icon star paper   Articles  
Back grey_arrow_rt.gif
 
 
Cross-Resistance Between Amprenavir and Kaletra
 
 
  "Improving lopinavir genotype algorithm through phenotype correlations: novel mutation patterns and amprenavir cross-resistance"
 
AIDS 2003; 17(7):955-961. Neil T. Parkin; Colombe Chappey; Christos J. Petropoulos
 
Introduction. Protease inhibitors (PI) are important components of many highly active antiretroviral therapy (HAART) combination regimens. Resistance to PI is one of the major challenges inherent in the selection of HAART regimens for treatment-experienced patients. Lopinavir (LPV, ABT-378), co-formulated with ritonavir (LPV/r; Kaletra) has shown great promise in this area. LPV plasma levels are increased greatly by ritonavir, resulting in antiviral activity against viruses with reduced PI susceptibility.
 
Genotypic (GT) and phenotypic (PT) correlates of resistance to LPV/r have been investigated in vitro and in vivo. In vitro passage of HIV-1 in the presence of LPV resulted in the selection of viruses with reduced drug susceptibility and several mutations in the HIV protease. Analysis of viruses from PI-experienced patients (n = 112) entering phase III clinical trials of LPV/r in combination with other agents identified 23 mutations at 11 positions in protease associated with decreased LPV susceptibility (referred to here as LPV mutations): L10F, I, R, or V; K20M or R; L24I; M46I or L; F53L; I54L, T, or V; L63P; A71I, L, T, or V; V82A, F, or T; I84V; and L90M. Subsequent analyses linked a reduction in LPV susceptibility, as measured by the fold change (FC) in 50% inhibitory concentration (IC50) in vitro relative to a wild-type reference, of > 10-fold, or the presence of six or more LPV mutations, with diminished viral RNA responses during treatment with LPV/r. These clinical data were used to establish the LPV susceptibility cut-off in PT tests (FC > 10) and of the LPV mutation score (six or more mutations) for predicting resistance on GT tests.
 
Since it is likely that there are multiple mutational pathways that could result in reductions in susceptibility > 10-fold, we wished to determine how well this genotypic interpretation algorithm predicts reduced susceptibility to LPV in a larger and more diverse sample set than previously examined. The present study compares genotypes and phenotypes from a database of patient samples that were submitted for routine drug resistance testing. The analysis identified numerous mutations associated with reduced LPV susceptibility that are not included in the initial genotypic algorithm. Importantly, mutations that are known to contribute to amprenavir (APV) resistance were among these newly identified mutations. Current genotypic algorithms suggest that the HIV-1 protease inhibitors (PI) lopinavir (LPV) and amprenavir (APV) have distinct resistance profiles. However, phenotypic data indicate that cross-resistance is more common than expected.
 
ABSTRACT. Protease genotype (GT) and phenotype (PT) from 1418 patient viruses with reduced PI susceptibility and/or resistance-associated mutations (training data) were analyzed. Samples were classified as LPV resistant by GT (GT-R) if six or more LPV mutations were present, and by PT (PT-R) if the 50% inhibitory concentration (IC50) fold-change (FC) was over 10.
 
There were 182 samples (13%) that were GT-S but PT-R for LPV. A comparison of the mutation prevalence in PT-R/GT-S samples with that in PT-S/GT-S samples identified mutations associated with LPV PT-R. Several previously defined LPV mutations were found to have a stronger than average effect (e.g., M46I/L, I54V/T, V82A/F), and new variants at known positions (e.g., I54A/M/S, V82S) were identified. Other mutations, including known APV resistance mutations, were found to contribute to reduced LPV susceptibility. A new LPV genotypic interpretation algorithm was constructed that improved overall genotypic/phenotypic concordance from 80% to 91%. The algorithm demonstrated a concordance rate of 90% when tested on 523 new samples. Cross-resistance between APV and LPV was greater in samples with primary APV resistance mutations than in those lacking them.
 
The current LPV mutation score does not fully account for many resistant viruses. Consequently, cross-resistance between LPV and APV is underappreciated. Phenotypic results from large and diverse patient virus populations should be used to guide the development of more accurate GT interpretation algorithms.
 
Results
 
To obtain an initial estimate of the accuracy of the LPV mutation score [7, 8] as a predictor of reduced LPV susceptibility, LPV FC was plotted against the mutation score for 1418 viruses derived from patient plasma samples. While there was clearly a trend for susceptibility to decrease as the mutation score increased (linear regression coefficient r2 = 0.50), there was considerable variability in LPV susceptibility within each mutation score category. Although the phenotypic assay provided a more quantitative assessment of drug susceptibility than the genotypic assay, both test results were categorized as either sensitive (S; FC < 10 or fewer than six LPV mutations) or reduced susceptibility (R, or resistant; FC > 10 or six or more LPV mutations) to simplify the determination of concordant or discordant test results. Based on this categorical analysis, the overall concordance between PT and GT was 79%. Another 13% of the samples were PT-R/GT-S and 8% were PT-S/GT-R.
 
Mutations in the current LPV mutation score that are over-represented in PT-R/GT-S viruses included L10F or I; K20M; M46I or L; I54T or V; and V82A or F. This suggests that these mutations, particularly I54T/V and V82A/F (OR > 9), may have a stronger effect on LPV phenotype than others in the algorithm. New amino acid variants at known positions included K20I; I54A, M, and S; L63T; and V82S. These mutations are likely to have a similar impact on LPV susceptibility as other previously identified resistance mutations at the same amino acid positions, but since they are less prevalent, they were not identified in earlier studies performed with smaller numbers of patient samples.
 
The list of LPV resistance-associated mutations at novel amino acid positions that are not part of the current LPV mutation score includes several recognized PI resistance mutations such as G16E, V32I, L33F, I47V, G48M/V, I50V and G73T. G16E, V32I, and I47V have been selected by in vitro passage of HIV in cell cultures in the presence of LPV. Other mutations in this category (E34Q, K43T, Q58E, T74S, and L89I/M) have not previously been associated with PI resistance.
 
Based on these findings, a modified LPV mutation score algorithm was developed that included the mutations in Table 2 (an additional 20 mutations and 12 positions) and increased the weight assigned to positions 50, 54, and 82. The optimum number of mutations required for GT-R was seven, determined by evaluating the overall concordance rate in the training dataset (without mixtures). The resulting algorithm demonstrated 91% concordance with the training data, compared with 80% using the original LPV mutation score. The new algorithm was also tested on a validation dataset of 523 previously unanalyzed samples after removing samples with mixtures of mutant and wild-type viruses at amino acid positions in the original LPV mutation score. In this dataset, the concordance increased from 84% with the original algorithm to 90% with the new algorithm.
 
To assess the accuracy of the new algorithm further, the training and validation datasets were combined and a scatter plot of LPV FC versus new mutation score was generated. Compared with the original LPV mutation score, less variation in LPV susceptibility was observed within any given mutation score, resulting in a higher correlation coefficient with the new algorithm (r2 = 0.80 compared with 0.62). The two prominent outlier viruses with high-level resistance to LPV and four mutations were found to contain the rare I47A mutation. Both samples also contained V32I and M46I, as well as several polymorphisms. I47A was selected in vitro in the presence of high LPV concentrations.
 
Discussion
 
The development of GT interpretation algorithms for drugs that require multiple mutations for the acquisition of clinically relevant levels of resistance is complex. Such algorithms are highly dependent on the number of viruses included in the analysis and the previous treatment experience of the patients from which the viruses are derived. Additionally, since cross-resistance within a drug class is a common phenomenon, mutation patterns that confer resistance to one drug may not be completely defined solely by evaluation of viruses from patients treated with that drug. This often leads to genotype interpretation rules that are limited in their ability to predict phenotypic resistance. If a clinically relevant threshold for defining phenotypic drug susceptibility is set and a large set of PT and GT data from samples with diverse patterns of mutations is available, as in the present study, significant improvements can be made to a GT interpretation algorithm that had been initially derived from a limited number of viruses and patients with limited types of treatment experience.
 
The analysis revealed three categories of mutations that were either underemphasized or not included in the original LPV GT interpretation algorithm. It was not surprising to find that some mutations in the current algorithm contribute to LPV resistance more strongly than others, since many of the positions in the LPV mutation score are not in close proximity to the active site where the drug binds (e.g., 63, 71). However, not all mutations near the active site are associated with PT-R/GT-S discordance (e.g., L24I or I84V). Other factors, such as how the drug interacts with various amino acid residues, are likely to influence the relative importance in determining resistance. New variants at known mutation positions, and other previously uncharacterized resistance mutations, are likely to have been missed during the development of the initial GT algorithm because their prevalence is low and a relatively small population was studied, or because they are selected by drugs such as APV, which was not widely utilized in treatment regimens at the time that the original LPV clinical studies were conducted.
 
The importance of the APV-selected mutation I50V in determining susceptibility to LPV has been recently reported. However, viruses from patients failing APV as their first PI, which develop I50V, do not always have LPV FC > 10-fold. This apparent discrepancy can be explained by the fact that the I50V-containing samples analysed in the present study also contained other primary PI mutations.
 
There are several limitations to this study. A treatment history was available for relatively few of the patients studied, making it difficult to examine the patterns of resistance in all patients who have been exposed to a particular PI. Since a second cut-off above which a sample can be classified as being truly resistant has yet to be firmly identified, the samples cannot be divided into more than two groups, such as 'susceptible', 'resistant' and 'partially resistant' by PT or GT. However, this approach has the potential of providing more information by distinguishing 'partial response' from 'no response' based on GT or PT and is currently being investigated. In addition, the cut-off used to define phenotypic resistance to APV (2.5-fold) was not derived from clinical outcome studies. However, this threshold is probably relevant for several reasons. Strong correlations exist between PT results using this cut-off and clinical response data in cohorts that included patients receiving APV. Reductions in susceptibility to APV observed in patients who experienced viral load rebound while using APV as their first PI are modest (as low as 2-3-fold). Finally, the 99th percentile for the distribution of APV FC in genotypically wild-type viruses using the PhenoSense assay is 2.1. The clinically relevant PT cut-off, as well as the number of PI mutations required for resistance, is expected to be higher for patients treated with ritonavir-boosted APV therapy.
 
Isaacson and colleagues recently presented an independent analysis on the impact of various protease mutations on viral load response to LPV in 792 heavily PI-experienced patients. This study provides complementary information and independent confirmation of the importance of mutations at positions 10, 20, 33, 47, 48, 50, 54, and 82 for LPV resistance. In addition, another recent study found that mutations emerging following viral load rebound in 21 patients undergoing salvage treatment with LPV included variants at positions 33, 47, 48, 50, 58, and 73. A third study also reported the emergence of mutations of positions 10, 20, 33, 46, 47, 50, 54, and 82 following salvage therapy with LPV. Although other positions identified in these three studies were not identified here (e.g., 36, 77), the significant overlap between studies strongly suggests that our results are clinically significant and that these additional mutations should be included in LPV genotypic interpretation algorithms. The analysis of large numbers of phenotypes and genotypes in the context of a clinically significant PT cut-off value is an important means of improving the accuracy of drug susceptibility predictions generated using genotypic algorithms that are based on limited data from clinical trials in relatively homogeneous patient groups.
 
 
 
 
  icon paper stack View Older Articles   Back to Top   www.natap.org