Abstract
Background
The association between low-frequency human immunodeficiency virus type 1 (HIV-1) drug resistance mutations (DRMs) and treatment failure (TF) is controversial. We explore this association using next-generation sequencing (NGS) methods that accurately sample low-frequency DRMs.
Methods
We enrolled women with HIV-1 in Malawi who were either antiretroviral therapy (ART) naive (cohort A), had ART failure (cohort B), or had discontinued ART (cohort C). At entry, cohorts A and C began a nonnucleoside reverse transcriptase inhibitor–based regimen and cohort B started a protease inhibitor–based regimen. We used Primer ID MiSeq to identify regimen-relevant DRMs in entry and TF plasma samples, and a Cox proportional hazards model to calculate hazard ratios (HRs) for entry DRMs. Low-frequency DRMs were defined as ≤20%.
Results
We sequenced 360 participants. Cohort B and C participants were more likely to have TF than cohort A participants. The presence of K103N at entry significantly increased TF risk among A and C participants at both high and low frequency, with HRs of 3.12 (95% confidence interval [CI], 1.58–6.18) and 2.38 (95% CI, 1.00–5.67), respectively. At TF, 45% of participants showed selection of DRMs while in the remaining participants there was an apparent lack of selective pressure from ART.
Conclusions
Using accurate NGS for DRM detection may benefit an additional 10% of patients by identifying low-frequency K103N mutations.
Keywords: drug resistance mutations, antiretroviral therapy, low-frequency mutation, treatment outcome, next generation sequencing
Carrying the drug resistance mutation K103N at either low or high frequency when starting a nonnucleoside reverse transcriptase inhibitor antiretroviral therapy regimen increases treatment failure risk.
In resource-limited settings, combination antiretroviral therapy (ART) regimens are commonly standardized and prescribed without testing for drug resistance mutations (DRMs). If people with human immunodeficiency virus (PWH) experience treatment failure (TF) on their original regimen (first-line therapy), they transition to a new regimen (second-line therapy). DRMs can be detected in the population and have the potential to impact future treatment options. An ever-increasing population of PWH are on ART and will need effective ART regimens for decades to come; as such, it is important to understand the prevalence and impact of DRMs in TF [1].
HIV type 1 (HIV-1) DRMs can either be selected within a person due to subinhibitory drug concentrations or they can be transmitted. Insufficient drug concentrations to fully suppress viral replication can lead to viral rebound and expand/fix existing DRMs or drive development of DRMs de novo. Some regimens are especially susceptible to DRM development; certain nucleoside reverse transcriptase inhibitor (NRTI) and nonnucleoside reverse transcriptase inhibitor (NNRTI) drugs have low genetic barriers to the development of DRMs [2]. DRMs can be high-frequency (defined for this study as present in ≥20% of the viral population within a patient) or low-frequency (<20%). High-frequency DRMs have been shown to correlate with TF, a pattern reliably demonstrated across many regimens and multiple drug classes. In contrast, the impact of low-frequency DRMs on TF is still debated. Some studies suggest that low-frequency pre-ART DRMs are associated with increased TF risk on some NNRTI-based regimens [3–10], while other studies find no correlation between low-frequency pre-ART DRMs and TF [11–17].
To investigate the role of pretherapy high- and low-frequency DRMs in TF, we worked with a large longitudinal cohort of women with HIV-1 in Malawi. This cohort was divided into 3 subcohorts based on treatment history and drug regimen. We used a unique molecular identifier (UMI)–based next-generation sequencing (NGS) strategy, multiplexed Primer ID MiSeq (MPID), for accurate detection of low-frequency DRMs [18]. We analyzed DRMs at ART initiation for all participants. For participants who experienced TF, we also analyzed DRMs at the point of TF and compared DRM frequencies pre- and post-ART to identify changes that occurred during therapy. Grouping the participants by their treatment regimen (NNRTI-based regimen or protease inhibitor [PI]–based regimen), we found that the NNRTI DRM K103N was correlated with increased risk of TF when it was present at either high or low frequency pre-ART in participants receiving an NNRTI-based regimen. Accurate NGS revealed the presence of low-frequency K103N in 10% of participants.
METHODS
Participants and Specimen Collection
All study participants were enrolled through an antenatal clinic in Lilongwe, Malawi, between 2015 and 2020. The participants were divided into 3 cohorts based on ART history: cohort A, ART-naive participants starting first-line ART; cohort B, ART-experienced participants transitioning to second-line ART after first-line ART failure; and cohort C, participants resuming first-line ART after treatment interruption. Cohorts A and C received a 2NRTI + NNRTI regimen of tenofovir disoproxil fumarate (TDF)/lamivudine (3TC)/efavirenz (EFV), while cohort B received a 2NRTI + PI regimen of TDF/3TC/atazanavir/ritonavir (ATV/r) or zidovudine/3TC/ATV/r. At study entry and every 6 months thereafter, participants donated plasma, which was tested to measure HIV-1 viral load (VL). TF was defined as either suppression followed by rebound, or failure to suppress (>500 copies/mL) by 6 months. A cutoff of >500 copies/mL was chosen to avoid misclassifying participants with brief blips in viremia who went on to have continued suppression. This study was approved by the institutional review board at the University of North Carolina at Chapel Hill and the national health sciences research committee in Malawi, and we obtained written consent from participants.
Sequencing and Bioinformatics
We sequenced viral RNA from entry pre-ART plasma specimens, and for participants who experienced TF we also sequenced the first available post-TF viremic plasma specimens. We used the previously published MPID-NGS protocol to sequence HIV-1 protease (PR), reverse transcriptase (RT), and integrase (IN) coding regions [18]. MiSeq sequencing data were processed by the tcs pipeline (v2.5.2) to construct a template consensus sequence (TCS) for each viral template [18]. Surveyed DRMs included the full extended lists of NNRTI, NRTI, PI, and IN surveillance DRMs from the Stanford HIV database. DRMs were listed as their frequencies in the viral population. To distinguish true DRMs from sequencing artifacts, we calculated the false discovery rate (FDR)–adjusted P value for each detected DRM [19]. DRMs with FDR P values ≤.05 were considered true DRMs. Precise sampling depth was calculated by counting the total number of TCSs. HIV subtyping was performed using RT region consensus sequences, based on the phylogenetic grouping of participant sequences with reference genomes of different HIV subtypes.
Statistical Analysis
Entry characteristics of the study population were summarized using descriptive statistics. Categorical variables were compared using Fisher exact test, and continuous variables were compared using Mann-Whitney U test (for comparing 2 groups) or Kruskal-Wallis test (for comparing >2 groups). We used a Cox proportional hazards model (CoxPH) to compare accumulated incidence of TF between cohorts. We built CoxPH models to explore the association of DRMs at high frequencies (≥20%) and low frequencies (3%–20%) in the viral RNA population with TF. The 3% minimum cutoff for low-frequency DRM detection was selected because this was the average minimum frequency for participants at which we could detect DRMs with 95% confidence. In the analysis of low-frequency DRMs, we only included participants with validated viral population sampling depth for the detection of DRMs as low as 3%. Stratification of viral load levels was used in the CoxPH model [20].
RESULTS
Participants
A total of 360 participants with adequate sequencing depth (defined as ≥10 TCS for the RT region) were included in the analysis: 195 in cohort A, 77 in cohort B, and 88 in cohort C. Further participant characteristics are summarized in Table 1. The median age of the participants was 27 years. VLs were comparable across all 3 cohorts. CD4+ T-cell counts were significantly lower in cohort B, but comparable between cohorts A and C. There were more late-stage HIV-1 infections in cohort B than in cohorts A or C. Participants were infected with subtype C HIV-1, with 3 exceptions: 2 subtype A (0.6%) and 1 subtype D (0.3%). The study follow-up period lasted approximately 3.2 years, with participants in cohort A having a significantly longer follow-up time compared to those in the other cohorts. Additional demographic data can be found in Supplementary Table 1.
Table 1.
Characteristics of Participants at Study Entry
| Characteristic | Cohort A: Treatment Naive, First-line (n = 195) | Cohort B: Treatment Experienced, Second-line (n = 77) | Cohort C: Restart, First-line (n = 88) | Total (N = 360) | P Value |
|---|---|---|---|---|---|
| Age at baseline, years, median (IQR) | 26 (23–30) | 30 (26–33) | 26 (23–31) | 27 (23–31) | <.001 |
| WHO stage, No. (%) | .002 | ||||
| 1 | 184 (94.4) | 63 (81.8) | 79 (89.8) | 326 (90.6) | |
| 2 | 8 (4.1) | 4 (5.2) | 5 (5.7) | 17 (4.7) | |
| 3 | 3 (1.5) | 6 (7.8) | 4 (4.5) | 13 (3.6) | |
| 4 | 0 (0.0) | 4 (5.2) | 0 (0.0) | 4 (1.1) | |
| CD4 cell count, cells/µL, median (IQR) | 342 (227–498) | 231 (113–431) | 340 (230–519) | 320 (196–498) | .002 |
| Viral load, log10 copies/mL, median (IQR) | 4.36 (3.86–4.76) | 4.25 (3.61–4.65) | 4.44 (3.97–4.88) | 4.35 (3.79–4.76) | .104 |
| Follow-up, days, median (IQR) | 1200 (1104–1231) | 928 (646–1184) | 620 (470–990) | 1134 (750–1216) | <.001 |
| Treatment failure, No. (%) | 50 (25.6) | 30 (39.0) | 42 (47.7) | 122 (33.9) | .002 |
| No. of TCSs at RT region, median (IQR) | 1147 (49–327) | 79 (39–166) | 94.5 (47–293) | 102 (46.5–300) | .237 |
| Detection sensitivity (%) at RT, median (IQR) | 3.1 (1.1–7.4) | 4.6 (2.2–9.0) | 3.5 (1.2–7.5) | 3.6 (1.2–7.5) | .228 |
| No. of TCSs at PR region, median (IQR) | 159 (44–686) | 137 (31–380) | 222 (64–476) | 164 (42.5–599) | .384 |
| Detection sensitivity at PR, %, median (IQR) | 1.9 (0.4–6.1) | 1.90 (0.4–8.4) | 1.4 (0.7–5.40) | 1.80 (0.50–5.80) | .98 |
Demographic, clinical, and sequencing characteristics of participants in all 3 cohorts separately and in aggregate. Detection sensitivity is the minimum drug resistance mutation (DRM) frequency at which there is 95% chance of detecting a DRM in a viral population given the sequencing depth achieved. Continuous variables were compared using Kruskal-Wallis rank-sum test. Categorical variables were compared with Fisher exact test with simulated P value (based on 500 replicates).
Abbreviations: IQR, interquartile range; PR, protease; RT, reverse transcriptase; TCS, template consensus sequence; WHO, World Health Organization.
Entry DRMs
Sequencing depth varied between participant specimens but did not vary significantly between cohorts. Across cohorts, variation in TCS numbers between amplicon regions resulted in a median minimum detection sensitivity of 3.6% frequency for RT mutations and 1.8% frequency for PR mutations (Table 1). In this study, frequency always refers to abundance of a DRM in the viral population of 1 participant, while prevalence always refers to what percentage of participants carry a detected DRM.
To compare cohorts, we analyzed entry DRMs present in participants in each cohort before starting (or restarting) an ART regimen. Overall, DRMs present at entry differed between the cohorts in both frequency and prevalence. Participants in treatment naive cohort A tended to carry DRMs at lower frequencies compared to the participants in the drug-experienced cohorts B and C. Cohort B carried substantially higher prevalence of DRMs than either A or C (Figure 1). Entry DRMs relevant to the prescribed first- and second-line regimens are summarized in Supplementary Tables 2–4 by drug category.
Figure 1.
Prevalence and frequency of drug resistance mutations (DRMs) present at study entry. Selected DRMs are included based on clinical relevance or strong drug neutralization. For each DRM, drug class specificity is indicated. Frequency refers to abundance of a DRM within a participant, and prevalence refers to what precent of a population carries a DRM. Prevalence for each cohort is represented by columns. Median frequency for a given DRM across all participants with that DRM in each cohort is represented by color, with higher frequencies assigned darker colors. Abbreviations: DRM, drug resistance mutation; INSTI, integrase strand transfer inhibitor; NNRTI, nonnucleoside reverse transcriptase inhibitor; NRTI, nucleoside reverse transcriptase inhibitor; PI, protease inhibitor.
Three important NRTI and NNRTI entry DRMs showed differences between the cohorts. K65R was prevalent in the treatment experienced cohort B (31%) and fixed at high within-host frequencies (median frequency at 100% when detected). In cohorts A and C, K65R was detected in approximately 2% of participants, and at a much lower intra-host frequency. M184V was present at low frequency in only 1 cohort A participant, while it was fixed at a median 100% frequency in all cohort B and C participants carrying it (B: 44%; C: 7%). High-frequency entry K103N was prevalent across all 3 cohorts (Figure 1). Other NNRTI DRMs were also seen, especially in cohort B. Entry PI DRMs were rare. These mutation differences are consistent with the presumed previous drug exposure of the cohorts.
Accumulated Incidence of TF
We next asked how ART history impacted TF outcomes in the cohorts, comparing accumulated incidence of TF between cohorts A, B, and C using a Kaplan-Meier analysis (Figure 2). Approximately 10% of cohorts A and B and 20% of cohort C never suppressed (failure at 180 days). Cohort B was significantly more likely to experience TF than cohort A (multivariable hazard ratio [HR], 2.17 [95% confidence interval {CI}, 1.36–3.47]), as was cohort C (multivariable HR, 3.58 [95% CI, 2.31–5.54]). Cohort C was associated with higher risk of TF when compared with cohort B (multivariable HR, 1.76 [95% CI, 1.07–2.89]). Entry VL also impacted TF risk. Every 10-fold increase in entry VL increased the risk of TF by 1.69-fold (95% CI, 1.29–2.21). However, we did not see a significant association between entry CD4 value and the occurrence of TF (P = .37).
Figure 2.
Accumulated incidence of treatment failure (TF) over time. Kaplan-Meier analysis comparing incidence of TF for the 3 cohorts. TF was defined as either suppression followed by rebound or failure to suppress (>500 copies/mL) by 6 months. Participants who do not ever reach suppression are recorded as failing at day 180. Abbreviation: ART, antiretroviral therapy.
High-Frequency DRMs and Their Association With TF
High-frequency entry DRMs (≥20% intra-host frequency) against the drug regimen are known to increase TF risk; thus, our next step was to confirm this finding in the cohorts using the CoxPH model to analyze the impact of high-frequency DRMs. Because both cohorts A and C received the same regimen, we combined these 2 cohorts for the analysis. We found that K103N, which was present in 39 of 281 participants, was the only high-frequency entry DRM to significantly increase TF risk (multivariable HR, 2.62 [95% CI, 1.57–4.40]) (Table 2). Further analysis showed that the presence of dual-class DRMs (K103N and M184V or K65R in the same person) did not increase the risk for TF, although this analysis is limited due to the small sample size of the participants with dual-class DRMs (n = 4). All cohort B participants transitioned from a failing NNRTI-based regimen to a PI-based regimen immediately upon study entry, and entry PI DRMs were extremely rare (Supplementary Table 3), so the analysis focused only on NRTI DRMs K65R and M184V. However, neither of them were associated with TF in multivariable analysis (Table 2). Overall, high-frequency RT DRM K103N was the only DRM we observed to increase TF risk for an NNRTI-containing regimen.
Table 2.
Association of High-Frequency Nucleoside Reverse Transcriptase Inhibitor and Nonnucleoside Reverse Transcriptase Inhibitor Drug Resistance Mutations at Entry With Virologic Failure
| DRMs at Entry | Present | No. (%) | Univariable HR (95% CI) | Multivariable HR (95% CI) |
|---|---|---|---|---|
| Cohorts A and C combined | ||||
| K65R | No | 280 (99.6) | … | … |
| Yes | 1 (0.4)a | … | … | |
| M184V | No | 276 (98.2) | … | … |
| Yes | 5 (1.8) | 1.83 (.45–7.46), P = .40 | 11.53 (.33–7.24), P = .59 | |
| K103N | No | 240 (85.4) | … | … |
| Yes | 41 (14.6) | 2.87 (1.77–4.65), P < .001 | 2.62 (1.57–4.40), P < .001 | |
| Y181C | No | 274 (97.5) | … | … |
| Yes | 7 (2.5) | 1.33 (.42–4.19), P = .631 | 0.86 (.26–2.91), P = .814 | |
| Cohort B | ||||
| K65R | No | 55 (71.4) | … | … |
| Yes | 22 (28.6) | 0.54 (.22–1.32), P = .176 | 1.02 (.37–2.86), P = .965 | |
| M184V | No | 46 (59.7) | … | … |
| Yes | 31 (40.3) | 0.29 (.12–.71), P = .007 | 0.39 (.14–1.09), P = .072 | |
High-frequency DRMs are defined as DRMs present at ≥20% frequency in a viral population, cutoff chosen based on the limit of reliable detectability with Sanger sequencing. Included in analysis are nonnucleoside/nucleoside reverse transcriptase inhibitor DRMs defined as conferring high-level drug resistance in the Stanford Drug Resistance database to the current regimen. The HR for each DRM is calculated for both univariable analysis (not considering other DRMs) and multivariable analysis (taking incidence of other DRMs into account). All multivariable analyses are stratified by entry viral load (VL) into 3 categories: <10 000, 10–100 000, and >100 000 copies/mL, to account for entry VL impact on treatment failure risk. Values in bold are statistically significant. Cohort A: Treatment naive, with the regimen of tenofovir disoproxil fumarate (TDF)/lamivudine (3TC)/efavirenz (EFV). Cohort B: Treatment experienced, with the regimen TDF (or zidovudine)/3TC/atazanavir/ritonavir. Cohort C: Treatment experienced, with the regimen TDF/3TC/EFV.
Abbreviations: CI, confidence interval; DRM, drug resistance mutation; HR, hazard ratio.
aFor n = 1, we did not calculate HR.
Low-Frequency DRMs and Their Association With TF
The primary goal of this study was to determine whether low-frequency DRMs at entry could lead to TF or increase TF risk. Here we examined the relationship between TF risk and low-frequency entry DRMs using the CoxPH model. Analysis only included participants with sufficient sequencing depth to detect DRMs down to a 3% frequency (n = 137 participants). In this analysis, the presence of the DRMs was divided into 2 levels, high and low frequency, using the 20% frequency cutoff. As in the previous section, cohort B was analyzed separately. For cohort B, neither low-frequency entry K65R nor M184V DRMs were associated with TF. For cohorts A and C, K103N was the only low-frequency entry DRM that significantly increased TF risk (multivariable HR, 2.38 [95% CI, 1.00–5.67]) (Table 3). Thus, on an NNRTI-containing regimen, presence of K103N at any frequency correlated with increased TF risk. An important feature of this analysis is that K103N had the highest prevalence of any DRM in cohorts A and C, giving this analysis the greatest sensitivity. The use of an accurate deep sequencing method allowed the detection of low-frequency K103N in approximately 10% of the participants in cohorts A and C who had not previously experienced TF (Table 3).
Table 3.
Association of Low-Frequency Nucleoside and Nonnucleoside Reverse Transcriptase Inhibitor Drug Resistance Mutations at Entry With Treatment Failure
| DRMs at Baseline | Frequency categorya | No. (%) | Univariable HR (95% CI) | Multivariable HR (95% CI) |
|---|---|---|---|---|
| Cohorts A and C combined | ||||
| K65R | Absent | 133 (97.1) | … | … |
| Low-freq | 4 (2.9) | 1.53 (.37–6.30), P = .557 | … | |
| M184V | Absent | 134 (97.8) | … | … |
| High-freq | 2 (1.5) | 3.03 (.41–22.29), P = .276 | … | |
| Low-freq | 1 (0.7)b | … | … | |
| K103N | Absent | 100 (73.0) | … | … |
| High-freq | 24 (17.5) | 3.00 (1.55–5.80), P = .001 | 3.12 (1.58–6.18), P = .001 | |
| Low-freq | 13 (9.5) | 2.51 (1.10–5.76), P = .029 | 2.38 (1.00–5.67), P = .049 | |
| V106M | Absent | 125 (91.2) | … | … |
| High-freq | 2 (1.5) | 0.00 (.00–∞), P = .997 | … | |
| Low-freq | 10 (7.3) | 1.91 (.76–4.82), P = .172 | … | |
| Y181C | Absent | 130 (94.9) | … | … |
| High-freq | 5 (3.6) | 1.15 (.28–4.76), P = .842 | … | |
| Low-freq | 2 (1.5) | 1.03 (.14–7.46), P = .979 | … | |
| G190A | Absent | 134 (97.8) | … | … |
| High-freq | 2 (1.5) | 5.28 (1.27–22.03), P = .022 | … | |
| Low-freq | 1 (0.7)b | … | … | |
| Cohort B | ||||
| K65R | Absent | 16 (61.5) | … | … |
| High-freq | 8 (30.8) | 0.69 (.22–2.18), P = .532 | 1.10 (.23–5.12), P = .906 | |
| Low-freq | 2 (7.7) | 1.40 (.18–11.25), P = .749 | 5.29 (.32–86.93), P = .243 | |
| M184V | Absent | 16 (61.5) | … | … |
| High-freq | 8 (30.8) | 0.41 (.12–1.46), P = .169 | 0.47 (.09–2.50), P = .380 | |
| Low-freq | 2 (7.7) | 0.00 (.00–∞), P = .998 | 0.00 (.00–∞), P = .999 | |
In this analysis, only participants with sufficient sequencing depth to detect DRMs down to 3% frequency were included (n = 137). Low-frequency DRMs are defined as DRMs between the 3% minimum detection limit and the 20% high-frequency cutoff. Selected DRMs are included due to possible relevance to treatment failure. For each included DRM, HRs were listed for participants who had that DRM at high or low frequency compared to participants without the DRM. Univariable HR is calculated for all included DRMs; multivariable HR is calculated to take incidence of other DRMs into account only for DRMs that have a significant HR in univariable analysis. All multivariable analyses are stratified by entry viral load (VL) into 3 categories: <10 000, 10–100 000, and >100 000 copies/mL, to account for entry VL impact on treatment failure risk. Values in bold are statistically significant. Cohort A: Treatment naive, with the regimen of tenofovir disoproxil fumarate (TDF)/lamivudine (3TC)/efavirenz (EFV). Cohort B: Treatment experienced, with the regimen TDF (or zidovudine)/3TC/atazanavir/ritonavir. Cohort C: Treatment experienced, with the regimen TDF/3TC/EFV.
Abbreviations: CI, confidence interval; DRM, drug resistance mutation; HR, hazard ratio.
a“Absent” refers to participants who do not carry the mutation, “High-freq” refers to participants who carry the DRM at high frequency, and “Low-freq” refers to participants who carry the DRM at low frequency.
bFor n = 1, we did not calculate HR.
DRM Frequency Changes From ART Initiation to TF
Next, in order to assess the possible confounding role of drug exposure in TF, we examined DRM frequency changes over the course of ART for participants who experienced TF. There are 2 main paradigms for TF, which drive different trends in the changes in DRM frequency. The first, TF due to a preexisting DRM even in the presence of sufficient drug levels, will likely result in a DRM becoming fixed to near 100% frequency at failure. The second, TF due to inconsistent drug exposure, will not fix DRMs [21, 22]. Very low levels or absence of drug exposure may even result in the apparent reversion and/or archiving of a DRM with wild type growing out, making it difficult to detect the entry DRM at TF.
Participants were divided into 3 categories based on changes in DRM profiles between entry and TF. Category 1 was defined as clear selective pressure with regimen-relevant DRMs newly fixed at 100% at TF, with failure clearly linked to drug resistance. Category 2 was defined as a clear absence of selective pressure from ART, with drug resistance not driving failure, and included cases of TF with an absence of high-frequency DRMs or loss of a fixed DRM. Category 3 included participants with trends that were indeterminant for category 1 or 2. The 90 participants for whom sequencing data were available at entry and TF were included in this analysis (Figure 3).
Figure 3.
Characterization of drug resistance mutation (DRM) profile changes over the course of treatment. DRM profiles from study entry and post–treatment failure (TF) were compared for each participant with sufficient next-generation sequencing data to detect DRMs down to 3% frequency (n = 90). For cohorts A and C, changes in DRMs relevant to the 2NRTI/NNRTI regimen were analyzed: K103N, K65R, M184V, G190A, V106M, and Y181C. For cohort B, DRM changes relevant to the 2NRTI/PI regimen were analyzed: K65R, M184V, G48VM, I50V, I84V, and N88S. Changes between pre–antiretroviral therapy (ART) and post-ART profiles were categorized as either 1, 2, or 3. Ninety-five percent was the cutoff for fixed DRMs, 20% was the cutoff between low and high-frequency DRMs, and 3% was the minimal detection limit. Category 1 includes TF participants who have clear drug regimen selective pressure at TF. Category 2 includes TF participants who have clear absence of drug regimen selective pressure at TF. Category 3 includes TF participants with inconclusive DRM change patterns. A, Chart depicting DRM change categories for all TF participants included in analysis. B, Table of category breakdowns for TF participants by cohort and further detail on category 3 subcategories. C, Category 1 breakdowns by cohort and fixed DRMs at TF.
Most participants had a clear primary driver of TF, either evidence of strong drug selective pressure or evidence of low drug exposure (Figure 3A). The most frequent driver of TF across all cohorts was low (or no) drug exposure (category 2), representing between 49% and 67% of participants experiencing TF across the cohorts (Figure 3B). Cohort A (therapy naive) participants had the highest proportion of participants who failed therapy with evidence of drug selective pressure (category 1) at 36%, compared to cohorts B and C, but the trend was not statistically significant (P = .20) (Figure 3B). Cohort C had the lowest impact of drug selection, suggesting that their TF might be largely driven by nonadherence. For category 1 DRM-driven TF participants, K103N was commonly seen for cohorts A and C, consistent with our finding that K103N conveys increased TF risk at both high and low frequency. In cohort B, 4 participants had the M184V mutation fixed at the time of TF (Figure 3C).
DISCUSSION
In this study we investigated the question of whether low-frequency DRMs at therapy initiation can increase TF risk, using a stringent method of sequencing to reliably detect low-frequency DRMs with a validated level of sensitivity. Most clinically relevant NNRTI/NRTI DRMs, when present at entry at low frequency, did not change TF risk, with the NNRTI DRM K103N as the notable exception—low-frequency entry K103N significantly increased TF risk in participants starting a 2NRTI/NNRTI-based ART regimen.
Previous studies addressing the role of low-frequency DRMs in TF have come to a variety of conclusions [23]. Some studies have found no association between low-frequency DRMs and TF, while others have reported a correlation between low-frequency DRMs and TF on some NRTI/NNRTI regimens. There are multiple factors that may contribute to these inconsistent outcomes, with an important difference being the utilization of different sequencing platforms that vary dramatically in their ability to accurately detect low-level DRMs [9]. Also, different studies utilize a range of cutoff benchmarks and stringencies, which confounds comparison [24]. With regard to study design and sequencing methods, few NGS studies leverage UMI methods, so reported sequencing depth estimates and detection sensitivity limits may be unreliable, and many studies did not include longitudinal sampling or had limited cohort size, hindering analysis of entry DRM impact on treatment outcome [4, 16, 25–28]. Additionally, in our cohorts, different major DRMs had diverse profiles, trends, and impacts on TF risk when present as low-frequency entry DRMs, which likely has contributed to the inconsistency of the literature. Despite ongoing studies in the field, these and other possible factors have undercut efforts to conclusively determine whether low-frequency DRMs impact treatment outcome.
We were able to work with longitudinal samples from a large cohort of women, enabling tracking of DRM changes to infer the cause of failure and the possible role of low-frequency DRMs. We determined that even low-frequency entry DRM K103N correlates with increased TF risk for participants taking an TDF/3TC/EFV (NRTI/NRTI/NNRTI) regimen, although other low-frequency NRTI/NNRTI DRMs were not associated with a detectable increased TF risk on the same regimen. This K103N finding agrees with several reports which also found that low-frequency NNRTI DRMs at ART initiation with an NRTI/NRTI/NNRTI regimen can increase TF risk [3, 5–10], but is at odds with other studies that did not find that low-level DRMs confer TF risk [11–17]. Our results agree with the findings of a meta-analysis [3], which found that study participants with low-frequency entry DRMs undetectable by Sanger sequencing (with 20% sensitivity of detection) had a >2-fold increase in TF risk. Another study, by Beck et al, examined different NNRTI regimens and found that low-frequency entry DRMs were associated with TF on a nevirapine regimen, but not for an EFV regimen [5], whereas we found that K103N associated with TF on the EFV regimen. Milne et al found that the TF risk conferred by entry DRMs was dose dependent; both low- and high-frequency entry DRMs significantly increased TF risk, but high-frequency entry DRMs much more so [8]. In our study, high-frequency K103N also had a greater HR than low-frequency K103N for the risk of TF (3.12 vs 2.38). It is important to highlight that the majority of the studies mentioned earlier did not specify the particular NNRTI DRM associated with TF. Instead, these studies investigated the overall association of all DRMs within the NNRTI category with TF. However, different NNRTI DRMs may impact the treatment outcome differently. For instance, the Y181C mutation confers >50-fold reduced susceptibility to nevirapine, but only results in <2-fold reduced susceptibility to EFV [29]. Thus the impact of Y181C on an EFV-based regimen may not be as severe as that of K103N. In the present study, NNRTI DRMs were frequently observed at entry, with K103N detected in approximately one-third of the participants. This provided us with an opportunity to examine this specific NNRTI DRM and its influence on TF.
ART adherence is a necessary part of successful therapy but is affected by many challenges [30]. In the 3 cohorts, >50% of participants experiencing TF had DRM changes consistent with failure not driven by DRMs, indicating high overall inconsistent adherence, as noted in previous observations [31]. We cannot completely deconvolute the impact of adherence on our results, which is a limitation of our study, but one that is ever-present for most studies [32–34]. Of note, the presence of DRMs in the treatment-experienced participants might indicate past adherence problems resulting in fluctuating drug levels. These participants might have experienced TF faster than others because of the combination of the presence of DRM at the switch/restart of the therapy, and the unaddressed adherence issues.
While our cohort is a specific population, our conclusions about the role of low-frequency DRMs in TF are likely to be broadly relevant. Our data support the ongoing transition from a K103N-susceptible 2NRTI/NNRTI first-line regimen to an integrase strand transfer inhibitor (INSTI)–based regimen [28]. INSTI-based regimens have a higher barrier to the development of resistance. However, widespread low adherence over time can still foster DRMs and undercut the regimen, and NNRTI DRMs can also impact INSTI-based regimens [15]. Overall, our results indicate that entry DRM K103N increases TF risk when present at both low and high frequency on an NRTI/NNRTI regimen, but that more than half of TF is not driven by resistance to the drugs, which points to the important real-world role of adherence in therapy success and failure. Overall in this setting, using NGS for DRM detection may benefit an additional 10% of the people receiving NNRTI-based regimens, as they have low-frequency NNRTI DRMs that would be missed by Sanger sequencing.
Supplementary Data
Supplementary materials are available at The Journal of Infectious Diseases online (http://jid.oxfordjournals.org/). Supplementary materials consist of data provided by the author that are published to benefit the reader. The posted materials are not copyedited. The contents of all supplementary data are the sole responsibility of the authors. Questions or messages regarding errors should be addressed to the author.
Supplementary Material
Contributor Information
Rachel M Burdorf, Lineberger Comprehensive Cancer Center; Department of Microbiology and Immunology, University of North Carolina at Chapel Hill.
Shuntai Zhou, Lineberger Comprehensive Cancer Center; Department of Microbiology and Immunology, University of North Carolina at Chapel Hill.
Claire Amon, Lineberger Comprehensive Cancer Center.
Nathan Long, Lineberger Comprehensive Cancer Center; Department of Microbiology and Immunology, University of North Carolina at Chapel Hill.
Collin S Hill, Lineberger Comprehensive Cancer Center.
Lily Adams, Lineberger Comprehensive Cancer Center; Department of Microbiology and Immunology, University of North Carolina at Chapel Hill.
Gerald Tegha, UNC Project–Malawi, Lilongwe.
Maganizo B Chagomerana, UNC Project–Malawi, Lilongwe; Department of Medicine, University of North Carolina at Chapel Hill.
Allan Jumbe, UNC Project–Malawi, Lilongwe.
Madalitso Maliwichi, UNC Project–Malawi, Lilongwe.
Shaphil Wallie, UNC Project–Malawi, Lilongwe.
Yijia Li, Department of Medicine, University of Pittsburgh Medical Center, Pennsylvania.
Ronald Swanstrom, Lineberger Comprehensive Cancer Center; Department of Biochemistry and Biophysics, University of North Carolina at Chapel Hill.
Mina C Hosseinipour, UNC Project–Malawi, Lilongwe; Department of Medicine, University of North Carolina at Chapel Hill.
Notes
Acknowledgments. The authors are grateful for the generosity of the participants who provided the blood samples used in this study. We also acknowledge the contribution of the University of North Carolina (UNC) high-throughput sequencing facility.
Financial support. This work was supported by the National Institute of Allergy and Infectious Diseases at the National Institutes of Health (grant number R01-AI140970 to R. S.) and the National Institute of Child Health and Human Development (grant number R01-HD080485 to M. C. H.). This research received infrastructure support from the UNC Center for AIDS Research (grant number P30-AI050410) and the UNC Lineberger Comprehensive Cancer Center (grant number P30-CA016086).
References
- 1. Hamers RL, Rinke de Wit TF, Holmes CB. HIV drug resistance in low-income and middle-income countries. Lancet HIV 2018; 5:e588–96. [DOI] [PubMed] [Google Scholar]
- 2. Perrier M, Visseaux B, Landman R, et al. No impact of HIV-1 protease minority resistant variants on the virological response to a first-line PI-based regimen containing darunavir or atazanavir. J Antimicrob Chemother 2018; 73:173–6. [DOI] [PubMed] [Google Scholar]
- 3. Li JZ, Paredes R, Ribaudo HJ, et al. Low-frequency HIV-1 drug resistance mutations and risk of NNRTI-based antiretroviral treatment failure: a systematic review and pooled analysis. JAMA 2011; 305:1327–35. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Ávila-Ríos S, García-Morales C, Matías-Florentino M, et al. Pretreatment HIV-drug resistance in Mexico and its impact on the effectiveness of first-line antiretroviral therapy: a nationally representative 2015 WHO survey. Lancet HIV 2016; 3:e579–91. [DOI] [PubMed] [Google Scholar]
- 5. Beck IA, Levine M, McGrath CJ, et al. Pre-treatment HIV-drug resistance associated with virologic outcome of first-line NNRTI-antiretroviral therapy: a cohort study in Kenya. EClinicalMedicine 2020; 18:100239. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Li JZ, Paredes R, Ribaudo HJ, et al. Impact of minority nonnucleoside reverse transcriptase inhibitor resistance mutations on resistance genotype after virologic failure. J Infect Dis 2013; 207:893–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Li JZ, Paredes R, Ribaudo HJ, et al. Relationship between minority nonnucleoside reverse transcriptase inhibitor resistance mutations, adherence, and the risk of virologic failure. AIDS 2012; 26:185–92. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Milne RS, Silverman RA, Beck IA, et al. Minority and majority pretreatment HIV-1 drug resistance associated with failure of first-line nonnucleoside reverse-transcriptase inhibitor antiretroviral therapy in Kenyan women. AIDS 2019; 33:941–51. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Simen BB, Simons JF, Hullsiek KH, et al. Low-abundance drug-resistant viral variants in chronically HIV-infected, antiretroviral treatment–naive patients significantly impact treatment outcomes. J Infect Dis 2009; 199:693–701. [DOI] [PubMed] [Google Scholar]
- 10. Hassan AS, Bibby DF, Mwaringa SM, et al. Presence, persistence and effects of pre-treatment HIV-1 drug resistance variants detected using next generation sequencing: a retrospective longitudinal study from rural coastal Kenya. PLoS One 2019; 14:e0210559. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Chimukangara B, Giandhari J, Lessells R, et al. Impact of pretreatment low-abundance HIV-1 drug-resistant variants on virological failure among HIV-1/TB-co-infected individuals. J Antimicrob Chemother 2020; 75:3319–26. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Messiaen P, Verhofstede C, Vandenbroucke I, et al. Ultra-deep sequencing of HIV-1 reverse transcriptase before start of an NNRTI-based regimen in treatment-naive patients. Virology 2012; 426:7–11. [DOI] [PubMed] [Google Scholar]
- 13. Porter DP, Daeumer M, Thielen A, et al. Emergent HIV-1 drug resistance mutations were not present at low-frequency at baseline in non-nucleoside reverse transcriptase inhibitor-treated subjects in the STaR study. Viruses 2015; 7:6360–70. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Raymond S, Nicot F, Pallier C, et al. Impact of human immunodeficiency virus type 1 minority variants on the virus response to a rilpivirine-based first-line regimen. Clin Infect Dis 2018; 66:1588–94. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Siedner MJ, Moorhouse MA, Simmons B, et al. Reduced efficacy of HIV-1 integrase inhibitors in patients with drug resistance mutations in reverse transcriptase. Nat Commun 2020; 11:5922. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Su B, Zheng X, Liu Y, et al. Detection of pretreatment minority HIV-1 reverse transcriptase inhibitor-resistant variants by ultra-deep sequencing has a limited impact on virological outcomes. J Antimicrob Chemother 2019; 74:1408–16. [DOI] [PubMed] [Google Scholar]
- 17. Zoufaly A, Jochum J, Hammerl R, et al. Virological failure after 1 year of first-line ART is not associated with HIV minority drug resistance in rural Cameroon. J Antimicrob Chemother 2015; 70:922–5. [DOI] [PubMed] [Google Scholar]
- 18. Zhou S, Jones C, Mieczkowski P, Swanstrom R. Primer ID validates template sampling depth and greatly reduces the error rate of next-generation sequencing of HIV-1 genomic RNA populations. J Virol 2015; 89:8540–55. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Zhou S, Hill CS, Spielvogel E, Clark MU, Hudgens MG, Swanstrom R. Unique molecular identifiers and multiplexing amplicons maximize the utility of deep sequencing to critically assess population diversity in RNA viruses. ACS Infect Dis 2022; 8:2505–14. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Armenia D, Di Carlo D, Cozzi-Lepri A, et al. Very high pre-therapy viral load is a predictor of virological rebound in HIV-1-infected patients starting a modern first-line regimen. Antivir Ther 2019; 24:321–31. [DOI] [PubMed] [Google Scholar]
- 21. von Wyl V, Klimkait T, Yerly S, et al. Adherence as a predictor of the development of class-specific resistance mutations: the Swiss HIV Cohort Study. PLoS One 2013; 8:e77691. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Rupérez M, Noguera-Julian M, González R, et al. HIV drug resistance patterns in pregnant women using next generation sequence in Mozambique. PLoS One 2018; 13:e0196451. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Mbunkah HA, Bertagnolio S, Hamers RL, et al. Low-abundance drug-resistant HIV-1 variants in antiretroviral drug-naive individuals: a systematic review of detection methods, prevalence, and clinical impact. J Infect Dis 2020; 221:1584–97. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24. Parkin NT, Avila-Rios S, Bibby DF, et al. Multi-laboratory comparison of next-generation to sanger-based sequencing for HIV-1 drug resistance genotyping. Viruses 2020; 12:694. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Mathobo P, Matume ND, Bessong PO. Application of next generation sequencing in HIV drug resistance studies in Africa, 2005–2019: a systematic review. Sci Afr 2021; 12:e00829. [Google Scholar]
- 26. El Bouzidi K, Datir RP, Kwaghe V, et al. Deep sequencing of HIV-1 reveals extensive subtype variation and drug resistance after failure of first-line antiretroviral regimens in Nigeria. J Antimicrob Chemother 2022; 77:474–82. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. Nicot F, Sauné K, Raymond S, et al. Minority resistant HIV-1 variants and the response to first-line NNRTI therapy. J Clin Virol 2015; 62:20–4. [DOI] [PubMed] [Google Scholar]
- 28. Nguyen T, Fofana DB, Lê MP, et al. Prevalence and clinical impact of minority resistant variants in patients failing an integrase inhibitor-based regimen by ultra-deep sequencing. J Antimicrob Chemother 2018; 73:2485–92. [DOI] [PubMed] [Google Scholar]
- 29. Melikian GL, Rhee S-Y, Varghese V, et al. Non-nucleoside reverse transcriptase inhibitor (NNRTI) cross-resistance: implications for preclinical evaluation of novel NNRTIs and clinical genotypic resistance testing. J Antimicrob Chemother 2014; 69:12–20. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30. Kweyamba M, Buregyeya E, Kusiima J, Kweyamba V, Mukose AD. Loss to follow-up among HIV positive pregnant and lactating mothers on lifelong antiretroviral therapy for PMTCT in rural Uganda. Adv Public Health 2018; 2018:7540587. [Google Scholar]
- 31. Haas AD, Msukwa MT, Egger M, et al. Adherence to antiretroviral therapy during and after pregnancy: cohort study on women receiving care in Malawi's option B+ program. Clin Infect Dis 2016; 63:1227–35. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32. Bezabhe WM, Chalmers L, Bereznicki LR, Peterson GM. Adherence to antiretroviral therapy and virologic failure: a meta-analysis. Medicine (Baltimore) 2016; 95:e3361. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33. Byrd KK, Hou JG, Hazen R, et al. Antiretroviral adherence level necessary for HIV viral suppression using real-world data. J Acquir Immune Defic Syndr 2019; 82:245–51. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34. Dalmat Ronit R, Makhsous N, Pepper Gregory G, et al. Limited marginal utility of deep sequencing for HIV drug resistance testing in the age of integrase inhibitors. J Clin Microbiol 2018; 56:e01443–18. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.