Detecting and understanding combinatorial mutation patterns responsible for HIV drug resistance

Abstract

We propose a systematic approach for a better understanding of how HIV viruses employ various combinations of mutations to resist drug treatments, which is critical to developing new drugs and optimizing the use of existing drugs. By probabilistically modeling mutations in the HIV-1 protease or reverse transcriptase (RT) isolated from drug-treated patients, we present a statistical procedure that first detects mutation combinations associated with drug resistance and then infers detailed interaction structures of these mutations. The molecular basis of our statistical predictions is further studied by using molecular dynamics simulations and free energy calculations. We have demonstrated the usefulness of this systematic procedure on three HIV drugs, (Indinavir, Zidovudine, and Nevirapine), discovered unique interaction features between viral mutations induced by these drugs, and revealed the structural basis of such interactions.Bayesian model selection | free energy calculation | Markov chain Monte Carlo | molecular dynamics | mutation interactions H IV drug-resistance, which is caused by mutations of viral proteins that disrupt the drugs' binding but do not affect the viral survival, is a major hurdle that hinders a successful treatment of AIDS (1, 2). Due to the high rate and low fidelity of HIV replication, resistant strains quickly become dominant in a viral population under the selection pressure of a drug. By sequencing viral strains in the treated-patient isolates, genotypic data have been accumulated for the drugs targeting two viral enzymes, protease and reverse transcriptase, that are essential to the virus's replication. Because each mutation of the viral protein is not equally important for drug resistance, the observed, complicated mutation patterns are difficult to interpret (3, 4) and are limited in helping physicians design the best therapeutic regimen for a patient (5) (Fig. 1A).In past decades, many statistical learning methods (3, 4, 67-8) have been employed to help predict phenotypes from genotypes. There are also rule-based systems that infer drug-resistance levels from sequence information such as the Stanford University HIV Drug Resistance Database (Stanford HIVdb). However, these methods provide little insight on the genetic and molecular basis of drug resistance and often give inconsistent results when analyzing the same input mutation data (4, 6).In the present study, we investigated the problem of mutation interactions of the HIV induced by a certain drug treatment. Using a unique probabilistic model, we first detect resistant mutation combinations (9) and infer the interaction dependence structure of these combinations. Then, we use molecular dynamics (MD) simulations to reveal the molecular basis of how these mutations interact with each other to interfere with the drugs' binding. We have shown that our procedure is applicable to different antiretroviral drugs treating different types of HIV infection by analyzing the sequence mutations induced by three different drug treatments: a...

Novel HBsAg markers tightly correlate with occult HBV infection and strongly affect HBsAg detection

“…Bayesian Variable Partition Model and Recursive Model Selection were carried out according to the methodology described in Zhang et al (2010). A detailed description of these models is reported in Supplementary Text 2.…”

Identification and Structural Characterization of Novel Genetic Elements in the HIV-1 V3 Loop Regulating Coreceptor Usage

“…Bayesian variable partition model and recursive model selection were carried out according to the methodology described in Zhang et al [20]. A detailed description of these models is reported in Additional file 2.…”

A Bayesian Method for Disentangling Dependent Structure of Epistatic Interaction

“…Although this type of Bayesian recursive partition idea has been successfully used in several difference scenarios (Zhang et al, 2010;Svicher et al, 2011a;2011b) the current RBP model can be further improved in several ways. First, this model does not consider missing genotypes and unobserved SNPs in the casecontrol sample, but this can be treated via imputation or EM algorithm incorporated in RBP model.…”

“…That is, only within each group, there are some interactions. Then, RBP uses ChainDependence Model (CDM) to recursively infer the conditional independence within each interaction group (Zhang et al, 2010).…”