Transmitted HIV-1 Drug Resistance in a Large International Cohort Using Next-Generation Sequencing

Results From the Strategic Timing of Antiretroviral Treatment (START) Study

JD Baxter; D Dunn; A Tostevin; RL Marvig; M Bennedbæk; A Cozzi-Lepri; S Sharma; MJ Kozal; M Gompels; AN Pinto; J Lundgren


HIV Medicine. 2021;22(5):360-371. 

In This Article


Study Participant Baseline Characteristics

The START study enrolled 4684 ART-naïve individuals from 35 countries between April 2009 and December 2013.[14] Europe had the highest number of participants (1539), followed by Latin America (1174), Africa (999), USA (507), Asia (356) and Australia (109). Of the 4072 individuals with viral load > 1000 copies/mL (and thus eligible for this study), 3785 (93%) had a baseline specimen analysed by NGS. Baseline characteristics of these individuals (see Table 1) showed a median CD4 count of 643 cells/μL, median HIV RNA of 18 105 copies/mL, and median time since diagnosis of 0.96 years. Using the criteria described in the Methods section, 2901 (77%) specimens produced an evaluable result for the assessment of protease TDR, 2180 (58%) for NRTI/NNRTI TDR, and 1338 (35%) for integrase TDR. Median (IQR) read depth was 13 134 (5644–26 671) in protease, 6891 (3078–14 194) in RT, and 2240 (987–5213) in IN. Subtype B was the most common subtype, followed by subtype C (Table 1).

Prevalence of Transmitted Drug Resistance

Figure 1 shows the prevalence of TDR, overall and by geographical region, at different detection thresholds. Overall, at the 2% threshold, TDR prevalence was 9.2% for NRTI DRMs, 9.2% for NNRTI DRMs, 11.4% for PI DRMs, and 3.5% for INSTI DRMs. Using a 5% threshold, the respective values were 5.6%, 6.6%, 5.5% and 1.6%; and using a 20% threshold the respective values were 3.2%, 4.9%, 2.4% and 0.1%. Comparing the 2% and 20% thresholds of detection for TDR DRMs by drug class, there was a 4.8 (11.4%/2.4%) and 35-fold (3.5% vs. 0.1%) greater proportion of viruses with PI and INSTI minor variants, respectively, whereas this ratio was 2.9 and 1.9 for TDR DRMs in the NRTI and NNRTI drug classes, respectively.

A comparatively high prevalence of NNRTI DRMs was observed in the USA (mainly mutations above the 20% threshold), as was a comparatively high prevalence of NRTI DRMs in Australia (mainly mutations above the 5% threshold). Otherwise, there was no clear geographical variation in the prevalence of DRMs for any drug class, with the apparent variation in integrase TDR probably due to the small number of cases. Nevertheless, there was some evidence of geographical variability in the level of DRM variants, conditional on being above the 2% threshold: P = 0.05 for NRTI TDR; P = 0.007 for NNRTI; P = 0.11 for PI TDR. The NRTI effect was mainly driven by a relative deficit of variants at 5–20% in Europe, and the NNRTI effect by a relative excess of variants at 2–5% in Asia and Europe, and a relative excess at 5–20% in Africa. There also was some evidence that TDR prevalence differed by HIV-1 subtype (data not shown), but these were reflective of the geographical distribution of different subtypes.

Table 2 shows the results of a logistic regression analysis of selected predictors for drug class-specific DRMs (2% detection threshold). This confirmed the visual impression of regional variability in the prevalence NRTI DRMs (P = 0.002) and NNRTI DRMs (P = 0.02) that was observed in Figure 1. The prevalence of PI DRMs was strongly associated with age (although lacking a clear trend) and weakly associated with gender; however, these associations were not observed in sensitivity analyses limited to mutant variants detected above the 5% or 20% threshold (results not shown). There was a weak suggestion of an overall increase in the prevalence of NNRTI DRMs (but not NRTI DRMs or PI DRMs) over the 4.5-year recruitment period to START. Calendar time trends stratified by geographical region were also examined but revealed no notable findings.

Specific DRMs observed in at least 0.5% of samples (at the 2% detection threshold) are shown in Table 3. The most common NRTI DRMs were T215 revertants (2.5%), M41L (2.2%), and K219QENR (1.7%). The T215 revertants and, to a lesser extent, M41L were mostly high-level variants representing the majority of the quasispecies (i.e. occurring in > 80% of the viral population), whereas a wider range of variant frequency was observed for K219QENR. Notably, the K65R mutation was not observed in any sample, even at the 2% threshold. M184V and M184I were detected in seven and 18 samples respectively, the latter generally as a low-level variant (2–5%). The most common NNRTI DRMs – K103NS (3.5%), G190ASE (3.1%), and E138K (1.6%) – showed diverse patterns. K103NS variants were mainly observed (60/77, 78%) in > 80% of the viral quasispecies, E138K mainly (20/34, 59%) in < 5%, and G190ASE displayed a more uniform spread. Estimates of PI TDR are strongly influenced by the M46IL mutation, which was observed in 5.5% of all samples, mostly as low-level variants (Table 3). Excluding M46IL variants would reduce estimated PI TDR (at the 2% threshold) from 11.4% to 6.6%. The D30N mutation was mainly (41/46, 89%) present as a low-level variant below 20%, whereas most (12/18, 67%) L90M were detected in > 80% of the viral quasispecies. All individual INSTI DRMs were observed below the 20% threshold, with the exception of G140S (detected at the 25% level, sample from Spain collected in 2012) and G140A (21%, sample from Peru in 2011).

The majority of DRMs detected at each threshold occurred as solitary mutations in a single drug class for individual participant samples. There were a relatively small number of samples that had more than one DRM within a drug class or mutations present in multiple drug classes. For individual samples where TDR DRMs were detected at the 2% threshold, multiple within-class DRMs occurred in 18.5% of those with NRTI DRMs, 13.4% with NNRTI DRMs, 13.5% with PI DRMs, and 4.3% with INSTI DRMs. Of those with TDR DRMs detected at the 2% threshold in both RT and PR, multi-class resistance with at least one NRTI, NNRTI and PI DRM occurred in only three participant samples.

We further examined (but did not formally analyse) specific DRMs by geographical region, which indicated marked variability (Table S4). For example, K103NS was the most common NNRTI mutation in the USA and Latin America, whereas G190ASE was the most common NNRTI mutation in Europe, Africa and Asia. This would appear to be the main explanation for the geographical variability in the level of DRM variants that was observed in Figure 1.

Predicted Phenotypic Drug Susceptibility

Analysis of predicted phenotypic resistance (low level, intermediate, and high level) for selected antiretroviral agents according to threshold of detection (Figure 2) resulted in several important findings. First, for certain drugs, particularly nelfinavir and raltegravir, using a 2% rather than a 20% variant threshold dramatically increases estimates of predicted phenotypic resistance (much more than drug class-specific TDR estimates). This is due to the wider range of mutations incorporated in the Stanford HIVdb algorithm than in the WHO or Stanford drug mutation surveillance lists. For example, the IN mutations 92G, 138K, and 163R, which were generally observed at the 2–5% variant level by NGS, individually predict low-level resistance to raltegravir, but are not counted on the Stanford drug mutation surveillance list. Conversely, resistance to efavirenz is mainly predicted by the K103NS mutations. As these mainly occur at high variant levels (Table 3), the frequency of predicted resistance is largely unaffected by the variant threshold that is used. Second, there was minimal predicted resistance to darunavir and dolutegravir regardless of the detection threshold.

Figure 2.

Predicted phenotypic drug susceptibility by detection threshold. (a) Above 2% detection threshold; (b) above 5% threshold; (c) above 20% threshold. Predicted susceptibility is based on Stanford HIVdb algorithm v.8.6. Data are also shown in tabular form in Table S2. 3TC, lamivudine; FTC, emtricitabine.

Finally, the frequency of tenofovir resistance was estimated to be 1.3% at the 20% threshold, 1.6% at the 5% threshold, and 2.4% at the 2% threshold, despite the complete absence of the K65R mutation. This is explained by the contribution of thymidine analogue mutations to predicted tenofovir resistance.