Identifying Genetic Correlates of HIV-1 Drug Resistance Using a Multifaceted Approach: Genotype-Treatment, Genotype-Phenotype, and Genotype-Clinical Outcome Correlations

HIV-1 drug-resistance knowledge is critical for antiretroviral (ARV) drug design, the management of HIV-1 infected patients, and estimating transmitted drug resistance (TDR). However, understanding the genetic basis of HIV-1 drug resistance is hindered by the high levels of naturally occurring polymorphisms and large numbers of mutation patterns associated with cross-resistance within each antiretroviral drug class.

I studied three types of data correlations to investigate the genetic mechanisms of HIV-1 drug-resistance: (i) genotype-treatment correlations: which mutations are selected in patients at treatment failure; (ii) genotype-phenotype correlations: which mutations are associated with reduced in vitro drug (phenotypic) susceptibility; (iii) genotype-virological response correlations: which mutations interfere with the virological response to a new ARV therapy.

Treatment-selected mutations (TSMs) provide Darwinian evidence suggesting that the mutations selected by a treatment are likely to reduce susceptibility to that treatment. Genotype-treatment correlation studies allow us to identify TSMs which then can be further investigated in drug susceptibility studies to inform genotypic resistance interpretation systems. Another application of genotype-treatment correlation studies is to identify TSMs that are indicative of TDR in ARV-naïve persons. We identified additional TSMs from large datasets and published a list of non-polymorphic TSMs suitable for surveillance of TDR. The individual-patient and sequence-level meta-analysis of TDR presented in the thesis underscores the importance of making sequence data publicly available. The availability of individual sequences and the associated data made it possible to apply uniform methods for analyzing geo-temporal trends of TDR.

Drug-resistance arises from complex patterns of mutations in patients failing ARV therapy. Understanding how much each individual mutation contributes to resistance against each ARV is essential for using genotypic resistance testing to select optimal therapies. Genotype-phenotype correlations were studied by performing statistical learning methods on a large set of clinical isolates for which phenotypic drug resistance test results were available. These studies allowed us to estimate the relative contribution of a drug-resistance mutation to reduced ARV susceptibility and the extent of cross-resistance, which further led us to improve genotypic drug-resistance interpretation.

Studying the direct correlation between mutant viruses and the virological response to a new ARV therapy is challenging because many factors can influence the clinical outcomes and multiple ARVs are used in each regimen in many different combinations. As a result, experts have developed algorithms for interpreting HIV-1 genotypic drug-resistance test results. These algorithms have been widely used for selecting optimal ARV combinations. As new ARVs are licensed and new knowledge about drug resistance accrues leading to change in the algorithms, the predictive value of these algorithms should be analyzed using contemporary genotype-clinical correlation datasets. We presented an analytical framework for validating the predictive values of interpretation algorithms, and demonstrated the need to facilitate sharing of clinical data for the testing of novel hypotheses pertaining to the optimal use of salvage ARV therapy.

Three types of correlation studies have contributed to HIV-1 drug resistance knowledge. Efforts by researchers in making HIV-1 drug resistance data publicly available has facilitated HIV-1 drug resistance knowledge discovery. The public availability of such data allows meta-analyses and application of other types of combined bioinformatics methods. The same methodologies used to gain HIV-1 drug resistance knowledge can also be applied to studying drug resistance in other viruses.