By analyzing a large number of patients with TDR, we were able to show that the M184V/I mutations become undetectable in population sequencing substantially faster than do other mutation groups, which are generally similar to one another. Although rates of replacement for mutations other than M184V/I were relatively low, there was still appreciable replacement. Although NNRTI mutations are assumed to have minimal effects on fitness, replacement of these mutations occurred over time. In addition, our results suggest that higher viral load may promote mutation loss, but this does not account for the very substantial person-to-person variation that we observed in the rate of mutation loss.
Prior studies of TDR mutations have reported that replacement with wild-type variants occurs but that most mutations are maintained for at least 1–2 years after transmission [11
]. The largest and most definitive study grouped all mutations together and estimated a median time-to-loss of detectable drug resistance (using population-based assays) ranging from 4.1 years, using a conservative estimate, to longer than the lifetime of the individual, using a less conservative estimate [12
]. Although our results might appear to differ from this estimate, the difference is primarily attributable to our ability, with larger numbers, to assess groups of mutations separately. The high rate of replacement that we observed among M184V/I mutations is consistent with prior reports [11
] and almost certainly reflects this mutation's association with reduced fitness in the absence of therapy [6
]. The relative stability of thymidine analog–associated, NNRTI, and PI mutations over time is also consistent with other reports [12
We had hypothesized that PI mutations would be replaced more rapidly than would NNRTI mutations. However, we found that the PI group had a rate of replacement similar to that of the NNRTI group, although CIs were wide (HR for replacement of PI vs NNRTI, 1.12; 95% CI, 0.3–4.0; P
= .9). The slow replacement of PI mutations is notable, because these mutations, similar to M184V/I, are known to affect viral fitness [31
]. Our group previously performed a partial treatment interruption study in which patients receiving a stable partially suppressive regimen discontinued one drug class while maintaining the other drug classes. Continuation of at least some ART prevented rapid rebound of archived variants, forcing HIV to back-mutate in the same way that occurs after the acquisition of a TDR variant. In the study, selective removal of NRTIs was associated with loss of the M184V/I mutation, and removing PIs was associated with limited mutation replacement, even after several years [33
]. The mechanism for the persistence of PI-resistant variants despite fitness costs was investigated by van Maarseveen et al [34
], who argued that viral variants with both PI mutations and compensatory mutations must travel through a “fitness valley” to lose both types of mutations— first reducing fitness in order to ultimately maximize it. Our findings are consistent with the theoretical framework advanced by these 2 studies and with prior reports of infrequent replacement of PI mutations [11
The clinical implication of our finding that PI mutations are replaced at a similar rate to NNRTI mutations is that one potential advantage of including PIs in first-line ART regimens in resource-limited settings—namely, faster replacement of transmitted mutations by wild-type and lower risk of spread of resistance—does not appear to be present. The risk of drug resistance propagation is further amplified by the substantial fraction of all HIV transmissions that occur from source patients who have acute and/or early HIV infection [35
], implying that, in many patients with TDR, the available time for mutation replacement before retransmission may be quite limited.
A second clinical implication of our findings relates to the rapid replacement of M184V that we observed. Because of the frequent use of drugs that select for this mutation, should a patient who has received a new diagnosis of HIV who has other mutations be assumed to have lamivudine/emtricitabine-resistant virus even if this is not present on genotyping? The rapid mutation loss that we observed suggests that the lack of M184V in this context may be attributable to replacement, which is supported by a prior study showing that very early replacement of M184V is common: 11% of persons with early HIV infection had M184V variants detected using a minor variant assay but not in population sequencing [37
]. Other studies, however, have shown that genotyping with population sequencing is usually a reliable guide to choosing therapy for patients with TDR [38
]. One way to reconcile these studies is that M184V variants may rapidly decrease to levels that are not clinically significant, leaving a relatively short window during which clinically important frequencies are missed. Precise quantitation of the prevalence of M184V variants after the loss of detection on population sequencing, along with correlation with treatment outcomes, could help better define the threshold for clinically significant levels of M184V.
In analyzing predictors of mutation replacement, we made 2 observations that together raise interesting questions about host-virus interactions. First, we hypothesized that higher viral load would predict faster mutation loss but found only a modest trend. Second, our study included many persons with multiple baseline mutations, allowing us to assess whether there are person-level factors influencing mutation replacement. We found that, after accounting for viral load and mutation drug class, there was marked person-to-person variability in the likelihood of mutation replacement. This suggests that there are additional patient-level factors driving mutation loss. Our observations about mutation replacement are consistent with the concept that viral evolution, rather than being driven exclusively by selective (deterministic) events based only on the fitness costs of mutations, may also be driven by stochastic forces. This idea that evolution is influenced by a combination of factors and that the effective viral population size is not simply reflected by the plasma HIV RNA level has been supported by seminal studies [40
] and helps contextualize our results.
Our study has several important limitations. First, we used population sequence genotypes rather than more sensitive methods capable of detecting minor variants. This limits our ability to assess mutations below the detection threshold of conventional sequencing, which may persist in the viral quasi-species in sufficient quantities to remain clinically significant. This phenomenon has been illustrated by studies of the K103N mutation among women who received single-dose nevirapine to prevent vertical HIV transmission [43
]. Our inability to detect M184V minor variants in particular suggests that our data may underestimate the rate at which replacement with wild-type virus occurs. Second, we included 11 individuals who received and subsequently interrupted ART. Because viral evolution is minimal or nonexistent during effective therapy [44
], we do not believe that this affected our central observations. Third, because of the challenges of identifying patients with very early HIV infection, we could not determine whether any of our participants acquired mutations and lost them in the earliest weeks of infection. Fourth, we did not assess all factors that might influence replacement rates, such as viral tropism. Although the effective viral population size discussed earlier is highly pertinent to analyzing rates of mutation replacement, estimating this factor is complex and requires measurement of multiple clonal populations that we did not perform. Finally, caution should be exercised in generalizing our results from sexual transmission cases to injection drug use transmission. Because intravenous infections have been shown to be caused by a multiplicity of viral variants in most cases [47
], whereas the majority of sexual transmissions are caused by single virions [48
], there may be more rapid replacement of TDR in transmissions related to injection drug use.
Overall, our data indicate that transmitted M184V/I mutations are unique in their high propensity to wane below the detection threshold of population genotyping over time. Clinicians assessing new patients should consider the possibility that mutations not present in baseline genotyping may have been present previously and may have been replaced with wild-type viral variants, particularly in the case of M184V/I. Further investigation is warranted on both host and viral factors that are influencing mutation replacement.