Towards assessing therapy failure in HIV disease: estimating in vivo fitness characteristics of viral mutants by an integrated statistical-mechanistic approach
Sathej Gopalakrishnan (1,2), Hesam Montazeri (3), Stephan Menz (4), Niko Beerenwinkel (3), Wilhelm Huisinga (4)
(1) Institute of Biochemistry and Biology, Universitaet Potsdam, Potsdam, Germany, (2) PharMetrX Graduate Research Training Program, Freie Universitaet Berlin and Universitaet Potsdam, Germany, (3) ETH Zurich, Department of Biosystems Science and Engineering (D-BSSE), Basel, Switzerland, (4) Institute of Mathematics, Universitaet Potsdam, Potsdam, Germany
Objectives: Mechanistic viral infection models have long been used to investigate in vivo HIV-1 dynamics [1], while statistical models [2] have been applied to learn mutational schemes from sparse clinical data. While the former approach generally uses highly simplified mutation schemes, the latter methods limit mechanistic understanding. Combining the two approaches would prove valuable to estimate viral fitness characteristics and to assess causes of therapy failure. The objective of this work was to link the two modelling strategies and to evaluate the integrated approach by comparing estimated in vivo fitness characteristics of various mutants arising under monotherapy with Zidovudine (AZT) and Indinavir (IDV) to values in literature.
Methods: We used a two-stage mechanistic HIV infection model [3] to predict in vivo viral dynamics. We utilized continuous time conjunctive Bayesian networks [4] to learn mutational schemes, resistances and average waiting times to mutations from clinical data (the Stanford HIV Database) [5]. Simulations and estimation procedures were carried out with MATLAB R2010b.
Results: We translated phenotypic IC50 values from in vitro assays to an in vivo setting that we use in our mechanistic model. We then developed an approach to compare the average statistical waiting times and mechanistically predicted waiting times to observe various mutations. Estimating in vivo fitness costs of different mutants arising under AZT (a reverse transcriptase inhibitor), we recovered the well-known TAM-1 and TAM-2 mutation pathways and observed excellent agreement with existing knowledge. We also reproduced the interesting observation that the rebound of the wild-type strain, in case of insufficient drug efficacy, contributes significantly to the initial rebound in the total viral load and the wild-type is out-competed by the mutants only after about 60-70 days. We finally estimated fitness costs of mutants arising under IDV (a protease inhibitor) therapy to demonstrate the generality of our approach.
Conclusions: Our scheme relies only on clinical data that is typically censored and is usually the most commonly available form of data. Such integrated mechanistic-statistical approaches provide a first step towards analysis of HAART regimens.
References:
[1] Ho DD, Neumann AU, Perelson AS, Chen W, Leonard JM, et al. (1995) Rapid turnover of plasma virions and CD4 lymphocytes in HIV-1 infection. Nature 373: 123-126.
[2] Beerenwinkel N, Daeumer M, Sing T, Rahnenfuehrer J, Lengauer T, et al. (2005) Estimating HIV evolutionary pathways and the genetic barrier to drug resistance. J Infect Dis 191: 1953-1960.
[3] von Kleist M, Menz S, Huisinga W (2009) Drug-class specific impact of antivirals on the reproductive capacity of HIV. PLoS Comput Biol 6(3): e1000720. doi:10.1371/journal.pcbi.1000720
[4] Beerenwinkel N, Sullivant S (2009) Markov models for accumulating mutations. Biometrika 96: 645-661.
[5] Rhee SY, Gonzales MJ, Kantor R, Betts BJ, Ravela J, et al. (2003) Human immunodeficiency virus reverse transcriptase and protease sequence database. Nucleic Acids Res 31: 298-303.
