Background:  Quantitative phenotypic resistance information interpreted via clinical cut-offs can facilitate optimization of combination antiretroviral therapy. Virtual phenotype-LM (Linear Model) predictions of atazanavir (ATV) and fos-amprenavir (fAPV) drug susceptibility and clinical cut-offs are described.

Methods:  Based on clinical isolates with both drug-susceptibility phenotypes (Antivirogram) and viral genotypes, linear regression models were developed to predict ATV or amprenavir (APV) fold change in IC₅₀ from the viral genotype (Virtual phenotype-LM). Using treatment response data from clinical trials and cohorts, separate linear regression models were developed to predict 8-week change in viral load for regimens including ritonavir-boosted ATV (ATV/r) or fos-amprenavir (fAPV/r) as a function of baseline viral load, protease inhibitor (PI) fold change, activity of the background regimen, enfuvirtide (T-20) use, and, for ATV, concomitant tenofovir (TDF) use. We defined 2 clinical cut-offs, corresponding to predicted fold change values associated with a 20% or 80% loss of the boosted PI response of subjects infected with wild type strains.

Results:  Virtual phenotype-LM predictions of ATV and APV fold change weigh the contributions of multiple protease gene mutations and mutation pairs, and provide an accurate prediction of the measured fold change (R = 0.94, n >15,000 for ATV; 0.92, n ≥43,000 for APV).  Virtual phenotype-LM predicted resistance ranged from 0.8-fold change for wild type isolates to 78.2-fold change for ATV and 38.5-fold change for APV.  We observed 20% (clinical cut-off 1) and 80% (clinical cut-off 2) loss of the predicted boosted-PI response in treated patients at 2.7 and 32.9 ATV fold change and at 1.3 and 11.4 APV fold change. In the analysis datasets, patients receiving ATV/r-based regimens with baseline ATV fold change smaller than clinical cut-off 1 (n = 216), between clinical cut-off 1 and clinical cut-off 2 (n = 31), and larger than clinical cut-off 2 (n = 11) had median viral load reductions of –2.0, –1.1, and –0.25 log₁₀ HIV RNA at 8 weeks, respectively. Patients receiving FPV/r-based regimens with baseline APV fold change less than clinical cut-off 1 (n = 69), between clinical cut-off 1 and clinical cut-off 2 (n = 47), and more than clinical cut-off 2 (n = 20) had median 8-week reductions in viral load from baseline of –1.77, –0.75, and –0.20 log₁₀ HIV RNA.

Conclusions:  We have integrated complex interactions among multiple protease gene mutations to provide accurate quantitative predictions of ATV and fAPV susceptibility, which can be interpreted utilizing drug-specific clinical cut-offs.