Detection of minority drug resistant mutations in Malawian HIV-1 subtype C-positive patients initiating and on first-line antiretroviral therapy

Background Minority drug resistance mutations (DRMs) that are often missed by Sanger sequencing are clinically significant, as they can cause virologic failure in individuals treated with antiretroviral therapy (ART) drugs. Objective This study aimed to estimate the prevalence of minor DRMs among patients enrolled in a Malawi HIV drug resistance monitoring survey at baseline and at one year after initiation of ART. Methods Forty-one plasma specimens collected from HIV-1 subtype C-positive patients and seven clonal control samples were analysed using ultra-deep sequencing technology. Results Deep sequencing identified all 72 DRMs detected by Sanger sequencing at the level of ≥20% and 79 additional minority DRMs at the level of < 20% from the 41 Malawian clinical specimens. Overall, DRMs were detected in 85% of pre-ART and 90.5% of virologic failure patients by deep sequencing. Among pre-ART patients, deep sequencing identified a statistically significant higher prevalence of DRMs to nucleoside reverse transcriptase inhibitors (NRTIs) compared with Sanger sequencing. The difference was mainly due to the high prevalence of minority K65R and M184I mutations. Most virologic failure patients harboured DRMs against both NRTIs and non-nucleoside reverse transcriptase inhibitors (NNRTIs). These minority DRMs contributed to the increased or enhanced virologic failures in these patients. Conclusion The results revealed the presence of minority DRMs to NRTIs and NNRTIs in specimens collected at baseline and virologic failure time points. These minority DRMs not only increased resistance levels to NRTIs and NNRTIs for the prescribed ART, but also expanded resistance to additional major first-line ART drugs. This study suggested that drug resistance testing that uses more sensitive technologies, is needed in this setting.


Introduction
Rapid scale-up of antiretroviral therapy (ART) over the past decade has remarkably reduced the mortality and morbidity of HIV-positive patients and decreased HIV transmission. Seventeen million HIV-1-positive patients around the world were receiving ART by the end of 2015. 1 However, the scale-up of ART in resource-limited settings without adequate treatment monitoring has raised concern about the development of HIV drug resistance. The quasi-species nature of HIV-1 makes the detection of drug resistant mutations (DRMs) more difficult, because the commonly-used Sanger sequencing for drug resistance testing is incapable of detecting these drug resistant HIV variants at a level of less than 20% of the viral population. 2,3,4,5 Minority drug resistant variants (also known as low-frequency mutants) that are not detected by Sanger sequencing are clinically important, as they can cause virologic failure in patients treated with ART for the first time. 6,7,8,9 Recent studies have demonstrated that particular drug resistant HIV mutant viruses are clinically significant at a level of 1% of the viral population, as the minority variants can replicate quickly and become the predominant viral population through the selective pressure of ART drugs, leading to treatment failure. 9,10 However, in the absence of drug pressure in treatment-naïve patients, the stability of transmitted DRMs is different. 11 For instance, a transmitted M184V mutation can quickly revert to wild-type due to diminished viral fitness. 12 http://www.ajlmonline.org Open Access In patients on ART, minority DRMs may persist for months or years during and post-ART. 13,14,15 These minority DRMs, not detected by Sanger sequencing, present a need for more sensitive methods to detect the minority DRMs in a clinical sample.
Deep sequencing or next-generation sequencing technologies are extensively used to examine HIV viral diversity and minority drug resistant variants. Next-generation sequencing is a highly sensitive and high-throughput sequencing platform. It can detect HIV variants that make up 0.05% to 1% of viral populations. 16,17,18,19,20,21 As part of HIV drug resistance surveillance by the Malawi Ministry of Health, a prospective cohort study to monitor ART outcomes and drug resistance development was conducted among patients from ART initiation to one year later. In this 2008 ART patient monitoring survey, 6.1% of the patients on ART for 12-15 months harboured drug resistant HIV. 22 The most common non-nucleoside reverse transcriptase inhibitor (NNRTI) mutations were K103N (58.1%), Y181C (41.9%) and G190A (6.5%), and the most frequent nucleoside reverse transcriptase inhibitor (NRTI) mutation was M184V (61.3%). The DRMs conferring resistance against NNRTI at baseline were associated with DRMs detected at 12-15 months on ART. 22 The present study aimed to evaluate parallel tagged deep sequencing primers on clinical samples and to investigate the profile of minority DRMs and their association with virologic failure in the same Malawi ART monitoring cohort.

Ethical considerations
The study protocol was approved by the National Health

Clinical samples
Between February and June 2008, HIV-1-positive patients aged 15 years or older, who initiated first-line ART at four ART clinics following the Malawi ART guidelines, were enrolled. Patients were treated with a first-line regimen combination of stavudine, lamivudine and nevirapine, or an alternative firstline regimen of stavudine to zidovudine substitutions in case of toxicity. Plasma specimens were collected before ART initiation and at 12-15 months on ART for viral load and HIV drug resistance testing. 22 In the present study, we selected plasma specimens that had enough volume available to evaluate the assay. These were 20 samples collected from participants before ART initiation with viral loads ranging from 10 471 to 2 041 738 copies/mL and 21 samples collected from ART patients at virologic failure (defined as viral load ≥1000 copies/mL) after 12-15 months on ART (viral load ranging from 1738 to 776 247 copies/mL). In addition, six plasmid clones and one mixed clone containing 1% mutant (2495 copies/µL) under the background of a wild-type clone were prepared and used to verify sequence accuracy in this study. All of these plasmid clones were derived from the Malawian cohort samples. All plasmids were constructed using TOPO TM 22 The viral RNA was subjected to one-step RT-PCR amplification as described previously. 11

Parallel deep sequencing
Degenerate primers, capable of amplifying multiple HIV-1 group M subtypes, were designed based on the HIV-1 pol gene sequences (www.hiv.lanl.gov) ( Table 1). Six overlapping primer sets (forward and reverse) were used for bidirectional coverage of protease amino acids 6 to 99 and reverse transcriptase amino acids 1 to 251. The size of the assembled gene fragment was 1035 base pairs. These six primers, tailed with Roche 454 adaptor and multiplex identifier sequences (tags), were synthesised at the CDC Biotechnology Core Facility. For PCR amplification, a 50µL reaction contained 1× AccuPrime PCR Buffer II (Thermo Fisher Scientific, Carlsbad, California, United States), 0.5 U AccuPrime Taq High Fidelity (Thermo Fisher Scientific, Carlsbad, California, United States), 13.5 µL water, 0.3 µM, each forward and reverse primer, and 2 µL DNA. All reactions were performed in 9700 thermal cyclers

Deep sequencing analysis
Sequence files generated by Roche 454 deep sequencing were analysed using GS Amplicon Variant Analyzer pipeline from Roche Applied Science (Indianapolis, Indiana, United States). The deep sequencing analysis process in the present study included quality restriction for base call setting at 60 for signal intensity. All amplicon reads of alignment and single nucleotide polymorphism (SNP), calling against the HXB2 reference sequence, were evaluated with a quality score ≥ 25 and read length ≥ 220 base pairs. Sequence accuracy of Roche 454 runs was evaluated using the sequences generated by Sanger sequencing of the seven plasmid clones. The minority variant was defined as a SNP detected at > 0.68% and < 20% of the frequency of mutations. For DRM analysis, mutations were called and grouped based on International AIDS Society (IAS)-USA 2011 recommendations. 23 Additional mutations for those samples collected before ART initiation were analysed based on the 2009 World Health Organization surveillance DRM list. 24

Standard Sanger sequencing
The standard genotyping of HIV drug resistance was performed on all plasma samples using an in-house population-based sequencing assay. 11 Raw sequencing data were analysed using the customised ReCall software, v.2.24 (provided by Dr. Richard Harrigan from British Columbia's Center for Excellence in HIV/AIDS Research, Vancouver, Canada). 25 The mixed mutation calling threshold was set at ≥ 20% of the main peak. The DRMs were analysed as stated above for parallel deep sequencing results.

Statistical analysis
All statistical analysis was performed using SPSS Statistics v19 (SPSS Inc., Chicago, Illinois, United States). The Wilcoxon Signed-Rank test was used to analyse the statistical differences in the number of DRMs detected between Sanger sequencing and deep sequencing. Fisher's exact test was used to compare the prevalence of HIV drug resistance by these two methods. A P-value of < 0.05 was considered statistically significant.

Parallel deep sequencing coverage and estimate of sequence errors
The GS-FLX deep sequencing of a single run yielded 246 849 raw sequence reads, of which 242 246 (98.1%) reads passed the quality restrictions with a mean read length of 270 nucleotides, while 4603 (1.9%) low-quality sequences were removed from the analysis pipeline. On average, 5490 reads per sample were obtained (range from 420 to 5673 reads). From the 1% mixed clone, K103N was not detected within the 520 sequence reads; Y181C and M184V were detected at the level of 1.03% with 1335 reads, while G190A was detected at 0.97% with 785 reads. The mean error rate plus two standard deviations was 0.43 from the six control plasmid clones. Sequence errors mostly occurred at the overlapping areas of amplicons in the RT gene. There was only one SNP showing an error rate of 0.68% at codon 17 of the PR gene in a polyG region (GGGGGGCA, nucleotide 35, Figure 1). This error at PR codon 17 was not in the DRM position according to IAS and Stanford HIV database definitions. Another high-error site was codons 63 of the RT gene (nucleotides 470 and 471, Figure 1), with a 0.61% error rate. Based on these background errors for each nucleotide position, the frequency > 0.68% error rate was used as the threshold for true SNPs when evaluating the minority DRMs in the clinical samples.

Clinical impact of minority drug resistance mutations detected by deep sequencing on virologic failure
To study the impact of minority DRMs detected by deep sequencing on the clinical outcome of patients on ART, we compared drug resistance levels (genotype susceptibility score) or expansion of drug resistance to additional drugs or drug classes from those 21 ART-failure patients against NRTIs and NNRTIs using the Stanford HIV drug resistance database tool. Among 19 patients with the additional minority DRMs detected, we found that seven (36.8%) of the patients had enhanced resistance levels to NRTIs. More importantly, three of the seven patients had gained lowto high-level resistance against tenofovir (Table 3), a key component of the current World Health Organizationrecommended first-and second-line regimens. 26 Fourteen (73.7%) out of the 19 patients also had an enhanced resistance level to NNRTIs and 13 of these 14 patients had intensified or expanded resistance profiles against the second generation of NNRTIs (etravirine and rilpivirine) ( Table 4). Furthermore, the drug resistance profile analyses revealed an intermediateor high-level resistance to the relevant first-line regimens (stavudine, lamivudine and nevirapine or zidovudine, lamivudine and nevirapine) that are prescribed to these Malawian patients and which might explain the virologic failures these patients experienced.       amplify subtype B, B/C, F and G samples using these primers which were confirmed by gel electrophoresis and successfully detect 5 of 5 subtype B samples using deep sequencing methods. 27 Previous studies reported the use of parallel tagged deep sequencing methods in detecting low levels of HIV-1 subtype B variants. 18,20,21,28,29 A study by Dudley et al. 20 evaluated a 454 GS Junior sequencer by multiplexing 48 samples collected from HIV-1 subtype B-positive individuals and obtained a sequencing success rate of 93% and an error rate of 0.71%. Our study using GS-FLX with multiplexing on 48 samples collected from HIV-1 subtype C-positive patients showed a 100% amplification rate and 0.265% mean error rate. Many studies have reported that the mean error rate of pyrosequencing techniques can be down to 0.05% to 1%. 17,19,21,30 The error rates for deep sequencing not only affect the accuracy of base calling, but also impact the sensitivity of minority variant detection. It has been reported that factors, such as the input number of template molecules, sequence primers, amplicon length, nucleotide sequences, PCR errors and operational procedures, might contribute to deep sequencing assay sensitivity and accuracy. 17,31,32,33 In addition, error rates are nucleotide position-dependent and Roche 454 deep sequencing methods are prone to have more errors at the homopolymeric regions. 17,31 In the present study, cross-over errors of major DRMs between samples were not found. To balance the detection sensitivity with detection accuracy, we set up the base calling threshold for low-frequency mutations at > 0.68% (mean error rate + 2 standard deviations) which was calculated based on the error rates of individual nucleotide positions of six control plasmid samples. In the present study, the K103N mutation was not detected from the mixed clone at 1% of minority variant level of control plasmids from 520 reads. This was likely due to not having enough reads amplified for this mixed plasmid as a previous study demonstrated that at least 1950 reads are required for detecting a minority variant for K103N mutations at about the 1% level. 17 One limitation of our study was that the primer pair for amplifying amplicons containing codon 103 of RT gene was not optimal for the depth of coverage. Some minority K103N mutations could have been missed in this study. Thus, the number of sequence reads obtained for each nucleotide position and errors at the homopolymeric regions played an important role in the depth of the next-generation sequencing.

Sample ID Mutations by Sanger Drug resistance level (score) † Mutations by NGS
Our results from samples collected from patients failing ART and initiating ART demonstrate that deep sequencing has the added benefit of detecting low-frequency mutations in this Malawian cohort. Overall, deep sequencing detected significantly more DRMs than Sanger sequencing. Of those specimens collected from patients initiating ART, more DRMs against NRTI were detected by deep sequencing. Among the minority DRMs, detected by 454 deep sequencing, K65R and M184I were the most common and may compromise the effectiveness of both first-and secondline drugs used according to the Malawi ART guidelines.
The K65R mutation can confer resistance to stavudine and cross-resistance to lamivudine, abacavir, emtricitabine and tenofovir, 23,34,35 and is more frequently identified in HIV-1 subtype C. 30,35,36 Similar to previous studies, the K65R mutation was seen in both treatment-naïve patients and patients failing ART in this cohort. Several studies have reported that increased presence of K65R mutations is caused by pyrosequencing errors or by the nucleotide template of subtype C viruses (such as the ATA sequence at codon 63 of the RT gene). 30,32,37 Even though no errors at codon 65 of the RT gene were found by deep sequencing in the current study, we did find a relatively higher error rate at codon 63 of the RT gene in the subtype C plasmid sequences. However, we did not find higher error rates for K65R compared with other DRM sites in these patients. The M184I mutations were only detected at low frequencies by deep sequencing in pre-ART patients and patients with treatment failure. M184I was considered to be a transient mutation before being replaced by M184V. 19,38,39 No detectable levels of M184V mutations were found in pre-ART samples using deep sequencing, but M184V mutation was detected in over three-quarters of samples from patients failing ART. Taken together, the higher NRTI resistance mutations of M184V and K65R in patients failing ART were more likely acquired by selective drug pressure in this Malawian cohort treated with a regimen containing stavudine and lamivudine. 22 In this study, most samples collected from virologic failure patients had detectable DRMs to NNRTI by both Sanger sequencing and deep sequencing. The mutations K101E, K103N, V106A/M, V179D/T, Y181C, G190A/E and H221Y to NNRTI were the most common minority mutations detected in these patients. Virus with K101E/Y181C/G190A and other mutations could increase levels of resistance to nevirapine 893-fold. 40 The H221Y mutation could also impact clinical outcomes as Y181C/H221Y along with the K103N or K101Q mutations could increase resistance levels to nevirapine over 100-fold (K103N) or 3000-fold (K101Q). 41 The drug resistance profile generated by deep sequencing revealed that these mutations were associated with the firstline regimen (stavudine, lamivudine and nevirapine). The DRMs to NNRTI in pre-ART samples were also relatively high in these patients and were likely due to single-dose nevirapine used in the prevention of mother-to-child transmission program in Malawi. 22 However, our results could not rule out the presence of transmitted drug resistance to NNRTI in these pre-ART patients.
Although DRMs against PIs were not detected using Sanger sequencing, they were detected by deep sequencing in this cohort. M46I/L is considered a major PI mutation and would increase drug resistance levels to PIs along with other mutations. 34,42 Because no PI drugs were used in the firstline ART in this cohort, these PI mutations were likely natural polymorphisms of HIV-1. The natural polymorphism of M46I has been reported to have a replicative advantage for subtype B, 43 while the impact of M46I/L natural polymorphisms on the development of drug resistance in patients is unknown. As Malawi has started to use lopinavir/ ritonavir for second-line regimens, 44 the emergence of DRMs to PIs should be closely monitored.
Evidence is lacking in understanding the real clinical impact of minority DRMs. Clinical trials are needed to accurately evaluate the clinical consequences of these DRMs. However, our results indicate that minority mutations detected by barcoded deep sequencing show an increased or expanded level of resistance to NRTIs and NNRTIs (see mutation scores in Tables 3 and 4). For instance, increased M184IV/I mutations would reduce susceptibility to lamivudine and emtricitabine (scores from 0 to 60, from susceptible to high-level of resistance). K65R+ M184V/I would reduce susceptibility to tenofovir and didanosine from low-level (scores from 0 to 60) to highlevel resistance (scores from 15 to 75). Additionally, other individual thymidine-analog mutations showed an intermediate or high level of resistance to Malawi's firstline regimens (stavudine, lamivudine and nevirapine). These low-frequency mutations detected by the barcoded parallel sequencing added significant values to the resistant reservoir in the HIV-positive population. All mutation data, both majority and minority mutations, can be used by doctors or policy makers as a reference when changing or revising treatment therapy for patients with virologic failure or country first-line regimens in Malawi.

Limitations
This study had its limitations. First, the sample size was small due to the availability of remnant samples and budget constraints. A statistically-appropriate sample size should be used for population-level estimations of DRMs in order to make a meaningful statement. Second, although significantly increased minority DRMs were observed in the samples collected from pre-ART and patients with virologic failure using Roche 454 barcoded deep sequencing, lack of proper plasmid mutant K65R control in the test might have compromised the accuracy of calculating the K65R mutation rate. Third, some methods in the current study could not be applied further as a result of the 454 platform and technologies being discontinued due to high cost and errors at some homopolymeric regions. However, the designed primers and barcoded strategy in the current study could be applied to other deep sequencing platforms for HIV drug resistance testing or studies.

Conclusion
In conclusion, our study confirmed that barcoded parallel deep sequencing technology is capable of detecting minority DRMs from clinical patient samples. These minority DRMs not only increased resistance levels to the antiretroviral drugs that are being prescribed, but they also expanded resistance to additional major first-line antiretroviral drugs such as tenofovir. The minority DRMs detected by deep sequencing may be helpful for selecting the optimal regimens for patients initiating ART and for patients who fail first-line regimens.