Diagnostic Accuracy of Pan-Degenerate Amplification and Adaptation Assay for HIV-1 Drug Resistance Mutation Analysis in Low- and Middle-Income Countries

HIV drug resistance (HIVDR) is a barrier to sustained virologic suppression in low- and middle-income countries (LMICs). Point mutation assays targeting priority drug resistance mutations (DRMs) are being evaluated to improve access to HIVDR testing. In a cross-sectional study (June 2018 to September 2019), we evaluated the diagnostic accuracy of a simple and rapid HIVDR assay (the pan-degenerate amplification and adaptation [PANDAA] assay targeting the mutations K65R, K103NS, M184VI, Y181C, and G190A) compared to Sanger sequencing and next-generation sequencing (NGS).

ABSTRACT

HIV drug resistance (HIVDR) is a barrier to sustained virologic suppression in low- and middle-income countries (LMICs). Point mutation assays targeting priority drug resistance mutations (DRMs) are being evaluated to improve access to HIVDR testing. In a cross-sectional study (June 2018 to September 2019), we evaluated the diagnostic accuracy of a simple and rapid HIVDR assay (the pan-degenerate amplification and adaptation [PANDAA] assay targeting the mutations K65R, K103NS, M184VI, Y181C, and G190A) compared to Sanger sequencing and next-generation sequencing (NGS). Plasma samples from adolescents and young adults (aged 10 to 24 years) failing antiretroviral therapy (viral load, >1,000 copies/ml on 2 consecutive occasions 1 month apart) were analyzed. Sensitivity and specificity of the PANDAA assay were determined by a proprietary application designed by Aldatu Biosciences. Agreement between genotyping methods was evaluated using Cohen’s kappa coefficient. One hundred fifty samples previously characterized by Sanger sequencing were evaluated using PANDAA. For all DRMs detected, PANDAA showed a sensitivity and specificity of 98% and 94%, respectively. For nucleotide reverse transcriptase inhibitor DRMs, sensitivity and specificity were 98% (95% confidence interval [CI], 92% to 100%) and 100% (94% to 100%), respectively. For non-nucleotide reverse transcriptase inhibitor DRMs, sensitivity and specificity were 100% (97% to 100%) and 76% (61% to 87%), respectively. PANDAA showed strong agreement with Sanger sequencing for K65R, K103NS, M184VI, and G190A (kappa > 0.85) and substantial agreement for Y181C (kappa = 0.720). Of the 21 false-positive samples genotyped by PANDAA, only 6 (29%) were identified as low-abundance variants by NGS. With the high sensitivity and specificity to detect major DRMs, PANDAA could represent a simple and rapid alternative HIVDR assay in LMICs.

INTRODUCTION

The increasing prevalence of human immunodeficiency virus type 1 (HIV-1) drug resistance (HIVDR) remains a major threat to HIV-1 treatment and prevention globally, particularly in many low-and middle-income countries (LMICs) impacted by the HIV-1 pandemic (1, 2). High rates of acquired drug resistance to nucleotide reverse transcriptase inhibitors (NRTIs) and non-NRTIs (NNRTIs) among HIV-infected individuals failing first-line antiretroviral therapy (ART) have been reported in many LMICs (1, 3, 4). We previously reported a high level of acquired drug resistance mutations (86%) among young people failing ART in Zimbabwe (5). Similarly, low-abundance variants (viral populations that occur at less than 20% of the total population) particularly involving NNRTI resistance have been shown to adversely affect subsequent ART (6, 7). Thus, HIVDR monitoring remains critical to achieving the third of the 90-90-90 UNAIDS targets (8–10) for maximal viral load (VL) suppression and elimination of AIDS by 2030.

Standard HIV genotypic resistance testing (GRT) using Sanger sequencing to guide the selection of initial and subsequent ART is highly recommended (11–13). This has been proven to be cost-effective in high-income countries (14) but practically limited and not feasible in many LMICs, as it remains complex and costly (15, 16). In Zimbabwe, although a validation of an in-house genotyping method at the African Institute for Biomedical Sciences and Technology laboratory showed good-quality HIVDR results, Chimukangara et al. concluded that the implementation of such techniques in Zimbabwe was too costly (17). Additionally, Phillips et al. showed that the use of GRT at the time of first-line-ART failure as part of the decision whether to switch to second-line therapy was not cost-effective in LMICs (16). Meanwhile, in Zimbabwe, drug resistance surveillance (by Sanger sequencing) for patient management costs $300 to $600 per sample in the private sector. The World Health Organization (WHO) has prioritized increasing laboratory capacity and access to HIVDR testing in LMICs (18, 19). An affordable HIVDR monitoring method (Southern African Treatment and Resistance Network [SATuRN]/Life Technologies) for LMICs has been implemented in Zimbabwe (20). However, the amplicons generated are shipped to commercial laboratories (Molecular Cloning Laboratory, USA, and Inqaba, South Africa) for GRT. This has contributed to delayed turnaround time (TAT) of results, and consequently, switches to more expensive and unnecessary ART may occur or individuals may be switched to suboptimal treatment.

The 2017 and most recently the 2019 WHO HIV Resistance Network (HIVRESNET) annual meeting advocated for the implementation of HIVDR point mutation assays (PMAs) for testing in LMICs (10). Point mutation assays (oligonucleotide ligation, allele-specific primer extension, multiplex melting curve analysis, and pan-degenerate amplification and adaptation assays) are being developed and validated, and some have been implemented to improve access to HIVDR testing. Increased sensitivity for low-abundance drug-resistant variants, lower cost, simpler procedures, fewer equipment requirements, and faster TAT provide potential advantages compared to standard Sanger sequencing (21). Thus, PMAs may be suitable and convenient assays for individual-level HIVDR testing and clinical management in many LMICs.

Pan-degenerate amplification and adaptation (PANDAA) is a genotyping technology designed to provide inexpensive and high-throughput focused HIVDR testing in LMICs. The PANDAA assay is a focused genotyping that identifies specific mutations affecting susceptibility to NRTIs and NNRTIs (K65R, M184VI, K103NS, Y181C, and G190A) which are present in >98% of patients who fail an NNRTI first-line regimen (22). The assay can quantify key HIV DRMs present at ≥ 5% with a diagnostic sensitivity and specificity of 96.9% and 97.5%, respectively (23). PANDAA is based on quantitative real-time PCR (qPCR) technology, a well-established gold standard technique for low-cost and sensitive genotyping analysis. To date, there are no data evaluating the PANDAA assay for detecting HIV-1 DRMs in LMICs. Therefore, we assessed the diagnostic accuracy of PANDAA for detecting major DRMs among adolescents and young adults failing ART in Zimbabwe.

MATERIALS AND METHODS

Study design.

In a cross-sectional study conducted between June 2018 and September 2019, we evaluated the sensitivity and specificity of the PANDAA assay in detecting major DRMs (≥20%) compared to standard GRT by Sanger sequencing and in detecting low-level (<20%) major DRMs compared to next-generation sequencing (NGS). We used plasma samples collected during a randomized clinical trial at the Parirenyatwa Hospital HIV ART treatment clinic (OI clinic) in Harare, Zimbabwe, as previously described (5).

Study population, settings, and procedures.

Patients eligible for inclusion were consenting HIV-1-infected adolescents and young adults aged 10 to 24 years, with confirmed virological failure (VL > 400 copies/ml on 2 consecutive occasions 1 month apart). We recruited participants at the Parirenyatwa Hospital HIV ART treatment clinic, a tertiary-level setting, in Harare, Zimbabwe. Participants were receiving either first-line ART (2 NRTIs plus 1 NNRTI) or second-line ART (2 NRTIs plus a protease inhibitor) and had been on ART treatment for at least 6 months. Written informed consent was obtained from eligible participants aged 18 to 24 years. Assent, as well as informed consent from their legal guardians, was obtained from children and adolescents aged 10 to 17 years. Age, gender, and ART history data (treatment initiation date, treatment regimens, and ART duration) were extracted from the medical records.

Genotyping by Sanger sequencing (reference assay with a cutoff of ≥20%).

Whole blood was collected in an EDTA tube, and plasma was harvested after centrifugation at 1,000 × g for 10 min. The harvested plasma was then stored at −80°C in the Infectious Diseases Research Laboratory (IDRL), College of Health Sciences, University of Zimbabwe. In preparation for HIV pol gene genotyping, HIV-1 viral RNA was isolated from 200 μl of the stored plasma samples using a column-based extraction kit, the PureLink mini-viral RNA/DNA minikit (Thermo Fisher Scientific, Carlsbad, CA, USA) in accordance with the manufacturer’s instructions. The RNA was eluted in 30 μl of RNase-free water and stored at −80°C when not used for reverse transcription-PCR (RT-PCR) immediately. For the amplification, the Southern African Treatment Resistance Network (SATuRN) protocol was used as previously described (20). Briefly, this is a 2-step RT-PCR protocol, followed by nested PCR, which generates an amplicon of 1,197 bp covering all 99 HIV-1 protease codons and the first 300 codons of the reverse transcriptase (RT) of the HIV-1 pol gene. All amplicons were sequenced using commercial Sanger sequencing services accessed at Molecular Cloning Laboratories, San Francisco, CA. The chromatograms generated were assembled using Geneious software, version 8 (24), and HIV DRMs were determined using the online Stanford HIVDB program (25). The remaining HIV-1 RNA and PCR amplicons generated for Sanger sequencing were batched and stored at −80°C and −20°C, respectively, for 7 to 12 months prior to genotyping by NGS and PANDAA, respectively.

Genotyping by PANDAA (index test with a cutoff of ≥5%).

(i) Overview of the PANDAA assay validation. The PANDAA assay is an allelic discrimination test designed to mitigate the negative impact of DRM-proximal sequence variability on qPCR performance by using extremely degenerate primers that overlap with the probe-binding site (23). The assay was designed with allele 1 representing the wild-type codon and allele 2 the mutant codon. PANDAA was validated using two sets of five synthetic DNA templates incorporating probe-binding site alleles for RT codons 65, 103, 181, 184, and 190. One set contained the wild-type codon (integrated set A, templates 001 to 005), and the second contained the DRM-conferring nucleotide substitutions for K65R, K103N, Y181C, M184VI, and G190A (integrated set B, templates 001 to 005). The PANDAA validation was performed with differentially labeled TaqMan probes to discriminate wild-type DNA (VIC labeled [green], allele 1) from the DRM types (6-carboxyfluorescein [FAM] labeled [red], allele 2) for each drug resistance codon. Thus, the PANDAA selectivity (i.e., the detectable proportion of DRM on a wild-type background) was assessed using mixed ratios of integrated set A (wild-type) and B (DRM) DNA templates 001 to 005 down to a DRM proportion of 1% (23). Four sets of controls with different proportions of wild-type and DRM sequences were provided with the kit used in this study. These controls served as quality controls for the PANDAA reagents and were also used to verify the performance of the real-time thermocycler, as they provided a baseline data set for preliminary analysis. The controls included 2 sets of 1a/2a (50%/50%) containing 50% DRM with the resistance profiles K65R, K103N, Y181C, M184V/I, and G190A/S at 1.0 × 10⁵ copies/μl and 1.0 × 10⁴ copies/μl; 1 set of 1b/2b (80%/20%) containing 20% DRM with the resistance profiles K65R, K103N, Y181C, M184V/I, and G190A/S) at 1.0 × 10⁵ copies/μl; and finally 1 set of 1b (100%) containing 0% DRM or a wild type at 1.0 × 10⁵ copies/μl.

(ii) Reaction set up in this study. One microliter of the stored PCR amplicons (10 to 20 ng/μl) was added to 249 μl Tris-EDTA buffer to generate a master stock dilution. This was diluted twice in a 1/50 dilution each time with nuclease-free water to make a working stock amplicon containing approximately 2.5 × 10⁴ copies/μl prior to PANDAA genotyping. The working stock amplicons were added to the qPCR master mix, which included forward and reverse PANDAA primers as well as probes (VIC-labeled wild-type and FAM-labeled DRM-specific probes). The codons corresponding to K65R, K103N, Y181C, M184V/I, and G190A/S were tested individually. Thus, for every sample, five different runs were carried out to get the full DRM profile. In each run, the samples were tested in triplicate, to assess reproducibility (replicate group cycle threshold [C_T] standard deviation < 0.2). The final reaction volume of 10 μl was then added to a 96-well plate containing the samples, at least 2 sets of controls and the no-template control (NTC), which was nuclease-free water used for sample amplicon dilution. The reaction was done under the following PCR conditions: 95°C for 3 min for an initial reaction incubation, followed by 10 three-step adaptation cycles of 95°C for 3 s, 50°C for 60 s, and 60°C for 30 s, and finally 35 two-step amplification cycles of 95°C for 3 s and 60°C for 60 s, during which fluorescence data were captured. A CFX96 thermocycler (Bio-Rad Laboratories, Inc., CA, USA) was calibrated for the two distinct fluorophores used in PANDAA, FAM (excitation maximum = 494 nm and emission maximum = 518 nm) and VIC (excitation maximum = 538 nm and emission maximum = 552 nm).

All data generated by the CFX96 PCR thermocycler were analyzed using a proprietary application designed by Aldatu Biosciences. The relative abundance of the mutant codons versus the wild type was calculated based on the C_T values of both. The PANDAA assay was performed at the IDRL, University of Zimbabwe.

Genotyping by next-generation sequencing (low-abundance variants’ reference assay with a cutoff of ≥2%).

Paired-end libraries were generated using the Illumina Nextera XT DNA library prep index kit (26) with 96 indices. Following the manufacturer’s protocol, the paired-end libraries generated were amplified and purified using AMPure XP beads. Sequencing was performed on the Illumina MiSeq using the MiSeq reagent, version 2 (500 cycles), at the Kwazulu Natal Research Innovation and Sequencing Platform (KRISP), Durban, South Africa. Upon sequencing, the raw reads (FastQ files) generated were filtered for low-quality reads and de novo assembled into contigs using the online Genome Detective tools (27). Finally, the detection of low-abundance variants was done at ≥2% using Geneious software, version 8 (Biomatters, Ltd., Auckland, New Zealand) (24). Next-generation sequencing can detect alterations that are present at levels as low as 2% to 5% (28).

Analysis.

Samples with no results (overdiluted samples and samples with low viral loads) were considered indeterminate. These indeterminate samples were repeat-tested (with a less diluted sample), and the results were incorporated into the final analysis. Samples not amplified were excluded from the analysis. Results were categorized by each DRM detected (K65R, M184VI, Y181C, K103NS, or G190A), NRTI DRMs, NNRTI DRMs, and overall reverse transcriptase inhibitor (RTI) DRMs. False-positive samples were defined as samples with DRMs detected by the PANDAA assay and not detected by Sanger sequencing (the reference standard test). False-negative samples were samples with DRMs confirmed by Sanger sequencing but not detected by the PANDAA assay.

Sensitivity and specificity were conducted by a proprietary application designed by Aldatu Biosciences and by the diagnostic algorithm in Stata version 14 (StataCorp LP, College Station, TX, USA). Confidence intervals for sensitivity, specificity, and accuracy were “exact” Clopper-Pearson confidence intervals (http://www.real-statistics.com/binomial-and-related-distributions/beta-distribution/). Accuracy was defined as the overall probability that a patient’s sample would be correctly classified. Significance levels were set at a P value of 0.05. Phylogenetic tree and nucleotide ambiguity percentages were constructed with Geneious software to check for any contamination and ensure the quality of sequences generated by the different assays. The presence or absence of DRMs detected by either assay was checked by 2 neutral research scientists (blinded to the original results). All agreement between genotyping methods was determined by Pearson’s correlation coefficient. Cohen’s kappa, which is a robust statistic useful for either interrater or intrarater reliability testing, was used for determining the level of agreement between PANDAA and Sanger sequencing. The strength of the kappa coefficient was interpreted as follows: ≤0, no agreement; 0.01 to 0.20, slight agreement; 0.21 to 0.40, fair agreement; 0.41 to 0.60, moderate agreement; 0.61 to 0.80, substantial agreement; 0.81 to 1.00, strong or almost perfect agreement (29).

Ethics.

The study was reviewed and approved by the local institutional review board of the Joint Research and Ethics Committee of the University of Zimbabwe (JREC/185/15), by the Medical Research Council of Zimbabwe (MRCZ/A/1992), and by the Research Council of Zimbabwe (RCZ/A/1992). Approval and shipment of the samples for NGS testing were obtained from the institutional review board of the Biomedical Research Ethics Administration of the University of Kwazulu Natal, Durban, South Africa (BE 320/19).

Data availability.

Protease and partial reverse transcriptase sequences in this study are available in GenBank under the accession numbers MK583768 to MK583927.

RESULTS

Specimen overview.

Of the 212 participants enrolled in the study, samples from 185 receiving either first-line or second-line ART with VL of >1,000 copies/ml were genotyped initially by Sanger sequencing. Of the 185, 160 (86%) were successfully genotyped, and of the 160 successfully genotyped, 151 (82%) samples with sufficient volume were sequenced by the PANDAA assay (the index test) and NGS assay (Fig. 1).

FIG 1.

Specimen overview. VL, viral load; NGS, next-generation sequencing; index assay, test whose accuracy is evaluated; reference assay, the best available sequencing method for establishing the presence or absence of drug resistance mutations and comparing distribution of the index test results. The PANDAA assay was assessed against Sanger sequencing for drug resistance mutations detected at ≥20% and NGS for low-abundance variants (≥2%).

Baseline demographic and clinical characteristics of patients with successfully genotyped samples.

The PANDAA assay time was approximately 2 h per batch of 24 samples from the amplicons input. The final result could be available within 2 to 5 days from sample collection. Overall, 150/151 samples were included in the data analysis. One sample was excluded as it did not amplify. The median age (interquartile range [IQR]) of the 150 participants successful genotyped was 18 (15 to 19) years. The median (IQR) log₁₀ VL was 4.51 (4.05 to 4.93) copies/ml. Besides lamivudine (3TC) in the participants’ NRTI backbone, 70% (105/150) were on tenofovir disoproxil fumarate (TDF), 21% (32/150) were on zidovudine (AZT), and the remaining 9% (13/150) were on abacavir (ABC) (Table 1).

TABLE 1.

Baseline demographic and clinical characteristics of the 150 participants^a

Characteristic	Value (n = 150)
No. (%) of males	81 (54)
Median (IQR) age at study enrollment, yrs	18 (15–19)
Median (IQR) age at HIV diagnosis, yrs	10 (7–13)
Median (IQR) age at ART initiation, yrs	11 (9–14)
Median (IQR) plasma VL, log10 copies/ml	4.50 (4.05–4.90)
Median (IQR) CD4 count at study enrollment, cells/ml	197 (50–360)
Median (IQR) duration on ART prior to VF, yrs	6 (4–9)
No. (%) on art regimen (n = 150)
TDF	105 (70)
AZT	32 (21)
ABC	13 (9)

^aIQR, interquartile range; ART, antiretroviral therapy; VL, viral load; VF, virologic failure; TDF, tenofovir disoproxil fumarate; AZT, zidovudine; ABC, abacavir.

Estimates of sensitivity and specificity of the PANDAA assay and its precision (95% CI) in detecting DRMs.

The detection of DRMs or wild-type viruses by the PANDAA assay was measured and used to calculate its sensitivity and specificity compared to the gold standard, Sanger sequencing. Overall, for all DRMs detected, the PANDAA assay showed a sensitivity (95% confidence interval [CI]) and a specificity (95% CI) of 98% (95% to 99%) and 94% (91% to 96%), respectively. Sensitivity and specificity were also reported by drug class. For NRTI DRMs, sensitivity (95% CI) and specificity (95% CI) were reported as 98% (92% to 100%) and 100% (94% to 100%), respectively, with an accuracy of 99%. For NNRTI DRMs, the sensitivity (95% CI) and specificity (95% CI) were reported as 100% (97% to 100%) and 76% (61% to 87%), respectively, both with an accuracy of 93%. Details of sensitivity and specificity in detecting individual DRMs can be found in Table 2.

TABLE 2.

Sensitivity, specificity, and agreement of the PANDAA assay versus Sanger sequencing^a

Mutation	TP	TN	FP	FN	% Sensitivity (95% CI)	% Specificity (95% CI)	Kappa value (95% CI)
K65R	38	109	1	2	95 (83–99)	99 (95–100)	0.95 (0.89–1.00)
M184V/I	80	69	0	1	99 (93–100)	100 (95–100)	0.99 (0.96–1.01)
NRTI DRMs	87	61	0	2	98 (92–100)	100 (94–100)	0.90 (0.83–0.97)
K103NS	54	93	2	1	98 (90–100)	98 (93–100)	0.96 (0.91–1.01)
Y181C	41	90	18	1	98 (87–100)	83 (75–90)	0.72 (0.60–0.84)
G190A	47	93	9	1	98 (89–100)	91 (84–96)	0.85 (0.77–0.94)
NNRTI DRMs	105	34	11	0	100 (97–100)	76 (61–87)	0.85 (0.75–0.95)

^aTP, true positive; TN, true negative; FP, false positive; FN, false negative; CI, confidence interval; NRTI DRMs, mutations resulting in resistance to nucleotide reverse transcriptase inhibitors; NNRTI DRMs, mutations resulting in resistance to non-NRTIs. The row “NRTI DRMs” presents the frequency of NRTI mutations detected either individually or together (K65R and/or M184V). Similarly, the row “NNRTI DRMs” presents the frequency of NNRTI mutations detected either individually or together (K103N and/or Y181C and/or G190A).

Drug resistance mutations detected by Sanger sequencing.

For the 150 individuals whose samples were successfully genotyped by Sanger sequencing, additional major NRTI and NNRTI DRMs besides those assessed by PANDAA (K65R, M184V, K103N, Y181C, and G190A) using the online Stanford HIVDB program were present in 60% of sequences. Mutations affecting susceptibility to NRTIs were present in 34% (51/150), as follows: Y115F (which synergistically reduces TDF susceptibility only in combination with K65R) was found in 8%; L74VI (which reduces ABC susceptibility) was found in 5%; thymidine analog mutations (which reduce susceptibility to AZT) were found in 24%; and finally, K70E (which reduces TDF susceptibility) was found in only 2%. Mutations affecting susceptibility to NNRTIs were found in 48% (72/150), as follows: L100I in 5%, K101EP in 20%, V106AM in 29%, Y188LCH in 4%, and M230L in 3%. All these mutations decrease susceptibility to efavirenz (EFV) and nevirapine (NVP).

Levels of agreement between PANDAA and Sanger sequencing.

For all DRMs detected, the PANDAA assay showed a strong agreement with the reference assay, i.e., Sanger sequencing (kappa = 0.90; 95% CI, 0.86 to 0.93). The PANDAA assay also showed a strong agreement with Sanger sequencing for the individual DRMs K65R (kappa = 0.95), M184VI (kappa = 0.99), K103NS (kappa = 0.96), and G190A (kappa = 0.85) and a substantial agreement for Y181C (kappa = 0.72) (Table 2).

Evaluation of the PANDAA assay in detecting low-abundance DRMs.

Discrepancies between the PANDAA assay and Sanger sequencing were reported for 23% (34/150) of the samples. Of these 34 samples, 30 were false positives (FP), mostly those with the Y181C mutation (FP = 18), followed by G190A (FP = 9), K103NS (FP = 2), and K65R (FP = 1). False-negative (FN) results were obtained with only 4 samples. To verify that the FP and FN results recorded for the PANDAA assay were not based on data interpretation, both the original Sanger sequencing chromatograms and the NGS data were reviewed by a third neutral scientist. Similarly, the quality of the consensus sequences generated was verified by phylogenetic tree reconstruction in Geneious software, version 8. These samples that were FP and FN by PANDAA (cutoff, ≥5%) were assessed by NGS with the Illumina MiSeq platform (cutoff, ≥2%). Of the 30 FP samples, 9 (30%) failed sequencing by NGS. Of the 21 false-positive samples successfully genotyped, only 6 (29%) samples with mutations detected by the PANDAA assay also had mutations detected (at 2 to 5%) by NGS. However, differences in the detection of false-positive DRMs affecting susceptibility to NNRTIs were clinically seen with Y181C versus G190A when PANDAA and NGS results were compared to Sanger sequencing results. The G190A detected by PANDAA was also detected by Sanger sequencing and NGS as G190E/S (which causes high-level resistance to EFV and NVP) (Table 3).

TABLE 3.

Evaluation of the discrepancy results^a

Discrepancy type	Sample name	Sequence detected by:
		PANDAA (cutoff, ≥5%)	Sanger sequencing (cutoff, ≥20%)	NGS (cutoff, ≥2%)
False positive	P 053	K65R	WT	Failed
	P 082	K103N	WT	Failed
	P 096	K103N	WT	K103N
	P 025	Y181C	WT	WT
	P 033	Y181C	WT	WT
	P 040	Y181C	WT	WT
	P 061	Y181C	WT	Failed
	P 073	Y181C	WT	Failed
	P 094	Y181C	WT	WT
	P 098	Y181C	WT	Failed
	P 107	Y181C	WT	Failed
	P 113	Y181C	WT	Failed
	P 120	Y181C	WT	WT
	P 121	Y181C	WT	WT
	P 122	Y181C	WT	Y181C
	P 124	Y181C	WT	WT
	P 129	Y181C	WT	Y181C
	P 130	Y181C	WT	WT
	P 132	Y181C	WT	WT
	P 140	Y181C	WT	WT
	P 147	Y181C	WT	WT
	P 002	G190A	G190S	G190S
	P 023	G190A	G190E	Failed
	P 046	G190A	G190S	G190S
	P 060	G190A	G190S	G190AS
	P 062	G190A	WT	Failed
	P 121	G190A	G190E	G190A
	P 127	G190A	G190E	G190E
	P 129	G190A	WT	WT
	P 149	G190A	G190S	G190S
False negative	P 008	WT	K65R	K65R
	P 107	WT	K65, M184V	Failed
	P 132	WT	K103N	K103N
	P 082	WT	Y181C, G190A	Failed

^aNGS, next-generation sequencing; WT, wild-type virus. The PANDAA assay (≥5%) was assessed against Sanger sequencing for drug resistance mutations detected at ≥20%. We hypothesized that mutations detected by PANDAA and not by Sanger sequencing could represent low-abundance variants. These were later verified by NGS (≥2%). G190A/S/E mutations are nonpolymorphic mutations selected by EFV and NVP.

DISCUSSION

Despite improved ART, virological failure and the emergence of DRMs remain a challenge in many LMICs. In such countries, resistance testing still remains expensive (15, 16). Following the 2018 WHO HIVRESNET recommendations (10), considerable advances have been made in HIVDR PMAs. These technologies have been developed, and some are now being evaluated in LMICs, to address the growing problems of HIVDR. We focused on evaluating the diagnostic accuracy of a novel HIVDR assay, the PANDAA assay, in detecting major DRMs among adolescents and young adults failing NNRTI-based first-line ART.

Several PMAs have been developed to detect HIVDR. These include the oligonucleotide ligation assay (OLA; University of Washington, Seattle, WA, USA), allele-specific primer extension (U.S. Centers for Disease Control and Prevention [CDC]), and multiplexed melting curve analysis (InSilixa, Sunnyvale, CA, USA). Just as the PANDAA assay compared favorably with the reference assay (i.e., Sanger sequencing) and demonstrated an overall sensitivity and specificity of 98% and 94%, respectively, the OLA-Simple kit, developed for detection of HIVDR against first-line NNRTIs, was previously implemented successfully in Thailand and Zimbabwe (30, 31). The valid OLA results obtained in the Zimbabwe laboratory were comparable with high concordance to those from a validated and accredited laboratory in Seattle, WA, USA, for K103N (100%), V106M (39/40; 97.5%), Y181C (38/40; 95%), and G190A (100%). Hence, Mutsvangwa et al. concluded that the use of this low-cost assay for detection of HIVDR in virus from dried blood spots in LMICs could improve HIV care for infected children (30).

Potential advantages of PMAs compared to standard Sanger sequencing-based HIVDR genotyping include assay simplicity, faster turnaround time, lower cost, and more manageable equipment requirements. The PANDAA assay evaluated here required quantitative real-time PCR (qPCR) technology, a well-established gold standard technique for low-cost and sensitive genotyping analysis. The assay and analysis software are user-friendly. In Kenya, the OLA assay was successfully implemented; this assay was performed on a weekly batch of specimens, and the HIVDR results were provided within 10 to 14 days of sample collection, to guide the choice of ART regimen at treatment initiation (32).

Due to the high prevalence of NNRTI DRMs, integrase strand transfer inhibitors are recommended as first-line treatment (9). The PANDAA assay may therefore become less relevant for detecting NNRTI DRMs, as most individuals failing NNRTI will be switched to dolutegravir (DTG). However, the assay will remain critical for the detection of NRTI DRMs associated with the highest levels of reduced susceptibility to TDF and 3TC (K65R and M184V, respectively). In particular, the sensitivity and specificity of the K65R assay were excellent. The high rate of TDF-associated DRMs (57 to 60%) found in sub-Saharan Africa (33) highlights the need for ongoing surveillance.

The PANDAA assay may remain relevant for women living with HIV initiating ART during pregnancy, as DRMs pose significant challenges to the prevention of mother-to-child transmission of HIV and to maternal and child health outcomes (34). The assay could also be relevant for infants and young children with HIV, for whom significant loss in susceptibility to the NRTI class is of great concern.

The clinical significance of low-abundance variants for clinical outcomes has been an open research question for several years. Published data show that low-abundance variants have no clinical significance for NRTIs and protease inhibitors (PIs) (35–39). For NNRTIs, available data are inconsistent regarding whether low-abundance variants predict viral failure (6, 7, 40–43). Because of the inconsistency in results, the WHO strongly recommended that data derived from NGS be interpreted using a Sanger sequencing-like threshold of 15% and that NGS and PMAs remain useful for research studies until the clinical relevance of low-abundance variants has been established for all drug classes. In this study, we speculated that the decreased specificity of the PANDAA assay (cutoff of ≥5%) for the detection of NNRTI DRMs (76%) could be attributed to the presence of low-abundance variants not detected by Sanger sequencing. Hence, the samples that were false positive for Y181C and G190A (by PANDAA) were interrogated by NGS, and the mutations were verified by NGS in only 6/21 (29%). However, in the case of the 9 samples with the G190A mutation considered FP as assessed by PANDAA, the Sanger sequencing and NGS results demonstrated the presence of G190S/E NNRTI resistance mutations in 7/9 of the samples. These mutations (G190A/S/E) decrease susceptibility to EFV and NVP (https://hivdb.stanford.edu/dr-summary/resistance-notes/NNRTI/). The decreased agreement between the PANDAA and NGS may suggest that optimization of the PANDAA primers for accurate detection of major and low-abundance variants may be needed. However, as per the 2018 WHO report (10), additional studies are still needed to determine the clinical application of these new technologies to detect HIV-1 low-abundance variants.

The limitations of our study include the fact that the PANDAA assay requires real-time thermal cycler PCR and a stable power supply, water supply, and infrastructure, which may be difficult to sustain economically in district or rural laboratories. Another limitation of our analysis is the 30% failure rate of NGS, which might have been caused by the degradation of the HIV RNA samples during shipment to Kwazulu Natal, South Africa, for retrospective testing. This failure rate of NGS did not allow thorough evaluation of the performance of PANDAA in detecting low-abundance variants. However, these limitations are balanced by the WHO recommendations that genotyping results for clinical management be interpreted at the Sanger sequencing cutoff (15% to 20%). Additionally, this PANDAA assay evaluation did not include infants or children <10 years old, for whom the assay might have significant benefits.

Conclusion.

The sensitivity and specificity of the PANDAA assay in detecting DRMs showed favorable results and an almost perfect agreement with Sanger sequencing for important DRMs. The PANDAA assay could represent a simple and rapid alternative approach to HIVDR assay in LMICs.