Pharmacogenomics-Based Drug Response Prediction Model for Acute Myeloid Leukemia with Normal Karyotype

Abstract 2698 Introduction: AML is a heterogeneous disease characterized by various recurrent cytogenetic abnormalities, which provide the most important prognostic information. Although AML with normal karyotype (AML-NK) constitutes the largest subgroup representing 45–50% of adult AMLs, the prognosis of AML-NK is quite heterogeneous and their clinical outcomes are diverse. Several model capable of identifying non-responders prior to induction treatment has been proposed in AML-NK. While molecular-guided risk assessment (i.e. FLT3/ITD mutation or NPM1 mutation) and stratification are needed to confirm their prognostic significances, it is quite enthusiastic that genetic variations associated with ADMET (adsorption, desorption, metabolism, excretion, and transport) genes predict the responsiveness of anticancer drugs with better predictive power compared to other genomic variations based stratification. It is also expected SNP markers that reside within the loci of these genetic pathways may play an important role to classify responders and non-responders in AML-NK. Methods and materials: A high density SNP genotyping microarray designed by our group (custom PGX chip; Samsung Electronics Co., Suwon, Korea) was evaluated in 139 patients with AML-NK as a discovery set using both custom PGX chip and Affymetrix SNP array 6.0 (Affymetrix Inc., Santa Clara, CA USA). The DNA samples were obtained from the marrow samples at the time of diagnosis. The custom PGx chip included more extensive SNP markers of ADMET genes and cancer-related genes. Genotypes were determined by Birdseed v2 algorithm for Affymetrix SNP 6.0 and by BRLMM-p for custom PGx chip. Genotypes from 906,600 SNPs were obtained in Affymetrix SNP 6.0 chip while custom PGx chip provided genotype information of 90,647 SNPs. Between Affymetrix SNP 6.0 and custom PGx chip, 67,251 SNPs were shared. ADMET and cancer-related SNPs (n=23,396) were only available on custom PGx chip. The predictive model for the achievement of complete remission was developed using Max Test method with leave-one-out LDA cross validation (repeating 10,000 times of permutation test). The analysis compared the data of PGx chip with data from Affymetrix SNP 6.0 microarray. Results: Genotyping results from custom PGx chip was compared to results from Affymetrix SNP 6.0 showing 94.33% of concordance rate. Among the 164,578 SNP markers included in the original design, 90,647 of them were finally chosen to generate reliable genotype calls. Additional SNP markers including missing genotypes were again removed before the construction of prediction model. Finally, the total number of available SNP markers was reduced to 37,676 for the analysis and model contruction. Max Test provided p-values by performing permutation test. Significant SNP markers with p-values less than 0.01 were chosen to construct a prediction model using leave-one-out LDA (linear discriminate analysis) method. The numbers of SNP markers were narrowed down as p-values became more significant (No. of SNPs in p-value): n=329 in p The selection of SNP markers could be optimized and further narrowed down by extensive search. The accuracy of the model improved as more SNP markers were included in the model: the accuracy reaches 100% when N was 59. Followings are the example of the significant SNP markers included in the genetic predictive model for complete remission after induction therapy for AML-NK: CCDC93, LRP1B, LASS6, ATF2 , and PDE11A on chr 2, PDE4D and PPP2R2B on chr 5, PARK2 on chr 6, PTPRD on chr 9, ABCC4 on chr 13, SLC25A21 on chr 14, ABCC1, PRKCB, ITGAX , and CNOT1 on chr 16, and other SNP markers. The biologic significances of these SNP markers are under investigation. Conclusion: The clinical relevance of the predictive model based on pharmacogenomic informations will be further clarified with external replication in the independent cohorts of AML. In patients with AML-NK which has diverse and heterogeneous clinical characteristics, the predictive model could identify non-responders who might benefit from alternative treatments prior to commencement of conventional induction therapy based on idarubicin plus cytarabine. In addition, it will also reduce the risk of adverse events in the non-responders and eventually improve overall survival in AML-NK. Disclosures: No relevant conflicts of interest to declare.