Background: The concordance between drug resistance measured directly by phenotypic susceptibility assays or predicted by interpretive genotyping methods is weakest for newly introduced drugs with limited clinical experience.  For LPV, a reduction in susceptibility of ≥ 10-fold is associated with decreased likelihood of clinical response, and 23 mutations at 11 positions in protease (10, 20, 24, 46, 53, 54, 63, 71, 82, 84, and 90) have been associated with decreased response when combinations of 6 or more mutations are present. However, these genotypic correlates are likely to be incomplete and may not be applicable to all patient populations.

Methods: A training data set of 2038 patient samples was analyzed.  2 groups of viruses with discordant genotypic and phenotypic results were defined: a false negative group with a LPV fold-change > 10-fold but < 6 mutations, and a false positive group with a LPV fold-change < 10-fold but 6 or more mutations.  The prevalence of specific mutations in the false negative group was compared to that in the true negative (LPV fold-change < 10 and < 6 mutations) group using Fisher’s exact test; comparisons with p<0.001 were considered significant. Only positions with changes that occurred in >1% of the samples and recognized resistance mutations were considered.

Results: Using the existing list of LPV mutations, there were 9% false negative and 5% false positive samples (overall concordance 86%). When samples containing mixtures in at least one LPV resistance-associated position were excluded, the number of false positive samples decreased (2% false positive, 10% false negative; remaining n=1402). Mutations significantly associated with the false negative group fell into 1 of 3 categories: those already in the algorithm (I54V/T, V82A/F, K20M, L10F), new amino acids at known positions (I54A/M/S, V82S, K20I), or new mutations (including I50V, G48V, I47V, E34D/K/Q, G73C/T, V32I, L33I/F/V, K55R). A new algorithm was designed based on these findings that will be used to analyze a validation data set of approximately 1000 previously unseen samples.

Conclusions: Accurate and comprehensive genotypic interpretation systems require extensive analysis of paired phenotypic and genotypic data, guided by phenotypic clinical cut-offs. In the absence of complex predictive algorithms, genotyping may underscore viruses with clinically significant reductions in drug susceptibility.