Background:  Current prediction of tipranavir (TPV) susceptibility based on genotype uses an algorithm that counts mutations associated with reduced in vitro susceptibility or in vivo virological response. This algorithm (TPV mutation score) was derived from analysis of a limited number of samples from phase 2 and 3 clinical trials and considers the following mutations: L10V, I13V, K20M, R, or V, L33F, E35G, M36I, K43T, M46L, I47V, I54A, M, or V, Q58E, H69K, T74P, V82L or T, N83D, I84V. We sought to assess the accuracy of the TPV mutation score in an unrelated collection of clinical samples from the Monogram database.

Methods:  A dataset consisting of samples containing at least 1 protease inhibitor (PI)-resistance mutation, but no more than 1 mixture at a position in the existing TPV mutation score (n = 1411), was analyzed. Samples with off-scale TPV resistance were assigned a fold-change (FC) value of 100. The median TPV FC for samples in each category defined by TPV ms was calculated, and samples were grouped based on measured TPV FC being higher or lower than the median for each group. Fisher’s exact test with the Benjamini correction for multiple comparisons was used to determine which PR mutations were associated with higher or lower TPV FC. The odds ratio (OR) for each mutation (percentage of higher samples with the mutation to percentage of lower samples with the mutation) was calculated.

Results:  TPV FC (log-transformed) was correlated with TPV mutation score, with a linear regression coefficient of 0.51 (p <0.0001). Median TPV FC values are given in the table. The following mutations were over-represented (OR >1) in samples with higher TPV FC (adjusted p <0.05):  L10I, V11L, V32I, M36L, M46I, I47V, I54A, K55R, D60E, A71L, G73T, V82T, I84V, L89V, L90M (underlined mutations in existing TPV mutation score). Conversely, L10F or V, I13V, K20R, L24I, D30N, M36I, M46L, I50L or V, I54L, L76V, V82I, and N88D were associated with lower TPV FC (OR <1). A revised scoring algorithm incorporating this information had a linear regression coefficient of 0.66.

Conclusions:  The TPV FC variability within a given group defined by mutation score is high, especially in ranges thought to be clinically relevant. Currently available genotypic algorithms do not capture all mutations which impact susceptibility. We have identified additional PR mutations which contribute to this variability. Identification of these mutations and association with susceptibility are necessary to derive a more advanced TPV genotype interpretation algorithm.