Evaluation of a molecular modelling decision support system for resistance testing interpretation applied to ritonavir-related

Molecular Tridimensional Phenotyping Predicts HIV-1 Protease Inhibitor
Resistance

G Pèpe1,2, J Courcambeck2, J Durant3, P Dellamonica3, R Boulme4 and P Halfon2,5

1 CNRS, Marseille, France; 2 Molecular Modelling Unit, Genoscience, Marseille,
France; 3 Dept. Infectious Diseases, Nice, France; 4 ABL, Luxembourg; 5
Alphabio laboratory, Marseille, France
BACKGROUND: Drug resistance development is a major limitation for successful highly
active antiretroviral therapy. HIV genotyping and phenotyping resistance testing are the
most commonly used methods to guide treatment decisions. Clinical interpretation of
genotypic tests and determination of the phenotyping test clinical cut off are still limited
regardless time consuming and costly. We developed a 3Dimensional molecular computer-
based modelling system which rapidly predicts HIV-1 protease drug resistance by measuring
the binding energies between a protease mutant (PM) and a protease inhibitor (PI).
METHODS: HIV-1 protease mutants were built from X-ray crystal structure corresponding
to each PI complex. The GenMol™ (GM) program (www.3dgenoscience.com) was used to
generate amino-acid mutations by energy minimisation. The PI-PM binding resulting energy
complex was determined after energy refinement. We applied the methodology to ritonavir.
(i) We compared the actual predicted GM ritonavir result with genotypic data interpretation
(ABL ViroScorer™) for cut-off determination and concordance. (ii), We assessed the
correlation between GM ritonavir values and phenotyping results (PhenoSense™, resistance
cut-offs of >2.5 fold-resistance (FR) and >10). (iii), We evaluated the baseline GM
predictive value on virological outcome at month 3 and 6 among 48 patients from the
Viradapt trial who had been adapted with ritonavir.
RESULTS: Based on 129 HIV-1 protease sequences, an energy cut-off value (-91.1) was
determined for the GM analysis. (i) Individual predictive interpretation as resistant or
susceptible were determined with a good agreement (k>0,80). (ii) Significant correlations
between binding energies and ritonavir phenotyping results were found without excluding
outliers. Kappa scores were 0.55 and 0.67 for FR>2.5 and FR>10, respectively. The
association between log10(FR) and GM value was Rs=0.61 (Spearman’s correlation). (iii) At
M3, median GM value of –90.21 and –92.01 was significantly different (p<0.01, Wilcoxon)
for virological failure and success, respectively. Identically at M6, median GM value of –
90.78 and –92.03 was significantly different (p<0.01). GM values were entered in a logistic
model with baseline viral load (bvl) and genotypic-guided treatment covariates. Using cut-
offs –91.0 and –90.5, GM predicts a virological outcome in the uni- (odds-ratio=5.09, 95%
confidence interval (CI)=1.55-22.58, p<.01 (Wald)) and the multivariate (OR=5.89,
CI=1.53-22.60, p<.01 (bvl-GM) model.
CONCLUSIONS: GenMol™, a novel 3Dimensional molecular modelling method was
developed to predict HIV-1 ritonavir drug resistance. 2) Good agreement was reported with
genotypic and phenotypic interpretation. 3) Other PIs GM validation is underway 4) Further
work is required to empower first findings about GM as an independent predictor of
virological outcome in prospective clinical trials.

Source: http://www.3dgenoscience.com/molecular_modeling/3dgm_rtv_seville/3dgm_rtv_seville.pdf