Model-based anti-HIV therapy [article]

Tobias Sing, Universität Des Saarlandes, Universität Des Saarlandes
2009
Modern combination drug therapy has substantially improved the clinical management of HIV-1 infection. Still, the emergence of drug-resistant variants eventually leads to therapy failure in most patients. The selection of an optimal follow-up regimen is complicated by an ever-increasing range of possible drug combinations. In this thesis, we present foundations for rational, model-based treatment strategies. Firstly, we study viral evolution. In a simulation study, we establish a general link
more » ... tween the shape of a fitness landscape and population dynamics, using an idealized population undergoing mutation, recombination and selection at three biallelic loci as an example. Using techniques from survival analysis, a model of mutation dynamics in the absence of drug is proposed. Differently from mutation accumulation, mutations are found to disappear independently from each other, but with individual survival probabilities. A Fisher kernel for mixtures of mutagenetic trees is derived, quantifying the similarity of evolutionary escape from drug pressure between two viral sequence samples. Kernel-based prediction of drug resistance leads to significant improvements over an evolution-agnostic approach. Secondly, the controversial interplay between genotypic, phenotypic, and clinical resistance is analyzed. Methods for identifying resistance mutations from either in vitro or in vivo data, and for characterizing mutational covariation patterns are described. A case study focusing on reverse transcriptase inhbitors yields over 20 previously undescribed mutations, most of them extending classical resistance pathways. Finally, the widely held view of an ill-defined relation between phenotypic and clinical resistance is challenged, and a hybrid model incorporating both genotypic and inferred phenotypic information is shown to outperform its components in predictivity. Surprisingly, the incorporation of viral fitness does not lead to any further improvements. Thirdly, genotypic, clinical, host-specific, and structural determinants of viral cell entry via the human receptors CCR5 and CXCR4 are investigated. Previously undescribed mutations in the third hypervariable region of the viral envelope protein, CD4 + cell counts, host heterozigosity for the CCR5-∆32 allele, number of sequence ambiguities arising from population sequencing, and presence of indels are shown to be predictive of coreceptor choice alone and in combination, as are changes in the predicted side-chain conformation in a three-dimensional model of the V3 loop. vi This work was carried out in the
doi:10.22028/d291-25925 fatcat:gs4dsgnzg5czzfanlcx7fdqpu4