Sathej Gopalakrishnan (1,2), Hesam Montazeri (3), Stephan Menz (4,5), Niko Beerenwinkel (3), Wilhelm Huisinga (4)
(1) Institute of Biochemistry and Biology, Universität Potsdam, (2) Graduate Research Training Program PharMetrX, Freie Universität Berlin and Universität Potsdam, (3) ETH Zurich, Department of Biosystems Science and Engineering (D-BSSE), Basel, (4) Institute of Mathematics, Universität Potsdam, (5) Current address: Bayer HealthCare Pharmaceuticals, Berlin
Objectives: The emergence of drug-resistant mutants remains a challenge to successful anti-human immunodeficiency virus (HIV) therapy [1]. Upon treatment failure, the choice of an optimal salvage therapy depends on the dominant viral mutant genotypes present. Our objective were (i) to assess genotypic reasons for therapy failure with multi-drug regimens and to determine how each drug in a combination-regimen individually affects treatment outcome; (ii) to demonstrate the benefit of regularization techniques developed for linear systems, in stabilizing parameter estimation for our nonlinear viral dynamics model.
Methods: We used a validated two-stage viral infection model [2] to predict in vivo HIV dynamics. We incorporated drug-specific mutation pathways and resistance factors estimated from clinical data (HIV Stanford drug resistance database [3]). See [4] for details. The fitness costs of different mutant genotypes that arose under anti-retroviral monotherapy with two different drugs – zidovudine (ZDV), a nucleoside reverse transcriptase inhibitor (NRTI) and indinavir (IDV), a protease inhibitor (PI), were estimated by simulated annealing [4]. To study the impact of regularization on parameter estimation, we used ridge regression [5]. Simulations and estimation procedures were carried out with MATLAB R2010b.
Results: We examined the fitness of the viral population during the course of treatment with ZDV and IDV and identified evolutionary bottlenecks in the path to full resistance. Simulations of the two-drug regimen showed that the viral composition at therapy failure strongly depended on drug efficacies. We identified combinations of individual drug efficacies where virological failure could be due to a) wild-type; b) PI-resistant mutations; c) NRTI-resistant mutations; or d) both NRTI and PI resistant mutations. Importantly, the times to failure with monotherapy regimens were not additive when predicting the time to failure with combination therapy. Finally, we showed that regularization techniques are beneficial in stabilizing estimates of fitness costs in our nonlinear model.
Conclusions: We demonstrated how clinical data from monotherapy can be leveraged to predict viral dynamics under combination regimens. This is a step towards studying potential salvage regimens upon treatment failure.
References:
[1] Ho DD et al. (1995) Rapid turnover of plasma virions and CD4 lymphocytes in HIV-1 infection, Nature 373: 123-126,[2] von Kleist M et al (2009) Drug-class specific impact of antivirals on the reproductive capacity of HIV, PLoS Comput Biol 6(3): e1000720, [3] Rhee SY et al (2003) Human immunodeficiency virus reverse transcriptase and protease sequence database, Nucleic Acids Res 31: 298-303. [4] Sathej Gopalakrishnan et al (2013) Towards assessing therapy failure in HIV disease: estimating in vivo fitness characteristics of viral mutants by an integrated statistical-mechanistic approach, PAGE Poster 2013, Abstr. 2909 [5] Hoerl AE et al (1970), Ridge regression: Biased estimation for nonorthogonal problems, Technometrics 12: 55-67.
Reference: PAGE 23 () Abstr 3274 [www.page-meeting.org/?abstract=3274]
Poster: Methodology - New Modelling Approaches