Quality Assessment Program for Genotypic Antiretroviral Testing Improves Detection of Drug Resistance Mutations

Abstract

Genotypic antiretroviral testing is now widely used for the management of patients who are undergoing antiretroviral therapy for human immunodeficiency virus infection. The assays are complex, and there is considerable potential for variation between laboratories. Informative and ongoing quality assessment programs (QAPs) which address all aspects of testing are required. The panel distribution of clinical material is a critical component of QAPs. We report on the results and data from a recent panel. Four cryopreserved plasma samples from treated donors were distributed to nine laboratories. Three laboratories performed testing by commercial assays, and six laboratories used in-house assays, with one laboratory reporting results from two in-house assays. There was complete concordance between results for 95.9% of the nucleotide sequence and 94.5% of the amino acid sequence. Despite this overall high level of concordance, the degree of concordance at drug resistance mutation (DRM) sites when DRMs were present was considerably less (38% of DRM sites). Consequently, only 3 of the 10 methods reported 100% of DRMs as present. This elevated discrepancy rate is almost certainly a result of variability in the identification of mixtures of nucleotides (mixtures) at any site within the sequence. In addition, laboratories differed in the number of codons in the reverse transcriptase gene that were sequenced and their ability to amplify all samples. This panel distribution demonstrated a requirement for laboratory participation in ongoing QAPs and the optimization of assays with standards that contain mixtures.

The rapid replication rate of human immunodeficiency virus (HIV) and the lack of proofreading of the HIV reverse transcriptase (RT) enzyme results in a high rate of mutation during HIV replication (3). This creates diversity in the HIV genome and results in the development of quasispecies.

The use of antiretroviral drugs that inhibit the HIV RT and protease enzymes results in reductions in the level of HIV RNA detectable in plasma, increases in CD4⁺-cell counts, and improved clinical outcomes for HIV-infected patients. However, ongoing HIV replication, even at low levels, in the presence of antiretroviral therapy results in the selection of virus quasispecies that are resistant to therapy. The development of HIV drug resistance is associated with treatment failure.

Present approaches to therapy include the simultaneous use of drugs of different classes (highly active antiretroviral therapy). This may result in a complex array of drug resistance mutations (DRMs) that encode resistance to one or more of the drugs. In addition, DRMs that encode resistance to one drug may confer cross-resistance to drugs that have not been used by the patient and are being considered as potential alternatives. Testing for DRMs is required to confirm that treatment failure is the result of the development of DRMs and not the result of low drug potency, poor pharmacokinetics, high levels of plasma protein binding, poor patient adherence, and other factors (9). Testing for drug resistance also allows the identification of drugs that can be used by the patient to replace those to which the virus has developed resistance.

There are two approaches to testing for drug resistance. One is to determine the ability of the virus to replicate in the presence of drugs. These so-called phenotypic assays are complex and are generally not routinely used (10). The second is to determine the presence of mutations directly by one of a number of genotypic assays. Genotypic antiretroviral testing (GART) is used in clinical laboratories worldwide. Five randomized studies have assessed the value of GART in the clinical management of HIV infection. When coupled with expert interpretation, GART can result in modest but significant improvements in virological outcomes in patients who are changing therapy compared with those in patients in whom changes were made on the basis of clinical judgment (1, 2, 5). DNA sequencing is the “gold standard” and the most widely adopted method. Other approaches to genotyping, such as the use of mutation-specific oligonucleotide probes, have also been developed, and although they are sensitive, they may be unreliable because of the extensive polymorphic nature of the protease and RT genes and the likelihood of polymorphisms in the probe annealing sites (15).

GART by DNA sequencing requires the extraction of RNA from plasma, transcription to DNA, PCR amplification (usually by a nested PCR approach), DNA sequencing, sequence editing and analysis, and interpretation. Many laboratories use protocols developed in-house, and a variety of approaches at any of the steps may be used. Two commercially available GART kits are available: TRUGENE HIV-1 from Visible Genetics Inc. (Toronto, Ontario, Canada) and the ViroSeq HIV-1 genotyping system from Applied Biosystems (Foster City, Calif.). These assays have been shown to provide highly concordant sequences when the assays are tested in parallel (6).

Given the complexities of the procedures and the variety of approaches used, there is considerable potential for results to vary between laboratories. Several studies have indicated a high level of sequence concordance between participating laboratories when various aspects of GART are examined (4, 6, 14). However, there have been few reports of studies that have considered the entire GART procedure (16). High levels of concordance have been reported. However, discrepancies at some sites that contain two or more nucleotides (mixtures) are common. Sequencing of sites that encode drug resistance frequently reveals a mixture of mutated and wild-type sequences. Because relative sequence peak heights do not necessarily reflect the relative proportions of mutant and wild-type viruses, a mutation that is present as a mixture must be considered significant. It is possible, therefore, that despite a high level of concordance throughout the entire sequence, variability may exist at DRM sites. The reasons for the discrepancies at sites containing mixtures have not been fully examined, although it has been reported that sampling variability may be a significant contributor (16).

Given the importance of GART and the potential for variability between laboratories, it is critical that laboratories participate in ongoing, informative quality assurance programs (QAPs). It is also important that such QAPs include the material used for routine testing. The earlier studies, while informative, distributed cultured infected lymphocytes to multiple laboratories (4), plasma samples spiked with cloned HIV RT to multiple laboratories (14), or multiple clinical samples to just two laboratories (16). Only one appears to be an ongoing QAP (14), and only the last one included both the RT and protease genes.

We report here the results from the second panel distribution (PD) of four clinical samples sent to nine laboratories that participate in an Australasian QAP. The PD is one of a number of activities that make up this program. Educational workshops and PDs of raw data and nucleotide sequences for analysis are also included in the program. This is the first report to describe a multicenter QAP of clinical samples for GART that include both the RT and the protease genes. A comparison between this and the first PD held the previous year demonstrates an improvement in the detection of DRMs.

MATERIALS AND METHODS

Cryopreserved plasma from four HIV-infected patients were distributed in 2-ml aliquots to nine laboratories in Australia and New Zealand. The criteria for sample selection were that the donor had undertaken, or was undertaking, antiretroviral therapy and had a viral load of >10,000 copies per ml. Samples were not selected on the basis of prior genotyping results. The treatment details for the donors and the viral loads of the samples used in this QAP are shown in Table 1. Each laboratory was requested to perform HIV genotyping by its routine procedures and provide the resulting DNA sequence in text form for analysis.

TABLE 1.

Treatment histories of sample donors

Sample no.	Viral load (no. of HIV RNA copies/ml)	Treatment at time of donation	Previous treatment
1	25,800	Indinavir, ritonavir didanosine, stavudine	Lamivudine
2	50,000	Ritonavir, lopinavir, didanosine	Indinavir, amprenavir, nelfinavir, abacavir, lamivudine, stavudine, saquinavir, zidovudine, zalcitabine, nevirapine
3	11,600	Didanosine, lamivudine, efavirenz	Stavudine, nevirapine, saquinavir, zidovudine
4	44,300	Lamivudine, zidovudine, nevirapine

The ambiguity codes defined by the International Union of Biochemistry and Molecular Biology were interpreted to indicate mixtures. For example, if a laboratory reported a “W,” the interpretation was that the position had a mixture of adenine (A) and thymine (T).

The DNA sequence obtained for each sample by each laboratory was aligned, and a target genotype (TG) for each sample was deduced (see below). The TG and the DNA sequences from each laboratory were translated into amino acid sequences for additional analysis.

An HIV genotype is a consensus of the predominant populations within a quasispecies. The TG was created to represent as close as possible the most likely consensus HIV population detectable by DNA sequencing for each sample. The rules that we have applied to define the TG have a strong bias toward the identification of mixtures. We made the assumption that it is extremely unlikely for two laboratories to incorrectly sequence the same nucleotide at any one position. It was more likely that these laboratories were sequencing an alternative population(s). Table 2 summarizes the rules for defining the TG.

TABLE 2.

Deduction of TG^a

Laboratory	Sequence at positions 1 to 7
A	AAAGCTT
B	ATAGCTA
C	AAWGCCW
D	AAAACTT
E	AAAGYTT
F	AAAACTT
G	AAAGYAT
Target sequence	AAARYTW

^aThe TG represents the most likely consensus quasispecies detectable by DNA sequencing for each of the samples of this PD. If a single laboratory sequences a nucleotide different from that sequenced by the other laboratories, either as a partial or as a complete difference, then the TG is the conscensus sequence sequenced by the other laboratories (positions 2 and 3). If two or more laboratories report an identical nucleotide which is different from the sequence reported by the other laboratories, the TG is a mixture of the two differently reported sequences (positions 4 and 5). If two laboratories report nucleotides different from those reported by the other laboratories and from those reported by each other, the TG is the sequence reported by the majority (position 6). However, if two laboratories report sequences different from those reported by the majority and one is a mixture that is present and contains the sequence reported by the majority and the other laboratory, the TG is the mixture (position 7).

Because laboratories differ in the length of the RT gene that they sequenced, the TG included sequences provided by more than half of the laboratories. Any sequence provided by any of the laboratories beyond this point was not included in the analysis. Thus, the TG of samples 1 and 3 is 1,203 nucleotides (nt) of contiguous pol sequence starting from the first codon of the protease gene to codon 302 of the RT gene. The TG of sample 2 is 1,140 nt (to codon 281 of RT), and the TG of sample 4 is 1,119 nt (to codon 274 of RT).

Sequence alignments were performed by using the programs readseq, dnapars, and protpars from the Phylip package (7).

CLUMP software was used to perform statistical analysis (18).

RESULTS

Laboratories differed in the protocols that they used for GART by DNA sequencing.

All laboratories performed genotyping by DNA sequencing using automated, dye-labeled sequencing technology. Six of the nine laboratories performed genotyping by assays developed in-house. These are indicated by the letter H in the laboratory name (laboratories 1H, 2H, 3BH, 3GH, 4H, 6H, and 9H). Laboratory 3 provided results from two in-house assays (3BH and 3GH) that differed in the locations of the amplification and sequencing primers. Precise assay details including PCR and sequencing conditions were not requested. However, all in-house assays used Applied Biosystems automated DNA sequencers. In addition, all laboratories that performed in-house assays except laboratory 2H used dye-terminator sequencing chemistry from Applied Biosystems. Laboratory 2H used dye primer sequencing chemistry from Applied Biosystems. Three laboratories used commercial kits. Laboratory 7V used TRUGENE HIV-1 from Visible Genetics Inc., and laboratories 5A and 8A used the Applied Biosystems ViroSeq HIV-1 genotyping system. All laboratories sequenced the entire protease gene, but the laboratories varied in the amount of the RT gene that they sequenced. All laboratories sequenced to amino acid 236, but only laboratory 2H included the DRMs at amino acids 318 (11) and 333 (12).

There was a high level of sequence concordance between laboratories.

cDNA from three of the four samples was successfully amplified and sequenced by all laboratories. Four laboratories had difficulties amplifying cDNA from one sample, sample 3, even though this sample had a significant viral load (11,600 copies per ml). Laboratories 1H, 3GH, and 9H were unable to amplify any of the protease and RT genes from this sample, and laboratory 6H was unable to amplify a section of RT that included all reported DRMs up to and including amino acid 118.

The differences between the reported sequence and the TG were interpreted as partial or complete. A partial difference is present when the test sequence contains a mixture of nucleotides or amino acids and the TG contains a single nucleotide or amino acid that is present as part of the mixture in the test sequence. The same applies if the TG contains the mixture and the test sequence contains a single nucleotide or amino acid that is present within the mixture. A complete difference is present if the reported sequence and the TG do not share a common nucleotide or amino acid at the same site.

Of the 4,665 nt (1,555 codons) within the TGs for all samples, there was complete concordance at 4,473 sites (95.9%). One hundred fifty-eight of the 192 discrepant sites were partial differences (82%) and 34 (18%) were complete differences. A total of 106 of the 192 discrepant sites (55.2%) were synonymous differences, and 86 (44.8%) were nonsynonymous differences. Thus, of the 1,555 codons sequenced, there were differences between laboratories at 86 positions (5.5%).

A comparison of the nucleotide and amino acid sequences reported by each laboratory and the TG for each sample are shown in Fig. 1. High-level concordance (97.5%) was found between the nucleotide sequences derived by all laboratories and the TGs of all samples. Some laboratories (laboratories 5A, 3BH, 2H, and 8A) reported greater than 99% concordance for at least three samples. Most differences were partial, with 5 of the 10 assays reporting no complete nucleotide differences with the TG for any of the samples.

FIG. 1.

Concordance between the nucleotide and amino acid sequences reported by each laboratory and the TG for each sample. All laboratories reported >97.5% complete nucleotide sequence identity (shaded bars) with the target sequence, with some laboratories reporting >99% complete nucleotide sequence identity for some samples. Most discrepancies are partial differences (white bars). Complete nucleotide sequence differences (black bars) were reported by 5 of the 10 laboratories. No sequence was reported for sample 3 by laboratories 1H, 3GH, and 9H.

The sequence differences (expressed as a percentage) observed between the reported nucleotide sequence and the TG appeared to be sample dependent. Most sequence differences were found in sample 1 by 7 of the 10 assays used by the nine laboratories. This may reflect the heterogeneity of this sample. Sample 1 contained the largest number of mixtures (n = 26 in the TG), whereas sample 4 contained 22 mixtures, sample 2 contained 18 mixtures, and sample 3 contained 9 mixtures.

The degree of concordance was lower at DRM sites than at non-DRM sites when mutations were present as mixtures.

It is critical to determine if the high level of concordance of the sequence between laboratories is consistent throughout the entire sequence and is not greater at DRM sites. The amino acid sequences at 47 DRM sites in the RT and protease genes for all samples are shown in Fig. 2. There were discrepancies at 18 (10%) of the 188 DRM sites (four samples × 47 DRM sites), whereas there were discrepancies at 86 (5.5%) of the 1,555 amino acids of the entire sequence (P = 0.01 chi-square). Thirty-seven DRMs were present in the TGs for all samples at the DRM sites. There were discrepancies between laboratories at 38% (14 of 37) of these positions, whereas there were discrepancies at 2% (3 of 151) of the positions that contained the wild-type sequence (P = 1.25 x 10⁻⁸ by Fisher's exact test). Twelve of the 14 discrepant sites contained mixtures of wild-type and mutant sequences in the TG. At 7 of the 14 sites (50%) all laboratories reported a DRM as either a mixture or a complete DRM. However, in the other 50% of cases, some laboratories reported the wild-type sequence only. Thus, at 7 of 37 (18.9%) sites that contain DRMs in the TG, some laboratories failed to detect a DRM. At three DRM sites the TG was the wild type and there was a discrepancy between the laboratories. It is not possible to say if this was a result of increased sensitivity for the detection of DRMs.

FIG. 2.

Concordance between laboratories at DRM sites. Some laboratories did not detect some DRMs. The deduced amino acid sequence from each laboratory is compared to the TG for each sample at DRM sites for the protease and RT genes. Dots indicate identity with the TG. However, for the purposes of clarity, the sequences from all laboratories at sites where at least one laboratory reported a mixture are shown, even if they are identical to the TG. All DRMs are boxed and are also indicated under the sequence for clarification. The drugs used to treat the donors of the plasma samples are also included. Most discrepancies are at sites containing DRMs. Some laboratories did not identify DRMs. Accepted abbreviations for the drugs used by the sample donors have been used: ddI, zalcitabine; 3TC, lamivudine; d4T, stavudine; ABC, abacavir; ddC, didanosine; AZT, zidovudine; NVP, nevirapine; EFV, efavirenz. The black bars indicate regions which were not sequenced.

DRMs were not detected by all laboratories.

None of the laboratories reported a complete difference in amino acids from the TG sequence at the DRM sites (Fig. 3A). However, many of the laboratories sometimes reported only the wild-type sequence when the TG contained a mixture of wild-type and mutant sequences (Fig. 3B). Only 3 of the 10 assays from the nine laboratories reported that 100% of the DRMs were present in the TG. The two laboratories that reported the largest number of partial differences (laboratories 3GH and 9H; Fig. 3A) also reported the fewest number of DRMs (Fig. 3B). Of note, laboratory 9H did not report any DRMs as mixtures of wild-type and DRM sequences.

FIG. 3.

Concordance of amino acid sequences at DRM sites (A) and numbers of DRMs detected by each laboratory (B). (A) Degree of concordance of the amino acid sequences at all positions containing DRMs within the target sequence for all samples. None of the laboratories reported complete amino acid sequence differences at DRMs with the TG. However, laboratories did vary in the percentage of partial differences (black bars) with the TG. The percentage of amino acids that were identical to the TG are indicated by the shaded bars. (B) Percentage of DRMs that were sequenced by each laboratory. The shaded bars are the percentage of DRMs that were reported as DRM only. The black bars are the percentage of DRMs that were reported as mixtures of DRM and wild type, and the white bars indicate the percentage of sites containing DRM within the TG that were reported as wild type only. The numbers above the bars indicate the number of DRMs sequenced by each laboratory.

Laboratories differed in their abilities to detect mixtures: laboratories that reported the fewest mixtures also reported the fewest DRMs.

The entire sequence was analyzed to determine if laboratories differed in their abilities to detect mixtures. Mixtures were present at 75 codons of the TGs for all samples: 26 codons for sample 1, 18 codons for sample 2, 9 codons for sample 3, and 22 codons for sample 4. None of the 75 sites was reported identically by all laboratories. The number of mixtures reported by each laboratory for all samples is shown in Fig. 4. It appeared that laboratories varied in their abilities to identify mixtures. Some laboratories reported a high percentage of mixtures for all samples, and others reported only a few. Furthermore, the inability to identify mixtures affected a laboratory's ability to identify DRMs. All four laboratories that reported 20% or fewer mixtures for any of the samples in this QAP reported less than 100% of the DRMs present in the TGs for all samples, whereas three of six laboratories that reported >20% mixtures for any sample reported less than 100% of the DRMs present in the TGs for all samples (P = 0.08 by Fisher's exact test).

FIG. 4.

The ability to identify mixtures varies between laboratories. The percentages of mixtures present in the TGs that were correctly identified by the participating laboratory are shown for each sample. Some laboratories (laboratories 3BH and 2H) consistently reported a high percentage of mixtures for each sample, while other laboratories (laboratories 6H and 9H) reported few mixtures for each sample.

There has been an improvement in the detection of DRMs compared to the previous PD.

Calculation of the percentage of DRMs reported by each laboratory for all samples allows a comparison of the performance of each laboratory between QAPs. A comparison of the percentage of DRMs detected by each laboratory for this and the first PD is shown in Fig. 5. Eight laboratories participated in both PDs. The eight laboratories sequenced a total of 263 positions that contained DRMs in the TGs for all samples in the second PD. Of these, 254 were correctly reported (96.6%). The same eight laboratories had sequenced a total of 116 DRMs in the first PD. Of these, 91 (78%) were correctly reported. This indicates a significant improvement (P = 7.6 × 10⁻⁸ by Fisher's exact test) in the detection of DRMs since the first PD. One laboratory reported 100% of mutations in both exchanges, and all other laboratories reported a greater percentage of DRMs in this exchange than they reported in the first PD.

FIG. 5.

Improvement in detection of DRMs by participating laboratories in two PDs of clinical samples. A comparison of the percentage of DRMs reported by each of the laboratories that participated in the first and second PDs of clinical material reveals an increase in the percentage of DRMs reported in this (second) PD by most laboratories. The samples were different between the two PD. Black bars, DRMs that were reported unambiguously; bars with shading, DRMs that were reported as mixtures of mutant and wild-type sequences; white bars, DRMs that were reported as the wild-type sequence only. Laboratory 5A participated in the second PD only, and the data for that laboratory are not included. Method B from laboratory 3 (laboratory 3BH) was chosen from the second PD. Laboratory 9H did not report any drug resistance mutations as mixtures from either PD. The numbers 1 and 2 next to the laboratory designations indicate the results from the first and second (the present) PD, respectively.

DISCUSSION

HIV drug resistance genotyping by DNA sequencing is a complex procedure requiring several sophisticated molecular and analytical procedures. This complexity means that there is potential for variability to exist between laboratories. Thus, effective and informative QAPs, which include the exchange of material that best represents the material that is tested by the participating laboratories, must be used. QAPs must be ongoing and provide laboratories with rapid and informative feedback regarding the performance for each PD.

In the present QAP, four samples were distributed to participating laboratories. The number of DRMs in these samples provided results sufficiently variable to analyze the performances of the participating laboratories and draw conclusions based on those performances. Sample 3 could not be genotyped successfully by three of the nine laboratories. The reasons for this are not clear. There was adequate viral load (11,600 copies/ml), and analysis of the TG sequence against the sequence in the Stanford database (http://hiv-4.stanford.edu/cgi-bin/hivseqweb.pl) indicates that the most likely subtype is subtype B. The TG of sample 3 had the fewest number of mixtures of the samples tested, indicating that the quasispecies may have been less diverse. One explanation may be that the predominating population contained sequence differences at amplification and/or sequencing primer sites, with the result being an inability of some laboratories to amplify this sample.

Laboratories also differed in the amount of the RT gene that they sequenced. All laboratories included codon 236, but only laboratory 2H included DRMs at codons 318 and 333. The Y318F mutation has been shown to predict a 10-fold increase in resistance to nonnucleoside reverse transcriptase inhibitors (11), regardless of the presence of other nonnucleoside reverse transcriptase inhibitor resistance mutations, and may be of significance. The effect of G333E may be of less clinical significance (17).

Overall, there was a high level of concordance between the sequences reported by each laboratory and the TG for each sample (>97.5% for both nucleotide and amino acid sequences). The number of nucleotides for which there was complete concordance between all laboratories was also high (95.9%). However, the level of concordance at DRM sites was less than that at non-DRM sites, and the level of concordance at DRM sites when DRMs were present was less than that when the wild-type sequence was present. Some laboratories did not report DRMs that were clearly present. This could have clinical implications. Laboratory 3H reported the wild-type sequence for amino acids 69 and 70 (threonine at amino acid 69 and lysine at amino acid 70) in the RT gene of sample 4 by using method G, while some laboratories, including laboratory 2H, reported DRMs for didanosine (serine at amino acid 69) and zidovudine (arginine at amino acid 70). The sequences for sample 4 from laboratories 3GH and 2H were analyzed by using the Stanford Drug Resistance Interpretation Beta Test software (http://hiv-4.stanford.edu/cgi-bin/hivtestweb.pl). The results for the sequence reported by laboratory 3GH indicated susceptibility to zidovudine, while potential low-level resistance was indicated for the sequence reported by laboratory 2H (data not shown). This may influence a clinical decision as to whether zidovudine should be used. Resistance to didanosine was indicated for both sequences as a result of the 184V mutation reported by both laboratories. Therefore, in this case, missing the DRM at amino acid 69 may not influence a decision on the use of didanosine.

There were clear differences in the abilities of the laboratories to detect mixtures of nucleotides. An inability to sequence mixtures of nucleotides at any position within a sequence correlates with an inability to detect all DRMs. It is difficult to determine from these data if the inability to detect mixtures was because the laboratories did not amplify mixtures or because mixtures were not detected during the sequence editing procedure. The data provided by the two methods used by laboratory 3 (3GH and 3BH) indicate that method G selectively amplified one of the predominating populations over the other(s) in most samples. The two methods differ in the locations of amplification and sequencing primers. The PCR amplification of HIV quasispecies can be regarded as numerous PCRs occurring concurrently. The PCR and sequencing conditions must be such that the genotype reflects the quasispecies. While this may be difficult in some cases, the use of appropriate standards that contain mixtures would be valuable during method optimization.

Accurate sequence editing is of utmost importance. Most sequence editing software programs that are available are unreliable for the detection of mixtures and are unable to differentiate between a true mixture and an artifact due to a high background. They indicate a mixture when an additional peak is present at a preselected threshold percentage of the major peak without considering the fact that the presence of an additional peak must correspond to a reduction of the major peak (20). In addition, some laboratories may not consider anything below the threshold as significant. However, given the within-sample variability, as shown by Frater et al. (8), the demonstration of an additional peak at any site must be considered significant. Furthermore, confusion exists when the sequence from one strand indicates a mixture and the sequence from the opposite strand is unambiguous. Our contention is that if a mixture is present in one orientation, it must be reported in the consensus sequence. It is not unusual for a mixture to be detected by sequencing one strand and for it not to be detected by sequencing the complementary strand. The reason for this is that the rate of incorporation of dye-labeled dideoxynucleotides during the sequencing reaction varies from nucleotide to nucleotide within a sequence (19). If a mutant nucleotide is present at low levels and the complementary dideoxynucleotide is incorporated at a low rate at that site within the sequence during the sequencing reaction, it may not be detected. However, by sequencing in the opposite direction, the incorporation of the complementary nucleotide may occur at a much higher rate and therefore be detected. This may also explain why the level of sensitivity of mutation detection varies from site to site within a sequence. In our laboratory the limit of detection of mutations varies between 2 and 20%, depending on the site of the mutation within the sequence (data not shown). Other groups have reported detection of mutations at levels of 5 to 10% by automated DNA sequencing (13).

It is difficult to compare the performances of commercial and in-house assays. Five of the seven assays that did not report 100% of the DRMs were in-house assays, and both laboratories that used the Applied Biosystems assay did not report the DRM at amino acid 69 of RT in sample 4. None of the three commercial assays had problems with the amplification of sample 3, whereas three of seven in-house assays did (P = 0.5 by chi-square analysis). In addition, the results from the two laboratories (laboratories 5A and 8A) that used the same commercial assay from Applied Biosystems were different. Laboratory 5A reported more mixtures throughout the sequence, including at DRM positions. Once again, this variability may be related to the approaches that individual laboratories apply to sequence editing. Nevertheless, two of the three laboratories that reported 100% of the DRMs were in-house assays. The laboratory that reported the fewest number of mixtures and the fewest number of DRMs (laboratory 9H) used an in-house assay. Furthermore, that laboratory did not amplify sample 3. It would seem that both in-house and commercial assays are capable of giving complete results.

This PD of clinical samples represents a component of an ongoing QAP to improve the concordance of GART between laboratories. There was evidence to suggest an improvement in performance by most laboratories since the first PD. Because GART is a complex series of procedures, our approach to QAP includes an evaluation of each component of GART. This will include the distribution of raw sequence data for editing and an edited sequence for genotypic interpretation. Future exchanges will also include samples with different subtypes, samples with low copy numbers, replicates from previous panels, and proviral DNA.