The main finding from this study supports the idea that the class of antiretroviral agents used in sequential therapy determines the viral diversity. By using the key resistance mutations as markers for determining the persistence of viral species, the data demonstrate that initial therapy with efavirenz-based regimens produce robust resistant species onto which failure from a nonboosted PI-based regimen adds. This is in distinction to individuals who fail a nonboosted PI-based regimen followed by a second regimen containing efavirenz. In this case, the data show in most cases EFV mutations add to a dominant viral species that no longer contained the key PI mutation. These data do not rule out the possibility that other minority species exist that carry both the PI and NNRTI mutations, or other mutations that genotyping may miss. Alternatively, because of how genotyping is performed, the existence of two mutations does not mean that they are necessarily on the same viral species when there is no drug pressure. However, in the current study, the fact that the viral outgrowth occurred on a suppressive regimen that contained both mutations suggests they are on the same virus. If they were not, the viral species that did not contain the resistance mutation would be suppressed.
The data from the ACTG-384 study in the Stanford HIV Database provided an opportunity to generate phylogenetic trees that permitted further investigation of the drug-sequencing hypothesis. The branch length differences shown in strongly suggest that the sequence of therapy impacts on viral diversity of the dominant breakthrough virus. We note that the consistency of the results of the two approaches we considered—one based only on the proportion of patients who carried a mutation from the first failure to the second, and the other based on phylogenetic analysis—strengthens these findings despite the relatively small sample size.
One explanation for the impact of sequence of therapy on diversity is the fitness of the dominant viral species. A number of studies have demonstrated that a K103N mutation has little effect on viral fitness,7,8
whereas protease inhibitor signature mutations exact a fitness cost.9,10
Our results would arise from addition of the new mutations at failure of the second regiment to the most fit virus. Numerous in vitro
studies have looked at the fitness of an HIV virus with mutations that occur after the failure for the therapies shown in .9–15
In general, these studies support the idea that mutations from the second failure did add to the most fit virus. Time from first to second failure was not significantly different between the two arms and would not account for the differences in diversity (data not shown).
Although the current preferred PI-containing therapies are ritonavir-boosted, these data demonstrate that the order of different antiretroviral class regimens in sequential therapeutic failures may be important in the development of resistant virus. Moreover, this study has relevance in developing countries since nonboosted PIs continue to be used (see, for example, Ref. 16
.) Because longitudinal data was available on these patients, we were able to control for a number of variables not available in large databases. Knowing the therapy history and whether the patient has been on continuous therapy is important in understanding evolutionary pressures. Evaluating genotypes after failure from a nondetectable viral burden provided an opportunity to study the addition of the signature mutations by diminishing the possibility of recombination.17–19
Predictive programs draw their power from statistical evaluation of large databases of point in time genotypes. Because the patient specific detail is not available from these databases, there will always be inherent scatter that may confound predictions. The possibility that resistant mutations can be dispersed among different virions by appropriate ordering of sequential therapy may preserve therapeutic options in salvage therapy that are not available when all the key mutations are on a single virion.20
Treatment of HIV continues to be complex and based on numerous factors. The role for informatics in guiding physicians in therapeutic decision-making will continue to expand. Combining across large longitudinal databases will grow in importance and will be needed in order to generate sufficient statistical power to address questions such as the one poised in this current work.