This study is the first to compare full-length HIV-1 RT sequences in paired plasma samples obtained pre-therapy and at first virologic failure in predominantly subtype B infected patients. The only mutations that were significantly more frequent at virologic failure than pre-therapy were K103N (p=0.001) and M184I/V (p=0.016) in the polymerase domain, which confer resistance to the study drugs EFV and 3TC, respectively. Other known polymerase domain mutations were not significantly associated with failure although, there were possible trends for K65R (p=0.13) and V106I/M (p=0.13). Mutations in the RT connection and RNase H domains were not significantly more frequent at virologic failure than at pre-therapy (p ≥0.25), and thus did not contribute to the virologic failure observed.
This study is the largest to date comparing pre-therapy and failure sequences from the same individuals (N=53). The only other similar study compared full-length RT sequences in one patient over three years4
. In this patient, the N348I connection domain mutation was selected by ZDV and/or didanosine therapy4
and confers resistance to ZDV, didanosine, NVP, EFV, delavirdine, tenofovir and etravirine.4,11,12,14,15
N348I was probably not identified in our study because efavirenz, not NVP, was the NNRTI used for initial randomized therapy in ACTG A5142 and virologic failure was detected early using a sensitive definition of failure.
Other prior analyses of RT connection or RNase H domain mutations have compared sequences from unrelated ART-naïve patients and ART-experienced patients.3–13
These studies have identified a number of mutations in the C-terminus of RT that are more frequent in ART-experienced patients compared to ART-naïve including E312Q, G333D/E, G335C/D, N348I, R356K, R358K, A360I/V, A365I, T369I, A371V, A376S and K451R. Our study did not identify these mutations at virologic failure. This may be due, in part, to the strict definition of virologic failure used in ACTG A5142 (see Methods). As a consequence, the failure samples analyzed were early in the course of virologic breakthrough with limited time on failing therapy for drug-resistant variants to emerge. Nevertheless, our study shows that the first mutations to arise in association with virologic failure in HIV-1 subtype B-infected patients are in the polymerase domain and not frequently in the connection or RNase H domains.
Although our study is the largest pre-therapy-failure comparison of RT sequences, the sample size (N=53 pairs) had limited power to detect mutations that emerge at low frequency. Specifically, the upper bound of the 95% confidence interval for the probability of a mutation that is not detected in 53 patients is 6.7%. Therefore, at alleles where no mutation was observed, there may have been up to 7% of emergent resistance in the study population. Larger sample sizes are needed to exclude connection and RNase H domain mutations that emerge infrequently.
Secondary analyses were performed to assess if pre-therapy RT polymorphisms predispose to virologic failure (). Polymerase, connection and RNase H domain mutations were not significantly associated with failure after correction for multiple comparisons. To achieve 80% power to detect such an association, assuming a low proportion (e.g. 2%) of the 144 patients that did not experience virologic failure had a specific polymorphism, would require that 14% of the 53 patients that experienced virologic failure have the polymorphism, i.e. a difference in proportion of 12%. As a consequence, our study had power to detect relatively large differences (>10%) in polymorphism frequency between patients experiencing virologic failure and those who did not. Such differences were not detected in our study.
In summary, this study of full-length RT did not identify mutations in the connection or RNase H domains associated with virologic failure in predominantly HIV-1 subtype B infection. These findings suggest that full-length RT sequencing is not essential for management of failure first-line efavirenz-containing regimens when failure is detected early, although analyses of larger datasets are needed before firm conclusions can be drawn.