Characterization of minority HIV-1 drug resistant variants in the United Kingdom following the verification of a deep sequencing-based HIV-1 genotyping and tropism assay

A total of 109 convenience plasma samples were obtained during routine patient monitoring from two well-characterized cohorts of HIV-infected individuals at St. George’s University Hospitals Healthcare NHS Foundation Trust (London, United Kingdom) and at NHS Lothian (Edinburgh, United Kingdom), with the written informed consent of each participant as described in local clinical research protocols. Clinical, virological, and demographics data, including HIV-1 drug resistance results based on standard Sanger sequencing performed locally and/or in reference laboratories testing, were obtained from patient care databases at the respective hospitals (Additional file 1: Table S1, Additional file 2: Table S2).

HIV-1 drug resistance and co-receptor tropism was determined using an all-inclusive deep sequencing-based assay, DEEPGEN™, as described [19]. Briefly, plasma viral RNA was purified and three RT-PCR products corresponding to the gag-p2/NCp7/p1/p6/pol-PR/RT- (1657 bp fragment), pol-IN- (1114 bp fragment), and env-C2V3- (480 bp fragment) coding region of HIV-1 amplified. These amplicons were purified, quantified, and used to construct a multiplexed library for shotgun sequencing on the Ion Personal Genome Machine (PGM, ThermoFisher Scientific). Reads were mapped and aligned against sample-specific reference sequences constructed for the gag-p2/NCp7/p1/p6/pol-PR/RT/IN or env-gp120 HIV-1 genomic regions using the DEEPGEN™ Software Tool Suite v2 (Alouani and Quiñones-Mateu, unpublished) as described [19]. Plasma samples were classified as containing non-R5 viruses if at least 2% of the individual sequences, as determined by deep sequencing, were predicted to be non-R5 [42, 43]. In this study, minority variants were defined as amino acid substitutions detected at ≥ 1% (based on the intrinsic error rate of the system [19]) and < 20% of the virus population, corresponding to those mutations that cannot be determined using population sequencing [14‐18].

Three consensus sequences, corresponding to the three amplicons (PR/RT, INT, and C2V3), were generated for each patient-derived virus, aligned using ClustalW [44] and their phylogeny reconstructed using the neighbor-joining statistical method as implemented within MEGA 6.06 [45]. HIV-1 subtype, initially predicted by phylogenetic analysis, was confirmed using pol-PR/RT/INT and env-V3 sequences with the DEEPGEN™ Software Tool Suite v2 and Geno2Pheno tools (http://​www.​geno2pheno.​org). Inter-patient genetic distances were determined using the Maximum Composite Likelihood model with bootstrap as the variance estimation method (1000 replicates) within MEGA 6.06 [45]. Intra-patient HIV-1 quasispecies diversity was determined using all three PR/RT-, INT-, and C2V3- coding regions based on the p-distance model as described for deep sequencing [46].

Descriptive results are expressed as median values, standard deviations, range, and confidence intervals. The non-parametric Kruskal–Wallis one-way analysis of variance test was used to compare the mutations detected among the different groups. All differences with a p value of < 0.05 were considered statistically significant. The kappa coefficient, calculated using ComKappa2 v.2.0.4 [47], was used to quantify the concordance between HIV-1 coreceptor tropism determinations. The kappa coefficient calculates a chance-adjusted measure of the agreement between any number of categories, in this case HIV-1 coreceptor tropism determined by the same assay in two different locations. All statistical analyses were performed using GraphPad Prism v.6.0b (GraphPad Software, La Jolla, CA) unless otherwise specified. gag-p2/NCp7/p1/p6/pol-PR/RT/IN and env-C2V3 nucleotide sequences obtained by deep sequencing in this study have been submitted to the Los Alamos National Laboratory HIV-db Next Generation Sequence Archive (http://​www.​hiv.​lanl.​gov/​content/​sequence/​HIV/​NextGenArchive/​Silver2018).

As described above, for this study we selected 109 plasma samples from HIV-1 patients being monitored at two hospitals in the United Kingdom: 59 from St. George’s University Hospitals Healthcare NHS Foundation Trust (St. George’s) and 50 from NHS Lothian (Table 1 and Additional file 1: Table S1, Additional file 2: Table S2). Participants from both institutions had similar median ages, i.e., 44 years (interquartile range, IQR: 35–51) and 42 years (IQR: 27–47) for St. George’s and NHS Lothian, respectively. Male gender was more frequent in the cohort of patients from NHS Lothian (68%) than in St. George’s (47%). Although not significant, median CD4⁺ T-cell counts were slightly lower in NHS Lothian patients (290 cells/mm³, IQR: 106-485) than in St. George’s patients (340 cells/mm³, IQR: 157–490), corresponding in both cases to relatively high median plasma HIV-1 RNA loads (4.62 log₁₀, IQR: 3.81–5.15 and 4.76 log₁₀, IQR: 4.36–5.19 for St. George’s and NHS Lothian patients, respectively). Most of the HIV-infected individuals from St. George’s were antiretroviral-experienced (45/59, 76%), while 56% (28/50) of NHS Lothian patients were antiretroviral-naïve. At the time of the study, most patients on treatment had received a median of 2.3 PIs (IQR: 0-5), all were treated with NRTIs (median 2.9 and 3.6 NRTIs for St. George’s and NHS Lothian patients, respectively), approximately half of the patients in each cohort were exposed to NNRTIs (24/45 and 12/22), and around one-third to INSTIs (17/45 and 6/22). Only four individuals from St. George’s were treated with the entry inhibitor maraviroc (Table 1, details in Additional file 1: Table S1, Additional file 2: Table S2).

Our deep sequencing-based HIV-1 genotyping and coreceptor tropism assay (DEEPGEN™) has been characterized and validated for clinical use in the US. since late 2013 [19]. Here we transferred the technology to two independent clinical laboratories in the U.K. (St. George’s and NHS Lothian) to evaluate assay performance and the feasibility to detect minority HIV-1 drug resistance variants in different populations of HIV-infected individuals. Each clinical laboratory multiplexed their respective samples into three Ion 318 chips with median loading efficiencies of 65% and 61%, generating a total of 11,831,865 and 11,542,222 quality reads, with equal median read lengths of 226 bp for St. George’s and NHS Lothian, respectively. Although comparable, the average sequencing coverage at each nucleotide position varied with each sample and HIV-1 genomic region analyzed, with no significant differences between laboratories, i.e., PR/RT/INT (mean 7205 and 8878 reads) and V3 (16,893 and 20,218 reads) for St. George’s and NHS Lothian, respectively (Fig. 1). More importantly, these metrics ensured the minimum coverage of 1000 per nucleotide position sequenced required guaranteeing the detection of a minor variant present at least at 1% of the population [48].

All 109 plasma samples from both cohorts of patients (St. George’s and NHS Lothian) were originally analyzed using standard Sanger-based HIV-1 genotyping in the respective clinical laboratories. Altogether a total of 157 mutations (129 and 28 in St. George’s and NHS Lothian cohorts, respectively) in positions associated with drug resistance were detected by Sanger sequencing (i.e., 44 in the protease, 93 in the RT, and 20 in the integrase) (Fig. 2a). As expected, all the drug resistance mutations identified by Sanger sequencing were also detected using DEEPGEN™, while 280 additional drug resistance mutations (120 and 160 in the St. George’s and NHS Lothian cohorts, respectively) were detected only by deep sequencing (i.e., 80 in the protease, 168 in the RT, and 32 in the integrase) (Fig. 2a). This difference in the numbers of drug resistance mutations detected by Sanger and deep sequencing—in both institutions—was significant, even when the mutations were quantified by drug class, ranging from 1.4- to 12-fold additional mutations detected by deep sequencing compared to Sanger sequencing (Paired t test, p < 0.0001 to p = 0.029) (Fig. 2a).

Overall, a similar number of drug resistance mutations were identified using Sanger sequencing and DEEPGEN™ with a mutation frequency threshold of ≥ 20%, resulting in comparable HIVdb mutation scores; however, the HIVdb scores were consistently higher for most antiretroviral drugs when DEEPGEN™ was used with a mutation frequency of ≥ 1% (Fig. 2b). In fact, no significant differences were observed in the HIVdb scores determined by Sanger or DEEPGEN™ using mutation frequencies ≥ 20% when all drug classes (PI, NRTI, NNRTI, and INSTI) were compared for both cohorts of patients (Fig. 2c). As expected, significantly higher HIVdb scores were obtained -for all antiretroviral drugs- using DEEPGEN™ with mutation frequencies ≥ 1% compared to Sanger or DEEPGEN™ with mutation frequencies ≥ 20%, i.e., 26- and twofold (PI), 16- and sixfold (NRTI), four- and twofold (NNRTI), and four- and twofold (INSTI) for St. George’s and NHS Lothian, respectively (Paired t test, p < 0.01 to p < 0.0001) (Fig. 2c).

Although the clinical significance of minority variants is still on debate, here we assumed equal impact on predicted drug resistance of a mutation detected by Sanger sequencing or DEEPGEN™ with mutation frequency thresholds of ≥ 20%, ≥ 5%, or ≥ 1%, the HIVdb scores to infer the levels of susceptibility to the different antiretroviral drugs. As observed in Fig. 3, similar drug resistance profiles (susceptible, low-level/intermediate, or high-level resistance) were obtained using Sanger and DEEPGEN™ with mutation frequencies ≥ 20%; however, a few additional mutations detected at frequencies between 5 and 20% increased the resistance level to certain NRTIs, NNRTIs and/or PIs in viruses from seven St. George’s and nine NHS Lothian’s patients. This effect was more dramatic when all mutations with frequencies ≥ 1% were included. A total of 25 (42.4%) and 21 (42%) HIV-infected individuals in the St. George’s and NHS Lothian cohorts, respectively, showed some kind of resistance HIV-1 genotype (reduced susceptibility) determined by Sanger or DEEPGEN™ with mutation frequencies ≥ 20%; however, these numbers increased to 44/59 (74.6%) and 42/50 (84%) using DEEPGEN™ with mutation frequencies ≥ 1% (Fig. 3), which correlated with the antiretroviral drugs listed in their treatment histories (Supp. Tables 1 and 2).

Finally, we used DEEPGEN™ to quantify the frequency of CCR5- or CXCR4-tropic variants and determine HIV-1 coreceptor tropism in both cohorts of patients. HIV-1 tropism based on Sanger sequencing had been determined in six patients from St. George’s, all classified as being infected with R5 viruses (Fig. 4). Interestingly, DEEPGEN™ was able to corroborate the R5 tropism in 5/6 viruses while in one patient X4 HIV-1 variants were detected at low frequency (i.e., 5.1%), changing the tropism determination to dual- or mixed-tropic (D/M-tropic). Overall, 35% of the St. George’s patients were infected with D/M-tropic viruses, with the frequency of X4 variants ranging from 5.1 to 100% within the HIV-1 population (Fig. 4). In the case of the NHS Lothian, only 16% (8/50) of the patients harbored D/M-tropic HIV-1 strains, with X4 variants ranging from 10.9 to 100% (Fig. 4).

Following the implementation of DEEPGEN™ in the clinical laboratories at St. George’s and NHS Lothian, and the successful test of 59 and 50 clinical HIV-1 samples in the respective institutions, 32 of these samples (16 from each group) were sent to the University Hospitals Translational Laboratory (UHTL, Cleveland, Ohio, USA) to complete the verification of the assay in the U.K. laboratories. After testing all 32 samples with DEEPGEN™ in the UHTL, we quantified the number of mutations -and their frequency in the population- determined in the UHTL and compared them to the values obtained in the U.K. (St. George’s and NHS Lothian). As expected, strong significant correlations were observed when the two sets of 16 sequences were compared, even after segregating the mutations per drug class, with r values ranging from 0.995 to 0.999 (p < 0.0001, Pearson coefficient correlation) (Fig. 5a). We next quantified the number of drug resistance mutations detected using different mutation frequency thresholds for DEEPGEN™ (≥ 1%, ≥ 5%, or ≥ 20%) in all three laboratories. With the exception of a slight difference between St. George’s and UHTL in the number of drug resistance mutations quantified at ≥ 1% (mean 4.1 vs. 6.2 mutations, p = 0.029, Mann–Whitney), no significant differences were observed when the number of mutations associated with drug resistance were quantified in the U.S. or in the U.K. (Fig. 5b). More importantly, no difference was observed in the drug resistance profiles determined using the HIVdb algorithm (Fig. 5c). Finally, a perfect agreement (100% concordance, κ = 1) was observed comparing HIV-1 coreceptor tropism determinations based on V3 sequences obtained in the U.K. (St. George’s and NHS Lothian) and in the U.S. (UHTL) (Fig. 5d).

As described above, we implemented and verified the use of our deep sequencing-based HIV-1 genotyping and coreceptor tropism assay (DEEPGEN™) in the U.K. to study minority HIV-1 drug resistant variants in patients from London (St. George’s) and Edinburgh (NHS Lothian). Figure 6 summarizes all the primary and compensatory drug resistance mutations, and their frequency in the HIV-1 quasispecies, identified by DEEPGEN™. As expected, the number and type of mutations with frequencies > 20% (those within Sanger sequencing-based detection level) matched the cART history of the patient, e.g., the virus from patient SG28 had mutations associated with resistance to ABC (L74I 97.6% frequency), 3TC (M184V 99.7%), and EFV (K103N 98.8%, P225H 99.5%) and had been exposed to RTV, DRV, ABC, 3TC, and EFV; while patient SG86 had a virus with resistance to FTC (M184V 98.9%) and RAL (L74M 97.9%, E92Q 86.1% and T97A 99.8%) after being treated over the years with RTV, LPV, DRV, FTC, TDF, and RAL (Additional file 1: Table S1). Similar drug resistance profiles were observed for individuals experiencing virologic failure from both cohorts of patients (data not shown). More interestingly, a number of minority mutations at frequencies below the Sanger threshold (~ 20%) were detected in a multitude of individuals (Fig. 6). In the 59 patients from St. George’s, 87, 126, 66, and 55 (334 total) mutations associated with resistance to PIs, NRTIs, NNRTIs, or INSTIs were detected, respectively; of those, 18, 68, 31, and 14 (131 total) were present at frequencies below 20%. That means that approximately 40% (131/334) of the total number of drug resistance mutations from these patients were present in minority HIV-1 variants. On the other hand, in the NHS Lothian cohort with a higher number of cART-naïve patients (Table 1), we observed a reduced number of drug resistance mutations compared to the St. George’s cohort (167 vs. 334, respectively), although with a slightly higher proportion of drug resistance minority mutations, i.e., 66% (111/167) (Fig. 6), most likely a consequence of the reduced number of majority (> 20% frequency) drug resistance mutations identified in this cohort of patients.

The distribution of minority drug resistance mutations per drug class was different in both cohorts of patients. A higher proportion of minority PI-, NRTI-, and INSTI-resistance mutations was detected in NHS Lothian patients compared to individuals from St. George’s, i.e., 33/41 (80.5%) vs. 18/87 (20.7%), 53/70 (75.7%) vs. 68/126 (54%), and 9/21 (42.9%) vs. 14/55 (25.5%), respectively (unpaired t test, p < 0.0001) (Fig. 6). The most prevalent primary minority drug resistance mutations observed in both cohorts of patients were the PIs M46I/L (in 3 and 9 SG and NHS individuals, respectively) and I50 V (2 and 5); the NRTIs D67N (23 and 20), K65R (0 and 9), L74I (2 and 0), M184V/I (6 and 0) and K219Q (9 and 10); and the NNRTI L100I (6 and 3). Although a few minority primary and compensatory INSTI mutations were observed in viruses from both cohort of patients, none of them were observed at a particularly relevant prevalence (Fig. 6). Most of these minority drug resistance mutations were detected in the 67 cART-experienced individuals, e.g., the HIV-1 genotype of the following patients consisted of a mixture of majority and minority drug resistance mutations: patient SG79 (PR V11I 99.9%, K20I 99.9%; RT E44D 1.3%, D67N 1.1%, L100I 1.7%, M184V 84.9%, M184I 14.8%, K219Q 2.1%, INT E157Q 99.8%), patient SG86 (RT M184V 99.8%, INT L74M 97.9%, L74I 1.9%, E92Q 86.1%, E92V 7.4%, T97A 99.8%, G163R 1.1%), patient NHS49 (PR E35G 1.3%, M46L 3.1%, I50 V 2.5%, F53L 1.4%, RT M41L 99.2%, D67N 2.3%, M184V 15.4%, T215Y 99.1%) and patient NHS72 (PR M46L 1.1%, RT D67N 1.4%, K103R 99.6%, M184V 99.8%, Y188C 19.3%, INT T66I 99.8%, L74I 99.8%, T97A 98.8%, E157Q 8.7%). On the other hand, 27 of the 42 cART-naïve patients carried viruses with a mixture of primary and compensatory minority drug resistance mutations. A range of 1–3 minority mutations were detected in half of the cART-naïve individuals in the St. George’s cohort (7/14). Interestingly, a higher proportion of cART-naïve NHS Lothian patients (20/28) had viruses with minority drug resistance mutations, ranging from 1 to 8 mutations per patient. Most of the minority mutations in viruses from both groups of naïve patients were observed in the RT, e.g., M41L, E44D, A62V, K65R, D67N, D67G, V75I, L100I, K103N, K103R, V188I, M184I, L210W, K219Q, Y318F, etc., although a number of minority mutations associated with resistance to PI (L10F, V11I, M46I/L, I50V, F53L, Q58E, T74S) or INSTI (L74I/M, E92G, E138K, S147G, G163R, Q148K) were also identified (Fig. 6). Finally, it is interesting to highlight that most HIV-1 drug resistance mutations were observed at the ends of the spectrum in the HIV-1 population, with only a small fraction of mutations detected at frequencies between > 5 and < 90% (Fig. 6). This was more accentuated in PI and NRTI mutations from both group of viruses, ranging from 1/87 to 4/41 mutations in this mutation frequency range, while a slightly higher number of NNRTI and INSTI mutations were observed in this “transitional phase”, ranging from 8/66 to 12/35, respectively (Fig. 6).

As described above, DEEPGEN™ is based on deep sequencing viral RNA extracted from plasma samples and optimized to accurately detect minority HIV-1 variants above a 1% frequency level in the HIV-1 population [19]. Moreover, this methodology is capable of generating over 10,000 HIV-1 sequences (reads) per patient that can be used to analyze inter- and intra-patient HIV-1 genetic diversity [11, 41, 49]. Phylogenetic analyses confirmed the HIV-1 subtype initially determined for each patient-derived virus with the DEEPGEN™ Software Tool Suite v2 and Geno2Pheno. HIV-1 subtyping classification varied slightly depending on the HIV-1 genomic region analyzed (pol or C2V3); however, based on the most broadly used C2V3 region [19, 50, 51], the HIV-1 subtypes identified in these patients included: A1 (20), B (15), C (7), CFR02_AG (6), G (4), A2 (3), D (2), and F1 (2) in St. George’s cohort and B (33), C (8), A1 (4), F1 (3), CRF02_AG (1), and CFR01_AE (1) in NHS Lothian’s patients (Fig. 7a and Additional file 1: Table S1, Additional file 2: Table S2).

As expected, mean inter-patient genetic distances calculated with the consensus C2V3 sequences were higher than those calculated with the PR/RT and INT sequences (Fig. 7b). Viruses from St. George’s patients were significantly more diverse than viruses from NHS Lothian’s individuals comparing C2V3 (mean 0.265 vs. 0.215 substitutions per site, p < 0.0001 unpaired t test) and PR/RT (0.111 vs. 0.098 s/site, p < 0.0001 unpaired t test) sequences but not when comparing INT sequences (0.084 vs. 0.081 s/site, p = 0.082 unpaired t test), respectively (Fig. 7b). Finally, intra-patient HIV-1 diversity was also determined using all three PR/RT-, INT-, and C2V3- coding regions based on the p-distance model as described for deep sequencing [46]. Although slightly higher in patients from NHS Lothian compared with viruses from St. George’s individuals, no significant difference was observed in the HIV-1 quasispecies diversity of these patients, i.e., PR/RT (1.212 vs. 0.968), INT (1.013 vs. 0.817), and C2V3 (2.761 vs. 2.446), respectively (Fig. 7c).

Given the considerable amount of data that we were able to accumulate from deep sequencing patient-derived HIV-1 sequences from these HIV-infected individuals, i.e., majority (frequency > 20%) and minority (frequency < 20% and > 1%) drug resistance mutations, susceptibility to antiretroviral drugs (HIVdb scores), coreceptor tropism, subtyping, inter-patient and intra-patient viral diversity, we decided to investigate potential associations among any of these metrics and clinical parameters, mainly plasma HIV RNA load, CD4⁺ T-cell counts, and antiretroviral therapy history. As expected, HIVdb scores determined using only majority, or including minority, drug resistance mutations correlated significantly with ART history in patients from St. George’s (r = 0.51, p < 0.0001 or r = 0.58, p < 0.0001 Pearson coefficient correlation) and NHS Lothian (r = 0.37, p = 0.007 or r = 0.45, p = 0.001). On the other hand, no significant association was observed in most pairwise comparisons of the multiple virological metrics and clinical parameters studied (data not shown). For example, no significant correlation was observed between HIVdb scores determined using only majority (r = 0.13, p = 0.30 or r = 0.01, p = 0.98) or including minority (r = 0.12, p = 0.38 or r = 0.10, p = 0.47) drug resistance mutations, ART history (r = 0.22, p = 0.08 or r = 0.13, p = 0.35) or intra-patient HIV-1 diversity (r = 0.11, p = 0.39 or r = 0.10, p = 0.46) with plasma HIV RNA load in individuals from St. George’s or Lothian cohorts. However, inverse significant correlations were observed between the number of drugs in the ART history (r = − 0.22, p = 0.01 and r = − 0.43, p = 0.001) or intra-patient HIV-1 diversity (r = − 0.30, p = 0.01 and r = − 0.26, p = 0.01) and CD4⁺ T-cell counts in St. George’s and Lothian patients, respectively (data not shown).