Validation of an Oligonucleotide Ligation Assay for Quantification of Human Immunodeficiency Virus Type 1 Drug-Resistant Mutants by Use of Massively Parallel Sequencing

Abstract

Global HIV treatment programs need sensitive and affordable tests to monitor HIV drug resistance. We compared mutant detection by the oligonucleotide ligation assay (OLA), an economical and simple test, to massively parallel sequencing. Nonnucleoside reverse transcriptase inhibitor (K103N, V106M, Y181C, and G190A) and lamivudine (M184V) resistance mutations were quantified in blood-derived plasma RNA and cell DNA specimens by OLA and 454 pyrosequencing. A median of 1,000 HIV DNA or RNA templates (range, 163 to 1,874 templates) from blood specimens collected in Mozambique (n = 60) and Kenya (n = 51) were analyzed at 4 codons in each sample (n = 441 codons assessed). Mutations were detected at 75 (17%) codons by OLA sensitive to 2.0%, at 71 codons (16%; P = 0.78) by pyrosequencing using a cutoff value of ≥2.0%, and at 125 codons (28%; P < 0.0001) by pyrosequencing sensitive to 0.1%. Discrepancies between the assays included 15 codons with mutant concentrations of ∼2%, one at 8.8% by pyrosequencing and not detected by OLA, and one at 69% by OLA and not detected by pyrosequencing. The latter two cases were associated with genetic polymorphisms in the regions critical for ligation of the OLA probes and pyrosequencing primers, respectively. Overall, mutant concentrations quantified by the two methods correlated well across the codons tested (R² > 0.8). Repeat pyrosequencing of 13 specimens showed reproducible detection of 5/24 mutations at <2% and 6/6 at ≥2%. In conclusion, the OLA and pyrosequencing performed similarly in the quantification of nonnucleoside reverse transcriptase inhibitor and lamivudine mutations present at >2% of the viral population in clinical specimens. While pyrosequencing was more sensitive, detection of mutants below 2% was not reproducible.

INTRODUCTION

HIV drug resistance is associated with an increased risk of therapeutic failure, with a decrease in therapeutic options, and transmission of HIV, including drug-resistant variants. Access to antiretroviral therapy (ART) in resource-limited settings has increased in the past decade, with concomitant increases in the transmission of drug-resistant HIV in some communities (1–3). Because alternative ART regimens are limited in low-income countries, the emergence and spread of high levels of HIV drug resistance could compromise the effectiveness of national HIV treatment programs. Increases in incident infections with drug-resistant viruses led HIV treatment guideline panels in the United States (4) and Europe (5, 6) to recommend that HIV-infected individuals be tested for antiretroviral drug-resistant virus prior to initiation or modification of ART. HIV genotyping is routinely performed by Sanger sequencing of plasma HIV RNA; however, this technology is too costly and technically demanding to implement in most low-resource laboratories and clinics. Furthermore, the sensitivity of consensus sequencing is limited for minor variants constituting <25% of the viral population (7, 8), although low-frequency HIV drug resistance mutations, particularly involving resistance to nonnucleoside reverse transcriptase (RT) inhibitors (NNRTIs), have been associated with increased risks of virological failure with first-line ART (9–15). Therefore, sensitive and affordable assays are needed to monitor HIV drug resistance as part of global HIV treatment programs.

The oligonucleotide ligation assay (OLA) is a sensitive and specific point-mutation assay for detection of HIV drug resistance to protease inhibitors (PIs), nucleoside reverse transcriptase inhibitors (NRTIs), and NNRTIs (16–19). Because mutant- and wild-type-specific probes are hybridized and ligated at a relatively low temperature, the OLA accommodates viral polymorphisms in the regions of the probes to a greater degree than do allele-specific PCR assays (20, 21), and the assay is technically less complex than the LigAmp assay (22). The OLA is especially well suited for use in resource-limited areas due to its low cost and its use of relatively inexpensive laboratory equipment and because only a limited number of drug resistance mutations need to be assessed.

Massively parallel sequencing technologies, such as 454 pyrosequencing (23), also can detect and quantify low-frequency drug resistance variants in HIV populations from individuals (24–26). In this study, we compared the detection of drug resistance mutations by the OLA versus 454 pyrosequencing, to evaluate the sensitivity of the OLA to detect and to quantify mutants in clinical specimens across different HIV subtypes and across HIV codons associated with NNRTI and lamivudine (2′,3′-dideoxy-3′-thiacytidine [3TC]) resistance.

MATERIALS AND METHODS

Specimens.

Leftover specimens from two separate studies conducted in Kenya and Mozambique were selected based on the availability of ∼1,000 viral templates. Plasma specimens from 51 ART-naive Kenyans were collected in 2006 (27). Dried blood spots (DBS) or whole-blood specimens (n = 60) from 14 infants and 22 women exposed to single-dose nevirapine, with or without maternal zidovudine treatment, were collected in Mozambique between 2005 and 2008 (28). To analyze a wide range of drug-resistant mutant concentrations, follow-up specimens collected over a period of 4 to 12 months from 19/36 Mozambican subjects were included in this study.

Nucleic acid extraction and PCR amplification of HIV pol.

HIV RNA from Kenyan subjects was extracted from 0.5 to 1 ml of pre-ART plasma using the NucliSENS miniMAG kit (bioMérieux, Durham, NC, USA). cDNA was generated by reverse transcription using random hexamers and Superscript III (Invitrogen, Grand Island, NY, USA), and the amplifiable copies of HIV-1 were quantified with an HIV-1 long terminal repeat (LTR) real-time PCR assay (29). HIV DNA from Mozambican subjects was extracted from whole-blood samples (DNA ArchivePure DNA purification system; 5 PRIME, Inc., Gaithersburg, MD) or was amplified directly from dried whole-blood specimens spotted on FTA filter paper cards (Whatman, Inc., Florham Park, NJ) (30). Amplifiable copies of HIV DNA were quantified from either extracted DNA or DBS specimens with an HIV gag real-time PCR assay adapted to FTA paper (28).

Approximately 1,000 viral templates from each specimen were subjected to nested PCRs for pol using the first-round primers GARAGACAGGCTAATTTTTTAGGGA (forward, HXB2 positions 2071 to 2059) and AAYTTCTGTATATCATTGACAGTCCA (reverse, HXB2 positions 3303 to 3328) for the Kenyan specimens or CCTACACCTGTCAACATAATTGG (forward, HXB2 positions 2487 to 2509) and the same reverse primer for the Mozambican specimens. First-round PCR mixtures included 400 nM each primer, 200 μM each deoxynucleoside triphosphate (dNTP), 1.8 mM MgCl₂, and 2.5 U of FastStart High Fidelity DNA polymerase (Roche Diagnostics, Mannheim, Germany). Reaction conditions consisted of 2 min at 95°C and 35 cycles of 95°C for 30 s, 55°C for 30 s, and 72°C for 1 min, followed by a final extension of 7 min at 72°C.

Aliquots of the first-round product were used in separate second-round PCRs for OLA and 454 pyrosequencing. For OLA, 2 μl of the first-round reaction mixture was diluted 1:20 with water and added to a mixture containing 200 μM dNTPs, 3 mM MgCl₂, 2.5 U of MyTaq DNA polymerase (Bioline USA, Inc., Taunton, MA), and 400 nM inner primers to amplify codons 1 to 239 of HIV-1 reverse transcriptase (RT) (forward, Kenya, CAAATCACTCTTTGGCARCGACC [HXB2 positions 2556 to 2278]; Mozambique, AATTAAAGCCAGGAATGGATGG [HXB2 positions 2581 to 2602]; reverse, CAYTTGTCAGGATGGAGTTCATA [HXB2 positions 3243 to 3265]). PCR conditions were as described for first-round assays except that extension was at 72°C for 30 s. For 454 pyrosequencing, two regions of HIV-1 pol were amplified in separate second-round PCRs with the FastStart High Fidelity PCR system and one set of the following fusion primers containing the 454 universal adaptors A and B, with or without one of 14 multiplex identifiers (MIDs) (454 Sequencing technical bulletin 013-2009; Roche, Nutley, NJ): Kenya amplicon 1 (RT codons 21 to 134), CGTATCGCCTCCCTCGCGCCATCAG-MID-GTTAAACAGTGGCCATTGACAGA (forward) and CTATGCGCCTTGCCAGCCCGCTCAG-MID-ACTAGGTATGGTGAATGCAGTATA (reverse); Kenya amplicon 2 (RT codons 150 to 242), CGTATCGCCTCCCTCGCGCCATCAG-MID-CACAGGGATGGAAAGGATCAC (forward) and CTATGCGCCTTGCCAGCCCGCTCAG-MID-CTGGACTGTCCATYTGTCAGGATG (reverse); Mozambique amplicon 1 (RT codons 22 to 134), CGTATCGCCTCCCTCGCGCCATCAGAAACAATGGCCATTRACAGAAGA (forward) and CTATGCGCCTTGCCAGCCCGCTCAGACTAGGTATGGTRAATGCAGTATA (reverse); and Mozambique amplicon 2 (RT codons 150 to 242), CGTATCGCCTCCCTCGCGCCATCAGCACAGGGRTGGAAAGGRTCAC (forward) and CTATGCGCCTTGCCAGCCCGCTCAGCTGTACTGTCCATTTRTCAGGATG (reverse). Each second-round PCR mixture included 2 μl of first-round reaction mixture in a mixture containing 200 μM dNTPs, 1.8 mM MgCl₂, 400 nM each primer, and 2.5 U of FastStart High Fidelity DNA polymerase and was subjected to the following cycling parameters: 95°C for 2 min and 30 cycles at 95°C for 30 s, 55°C for 20 s, and 72°C for 30 s, followed by a final extension of 5 min at 72°C.

454 pyrosequencing.

Second-round amplicons were visualized on agarose gels, purified using a High Pure PCR product purification kit (Roche Applied Science, Mannheim, Germany), and quantified using a Quant-iT PicoGreen double-stranded DNA reagent (Invitrogen Life Technologies, Carlsbad, CA, USA). Amplicons diluted to 1 × 10⁷ molecules/μl were subjected to emulsion PCR and pyrosequencing on a 16-region gasket (Mozambican samples) or as pools of bar-coded amplicons from 14 subjects on a 2-region gasket (Kenyan samples) using a GS FLX Titanium system, according to the manufacturer's instructions (454 Life Sciences, Branford, CT, USA).

Sequence read-quality filtering and alignments for each sample were performed as described previously (31). Nucleotide frequencies at each position were calculated in forward and reverse reads to determine the frequency of mutations at codons conferring resistance to NRTIs and NNRTIs included in the Stanford HIV Drug Resistance Database (32). Errors introduced by PCR and 454 pyrosequencing were estimated by including an HIV subtype A or subtype C plasmid in each 454 pyrosequencing plate. The limits of mutant quantification were estimated based on the average pyrosequencing error rate observed across the control plasmids and the number of HIV templates from each specimen subjected to the assay.

Oligonucleotide ligation assay.

Second-round amplicons were tested by an OLA for point mutations conferring resistance to nevirapine (K103N, V106M, Y181C, and G190A) and 3TC (M184V) using probes optimized for HIV subtypes prevalent in Kenya (subtypes A, D, and C) and in Mozambique (subtype C) (Table 1). The OLA was performed as described previously (16), with the following modifications. Briefly, after visualization of the desired amplicon on agarose gels, 2 μl of the PCR mixture was diluted 1:4 with water and added to a 20-μl ligation mixture containing 0.67 U of Ampligase thermostable DNA ligase (Epicentre Technologies, Madison, WI) and 0.3 to 1.3 pmol of fluorescein-labeled mutant-specific and biotinylated common probes plus 0.08 to 0.3 pmol of digoxigenin-labeled wild-type-specific probe (Table 1). After 10 ligation cycles at 37°C, ligated products were captured on a streptavidin-coated 96-well plate (Roche Diagnostics, Mannheim, Germany) and incubated with alkaline phosphatase (AP)-labeled antifluorescein and horseradish peroxidase-labeled antidigoxigenin antibodies (Roche Diagnostics, Mannheim, Germany). Mutant genotypes were detected by addition of AP yellow liquid substrate (Sigma-Aldrich, St. Louis, MO) and measurement of the optical density (OD) at 405 nm. After 6 washes with Tris buffer (0.1 M Tris [pH 7.5], 0.15 M NaCl, 0.05% Tween 20), wild-type genotypes were detected by addition of 3,3′,5,5′-tetramethylbenzidine (TMB) (Promega, Madison, WI) followed by 0.3 N H₂ SO₄ and measurement of OD at 450 nm. Standards of plasmid mixtures containing 0%, 2%, 5%, 10%, 25%, 50%, 75%, or 100% mutant in a wild-type background (17) were included in each assay plate. All specimens and standards were assayed in duplicate. The OD value of the 2% mutant control was used as the cutoff value for mutant detection when it was at least twice the OD value of the wild-type (0% mutant) control. The mutant OD/(mutant OD + wild-type OD) ratio was plotted against the mutant concentration to generate a standard curve in Excel, which was used to quantify the proportion of mutant in the HIV population of each subject's specimen.

TABLE 1.

OLA probes optimized for detection of HIV NNRTI and lamivudine mutations in Mozambican (HIV subtype C) and Kenyan (HIV subtypes A, C, D, and AE) specimens

Mutation and OLA probea	Sequence (5′ → 3′)b
K103N
Moz 103 wild-type	Dig-ACATCCAGCAGGGTTAAAAAAGAAR
Moz 103 mutant	Fluo-ACATCCAGCAGGGTTAAAAAAGAAC
Moz 103 common	P-AAATCAGTRACAGTACTGGATGTGGGGGAT-Bio
Kenya 103 wild-type	Dig-CATCCAGCRGGCYTAAAAAAGAAR
Kenya 103 mutant	Fluo-CATCCAGCRGGCYTAAAAAAGAAY
Kenya 103 common	P-AAATCAGTRACAGTACTRGATGTGGG-Bio
V106 M
Moz 106 wild-type	Dig-CCAGCAGGGTTAAAAAAGAAAAAATCAG
Moz 106 mutant	Fluo-CCAGCAGGGTTAAAAAAGAAAAAATCAA
Moz 106 common	P-TRACAGTACTRGATGTGGGGGATGCATAT-Bio
Y181C
Moz 181 wild-type	Dig-CACAAAATCCAGAAATAGTCATCTA
Moz 181 mutant	Fluo-CACAAAATCCAGAAATAGTCATCTG
Moz 181 common	P-TCAATATATGGATGACTTGTATGTA-Bio
Kenya 181 wild-type	Dig-AAAAAATCCAGAAATARTTATYTA
Kenya 181 mutant	Fluo-AAAAAATCCAGAAATARTTATYTG
Kenya 181 common	P-YCAATACATGGATGAYTTGTATGTA-Bio
M184V
Kenya 184 wild-type	Dig-ATCCAGAAATARTTATCTATCAATAYA
Kenya 184 mutant	Fluo-ATCCAGAAATARTTATCTATCAATAYG
Kenya 184 common	P-TGGATGAYTTGTATGTAGGATCTGA
G190A
Moz 190 wild-type	Dig-TATGGATGACTTGTATGTAGG
Moz 190 mutant	Fluo-TATGGATGACTTGTATGTAGC
Moz 190 common	P-ATCTGAYTTAGAAATAGGGCAA-Bio
Kenya 190 wild-type	Dig-CATGGATGAYTTGTATGTRGG
Kenya 190 mutant	Fluo-CATGGATGAYTTGTATGTRGC
Kenya 190 common	P-ATCTGAYTTAGAAATAGGGCAGCA-Bio

^aMoz, Mozambique.

^bNucleotides constituting the codons of interest are in bold type. Dig, digoxigenin; Fluo, fluorescein; P, phosphate; Bio, biotin.

Statistical methods.

Fisher's exact test was used to compare the proportions of mutant codons detected by OLA and 454 pyrosequencing. The correlation of the concentrations of mutants quantified by OLA and pyrosequencing is reported as the coefficient of determination (R²), with posttreatment specimens selected for variations in mutant concentrations being evaluated as independent events in the analysis.

RESULTS

Sensitivity of OLA and 454 pyrosequencing for detection of drug-resistant mutants in controls and clinical specimens.

OLA standards containing the mutant plasmid at 0, 2, 5, 10, 25, 50, 75, or 100% in a mixture with wild-type plasmid were quantified for each mutant codon, to generate a standard curve. In each standard curve, ≥2% mutant in the HIV population was detectable.

By testing a median of 1,000 HIV templates per specimen (range, 163 to 1,874 templates per specimen) in pyrosequencing, with an average substitution error rate of 0.06 ± 0.01% in the plasmid controls, the limit of detection was set at >0.1%. Only substitutions present at frequencies of ≥0.1% in both the forward and reverse reads were considered for analysis.

Clinical specimens from Mozambican subjects (n = 60) infected with HIV-1 subtype C were tested by OLA for the NNRTI resistance mutations K103N, V106M, Y181C, and G190A, and Kenyans (n = 51) infected with HIV-1 subtypes A (n = 18), B (n = 1), C (n = 5), D (n = 11), and AE (n = 16) were tested for the mutations K103N, Y181C, and G190A plus the 3TC resistance mutation M184V. The same first-round amplicons from all 111 specimens were subjected to pyrosequencing of RT codons 30 to 234, and the genotyping results were compared. Of 444 codons analyzed by OLA (i.e., 111 specimens, with 4 codons each), three (0.7%) yielded indeterminate results, i.e., both the mutant and wild-type reactions were negative due to sequence polymorphisms near the ligation site of the OLA probes; therefore, a total of 441 codons with valid results from both methods were compared.

Pyrosequencing detected mutations in 94/237 and 31/204 codons in Mozambican and Kenyan specimens, respectively, while OLA detected mutations in 56/237 and 19/204 (Table 2). The majority of the mutations detected by pyrosequencing but not by OLA were present at levels below the 2.0% limit of detection established by the OLA plasmid standards. When only mutations present at ≥2.0% by pyrosequencing were compared, detection of mutants by pyrosequencing was similar to detection by OLA (56 Mozambican specimens and 15 Kenyan specimens; P = 0.78).

TABLE 2.

Comparison of detection of drug resistance mutations, at specific HIV pol sites encoding NNRTI and lamivudine resistance, by 454 pyrosequencing and OLA

Study group	No. of codons tested	Codon	No. of mutations detected by:
			Pyrosequencing at ≥0.1%a	Pyrosequencing at ≥2.0%a	OLA at ≥2.0%a
Mozambique	237	Any	94	56	56
		103	27	22	18
		106	20	6	8
		181	30	21	20
		190	17	7	10
Kenya	204	Any	31	15	19
		103	12	7	9
		181	7	4	5
		184	9	2	2
		190	3	2	3

^aCutoff value used for mutant detection.

Discrepancies in the detection of mutations included six mutations (four K103N, one V106M, and one Y181C in the Mozambican cohort and none in the Kenyan cohort) that were detected at >2.0% by pyrosequencing but were not detected by OLA. Of these, five were present at concentrations near the OLA cutoff value (2.1 to 2.7%) and one was present at 8.8%. Examination of the sequences generated by pyrosequencing from the latter subject's HIV population revealed the presence of two consecutive base substitutions, in the second and third positions from the ligation site, in a subset of the mutant and wild-type variants, which would interfere with ligation of the OLA probes. In addition, 10 mutations (three V106M and three G190A in the Mozambican cohort and two K103N, one Y181C, and one G190A in the Kenyan cohort) that were detected at <2.0% by pyrosequencing were detected at >2.0% by OLA. Of these 10 low-level mutations, the OLA detected nine at concentrations between 2.0 and 4.0% and one at 69.0%. In the latter case (subject 264), the K103N mutation detected by OLA was confirmed by consensus sequencing (see below).

Comparison of mutant concentrations quantified by OLA and 454 pyrosequencing.

The quantification of mutant concentrations by OLA compared favorably (R² > 0.8) with pyrosequencing results across the 94 codons from Mozambique (Fig. 1A) and 31 codons from Kenya (Fig. 1B). Discrepancies of more than 25% between the mutant levels quantified by pyrosequencing and OLA were noted for only two codons, both in subjects from Kenya and both involving a K103N mutation (Fig. 1B). One subject (subject 264) had K103N detected at 69.0% by OLA but at 0.7% by pyrosequencing. Consensus sequencing of the RT region in this subject's specimen clearly showed a mixture of the two genotypes at codon 103, confirming the OLA results, but also showed mixed bases at the third and sixth positions in the 3′ region of the pyrosequencing reverse primer for amplicon 1. Therefore, it is likely that this pyrosequencing primer preferentially amplified the wild-type variants in this specimen, which resulted in a majority of wild-type reads in the pyrosequences. The other subject (subject 292) had 93.0% K103N by OLA and 32.0% by pyrosequencing. Consensus sequencing of the RT region showed only sequence encoding K103N and revealed substitutions at three bases in the region of the pyrosequencing forward primer plus a mixture of bases on the reverse primer, again suggesting preferential amplification of the wild-type template for pyrosequencing. The levels of K103N quantified by pyrosequencing for these two specimens were confirmed by reamplification of the plasma RNA and performance of a second pyrosequencing run (Table 3, subjects 264 and 292), as well as testing of the shorter amplicon 1 fragment by OLA, suggesting that the discrepancies with OLA results obtained for the longer pol fragment were not due to pyrosequencing errors in the long homopolymeric region surrounding codon 103.

FIG 1.

Correlation of mutant concentrations at specific HIV pol sites encoding NNRTI and lamivudine resistance, quantified by OLA and 454 pyrosequencing. (A) Concentrations of NNRTI resistance mutations K103N, V106M, Y181C, and G190A detected in amplicons from HIV-1 subtype C specimens from Mozambique (n = 94 codons). (B) Concentrations of NNRTI resistance mutations K103N, Y181C, and G190A and lamivudine resistance mutation M184V in amplicons from HIV-1 subtype A, C, D, and AE specimens from Kenya (n = 31 codons).

TABLE 3.

Comparison of HIV pol resistance mutations detected at ≥0.1% by 454 pyrosequencing when plasma specimens were tested on two separate occasions

^aPID, patient identification number; P1, P2, P3, and P4, pyrosequencing plate numbers. pKS, a HIV-1 subtype A wild-type plasmid control, was evaluated in each of four 454 pyrosequencing plates with error frequencies detected at drug resistance sites below the 0.10% cutoff value.

^bMutations detected in HIV pol in duplicate 454 pyrosequencing runs. A lack of data indicates that mutants were not detected or were detected at frequencies of <0.1%.

Reproducibility of detection of low-frequency mutants by 454 pyrosequencing.

Among mutations detected by 454 pyrosequencing, many (43.2%) were detected at concentrations between 0.1 and 2.0% of the viral population. To confirm the presence of these low-level mutants, 13 specimens from Kenya were reamplified by PCR and subjected to a second pyrosequencing run. As shown in Table 3, of 24 mutations detected at <2.0% in the first pyrosequencing run, only 5 (20.8%) were detected in the second (subjects 169, 371, 492, 499, and 292), while 16 additional mutations were detected at <2.0% in the second run. Of note, the specimen from subject 264 had only 77 HIV templates available for pyrosequencing in the second run, compared with 794 templates evaluated in the first run, which likely decreased the sensitivity of detection of minority variants. In contrast, all 6 mutations present at >2.0% (range, 3.4 to 61.2%) were detected in both pyrosequencing runs.

DISCUSSION

In this study, quantities of HIV mutations associated with NNRTI and 3TC resistance, across HIV subtypes A, B, C, D, and AE, as determined by OLA or 454 pyrosequencing were highly correlated when the mutants constituted 2.0 to 100% of the viral population. Uncommonly, HIV polymorphisms in the region of the OLA probes or the pyrosequencing primers reduced mutant detection. While pyrosequencing detected a large number of mutants at levels below 2.0% of the viral population, the interassay reproducibility of detection between 0.1 and 2.0% was low. This suggests that, despite input of an estimated 1,000 viral templates and ∼10,000 pyrosequencing reads for each specimen, reliable detection and quantification by 454 pyrosequencing were similar to those with the OLA, with a cutoff value of ∼2.0% mutant.

The nonstringent conditions for hybridization and ligation in the OLA can usually tolerate target sequence polymorphisms in the region complementary to the ligation probes. However, assay failure may occur when polymorphisms are located within the 4 nucleotides flanking the ligation site, and occasionally when multiple target mismatches are in the region complementary to one of the probes (16–18). These usually produce an indeterminate OLA result, i.e., both the mutant and wild-type reactions are negative, thus reducing the possibility of false-negative genotypes. Because genetic polymorphisms in HIV variants are common, it has been difficult to design universal OLA probes that can effectively evaluate viruses from all HIV subtypes (18). In this study, optimization of the OLA probes for subtypes prevalent in Mozambique and Kenya, as well as the use of mixed bases at sites of frequent genetic polymorphisms, significantly reduced the rate of indeterminate results, in comparison with previous reports (18, 33). More importantly, HIV sequence variations did not adversely affect quantification of mutant concentrations by OLA, as demonstrated by the high level of correlation between OLA and pyrosequencing results among Kenyan specimens that included HIV subtypes A, C, D, and AE.

To facilitate high-quality pyrosequencing, total amplicon lengths should be less than 600 bp, a restriction that poses a challenge in the design of primers flanking the HIV drug resistance mutations of interest. HIV sequence heterogeneity in the PCR primer region can lead to failed amplification of all or some variants within a viral population and thus bias the representation of mutant sequences analyzed by pyrosequencing. The effect of biased amplification of wild-type variants was observed for two of the 111 specimens in this study, for which the concentrations of K103N mutants were greatly underestimated by pyrosequencing in comparison with OLA and consensus sequencing.

The limit of detection of the OLA was established at 2.0% mutant using amplicons from mixtures of mutant and wild-type plasmids for each codon tested (17), but variations in PCR amplification and ligation may affect the sensitivity for mutant detection in clinical specimens. Importantly, the sensitivity for mutant detection is ultimately determined by the number of HIV templates in the initial PCR. In this study, we aimed to amplify ∼1,000 HIV DNA or RNA templates, to ensure sampling of sufficient templates and an adequate sequence read coverage of each amplified template to allow accurate quantification of minor variants by pyrosequencing at a sensitivity of ∼0.1 to 1%. Using this relatively high HIV template input, the OLA detected all mutants quantified at >3% by pyrosequencing, except for one with evidence of interfering polymorphisms in the region critical for ligation of the OLA probes, but was less reliable at concentrations near the 2% cutoff value. Template quantification is not performed in routine clinical HIV drug resistance testing, and the sensitivity for mutant detection may be diminished in specimens with small sample volumes or fewer HIV templates, such as plasma with low HIV RNA viral loads or blood collected as dried spots on filter paper.

As reported in previous studies (34, 35), minority variants detected by pyrosequencing at frequencies of <1% of the viral population were not reproducibly confirmed in subsequent replicate experiments, raising the question of whether these low-frequency mutations were genuine or were sequencing artifacts. The resolution of pyrosequencing is primarily determined by the number of input DNA/cDNA templates and the error frequency introduced at multiple steps of the pyrosequencing method. Pyrosequencing processing introduces a significant number of insertion-deletions and point errors (23), and estimates of the average error rates in pyrosequencing range from 0.24% to 0.98%, with differences in homopolymer and nonhomopolymer regions (36–38). Different bioinformatic approaches to identify and remove or correct sequencing artifacts (31, 36, 39–41) can reduce error frequencies to ∼0.05%. However, a substantial proportion of errors that remain after data cleaning are substitution errors introduced during the PCR that precedes pyrosequencing (42, 43), which more often involve transitions versus transversions (42–44) and can vary by site depending on the sequence context (43). In sequencing of plasma specimens, reverse transcription of HIV RNA into cDNA can add another source of substitution errors, as reverse transcriptases have higher error rates than DNA polymerases (45). In this study, we attempted to include ∼1,000 amplifiable HIV DNA/cDNA templates in the PCR and to obtain ≥10,000 pyrosequencing reads per sample, to allow accurate quantification of minor variants at 1% of the population. The variability in detecting mutants at ≤2% by 454 pyrosequencing could have resulted from several phenomena, as follows: (i) the amplification of pol might have been less efficient than that of the LTR or gag region we used to quantify the templates, resulting in testing of fewer viral templates and stochastic sampling of mutant variants at low concentrations; (ii) primer biases might have variably amplified templates within a sample, as discussed above; or (iii) base substitutions during reverse transcription, PCR, or pyrosequencing might have produced mutants at low concentrations. To address some of these problems, others have reverse transcribed viral RNA with “primer ID,” which allows measurement of the number of viral templates sequenced (46), or have used bidirectional tags on both strands of duplex DNA to assess substitution errors by comparing the sequences derived from each strand of a DNA template (47). In addition, it is important to balance the oversampling of low-concentration viral templates in a specimen with oversampling of PCR amplicons for 454 system reads. While we estimated that an ∼10-fold increase in each would be adequate, greater oversampling may be needed at each step to achieve a reproducible sensitivity of 1% by 454 pyrosequencing. Use of sequencing techniques that accurately measure mutations at low concentrations will be necessary to better understand the role of low-abundance drug-resistant HIV variants in different clinical scenarios.

Limitations of this study include selection of two or three specimens from a subset of Mozambican subjects to increase the range of mutant concentrations compared. While the specimens were from independent time points collected over a period of 12 months, sequential specimens from a subject are likely to have similar genetic polymorphisms, thus limiting the range of variations evaluated in this study. In addition, because the HIV subtype of the Kenyan specimens was not known at the time of the study, an LTR real-time PCR assay shown to perform well across multiple subtypes was used for quantification of amplifiable HIV templates. Amplification of the longer region of pol might have been less efficient than amplification of the short LTR fragment in the real-time PCR assay, resulting in overestimation of the number of templates subjected to PCR and reduced sensitivity for detection of rare drug-resistant variants.

In summary, the OLA compared favorably to 454 pyrosequencing in the detection and quantification of drug-resistant HIV mutants present at a threshold of >2% of the viral population, and it performed similarly across diverse HIV subtypes prevalent in resource-limited regions. While this study focused only on NNRTI and 3TC resistance mutations, OLA probes have been designed to include additional antiretrovirals commonly used in first-line ART, such as tenofovir and stavudine-zidovudine. Compared to consensus sequencing, the OLA is more sensitive and economical and is easier to implement in resource-poor communities, and this or a similar assay could enhance the outcomes of ART programs.