Ultra-deep sequencing improves the detection of drug resistance in cellular DNA from HIV-infected patients on ART with suppressed viraemia

Abstract

Background

Standard genotypic tests performed on HIV DNA from patients on suppressive ART, with previous resistance-associated mutations (RAMs) detected in their plasma, underestimate resistance. We thus compared ultra-deep sequencing (UDS) with bulk sequencing of DNA to detect RAMs previously identified in plasma.

Methods

We sequenced the DNA of 169 highly treatment experienced patients with suppressed viraemia (ANRS 138-EASIER trial). Protease (PR) and reverse transcriptase (RT) genes from HIV DNA were sequenced by bulk sequencing and UDS, comparing 1% and 20% as thresholds of detection for UDS.

Results

Patients were highly treatment experienced (13.6 years). UDS of DNA was successful for the RT and PR genes in 133 (79%) and 137 (81%) patients, respectively. The detection of RAMs was similar by bulk sequencing and UDS with a 20% cut-off. However, the detection of RAMs by UDS with a 1% cut-off was significantly higher than that of bulk sequencing for RT codons D67N (65.4% versus 52.3%), M184V (66.2% versus 52.3%), L210W (48.9% versus 36.4%) and T215Y (57.9% versus 42.1%) and PR codons M46I (46% versus 26%), I54L (12.4% versus 3.9%), V82A (44.5% versus 29.9%) and L90M (57.7% versus 42.5%).

Conclusions

Genotypic resistance testing of cellular HIV DNA of well-controlled patients should use UDS technology with a sensitivity threshold of 1% to improve the detection of the resistant reservoir.

Introduction

In high-income countries, ART currently provides sustained control of HIV replication for ∼90% of patients.¹ However, patients or physicians may want to change therapy to improve long-term tolerability, quality of life or cost-effectiveness or to prevent toxicities.

Before switching to new therapeutic options, it is crucial to consider the resistance background acquired during previous periods of uncontrolled viraemia to limit the risk of treatment failure and the accumulation of additional resistance mutations.² Physicians currently use the patient’s treatment history and the cumulative results of plasma-based genotypic resistance tests from previous virological failures to assess drug resistance-associated mutations (RAMs) and identify switch regimens. However, it is sometimes difficult to collect past genotype resistance data if there are no electronic records, the patients have moved or the files have been lost.

The analysis of RAMs in cell-associated HIV DNA might be able to identify drug resistance when plasma viral load is undetectable, as viral variants, including circulating drug-resistant strains selected during ART, could be archived in the latent viral reservoir.^3–7 However, we and others have previously reported lower sensitivity in detecting mutations archived in HIV-1 cellular DNA when comparing the results of genotypic drug resistance assays on cellular HIV DNA at the time of suppressed viraemia with those of resistance genotyping of plasma HIV-1 RNA at the time of past treatment failures.^8–12 In addition, the duration of persistence of archived HIV mutations in cellular DNA is unknown. This information could have a potential impact on subsequent treatment and the possibility of drug recycling.

The underestimation of drug resistance mutations may be due to the lack of sensitivity of population-based sequencing (such as bulk sequencing), which is generally used in routine practice. The sensitivity of bulk sequencing in detecting resistant variants within quasi-species of WT viruses is ∼20%, whereas the sensitivity of the ultra-deep sequencing (UDS) assay allows the use of a sensitivity threshold as low as 1%.^13–16

Our previous study included 169 patients, enrolled in the ANRS EASIER trial, who had already been treated with three antiretroviral drug classes (NRTIs, NNRTIs and PIs) and had plasma HIV-1 RNA <400 copies/mL at inclusion.¹⁷ The results of previous resistance genotyping by population-based sequencing of plasma HIV-1 RNA in individual patients were compared with those of resistance genotyping of whole-blood HIV-1 DNA at inclusion.⁸ We reported that resistance to at least one NRTI was detected by RNA, but not DNA, genotyping for 63% of patients and by DNA, but not RNA, genotyping for 13% of patients. Resistance to at least one NNRTI was found by RNA, but not DNA, genotyping for 47% of patients and by DNA, but not RNA, genotyping for 1% of patients. Resistance to at least one PI was observed by RNA, but not DNA, genotyping for 50% of patients and by DNA, but not RNA, genotyping for 7% of patients. Our objective was to compare HIV DNA resistance mutations by UDS at two thresholds (>20%, considered to be the bulk-sequencing cut-off, and >1%, for the detection of minority resistant variants) with documented resistance mutations from previous genotype assays performed on plasma HIV RNA and cell-associated HIV DNA. We estimated the persistence of the resistance reservoir by analysing the detection of RAMs in HIV DNA by UDS, with the 1% cut-off, according to the timing of their last detection in plasma RNA genotype assays.

Materials and methods

Patients

We used DNA from the whole blood of patients who were enrolled in the previously published ANRS 138-EASIER study, a 48 week non-inferiority randomized multicentre trial that assessed the efficacy and safety of a switch from enfuvirtide to raltegravir in highly treatment-experienced patients receiving a suppressive PI and enfuvirtide-containing regimen in 2008.¹⁷ The main inclusion criteria were HIV-1-infected patients with: (i) failure on, or intolerance to, triple drug classes (NRTIs, NNRTIs and PIs); (ii) a stable enfuvirtide-based regimen for ≥3 months; and (iii) plasma HIV-1 RNA levels <400 copies/mL for at least the last 3 months.

Ethics

The institutional review boards of all participating institutions approved the parent study and substudy (ClinicalTrials.gov identifier NCT00454337) and all participants provided written informed consent.

HIV-1 RNA resistance genotyping

The results of all available HIV-1 RNA resistance tests performed by bulk sequencing of plasma during previous episodes of virological failure were collected from French virology laboratories of the ANRS network. We constructed the ‘cumulative’ RNA genotype for each patient by adding up all mutations found in previous genotypic tests for each drug class, as previously described.⁸

HIV-1 DNA resistance genotyping by population sequencing (bulk sequencing)

HIV-1 DNA resistance genotyping was performed in a central laboratory as previously described.⁸ DNA was extracted from 200 μL of frozen stored whole blood using an automatic nucleic acid extractor (MagnaPure; Roche, Meylan, France). PCR followed by nested PCR was used to amplify the reverse transcriptase (RT) and protease (PR) genes, according to the ANRS consensus methods (www.hivfrenchresistance.org). Population sequencing was performed on purified amplicons using Taq Dye Deoxy Terminator cycle sequencing kits and an ABI 3700 automated DNA sequencer (Applied Biosytems, Courtaboeuf, France).

HIV-1 DNA resistance genotyping by UDS

After DNA extraction, PCR amplification, similar to that used for the population sequencing, was performed. A GS FLX bead adaptor and a specific identity tag were added to the RT and PR primers. In parallel, 8E5 cells, containing one copy of HIV-1 DNA, were used as a control, in triplicate, for each sequencing run. Amplicons containing the bead adaptor and Multiplex Identifier were then purified in Nucleofast 96PCR plates (Clontech, Mountain View, CA, USA), according to the manufacturer’s instructions. Amplicons were then quantified with the Quant-iT PicoGreen dsDNA kit (Invitrogen), fixed to beads and amplified in a microemulsion with the GS FLX Titanium emPCR kit (454 Life Sciences; Roche Diagnostics). Amplified beads were purified and enriched according to the manufacturer’s instructions, counted with a Beckman Coulter Z1 particle counter (Beckman Coulter, Brea, CA, USA) and applied to a GS FLX Titanium PicoTiter Plate (454 Life Sciences; Roche Diagnostics). The pyrosequencing reaction was performed with the GSFLX Titanium sequencing kit on an FLX Genome Sequencer (454 Life Sciences; Roche Diagnostics).

Sequencing analysis

Bulk and UDS sequences were aligned with the HIV-1 subtype B HXB2 reference strain. UDS sequences were demultiplexed, filtered and aligned by two pipelines, Pyropack^18,19 and AVA software (Roche). UDS mutations were analysed using cut-offs of 1% (UDS 1%) and 20% (UDS 20%). Consensus sequences obtained by AVA were genotyped using the Stanford database (http://hivdb.stanford.edu/). Then, RT and PR mutations were identified using the Nov-2015 IAS-USA resistance list (www.iasusa.org/content/drug-resistance-mutations-in-HIV) and the 2015 ANRS v25 algorithm (www.hivfrenchresistance.org). Antiretroviral susceptibility was interpreted according to the 2015 ANRS v25 algorithm. For the recent antiretrovirals used at the time of the study (tipranavir, darunavir, rilpivirine and etravirine), we analysed resistance from HIV DNA (bulk sequencing and UDS) only and excluded data from previous plasma genotypes, as some RAMs were not screened (sequence FASTA files not available) and susceptibility based on plasma RNA could have been underestimated. Stop mutations were analysed independently of resistance mutations in amino acids and nucleotides to identify APOBEC3G-like G to A mutations.

The persistence of RAMs was defined by the detection of the mutation by UDS 1% in cellular HIV DNA. We estimated the persistence over time for each major resistance IAS mutation that was detected for at least 20 patients in cumulative RNA genotypes. For each patient, we recorded the time between the last detection of a mutation by the RNA genotype assay and detection by the DNA genotype assay. The number of patients with persistent RAMs was estimated according to the timing of their last detection in the plasma RNA genotypes.

Statistical analysis

All reported values are medians (with IQRs) for continuous variables and frequencies and percentages for categorical variables. ANOVA with a post hoc Tukey HSD test was used to compare the frequencies of RAMs detected in HIV DNA by UDS 1% with HIV DNA by bulk sequencing and cumulative RNA genotype. All analyses were performed using R software (https://cran.r-project.org/).

Results

Patient characteristics

The main characteristics of the 169 patients enrolled in the ANRS 138-EASIER trial are shown in Table 1. Most of the patients were men (85%) and the median age was 48 years. They were highly treatment experienced (median duration of ART of 13.6 years). At the time of the study (randomization in the trial), the regimens consisted of enfuvirtide plus at least one NRTI (95%), one or two PIs (99%) and one NNRTI (8%). The median CD4 cell count was 393 cells/mm³ and 86% of patients had a plasma viral load <50 copies/mL.

Cumulative HIV-1 RNA and DNA genotypes determined by bulk sequencing

We previously reported the results of cumulative HIV-1 RNA and DNA genotypes by bulk sequencing.⁸ Briefly, cumulative HIV-1 RNA genotypes were built using a total of 716 plasma HIV-1 RNA genotypes collected for the 169 patients, with a median (IQR) of 4 (3–5) tests per patient. The median interval between the last RNA genotype and randomization was 2.6 years (IQR 1.7–3.6). Population-based sequencing of cellular HIV-1 DNA was successful for the RT gene for 131 of 169 patients (78%) and for the PR gene for 127 of 169 patients (75%).

HIV-1 DNA genotypes determined by UDS

UDS of cellular HIV-1 DNA was successful for the RT gene for 133 of 169 patients (79%) and for the PR gene for 137 of 169 patients (81%). After trimming, we obtained 8634 ± 3175 sequences/sample for the RT domain, with a mean length of 267 ± 32 bp and a mean PHRED Q score of 35.4. For the PR domain, we obtained 4330 ± 1519 sequences/samples after quality trimming, with a mean length of 299 ± 18 bp and a mean PHRED Q score of 36.4. Analysis of 8e5 controls allowed us to establish an error rate per position of <1%. Further analysis was performed using cut-offs of 1% and 20%.

Resistance analyses

Major resistance mutations listed by the IAS are reported in Table 2 and Figure 1. The results were highly similar between the bulk sequencing and UDS 20% for almost all listed mutations. The detection of RAMs in DNA by UDS 1% was significantly higher than that by bulk sequencing for RT codons D67N (65.4% versus 52.3%), M184V (66.2% versus 52.3%), L210W (48.9% versus 36.4%) and T215Y (57.9% versus 42.1%) and PR codons M46I (46% versus 26%), I54L (12.4% versus 3.9%), V82A (44.5% versus 29.9%) and L90M (57.7% versus 42.5%) (Table 2). However, the detection of RAMs in cumulative RNA was significantly higher than in DNA, when using the 1% cut-off, for RT codons L74V (27.3% versus 11.2%), M184V (84.8% versus 52.3%), K103N (46.2% versus 22.4%) and Y181C (29.5% versus 16.8%).

We found similar drug resistance in HIV DNA, using the 2015 ANRS algorithm, by bulk sequencing and UDS 20%, but a higher frequency of drug resistance by UDS 1% (Figure 2).

STOP codons

We found STOP codons for 60 patients at amino acid positions W6, W42, W46, C67 and G86 in the PR gene and W24, W71, W88, W153, L210, W212, W229 and W239 in the RT gene. The percentage of detection was between 1% and 100%. Thirty-four patients harboured viruses with at least one STOP mutation as the most frequent mutation (>20%) and five with STOP mutations representing >80% of the viral population.

Decrease in resistance mutations in cellular HIV DNA

Most RAMs previously detected in cumulative RNA genotypes persisted over time in the HIV DNA, detected by UDS 1% (Figure 3).

Most NRTI mutations were slightly less frequent in the HIV DNA (M41L, D67N, K70R, M184V, L210W, T215F/Y and K219Q). All but one mutation (L74V) persisted in the HIV DNA reservoir for 80% of patients after 5 years since their last detection in the cumulative RNA. The mutation L74V persisted in 50% of patients after 5 years. NNRTI mutations (K103N, V108I, Y181C and G190A) appeared to be less persistent and were present in the cellular reservoir for 60% of patients after 5 years. Finally, the persistence of PI mutations (M46I/L, I54L, Q58E, V82A, I84V and L90M) was more heterogeneous: from 90% for Q58E to 70% for I54L and 60% for M46L.

Discussion

Archived resistance mutations in the cellular HIV DNA of ART-controlled patients, selected during previous virological failure, are a potential source of information when switching to a new regimen when prior genotypic data are not available. Studies have shown that resistance testing by standard sequencing of HIV DNA is less sensitive than accumulated drug resistance data from previous plasma genotypes. The lower detection by sequencing can be explained by two non-exclusive hypotheses: either population sequencing performed in routine practice cannot detect mutations present in <20% of the viral population (although this could be improved by examining cumulative genotypes) and/or the cellular clone harbouring the resistant virus becomes undetectable after many years of effective treatment.

We previously demonstrated that DNA genotyping by bulk sequencing detected significantly fewer resistance mutations in the RT and PR genes than previous cumulative RNA genotyping among 169 highly experienced patients included in the ANRS 138-EASIER switch trial,⁸ thus confirming data reported in other studies.^9,10,12 Here, we performed UDS sequencing on the same cellular HIV DNA and resistance was assessed using sensitivity detection thresholds of 20% and 1%. The detection of resistance mutations was similar for most mutations by bulk sequencing and UDS 20%, confirming the 20% threshold of detection for bulk sequencing. In contrast, UDS 1% showed better sensitivity in detecting mutations, confirming our first hypothesis. The increase was between 8 and 27% for major NRTI and PI RAMs. There was no improvement in detection for NNRTI RAMs.

In contrast, cumulative HIV RNA genotypes detected a higher number of NRTI and NNRTI RAMs than UDS 1%. We cannot exclude an overestimation of resistance detection, as some mutations may decrease over time (see below). For more recent antiretrovirals, the interpretation of past RNA genotypes may be less informative, because of absent information concerning new mutations, suggesting the need to reinterpret RT and PR sequences using more recent algorithms. UDS of cellular HIV DNA, using a 1% threshold, could improve the monitoring of resistance in controlled patients.

We found more STOP codons by UDS 1% than bulk sequencing (44% versus 26%) and two or more STOP codons were detected for half of the patients. Such information is important to determine the level of defective virus. However, Fourati et al.²⁰ demonstrated that HIV patients showing sustained undetectable viraemia may harbour a defective viral genome in PBMCs, possibly explaining the inability of resistant strains to successfully re-emerge.

We demonstrate that most RT and PR mutations formerly detected in plasma during virological failure persisted over time in the cellular HIV DNA, despite sustained virological control. We estimated the decrease in RAMs in the HIV DNA reservoir by UDS 1% for the different therapeutic classes. Twenty percent of patients had virus without resistance mutations for NRTIs, 40% for NNRTIs and 10%–40% for PIs 5 years after the last detection of RAMs in cumulative RNA. This suggests that NRTI and PI mutations are more likely to persist than NNRTI mutations. This result confirms that resistant variants that entered the cellular reservoir during past virological failures persisted and were not replaced over time. The long-term persistence of these archived variants can be explained by the characteristics of the target reservoir cells, particularly central memory CD4 T cells with a high proliferative capacity and long half-life. Moreover, the complete resistance profile was maintained because resistance mutations associated with each drug class were mostly linked to the same viral variants. In contrast, the apparent loss of archived NNRTI-associated resistance mutations could be explained by the lack of selective pressure with NNRTI drugs, as only 8% of patients were under NNRTIs at the time of DNA resistance analysis in our study.

Longitudinal studies addressing the dynamics of archived resistant quasi-species in cellular HIV DNA in heavily ART-experienced patients with controlled viraemia reported that most mutations persisted over time and also that archived HIV DNA quasi-species continued to evolve, depending on the selective pressure.^21–23 Moreover, several new resistance mutations have been shown to emerge from cellular HIV DNA, suggesting the persistence of residual viraemia below the limit of quantification of standard plasma HIV RNA tests in some patients.

Our study had some limitations. First, we analysed the persistence of resistant virus in HIV DNA in heavily pretreated patients and our conclusions may not directly apply to patients with less extensive exposure to ART and a shorter period since virological failure. Second, we compared cumulative RNA with a single DNA genotype. Cumulative HIV DNA genotypes could report the same rate of mutations. Third, determining the linkage of mutations, as shown by clonal analysis,²⁴ could be more informative in determining whether resistance mutations are dispersed between different clones and not associated with a single genome, possibly explaining the success of the current treatment regimen. Finally, our study did not analyse modern antiretroviral regimens, particularly those including integrase inhibitors.

In conclusion, genotypic resistance testing of cellular HIV DNA of drug-experienced patients should use UDS technology with a sensitive threshold to improve the detection of resistant virus in the cellular reservoir. To date, there are no definitive data on the ability of archived viral variants to re-emerge in case of suboptimal therapy. Further analyses are needed to prospectively evaluate the impact of the resistance detected in cellular HIV DNA on virological responses in clinical practice.

Acknowledgements

Presented in part at the XXV International HIV Drug Resistance Workshop, Boston, MA, USA, 2015. Abstract 43.

Funding

This work was supported by the French National Agency for Research on AIDS (ANRS 2014-AO1).

Transparency declarations

J. M. M. reports receiving support as an adviser for Gilead Sciences, MSD, Janssen, BMS and ViiV Healthcare, and research grants from Gilead Sciences and Merck. C. D. reports receiving financial support as an adviser for Gilead Sciences, Merck, Janssen and ViiV Healthcare, and research grants from Merck and ViiV Healthcare. All other authors: none to declare.

References