIV-1 Pol gene including reverse transcriptase (RT), protease (PR), and integrase (IN) regions, encoding the targets of the most commonly used HIV drugs: non-nucleoside reverse transcriptase inhibitors (NNRTIs), nucleoside reverse transcriptase inhibitors (NRTIs), protease inhibitors (PIs), and integrase strand transfer inhibitors (INSTIs). Conventional chain-terminating dideoxynucleotide population sequencing, or Sanger sequencing, is the primary sequencing method used to identify HIV drug resistance mutations (DRMs), and currently the only commercially available US Food and Drug Administration–approved HIV-1 genotyping kit, Viroseq HIV-1 Genotyping System version 2.0 (Abbott Molecular, Des Plaines, IL), also uses Sanger methodology.

The identification of HIV-1 DRMs have been studied extensively using next-generation sequencing (NGS) technologies.¹ NGS methods can reproducibly detect low-level DRMs at 5% of the population, or lower, compared with the 10% to 20% that can be detected by Sanger sequencing.² Furthermore, several studies have demonstrated that detection of low-level DRMs are associated with virologic failure,3, 4, 5, 6, 7 particularly NNRTI DRM. However, high-level DRMs, defined as mutations detectable by Sanger, are more likely to result in virologic failure.4, 7, 8

A recently developed semiautomated, NGS-based commercial platform for HIV-1 genotypic drug resistance testing is the Sentosa SQ HIV-1 Genotyping Assay, which utilizes a robotic liquid handling system for RNA extraction and library preparation (Vela Diagnostics, Singapore). RT, PR, and IN DRMs are identified via ion semiconductor sequencing coupled with integrated data analysis and reporting software. In this study, the clinical performance characteristics of the Sentosa SQ HIV-1 Genotyping Assay were investigated in a prospective cohort of HIV-1 infected patients undergoing HIV-1 genotypic evaluation at a US academic medical center.

Materials and Methods

Ethics Statement

This study was reviewed and waived by the institutional review board of Stanford University.

Specimen Selection

EDTA plasma samples collected between January and August 2018 submitted to the Stanford Health Care Clinical Virology Laboratory for HIV-1 genotypic resistance testing were aliquoted upon arrival and stored at −80°C until tested by NGS in batch. Inclusion criteria included sufficient volume for testing by both Sanger and NGS, and the presence of amplifiable RNA by the primers used for conventional RT-PCR and Sanger sequencing in the clinical laboratory.

A subset of specimens had concurrent virus load results available for analysis. Testing was performed using the cobas HIV-1 test on the cobas 8800 system (Roche Molecular Systems, Pleasanton, CA). Values were recorded in the Stanford Health Care Clinical Virology Database using 1.9 log₁₀ copies/mL EDTA plasma as the lower limit of quantitation, to maintain consistency with data collected using prior virus load assays.

Dideoxynucleoside Sequencing (Sanger)

Laboratory-developed dideoxynucleoside sequencing was performed for PR/RT and IN by the Stanford Health Care Clinical Virology Laboratory. The Stanford Health Care Clinical Laboratories are Clinical Laboratory Improvement Amendments (license 05D1038598) and College of American Pathologists (license CAP 2379301) accredited. For Sanger sequencing–based HIV-1 genotyping, the laboratory participates in both College of American Pathologists proficiency testing and the National Institute of Allergy and Infectious Diseases–sponsored Viral Quality Assurance Program, through which the Stanford Health Care Clinical Virology Laboratory is approved to perform PR/RT and IN genotypic drug resistance testing for National Institutes of Health–supported clinical trials.

Total nucleic acid extraction was performed using 400 μL of EDTA plasma with the EZ1 Virus Mini Kit version 2.0 on the EZ1 instrument (both from Qiagen, Germantown, MD). HIV-1 RNA was converted to cDNA and amplified using SuperScript One-Step RT-PCR System with Platinum Taq DNA Polymerase, and then further amplified in a second-round PCR using AmpliTaq DNA Polymerase (both from Thermo Fisher Scientific, Waltham, MA), as previously described.⁹ PR/RT sequences encompassed the entire PR gene and RT codons 1 to 300. IN sequences encompassed the entire IN gene. The primers used for amplification were as previously described.⁹ Bidirectional-sequencing was performed using BigDye Terminators (Thermo Fisher Scientific, Waltham, MA) with products resolved on an ABI 3730 capillary electrophoresis instrument. A mixture was defined as a position having a secondary peak at least 20% of the area under the curve and evidence of mixture in both directions.

Next-Generation Sequencing

The Sentosa SQ HIV Genotyping Assay is an NGS-based integrated workflow, comprising kits for RNA extraction HIV-1 library preparation and sequencing; a robotic liquid handling system for RNA extraction and library preparation; Ion Torrent instruments for sequencing; and data analysis and reporting software. The assay processes up to 15 samples (730 μL per sample) simultaneously. The system sequences the entire PR gene, the first 376 amino acids of RT, and the entire IN gene. For the purposes of this study, the assay exports a FASTA file containing a single consensus nucleotide sequence in which positions containing nucleotide mixtures with variants present at or above 3.2% are represented as International Union of Pure and Applied Chemistry (IUPAC) ambiguities. Each NGS sequence was truncated before comparison to its corresponding Sanger sequence: for the PR/RT amplicon, only the first 1300 nucleotides were used, and for the IN amplicon, only the first 1000 nucleotides were used, including the entire 864 nucleotides of the IN coding sequence.

The Sentosa SQ HIV-1 Genotyping Assay provides a system control for each run, processed from extraction as the 16th sample.¹⁰ The system control acts as both a positive and negative control. Criteria for system control acceptance include positive control assembly quality (completeness ≥95.0%, error rate ≤1.00%, median coverage ≥200 reads), no template control (HIV reads ≤50 or 0.0001 × the number of control amplicon reads), and detection of three low-level, non–HIV-1 variants (range 1.0% to 8.0%). The system control data are detailed in the run quality controls (QC) section of the QC report, which also includes run throughput (≥40,000,000 bp), loading % (≥15.00), key signal (≥30), invalid reads (≤10%), and invalid bases (≤10%). The QC report also contains a section entitled Sample Quality Controls that includes sample throughput (≥125,000 bp), integrase assembly (median coverage ≥50 reads), protease/RT assembly (median coverage ≥50 reads), and control amplicon median coverage (median coverage ≤200 reads). Sequencing failure occurred when one or more of the following criteria were not met: base call accuracy (≥97.5), key signal (≥0), mean AQ20 read length (≥122 bp), and percent reads (≥0.01). Each sample included in the analysis passed both run QCs and sample QCs.

Data Analysis

Separate analyses were performed for PR/RT and IN. Complete nucleotide concordance was defined as both Sanger and NGS identifying the same nucleotide or ambiguity code at a position. Partial nucleotide discordance was defined as one method identifying a nucleotide mixture and the other identifying one of the mixture's components. Complete nucleotide discordance was defined as both methods identifying different nonambiguous nucleotides or ambiguous nucleotides that are nonoverlapping.

Although the Sentosa assay has its own analysis pipeline, including a susceptibility report that uses three HIV drug resistance algorithms (Stanford HIVDB version 8.2, ANRS version 2016.26, and Rega version 9.1), Stanford HIVDB version 8.7 was used directly to allow for the most updated interpretation. For concatenated PR/RT sequences and for IN sequences, the Sanger and NGS FASTA files were submitted to the Stanford HIV Drug Resistance Database (HIVDB) genotypic resistance interpretation program, and the Spreadsheets (TSV) output option was selected. The rows in the formatted amino acid alignment files generated from the Sanger and NGS sequences were compared to determine concordance for detecting mutations, defined as amino acid differences from the subtype B consensus amino acid sequence (HIV Databases, https://www.hiv.lanl.gov/content/sequence/HIV/CONSENSUS/Consensus.html, last accessed February 6, 2019).

The Resistance Summary spreadsheets were used to compare DRMs, defined as mutations assigned a penalty score by the HIVDB interpretation system¹¹; DRMs detected by Sanger and NGS were considered concordant if they were present regardless of what other mutations were also present. The rows in the Resistance Summary spreadsheets generated from the Sanger and NGS sequences were used to determine the concordance for the detection of categorical drug resistance interpretations for the most commonly used antiretrovirals: the NRTIs lamivudine/emtricitabine (3TC/FTC; which were treated as one drug), abacavir (ABC), azidothymidine (AZT), and tenofovir (TDF); the NNRTIs efavirenz (EFV), etravirine (ETR), and rilpivirine (RPV); the PIs atazanavir (ATV), darunavir (DRV), and lopinavir (LPV); and the INSTIs dolutegravir (DTG), elvitegravir (EVG), and raltegravir (RAL). There were five predicted drug resistance interpretation levels: susceptible, potential low-level resistance, low-level resistance, intermediate resistance, and high-level resistance.¹¹ The penalty scores used to generate these levels were those in the HIVDB version 8.7 database, updated October 19, 2018. These levels were used to compare the drug resistance interpretations reported by NGS and Sanger sequencing.

The Sequence Summary spreadsheets were used to compare the number of signature APOBEC mutations defined as mutations specific for APOBEC-mediated RNA editing, with three or more being associated with a high probability of G-to-A hypermutation; and the proportion of positions with highly unusual mutations defined as having a prevalence <0.01% in HIVDB and not being a known DRM or signature APOBEC mutation.¹²

In addition to the HIVDB analysis, concordance was examined between NGS and Sanger sequence pairs for each individual patient sample. Sequence pairs were aligned using the Clustal X software version 2.1 (Clustal, http://www.clustal.org/download/current) and analyzed for concordance using R version 3.3.1 (R Foundation for Statistical Computing, Vienna, Austria), ignoring positions which lacked a nucleotide call in one or both assays. Finally, to identify the percent mixtures in samples with discrepant resistance profiles, raw NGS reads were aligned to the HIV-1 HXB2 reference genome (K03455) using Burrows-Wheeler Aligner's Smith-Waterman Alignment with default parameters,¹³ and mutant codons were counted using a custom python script (Supplemental File S1).

Sequences

The complete set of 261 PR/RT and 50 IN NGS sequences, annotated by sample ID and sequencing method, were submitted to the NCBI Sequence Read Archive BioProject ID PRJNA541016 (https://www.ncbi.nlm.nih.gov/bioproject/PRJNA541016; accession numbers SAMN11570569 to SAMN11570837). The corresponding set of Sanger sequences were submitted to GenBank (https://www.ncbi.nlm.nih.gov/genbank; accession numbers MK922662 to MK922970).

Results

Quality Control

A total of 308 unique samples met the inclusion criteria, comprising 358 Sanger sequences (302 RT/PR, 56 IN). Among these sequencing tests, no control failures were observed. For 88.5% (317/358), amplicon and sequence quality met acceptability criteria, and no repeats were required. However, 6.1% (22/358) required repeat Sanger sequencing, 3.4% (12/358) required repeat extraction/RT-PCR, and 2.0% (7/358) required both repeat extraction/RT-PCR and Sanger sequencing.

The 308 unique samples that met the inclusion criteria were tested using the Sentosa SQ HIV Genotyping Assay. Samples were divided into 23 batches. For 56% (13/23) of these batches, run QC was acceptable on the first run, and no repeats were required. However, for 44% (10/23), one or more repeats were necessary due to system control failure or sequencing failure. This included batches that underwent one (n = 5), two (n = 3), and three (n = 2) repeats, for a total of 17 additional runs. The system control failed eight times; these runs were repeated from extraction and included only the samples with sufficient residual volume. Sequencing failure occurred in nine runs, which were then repeated from the pooled library, starting at the emulsion PCR step. Two batches were repeated two and three times, respectively, without passing run QC; these batches were excluded from further testing.

Within successful Sentosa runs, no amplicon was sequenced by NGS in 24 samples and incomplete sequences were identified in five samples. Of these 29 samples, 44.8% (13/29) had concurrent virus loads in the quantifiable range, with a mean of 2.7 log₁₀ copies/mL (SD = 1.04). 34.5% (10/29) of these samples had concurrent virus loads below the lower limit of quantitation. Concurrent virus loads were not available for the remaining six samples.

Nineteen of the samples that generated no amplicon or a partial amplicon had sufficient volume for repeat testing. These samples were repeated along with one sample that failed due to low depth and 15 samples that generated a low depth warning but still produced a complete sequence. Of these 35 repeated samples, 20 were excluded due to system control failure, and four failed to produce sequence. Of the four samples that failed to produce sequence, three had concurrent virus load data: two within the quantifiable range, both at 2.1 log₁₀ copies/mL, and one below the lower limit of quantitation. The remaining 11 repeats were successful, and the new sequence replaced the original sequencing result for those samples. For these 11 repeated samples, 54.5% (6/11) had concurrent virus loads in the quantifiable range (mean = 3.7 log₁₀ copies/mL, SD = 1.62), and 9.1% (1/11) had a concurrent virus load below the lower limit of quantitation.

Samples Sequenced by Sanger and NGS

Two-hundred sixty-nine prospective samples were successfully sequenced by both NGS and Sanger. Of these samples, 70.3% (189/269) had concurrent virus loads in the quantifiable range, with a mean of 4.4 log₁₀ copies/mL (SD = 1.12). For 27.5% (74/269), concurrent virus loads were not available. The mean virus load for samples successfully sequenced by NGS and Sanger was significantly different from the virus loads of samples in which only Sanger sequence was obtained [4.4 ± 1.12 (n = 189) versus 2.7 ± 1.04 (n = 13); P < 0.001, t-test]. For six samples in which both NGS and Sanger sequence was obtained and 10 samples in which only Sanger sequence was obtained, the virus load was below the quantifiable range.

The 269 prospective samples included 219 that underwent PR/RT sequencing, 42 that underwent PR/RT and IN sequencing, and 8 that underwent just IN sequencing. Thus, 261 samples underwent PR/RT, and 50 underwent IN sequencing by both Sanger and NGS. For NGS, the median read coverages for prospective PR/RT and IN sequences were 10,032 and 6611, respectively (interquartile ranges: 3515 to 12,834; 1944 to 7934, respectively).

In all samples, Sanger and NGS were aligned, and only those positions that contained a nucleotide call by both pipelines were compared. There were 256 subtype B samples. Thirteen samples had non-B subtypes including subtypes A (n = 1), C (n = 6), CRF01_AE (n = 4), and CRF02_AG (n = 2). NGS and Sanger agreed on all subtype calls.

Table 1 lists each of the NRTI, NNRTI, PI, and INSTI DRMs detected by Sanger sequencing according to its frequency in the tested samples. The samples contained 75 distinct DRMs: 25 NRTI-, 23 NNRTI-, 16 PI-, and 11 INSTI-associated DRMs.

Table 1.

Prevalence of DRMs by Sanger in This Cohort

NRTI DRM	#	%	NNRTI DRM	#	%	PI DRM	#	%	INSTI DRM	#	%
M184V	16	6.1	K103N	24	9.2	L90M	5	1.9	E92Q	5	10.0
D67N	12	4.6	E138A	13	5.0	L10F	4	1.5	E157Q	3	6.0
M41L	9	3.4	V106I	11	4.2	M46I	3	1.1	G163R	2	4.0
M184I	7	2.7	V179D	9	3.4	Q58E	3	1.1	T97A	2	4.0
T69D	5	1.9	K101E	7	2.7	V82A	3	1.1	E138K	1	2.0
K219Q	4	1.5	G190A	6	2.3	I47V	2	0.8	G140C	1	2.0
K70R	4	1.5	Y181C	5	1.9	K43T	2	0.8	Q148R	1	2.0
L210W	4	1.5	A98G	4	1.5	V32I	2	0.8	T66I	1	2.0
T215E	4	1.5	V108I	4	1.5	F53L	1	0.4	Y143C	1	2.0
K65E	3	1.1	V179E	4	1.5	G73S	1	0.4	Y143H	1	2.0
T215D	3	1.1	Y188L	4	1.5	I50V	1	0.4	Y143R	1	2.0
T215S	3	1.1	E138K	3	1.1	I54V	1	0.4
A62V	2	0.8	K103S	3	1.1	K20T	1	0.4
D67G	2	0.8	P225H	3	1.1	L33F	1	0.4
K219E	2	0.8	H221Y	2	0.8	L89V	1	0.4
K65R	2	0.8	K238T	2	0.8	M46L	1	0.4
K70E	2	0.8	Y318F	2	0.8
K70N	2	0.8	E138G	1	0.4
T215A	2	0.8	E138Q	1	0.4
T215L	2	0.8	G190E	1	0.4
T215Y	2	0.8	Y181F	1	0.4
K70G	1	0.4	Y181I	1	0.4
K70Q	1	0.4	Y188H	1	0.4
T215N	1	0.4
T215V	1	0.4

DRM, drug resistance mutation; INSTI, integrase strand transfer inhibitor; NNRTI, non-nucleoside reverse transcriptase inhibitor; NRTI, nucleoside reverse transcriptase inhibitor; PI, protease inhibitor.

Nucleotide Ambiguities and Discordance

Pairwise analysis of NGS and Sanger sequences were highly concordant: 98.72% of positions were identical, 1.18% were partially discordant, and 0.10% were completely discordant. The median percentage of IUPAC ambiguities was significantly higher in the NGS compared with Sanger sequences: 1.6 (interquartile range: 0.5 to 2.9) versus 0.8 (0.2 to 1.4); P < 0.001 (paired t-test).14, 15 Furthermore, investigation of nucleotide mismatches revealed NGS had proportionately higher levels of G-to-A and C-to-T substitutions (Figure 1).

Figure 1.

Proportions of nucleotide mismatches between next-generation sequencing (NGS) and Sanger. Total number of nucleotide mismatches were counted and weighted by the abundance of A, T, G, and C in the HIV-1 subtype B sequence (NCBI accession number K03455). The size of each bubble represents the weighted percentage of the mismatch type out of all mismatches. IUPAC nomenclature was used for mixed bases: M (AC), R (AG), S (CG), W (AT), Y (CT).

Comparison of Amino Acid Mutations

Among the 261 PR/RT sequences, a mean of 9.5 mutations, defined as differences from the subtype B amino acid consensus sequence, were detected by both Sanger and NGS (Table 2). NGS identified a mean of 1.7 mutations not detected by Sanger, whereas Sanger identified a mean of 0.4 mutations not detected by NGS (P < 0.001; paired Student's t-test). Among the 50 IN sequences, a mean of 10.7 mutations were identified by both Sanger and NGS (Table 2). NGS identified a mean of 2.0 mutations not detected by Sanger, whereas Sanger identified a mean of 0.5 mutations not detected by NGS (P < 0.001; paired t-test).

Table 2.

Comparison of Amino Acid and DRMs Detected by Sanger and NGS

Mutations		PR/RT	IN
AA mutations	Shared mean	9.45	10.70
	Unique Sanger mean	0.36	0.46
	Unique NGS mean	1.72	1.96
	P value	<0.001	<0.001
DRMs	Shared mean	0.37	0.30
	Unique Sanger mean	0.03	0.00
	Unique NGS mean	0.08	0.08
	P value	0.007	0.04

AA, amino acid; DRM, drug resistance mutation; IN, integrase; NGS, next-generation sequencing; PR/RT, protease/reverse transcriptase.

Comparison of DRMs

Among the 261 PR/RT sequences, a mean of 0.37 DRMs were identified by both Sanger and NGS (Table 2). NGS identified a mean of 0.08 mutations not detected by Sanger, whereas Sanger detected a mean of 0.03 mutations not detected by NGS (P = 0.006; paired t-test). Among the 50 IN sequences, a mean of 0.3 DRMs were identified by both Sanger and NGS (Table 2). NGS identified a mean of 0.08 DRMs not detected by Sanger, whereas Sanger identified a mean of 0.0 DRMs not detected by NGS (P = 0.04; paired t-test).

Comparison of the Predicted Levels of Resistance

On average, there was no significant difference between Sanger and NGS in the predicted levels of resistance to any of the NRTIs, NNRTIs, PIs, and INSTIs (Table 3). Analysis of agreement in this cohort resulted in a Kappa score of 0.907 (95% CI, 0.883–0.932) and a weighted Kappa of 0.931. The strength of this agreement is considered very good.

Table 3.

Analysis of the Agreement between Predicted Levels of Resistance

Sanger	NGS
	1	2	3	4	5
1	2446	8	8	3	7
2	7	68	1	2	0
3	5	0	69	1	1
4	0	0	1	43	3
5	0	0	1	3	83

NGS, next-generation sequencing.

For the 261 PR/RT samples, NGS detected a higher level of NRTI resistance for 3.1% (n = 8) of samples, a higher level of NNRTI resistance for 2.3% (n = 6) of samples, and a higher level of PI resistance for 1.5% (n = 4) of samples compared with Sanger sequencing (Tables 4, 5, and 6). In By contrast, Sanger detected a higher level levels of NRTI, NNRTI, and PI resistance for 1.9% (n = 5), 1.1% (n = 3), and 0.7% (n = 2) of samples, respectively. Overall, 17 (6.5%) individuals with PR/RT sequences had a higher level of predicted resistance to one or more RTIs or PIs based on NGS. Ten (3.8%) had a higher level of predicted resistance based on Sanger sequencing. For the 50 IN samples, NGS detected a higher level of INSTI resistance for 4.0% (n = 2) of the samples (Table 7).

Table 4.

PI Resistance Mutations and Estimated PI Resistance Levels

SID	Sanger and NGS	Sanger alone	NGS alone	Sanger score → NGS score
				ATV/r	DRV/r	LPV/r
Samples for which NGS reported higher levels of drug resistance for one or more PIs
34_PR,RT			V32VI (58.5%), M46MI (56.5%), I47IV (57.0%), V82VA (54.5%), K43KT (57.4%)	1 → 5	1 → 4	1 → 5
177_PR,RT			F53FL (6.7%)	1 → 2	1	1
275_PR,RT			V32I (99.5%)	4 → 5	2 → 4	5
296_PR,RT			I50IV (5.1%)	1	1 → 3	1 → 4
Samples for which Sanger reported higher levels of drug resistance for one or more PIs
44_PR,RT	V82VA	I84IV,Q58QE	I54IV (56.4%)	5 → 4	3 → 1	5 → 4
84_PR,RT		M46MI		2 → 1	1	2 → 1
Samples for which NGS detected different DRMs, but for which levels of PI resistance were the same
20_PR,RT			L10LS (5.5%)	1	1	1
23_PR,RT			I50IN (4.9%)	1	1	1
72_PR,RT			I50IN (5.4%)	1	1	1
73_PR,RT			I50IN (5.8%)	1	1	1
78_PR,RT			G48GR (11.5%)	1	1	1
105_PR,RT			I50IN (4.4%)	1	1	1
122_PR,RT			K20KMT (40.9%; 26.0%)	1	1	1
193_PR,RT			L33LV (5.5%)	1	1	1
256_PR,RT			L10LS (4.3%)	1	1	1

Predicted levels of drug resistance according to the HIVdb genotypic resistance interpretation system. Susceptible (1); potential low-level resistance (2); low-level resistance (3); intermediate resistance (4); high-level resistance (5). Percent mixtures are shown in parentheses. All samples listed in this table are subtype B.

ATV/r, atazanavir/ritonavir; DRM, drug resistance mutation; DRV/r, darunavir/ritonavir; LPV/r, lopinavir/ritonavir; NGS, next-generation sequencing; PI, protease inhibitor; PR, protease; RT, reverse transcriptase; SID, sample ID.

Table 5.

NRTI Resistance Mutations and Estimated NRTI Resistance Levels

SID	Sanger and NGS	Sanger alone	NGS alone	Sanger score → NGS score
				FTC/3TC	ABC	AZT	TDF
Samples for which NGS reported higher levels of drug resistance for one or more NRTIs
30_PR,RT			K65KE (10.5%)	1	1 → 2	1	1 → 2
34_PR,RT			M41L (99.2%), D67N (90.9%), M184MV (50.1%), L210W (96.9%), T215Y (96.2%)	1 → 5	1 → 5	1 → 5	1 → 4
104_PR,RT			D67DN (7.8%)	1	1	1 → 3	1
111_PR,RT			T215TS (5.6%)	1	1	1 → 3	1
154_PR,RT	M41ML		M184MV (6.8%)	1 → 5	1 → 3	3 → 1	1
183_PR,RT			K65KE (4.4%)	1	1 → 2	1	1 → 2
184_PR,RT			L210LW (6.8%)	1	1	1 → 3	1
254_PR,RT			D67DN (5.4%)	1	1	1 → 3	1
Samples for which Sanger reported higher levels of drug resistance for one or more NRTIs
31_PR,RT		K219KR		1	1	2 → 1	1
50_PR,RT		T215TA		1	1	3 → 1	1
88_PR,RT		T215TILP		1	1	3 → 1	1
121_PR,RT		T215TADN		1	1	3 → 1	1
154_PR,RT	M41ML		M184MV (6.8%)	1 → 5	1 → 3	3 → 1	1
Samples for which Sanger or NGS detected different DRMs, but for which levels of NRTI resistance were the same
83_PR,RT	M41ML	T69D	T69DN (86.1%; 10.6%)	1	2	4	2
103_PR,RT	M184V		D67DN (5.0%)	5	3	1	1
105_PR,RT			A62AV (5.9%)	1	1	1	1
123_PR,RT	M41L,T215E		K65KE (3.4%)	1	3	4	3
157_PR,RT		M184IV	M184I (99.3%)	5	3	2	1
214_PR,RT	M184I	K70Q	K70NQ (27.2%; 70.1%)	5	4	1	3
281_PR,RT		M184V	M184IV (4.3%; 91.4%)	5	3	1	1

Predicted levels of drug resistance according to the HIVdb genotypic resistance interpretation system: susceptible, 1; potential low-level resistance, 2; low-level resistance, 3; intermediate resistance, 4; high-level resistance, 5. Percent mixtures are shown in parentheses. All samples listed in this table are subtype B.

ABC, abacavir; AZT, azidothymidine; DRM, drug resistance mutation; FTC/3TC, emtricitabine/lamivudine; NGS, next-generation sequencing; NRTI, nucleoside reverse transcriptase inhibitor; PR, protease; RT, reverse transcriptase; SID, sample ID; TDF, tenofovir.

Table 6.

NNRTI Resistance Mutations and Estimated NNRTI Resistance Levels

SID	Sanger and NGS	Sanger alone	NGS alone	Sanger score → NGS score
				EFV	ETR	RPV
Samples for which NGS reported higher levels of drug resistance for one or more NNRTIs
23_PR,RT	V108VI,V179VD,Y188HL		K103KE (5.4%), G190GE (4.7%)	5	3 → 5	5
39_PR,RT	Y181YC		K103KN (20.8%)	4 → 5	4	4
92_PR,RT			K103KN (13.6%)	1 → 5	1	1
166_PR,RT			V106IV (8.3%)	1	1 → 2	1 → 2
227_PR,RT	V179VAD		K103KR (14.6%)	2 → 4	2	2 → 3
299_PR,RT	V179D,Y188L		V106VI (6.0%)	5	3 → 4	5
Samples for which Sanger reported higher levels of drug resistance for one or more NNRTIs
50_PR,RT		V106IV		1	2 → 1	2 → 1
93_PR,RT	K101E	G190GA		5 → 3	4 → 3	5 → 4
269_PR,RT		V106IV		1	2 → 1	2 → 1
Samples for which Sanger or NGS detected different DRMs, but for which levels of NNRTI resistance were the same
18_PR,RT			K103KE (5.2%)	1	1	1
24_PR,RT			K103KE (6.3%)	1	1	1
25_PR,RT	Y188L		K103KE (8.4%)	5	2	5
27_PR,RT	V179D		K103KE (5.8%)	2	2	2
30_PR,RT			K103KE (6.6%)	1	1	1
46_PR,RT			K103KQ (5.1%)	1	1	1
57_PR,RT			K103KE (5.1%)	1	1	1
74_PR,RT			K103KE (4.7%)	1	1	1
103_PR,RT	A98AG,K101KE,E138Q,Y181YC,H221HY		V106VI (20.3%)	5	5	5
157_PR,RT	K103N	E138AT	E138A (99.6%)	5	4	5
161_PR,RT			K103KE (5.4%)	1	1	1
163_PR,RT			K103KE (5.5%)	1	1	1
190_PR,RT			K103KE (4.5%)	1	1	1
191_PR,RT			K103KE (4.6%)	1	1	1
192_PR,RT		V179E	K103KE (22.9%),V179DE (6.4%; 91.3%)	2	2	2
215_PR,RT	A98G		K238KT (5.8%)	5	2	3
221_PR,RT			K103KE (4.2%)	1	1	1
222_PR,RT∗	K103N		V108VI (10.0%),Y318YF (60.6%)	5	1	1
263_PR,RT∗	K101E,G190A		V179IT (92.9%; 5.7%)	5	4	5
273_PR,RT	K103KN		Y318YF (72.3%)	5	1	1

Predicted levels of drug resistance according to the HIVdb genotypic resistance interpretation system: susceptible (1); potential low-level resistance (2); low-level resistance (3); intermediate resistance (4); high-level resistance (5). Percent mixtures are shown in parentheses.

DRM, drug resistance mutation; EFV, efavirenz; ETR, etravirine; NGS, next-generation sequencing; NNRTI, non-nucleoside reverse transcriptase inhibitor; PR, protease; RPV, rilpivirine; RT, reverse transcriptase; SID, sample ID.

^∗Sample 222 is subtype C, and sample 263 is subtype CRF01_AE. All other samples listed in this table are subtype B.

Table 7.

INSTI Resistance Mutations and Estimated INSTI Resistance Levels

SID	Sanger and NGS	Sanger alone	NGS alone	Sanger score → NGS score
				DTG	EVG	RAL
Samples for which NGS reported higher levels of drug resistance for one or more INSTIs
103_IN			E138EK (5.5%)	1 → 2	1 → 3	1 → 3
281_IN	T66I, E92Q		G163GR (5.6%)	3	5	4 → 5
Samples for which Sanger or NGS detected different DRMs, but for which levels of INSTI resistance were the same
38_IN	G140C, Q148R		G163GR (6.5%)	4	5	5

Predicted levels of drug resistance according to the HIVdb genotypic resistance interpretation system: susceptible (1); potential low-level resistance (2); low-level resistance (3); intermediate resistance (4); high-level resistance (5). Percent mixtures are shown in parentheses. All samples listed in this table are subtype B.

DRM, drug resistance mutation; DTG, dolutegravir; EVG, elvitegravir; INSTI, integrase strand transfer inhibitor; NGS, next-generation sequencing; PR, protease; RAL, raltegravir; RT, reverse transcriptase; SID, sample ID.

For one sample, NGS demonstrated both higher and lower resistance to different drugs in a particular class compared to with Sanger. Namely, NGS showed a higher level of resistance to the NRTIs FTC/3TC and ABC in sample 154_PR,RT but identified lower level resistance to AZT.

Virus loads were not significantly different for those samples in which NGS and Sanger predicted concordant levels of resistance (mean = 4.4 log₁₀ copies/mL, SD = 1.11) compared to with those samples where the predicted levels of resistance were discordant (mean = 4.1 log₁₀ copies/mL, SD = 1.12; P = 0.3). Similarly, discordant predicted levels of resistance were not associated with non-B subtypes: 4.83% (13/269) of all samples were a non-B subtype, whereas 0.0% (0/27) of discordant samples were a non-B subtype.

Analysis of APOBEC and Highly Unusual Mutations

There was a significant difference in the mean number of signature APOBEC mutations per sequence between Sanger and NGS in PR/RT (0.02 for Sanger versus 0.07 for NGS; P = 0.006, paired t-test), but not in IN (0.08 for Sanger versus 0.10 for NGS; P = 0.32). One PR/RT NGS sequence had two signature APOBEC mutations, including a stop codon. By contrast, no Sanger sequence had more than one signature APOBEC mutation.

The mean number of highly unusual mutations per prospective sequence was also higher in NGS compared with Sanger sequences (PR/RT 0.48 versus 0.13; P < 0.001 :: IN 0.42 versus 0.12; P = 0.003, paired t-test). Among the 261 prospective PR/RT sequences, the highest number of highly unusual mutations was five for NGS. For Sanger, the highest number of highly unusual mutations was three.

Discussion

This study evaluated the clinical performance characteristics of the Sentosa SQ HIV-1 Genotyping Assay in a prospective cohort of HIV-1–infected patients undergoing HIV-1 genotypic evaluation at a US academic medical center. Overall, NGS was highly concordant with Sanger sequencing at the level of resistance interpretation, as has been previously demonstrated in the retrospective analysis of smaller numbers of highly selected samples.9, 16, 17 Herein, more samples were tested than in the previous three studies combined, and testing was performed prospectively to assess QC metrics in a real-world, clinical laboratory setting.

During the course of this study, 44% of the initial runs failed run QC, requiring 17 additional runs. Furthermore, within successful NGS runs, no PR/RT or IN sequence was obtained in 24 samples, and partial sequences were obtained in 5 samples, all of which were amplified, fully sequenced, and a drug resistance report generated via the conventional Sanger pipeline. Given that the Sentosa system automates a majority of the myriad steps involved in sample preparation, library generation, and sequencing, it is highly unlikely that these large numbers of run and sample failures were due to operator errors, particularly given the operators' significant experience performing high complexity molecular diagnostic testing, including other NGS technologies. Although there were both amplification and sequencing failures via the conventional Sanger approach, it was straightforward to identify the step at which the failure occurred. This was not the case with the Sentosa system, though a camera has been added to newer versions of the liquid handler to allow targeted troubleshooting of the pre-analytical process. Other changes were made in version 2.0 of the assay to address some of the QC issues observed in this study. For example, the system control, which in version 1.0 functioned as both a positive and negative control, now functions only as a negative control, and a separate positive control is required. Furthermore, the median coverage for the system control was reduced (from ≥200 to ≥50 reads), and the number of allowable HIV reads was increased (from ≤50 to ≤833 reads). Regarding the sequencing failures, the template preparation chemistry has changed from traditional emulsion PCR to isothermal amplification, and the sequencing reagent mixes and chip loading process is now fully automated. Nevertheless, future studies will be required to ensure that these improvements result in system performance that meets the robust demands of clinical diagnostic laboratories and high-throughput public health laboratories.

Importantly, however, in samples sequenced by both methods, the resistance interpretations obtained from the Sentosa SQ HIV-1 Genotyping Assay were highly concordant with those from Sanger sequencing. Though a number of studies have demonstrated that detection of low-level DRM are associated with virologic failure,3, 4, 5, 6, 7 there remains significant debate regarding the appropriate threshold for mutation calling and concern that overcalling low-level variants might prevent the use of effective antiretroviral regimens. The results presented here suggest that transition from Sanger to the Sentosa in clinical laboratories will not be disruptive from the standpoint of mutation detection and interpretation. Overall, 6.5% (17/261) of individuals with PR/RT sequences had a higher level of predicted resistance to one or more RTIs or PIs based on NGS, whereas 3.8% (10/261) had a higher level of predicted resistance based on Sanger.

Of interest, a single patient (sample 34) accounted for 34.5% (10/29) of the DRMs present in specimens in which NGS predicted a higher level of resistance. Further investigation revealed that the cohort included a second sample (275) from this patient who had returned for testing approximately 5 months later, but within the enrollment period of the study. In the first sample, NGS called high-level resistance to ATV, LPV, FTC/3TC, ABC, and AZT, and intermediate resistance to DRV and TDF. At follow-up, both Sanger and NGS identified resistance to this same set of antiretrovirals (Tables 4 and 5). These results indicate that the Sentosa HIV-1 Genotyping Assay is capable of identifying clinically relevant DRMs earlier than Sanger sequencing. Future studies with longitudinal samples from patients on ART will be required to provide additional evidence for early Sentosa DRM detection. Note that the percent variants detected by NGS only in samples 34 and 275 were >20%. This finding may be due to the relatively low virus load in these samples, 2.38 and 3.05 log₁₀ copies/mL, respectively.

Limitations of this study include the predominance of subtype B viruses, though this reflects the distribution of subtypes in the United States.¹⁸ Future studies in Africa or Asia or targeting specific populations in the United States or Europe will be necessary to confirm performance with non-subtype B viruses.¹⁹ In addition, the ability of the Sentosa SQ HIV-1 Genotyping Assay to sequence samples that were not amplifiable by in-house RT-PCR was not investigated. This question may need to be addressed in future studies, particularly as the manufacturer incorporates updated and improved primer sets. Finally, correlation of Sanger and NGS sequencing data with ART regimens was beyond the scope of this work and may be better addressed using samples from a laboratory with extensive local patient testing, or using archived specimens from therapeutic trials, such as from the AIDS Clinical Trial Group.

In conclusion, this study investigated the Sentosa SQ HIV-1 Genotyping Assay in a prospective cohort of HIV-1–infected patients undergoing HIV-1 genotypic evaluation at a US academic medical center. This assay provided highly concordant resistance interpretation compared with conventional Sanger sequencing in those samples in which data were available for both methodologies, though numerous run and sample QC issues were observed when the Sentosa platform was used in this clinical laboratory setting. However, with additional refinement, improvements in automation, and overall reduction in assay complexity, the Sentosa SQ HIV-1 Genotyping Assay may provide expanded access to HIV-1 genotyping to the 36.9 million individuals worldwide currently living with HIV,²⁰ including 1.1 million individuals in the United States (CDC, https://www.cdc.gov/hiv/pdf/statistics/overview/cdc-hiv-us-ataglance.pdf, accessed April 30, 2019). The availability of other semiautomated sequencing and virus load assays on the Sentosa platform, coupled with elimination of cold-chain dependence, provision of an independent power source, and cost sensitivity, may further facilitate access to HIV-1 genotyping in low- and middle-income countries,2, 21 and contribute to the global efforts to control HIV.